2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
218 struct packet_skb_cb {
220 struct sockaddr_pkt pkt;
222 /* Trick: alias skb original length with
223 * ll.sll_family and ll.protocol in order
226 unsigned int origlen;
227 struct sockaddr_ll ll;
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 static int packet_direct_xmit(struct sk_buff *skb)
248 struct net_device *dev = skb->dev;
249 netdev_features_t features;
250 struct netdev_queue *txq;
251 int ret = NETDEV_TX_BUSY;
253 if (unlikely(!netif_running(dev) ||
254 !netif_carrier_ok(dev)))
257 features = netif_skb_features(skb);
258 if (skb_needs_linearize(skb, features) &&
259 __skb_linearize(skb))
262 txq = skb_get_tx_queue(dev, skb);
266 HARD_TX_LOCK(dev, txq, smp_processor_id());
267 if (!netif_xmit_frozen_or_drv_stopped(txq))
268 ret = netdev_start_xmit(skb, dev, txq, false);
269 HARD_TX_UNLOCK(dev, txq);
273 if (!dev_xmit_complete(ret))
278 atomic_long_inc(&dev->tx_dropped);
280 return NET_XMIT_DROP;
283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
285 struct net_device *dev;
288 dev = rcu_dereference(po->cached_dev);
296 static void packet_cached_dev_assign(struct packet_sock *po,
297 struct net_device *dev)
299 rcu_assign_pointer(po->cached_dev, dev);
302 static void packet_cached_dev_reset(struct packet_sock *po)
304 RCU_INIT_POINTER(po->cached_dev, NULL);
307 static bool packet_use_direct_xmit(const struct packet_sock *po)
309 return po->xmit == packet_direct_xmit;
312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
314 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
319 const struct net_device_ops *ops = dev->netdev_ops;
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL,
324 __packet_pick_tx_queue);
325 queue_index = netdev_cap_txqueue(dev, queue_index);
327 queue_index = __packet_pick_tx_queue(dev, skb);
330 skb_set_queue_mapping(skb, queue_index);
333 /* register_prot_hook must be invoked with the po->bind_lock held,
334 * or from a context in which asynchronous accesses to the packet
335 * socket is not possible (packet_create()).
337 static void register_prot_hook(struct sock *sk)
339 struct packet_sock *po = pkt_sk(sk);
343 __fanout_link(sk, po);
345 dev_add_pack(&po->prot_hook);
352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
353 * held. If the sync parameter is true, we will temporarily drop
354 * the po->bind_lock and do a synchronize_net to make sure no
355 * asynchronous packet processing paths still refer to the elements
356 * of po->prot_hook. If the sync parameter is false, it is the
357 * callers responsibility to take care of this.
359 static void __unregister_prot_hook(struct sock *sk, bool sync)
361 struct packet_sock *po = pkt_sk(sk);
366 __fanout_unlink(sk, po);
368 __dev_remove_pack(&po->prot_hook);
373 spin_unlock(&po->bind_lock);
375 spin_lock(&po->bind_lock);
379 static void unregister_prot_hook(struct sock *sk, bool sync)
381 struct packet_sock *po = pkt_sk(sk);
384 __unregister_prot_hook(sk, sync);
387 static inline struct page * __pure pgv_to_page(void *addr)
389 if (is_vmalloc_addr(addr))
390 return vmalloc_to_page(addr);
391 return virt_to_page(addr);
394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
396 union tpacket_uhdr h;
399 switch (po->tp_version) {
401 h.h1->tp_status = status;
402 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 h.h2->tp_status = status;
406 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 WARN(1, "TPACKET version not supported.\n");
417 static int __packet_get_status(struct packet_sock *po, void *frame)
419 union tpacket_uhdr h;
424 switch (po->tp_version) {
426 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
427 return h.h1->tp_status;
429 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
430 return h.h2->tp_status;
433 WARN(1, "TPACKET version not supported.\n");
439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
442 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
445 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
447 return TP_STATUS_TS_RAW_HARDWARE;
449 if (ktime_to_timespec_cond(skb->tstamp, ts))
450 return TP_STATUS_TS_SOFTWARE;
455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
458 union tpacket_uhdr h;
462 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 switch (po->tp_version) {
468 h.h1->tp_sec = ts.tv_sec;
469 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
472 h.h2->tp_sec = ts.tv_sec;
473 h.h2->tp_nsec = ts.tv_nsec;
477 WARN(1, "TPACKET version not supported.\n");
481 /* one flush is safe, as both fields always lie on the same cacheline */
482 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
488 static void *packet_lookup_frame(struct packet_sock *po,
489 struct packet_ring_buffer *rb,
490 unsigned int position,
493 unsigned int pg_vec_pos, frame_offset;
494 union tpacket_uhdr h;
496 pg_vec_pos = position / rb->frames_per_block;
497 frame_offset = position % rb->frames_per_block;
499 h.raw = rb->pg_vec[pg_vec_pos].buffer +
500 (frame_offset * rb->frame_size);
502 if (status != __packet_get_status(po, h.raw))
508 static void *packet_current_frame(struct packet_sock *po,
509 struct packet_ring_buffer *rb,
512 return packet_lookup_frame(po, rb, rb->head, status);
515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
517 del_timer_sync(&pkc->retire_blk_timer);
520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
522 struct sk_buff_head *rb_queue)
524 struct tpacket_kbdq_core *pkc;
526 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
527 GET_PBDQC_FROM_RB(&po->rx_ring);
529 spin_lock_bh(&rb_queue->lock);
530 pkc->delete_blk_timer = 1;
531 spin_unlock_bh(&rb_queue->lock);
533 prb_del_retire_blk_timer(pkc);
536 static void prb_init_blk_timer(struct packet_sock *po,
537 struct tpacket_kbdq_core *pkc,
538 void (*func) (unsigned long))
540 init_timer(&pkc->retire_blk_timer);
541 pkc->retire_blk_timer.data = (long)po;
542 pkc->retire_blk_timer.function = func;
543 pkc->retire_blk_timer.expires = jiffies;
546 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
548 struct tpacket_kbdq_core *pkc;
553 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
554 GET_PBDQC_FROM_RB(&po->rx_ring);
555 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
559 int blk_size_in_bytes)
561 struct net_device *dev;
562 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
563 struct ethtool_cmd ecmd;
568 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
569 if (unlikely(!dev)) {
571 return DEFAULT_PRB_RETIRE_TOV;
573 err = __ethtool_get_settings(dev, &ecmd);
574 speed = ethtool_cmd_speed(&ecmd);
578 * If the link speed is so slow you don't really
579 * need to worry about perf anyways
581 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
582 return DEFAULT_PRB_RETIRE_TOV;
589 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
602 union tpacket_req_u *req_u)
604 p1->feature_req_word = req_u->req3.tp_feature_req_word;
607 static void init_prb_bdqc(struct packet_sock *po,
608 struct packet_ring_buffer *rb,
610 union tpacket_req_u *req_u, int tx_ring)
612 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
613 struct tpacket_block_desc *pbd;
615 memset(p1, 0x0, sizeof(*p1));
617 p1->knxt_seq_num = 1;
619 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
620 p1->pkblk_start = pg_vec[0].buffer;
621 p1->kblk_size = req_u->req3.tp_block_size;
622 p1->knum_blocks = req_u->req3.tp_block_nr;
623 p1->hdrlen = po->tp_hdrlen;
624 p1->version = po->tp_version;
625 p1->last_kactive_blk_num = 0;
626 po->stats.stats3.tp_freeze_q_cnt = 0;
627 if (req_u->req3.tp_retire_blk_tov)
628 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
630 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
631 req_u->req3.tp_block_size);
632 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
633 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
635 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
636 prb_init_ft_ops(p1, req_u);
637 prb_setup_retire_blk_timer(po, tx_ring);
638 prb_open_block(p1, pbd);
641 /* Do NOT update the last_blk_num first.
642 * Assumes sk_buff_head lock is held.
644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
646 mod_timer(&pkc->retire_blk_timer,
647 jiffies + pkc->tov_in_jiffies);
648 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
653 * 1) We refresh the timer only when we open a block.
654 * By doing this we don't waste cycles refreshing the timer
655 * on packet-by-packet basis.
657 * With a 1MB block-size, on a 1Gbps line, it will take
658 * i) ~8 ms to fill a block + ii) memcpy etc.
659 * In this cut we are not accounting for the memcpy time.
661 * So, if the user sets the 'tmo' to 10ms then the timer
662 * will never fire while the block is still getting filled
663 * (which is what we want). However, the user could choose
664 * to close a block early and that's fine.
666 * But when the timer does fire, we check whether or not to refresh it.
667 * Since the tmo granularity is in msecs, it is not too expensive
668 * to refresh the timer, lets say every '8' msecs.
669 * Either the user can set the 'tmo' or we can derive it based on
670 * a) line-speed and b) block-size.
671 * prb_calc_retire_blk_tmo() calculates the tmo.
674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
676 struct packet_sock *po = (struct packet_sock *)data;
677 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
679 struct tpacket_block_desc *pbd;
681 spin_lock(&po->sk.sk_receive_queue.lock);
683 frozen = prb_queue_frozen(pkc);
684 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
686 if (unlikely(pkc->delete_blk_timer))
689 /* We only need to plug the race when the block is partially filled.
691 * lock(); increment BLOCK_NUM_PKTS; unlock()
692 * copy_bits() is in progress ...
693 * timer fires on other cpu:
694 * we can't retire the current block because copy_bits
698 if (BLOCK_NUM_PKTS(pbd)) {
699 while (atomic_read(&pkc->blk_fill_in_prog)) {
700 /* Waiting for skb_copy_bits to finish... */
705 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
707 if (!BLOCK_NUM_PKTS(pbd)) {
708 /* An empty block. Just refresh the timer. */
711 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
712 if (!prb_dispatch_next_block(pkc, po))
717 /* Case 1. Queue was frozen because user-space was
720 if (prb_curr_blk_in_use(pkc, pbd)) {
722 * Ok, user-space is still behind.
723 * So just refresh the timer.
727 /* Case 2. queue was frozen,user-space caught up,
728 * now the link went idle && the timer fired.
729 * We don't have a block to close.So we open this
730 * block and restart the timer.
731 * opening a block thaws the queue,restarts timer
732 * Thawing/timer-refresh is a side effect.
734 prb_open_block(pkc, pbd);
741 _prb_refresh_rx_retire_blk_timer(pkc);
744 spin_unlock(&po->sk.sk_receive_queue.lock);
747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
748 struct tpacket_block_desc *pbd1, __u32 status)
750 /* Flush everything minus the block header */
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
757 /* Skip the block header(we know header WILL fit in 4K) */
760 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
761 for (; start < end; start += PAGE_SIZE)
762 flush_dcache_page(pgv_to_page(start));
767 /* Now update the block status. */
769 BLOCK_STATUS(pbd1) = status;
771 /* Flush the block header */
773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
775 flush_dcache_page(pgv_to_page(start));
785 * 2) Increment active_blk_num
787 * Note:We DONT refresh the timer on purpose.
788 * Because almost always the next block will be opened.
790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
791 struct tpacket_block_desc *pbd1,
792 struct packet_sock *po, unsigned int stat)
794 __u32 status = TP_STATUS_USER | stat;
796 struct tpacket3_hdr *last_pkt;
797 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
798 struct sock *sk = &po->sk;
800 if (po->stats.stats3.tp_drops)
801 status |= TP_STATUS_LOSING;
803 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
804 last_pkt->tp_next_offset = 0;
806 /* Get the ts of the last pkt */
807 if (BLOCK_NUM_PKTS(pbd1)) {
808 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
809 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
811 /* Ok, we tmo'd - so get the current time.
813 * It shouldn't really happen as we don't close empty
814 * blocks. See prb_retire_rx_blk_timer_expired().
818 h1->ts_last_pkt.ts_sec = ts.tv_sec;
819 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
824 /* Flush the block */
825 prb_flush_block(pkc1, pbd1, status);
827 sk->sk_data_ready(sk);
829 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
834 pkc->reset_pending_on_curr_blk = 0;
838 * Side effect of opening a block:
840 * 1) prb_queue is thawed.
841 * 2) retire_blk_timer is refreshed.
844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
845 struct tpacket_block_desc *pbd1)
848 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
852 /* We could have just memset this but we will lose the
853 * flexibility of making the priv area sticky
856 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
857 BLOCK_NUM_PKTS(pbd1) = 0;
858 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
862 h1->ts_first_pkt.ts_sec = ts.tv_sec;
863 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
865 pkc1->pkblk_start = (char *)pbd1;
866 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
868 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
871 pbd1->version = pkc1->version;
872 pkc1->prev = pkc1->nxt_offset;
873 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
875 prb_thaw_queue(pkc1);
876 _prb_refresh_rx_retire_blk_timer(pkc1);
882 * Queue freeze logic:
883 * 1) Assume tp_block_nr = 8 blocks.
884 * 2) At time 't0', user opens Rx ring.
885 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
886 * 4) user-space is either sleeping or processing block '0'.
887 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
888 * it will close block-7,loop around and try to fill block '0'.
890 * __packet_lookup_frame_in_block
891 * prb_retire_current_block()
892 * prb_dispatch_next_block()
893 * |->(BLOCK_STATUS == USER) evaluates to true
894 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
895 * 6) Now there are two cases:
896 * 6.1) Link goes idle right after the queue is frozen.
897 * But remember, the last open_block() refreshed the timer.
898 * When this timer expires,it will refresh itself so that we can
899 * re-open block-0 in near future.
900 * 6.2) Link is busy and keeps on receiving packets. This is a simple
901 * case and __packet_lookup_frame_in_block will check if block-0
902 * is free and can now be re-used.
904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
905 struct packet_sock *po)
907 pkc->reset_pending_on_curr_blk = 1;
908 po->stats.stats3.tp_freeze_q_cnt++;
911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
914 * If the next block is free then we will dispatch it
915 * and return a good offset.
916 * Else, we will freeze the queue.
917 * So, caller must check the return value.
919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
920 struct packet_sock *po)
922 struct tpacket_block_desc *pbd;
926 /* 1. Get current block num */
927 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
929 /* 2. If this block is currently in_use then freeze the queue */
930 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
931 prb_freeze_queue(pkc, po);
937 * open this block and return the offset where the first packet
938 * needs to get stored.
940 prb_open_block(pkc, pbd);
941 return (void *)pkc->nxt_offset;
944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
945 struct packet_sock *po, unsigned int status)
947 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
949 /* retire/close the current block */
950 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
952 * Plug the case where copy_bits() is in progress on
953 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
954 * have space to copy the pkt in the current block and
955 * called prb_retire_current_block()
957 * We don't need to worry about the TMO case because
958 * the timer-handler already handled this case.
960 if (!(status & TP_STATUS_BLK_TMO)) {
961 while (atomic_read(&pkc->blk_fill_in_prog)) {
962 /* Waiting for skb_copy_bits to finish... */
966 prb_close_block(pkc, pbd, po, status);
971 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
972 struct tpacket_block_desc *pbd)
974 return TP_STATUS_USER & BLOCK_STATUS(pbd);
977 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
979 return pkc->reset_pending_on_curr_blk;
982 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
984 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
985 atomic_dec(&pkc->blk_fill_in_prog);
988 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
989 struct tpacket3_hdr *ppd)
991 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
994 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
995 struct tpacket3_hdr *ppd)
997 ppd->hv1.tp_rxhash = 0;
1000 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1001 struct tpacket3_hdr *ppd)
1003 if (skb_vlan_tag_present(pkc->skb)) {
1004 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1005 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1006 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1008 ppd->hv1.tp_vlan_tci = 0;
1009 ppd->hv1.tp_vlan_tpid = 0;
1010 ppd->tp_status = TP_STATUS_AVAILABLE;
1014 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1015 struct tpacket3_hdr *ppd)
1017 ppd->hv1.tp_padding = 0;
1018 prb_fill_vlan_info(pkc, ppd);
1020 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1021 prb_fill_rxhash(pkc, ppd);
1023 prb_clear_rxhash(pkc, ppd);
1026 static void prb_fill_curr_block(char *curr,
1027 struct tpacket_kbdq_core *pkc,
1028 struct tpacket_block_desc *pbd,
1031 struct tpacket3_hdr *ppd;
1033 ppd = (struct tpacket3_hdr *)curr;
1034 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1036 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038 BLOCK_NUM_PKTS(pbd) += 1;
1039 atomic_inc(&pkc->blk_fill_in_prog);
1040 prb_run_all_ft_ops(pkc, ppd);
1043 /* Assumes caller has the sk->rx_queue.lock */
1044 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1045 struct sk_buff *skb,
1050 struct tpacket_kbdq_core *pkc;
1051 struct tpacket_block_desc *pbd;
1054 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1055 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057 /* Queue is frozen when user space is lagging behind */
1058 if (prb_queue_frozen(pkc)) {
1060 * Check if that last block which caused the queue to freeze,
1061 * is still in_use by user-space.
1063 if (prb_curr_blk_in_use(pkc, pbd)) {
1064 /* Can't record this packet */
1068 * Ok, the block was released by user-space.
1069 * Now let's open that block.
1070 * opening a block also thaws the queue.
1071 * Thawing is a side effect.
1073 prb_open_block(pkc, pbd);
1078 curr = pkc->nxt_offset;
1080 end = (char *)pbd + pkc->kblk_size;
1082 /* first try the current block */
1083 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1084 prb_fill_curr_block(curr, pkc, pbd, len);
1085 return (void *)curr;
1088 /* Ok, close the current block */
1089 prb_retire_current_block(pkc, po, 0);
1091 /* Now, try to dispatch the next block */
1092 curr = (char *)prb_dispatch_next_block(pkc, po);
1094 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1095 prb_fill_curr_block(curr, pkc, pbd, len);
1096 return (void *)curr;
1100 * No free blocks are available.user_space hasn't caught up yet.
1101 * Queue was just frozen and now this packet will get dropped.
1106 static void *packet_current_rx_frame(struct packet_sock *po,
1107 struct sk_buff *skb,
1108 int status, unsigned int len)
1111 switch (po->tp_version) {
1114 curr = packet_lookup_frame(po, &po->rx_ring,
1115 po->rx_ring.head, status);
1118 return __packet_lookup_frame_in_block(po, skb, status, len);
1120 WARN(1, "TPACKET version not supported\n");
1126 static void *prb_lookup_block(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1131 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1132 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1134 if (status != BLOCK_STATUS(pbd))
1139 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1142 if (rb->prb_bdqc.kactive_blk_num)
1143 prev = rb->prb_bdqc.kactive_blk_num-1;
1145 prev = rb->prb_bdqc.knum_blocks-1;
1149 /* Assumes caller has held the rx_queue.lock */
1150 static void *__prb_previous_block(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1154 unsigned int previous = prb_previous_blk_num(rb);
1155 return prb_lookup_block(po, rb, previous, status);
1158 static void *packet_previous_rx_frame(struct packet_sock *po,
1159 struct packet_ring_buffer *rb,
1162 if (po->tp_version <= TPACKET_V2)
1163 return packet_previous_frame(po, rb, status);
1165 return __prb_previous_block(po, rb, status);
1168 static void packet_increment_rx_head(struct packet_sock *po,
1169 struct packet_ring_buffer *rb)
1171 switch (po->tp_version) {
1174 return packet_increment_head(rb);
1177 WARN(1, "TPACKET version not supported.\n");
1183 static void *packet_previous_frame(struct packet_sock *po,
1184 struct packet_ring_buffer *rb,
1187 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1188 return packet_lookup_frame(po, rb, previous, status);
1191 static void packet_increment_head(struct packet_ring_buffer *buff)
1193 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1196 static void packet_inc_pending(struct packet_ring_buffer *rb)
1198 this_cpu_inc(*rb->pending_refcnt);
1201 static void packet_dec_pending(struct packet_ring_buffer *rb)
1203 this_cpu_dec(*rb->pending_refcnt);
1206 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1208 unsigned int refcnt = 0;
1211 /* We don't use pending refcount in rx_ring. */
1212 if (rb->pending_refcnt == NULL)
1215 for_each_possible_cpu(cpu)
1216 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1221 static int packet_alloc_pending(struct packet_sock *po)
1223 po->rx_ring.pending_refcnt = NULL;
1225 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1226 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1232 static void packet_free_pending(struct packet_sock *po)
1234 free_percpu(po->tx_ring.pending_refcnt);
1237 #define ROOM_POW_OFF 2
1238 #define ROOM_NONE 0x0
1239 #define ROOM_LOW 0x1
1240 #define ROOM_NORMAL 0x2
1242 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1246 len = po->rx_ring.frame_max + 1;
1247 idx = po->rx_ring.head;
1249 idx += len >> pow_off;
1252 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1255 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1259 len = po->rx_ring.prb_bdqc.knum_blocks;
1260 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1262 idx += len >> pow_off;
1265 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1268 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1270 struct sock *sk = &po->sk;
1271 int ret = ROOM_NONE;
1273 if (po->prot_hook.func != tpacket_rcv) {
1274 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1275 - (skb ? skb->truesize : 0);
1276 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1284 if (po->tp_version == TPACKET_V3) {
1285 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1287 else if (__tpacket_v3_has_room(po, 0))
1290 if (__tpacket_has_room(po, ROOM_POW_OFF))
1292 else if (__tpacket_has_room(po, 0))
1299 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1304 if (po->prot_hook.func == tpacket_rcv) {
1305 spin_lock(&po->sk.sk_receive_queue.lock);
1306 ret = __packet_rcv_has_room(po, skb);
1307 spin_unlock(&po->sk.sk_receive_queue.lock);
1309 ret = __packet_rcv_has_room(po, skb);
1312 has_room = ret == ROOM_NORMAL;
1313 if (po->pressure == has_room)
1314 xchg(&po->pressure, !has_room);
1319 static void packet_sock_destruct(struct sock *sk)
1321 skb_queue_purge(&sk->sk_error_queue);
1323 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1324 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1326 if (!sock_flag(sk, SOCK_DEAD)) {
1327 pr_err("Attempt to release alive packet socket: %p\n", sk);
1331 sk_refcnt_debug_dec(sk);
1334 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1336 int x = atomic_read(&f->rr_cur) + 1;
1344 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1345 struct sk_buff *skb,
1348 return reciprocal_scale(skb_get_hash(skb), num);
1351 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1352 struct sk_buff *skb,
1357 cur = atomic_read(&f->rr_cur);
1358 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1359 fanout_rr_next(f, num))) != cur)
1364 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1365 struct sk_buff *skb,
1368 return smp_processor_id() % num;
1371 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1372 struct sk_buff *skb,
1375 return prandom_u32_max(num);
1378 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1379 struct sk_buff *skb,
1380 unsigned int idx, bool try_self,
1383 struct packet_sock *po, *po_next;
1386 po = pkt_sk(f->arr[idx]);
1387 if (try_self && packet_rcv_has_room(po, skb) != ROOM_NONE)
1390 i = j = min_t(int, po->rollover->sock, num - 1);
1392 po_next = pkt_sk(f->arr[i]);
1393 if (po_next != po && !po_next->pressure &&
1394 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1396 po->rollover->sock = i;
1407 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1408 struct sk_buff *skb,
1411 return skb_get_queue_mapping(skb) % num;
1414 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1416 return f->flags & (flag >> 8);
1419 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1420 struct packet_type *pt, struct net_device *orig_dev)
1422 struct packet_fanout *f = pt->af_packet_priv;
1423 unsigned int num = f->num_members;
1424 struct packet_sock *po;
1427 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1433 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1434 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1439 case PACKET_FANOUT_HASH:
1441 idx = fanout_demux_hash(f, skb, num);
1443 case PACKET_FANOUT_LB:
1444 idx = fanout_demux_lb(f, skb, num);
1446 case PACKET_FANOUT_CPU:
1447 idx = fanout_demux_cpu(f, skb, num);
1449 case PACKET_FANOUT_RND:
1450 idx = fanout_demux_rnd(f, skb, num);
1452 case PACKET_FANOUT_QM:
1453 idx = fanout_demux_qm(f, skb, num);
1455 case PACKET_FANOUT_ROLLOVER:
1456 idx = fanout_demux_rollover(f, skb, 0, false, num);
1460 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1461 idx = fanout_demux_rollover(f, skb, idx, true, num);
1463 po = pkt_sk(f->arr[idx]);
1464 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1467 DEFINE_MUTEX(fanout_mutex);
1468 EXPORT_SYMBOL_GPL(fanout_mutex);
1469 static LIST_HEAD(fanout_list);
1471 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1473 struct packet_fanout *f = po->fanout;
1475 spin_lock(&f->lock);
1476 f->arr[f->num_members] = sk;
1479 spin_unlock(&f->lock);
1482 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1484 struct packet_fanout *f = po->fanout;
1487 spin_lock(&f->lock);
1488 for (i = 0; i < f->num_members; i++) {
1489 if (f->arr[i] == sk)
1492 BUG_ON(i >= f->num_members);
1493 f->arr[i] = f->arr[f->num_members - 1];
1495 spin_unlock(&f->lock);
1498 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1500 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1506 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1508 struct packet_sock *po = pkt_sk(sk);
1509 struct packet_fanout *f, *match;
1510 u8 type = type_flags & 0xff;
1511 u8 flags = type_flags >> 8;
1515 case PACKET_FANOUT_ROLLOVER:
1516 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1518 case PACKET_FANOUT_HASH:
1519 case PACKET_FANOUT_LB:
1520 case PACKET_FANOUT_CPU:
1521 case PACKET_FANOUT_RND:
1522 case PACKET_FANOUT_QM:
1534 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) {
1535 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1540 mutex_lock(&fanout_mutex);
1542 list_for_each_entry(f, &fanout_list, list) {
1544 read_pnet(&f->net) == sock_net(sk)) {
1550 if (match && match->flags != flags)
1554 match = kzalloc(sizeof(*match), GFP_KERNEL);
1557 write_pnet(&match->net, sock_net(sk));
1560 match->flags = flags;
1561 atomic_set(&match->rr_cur, 0);
1562 INIT_LIST_HEAD(&match->list);
1563 spin_lock_init(&match->lock);
1564 atomic_set(&match->sk_ref, 0);
1565 match->prot_hook.type = po->prot_hook.type;
1566 match->prot_hook.dev = po->prot_hook.dev;
1567 match->prot_hook.func = packet_rcv_fanout;
1568 match->prot_hook.af_packet_priv = match;
1569 match->prot_hook.id_match = match_fanout_group;
1570 dev_add_pack(&match->prot_hook);
1571 list_add(&match->list, &fanout_list);
1574 if (match->type == type &&
1575 match->prot_hook.type == po->prot_hook.type &&
1576 match->prot_hook.dev == po->prot_hook.dev) {
1578 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1579 __dev_remove_pack(&po->prot_hook);
1581 atomic_inc(&match->sk_ref);
1582 __fanout_link(sk, po);
1587 mutex_unlock(&fanout_mutex);
1589 kfree(po->rollover);
1590 po->rollover = NULL;
1595 static void fanout_release(struct sock *sk)
1597 struct packet_sock *po = pkt_sk(sk);
1598 struct packet_fanout *f;
1604 mutex_lock(&fanout_mutex);
1607 if (atomic_dec_and_test(&f->sk_ref)) {
1609 dev_remove_pack(&f->prot_hook);
1612 mutex_unlock(&fanout_mutex);
1614 kfree(po->rollover);
1617 static const struct proto_ops packet_ops;
1619 static const struct proto_ops packet_ops_spkt;
1621 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1622 struct packet_type *pt, struct net_device *orig_dev)
1625 struct sockaddr_pkt *spkt;
1628 * When we registered the protocol we saved the socket in the data
1629 * field for just this event.
1632 sk = pt->af_packet_priv;
1635 * Yank back the headers [hope the device set this
1636 * right or kerboom...]
1638 * Incoming packets have ll header pulled,
1641 * For outgoing ones skb->data == skb_mac_header(skb)
1642 * so that this procedure is noop.
1645 if (skb->pkt_type == PACKET_LOOPBACK)
1648 if (!net_eq(dev_net(dev), sock_net(sk)))
1651 skb = skb_share_check(skb, GFP_ATOMIC);
1655 /* drop any routing info */
1658 /* drop conntrack reference */
1661 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1663 skb_push(skb, skb->data - skb_mac_header(skb));
1666 * The SOCK_PACKET socket receives _all_ frames.
1669 spkt->spkt_family = dev->type;
1670 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1671 spkt->spkt_protocol = skb->protocol;
1674 * Charge the memory to the socket. This is done specifically
1675 * to prevent sockets using all the memory up.
1678 if (sock_queue_rcv_skb(sk, skb) == 0)
1689 * Output a raw packet to a device layer. This bypasses all the other
1690 * protocol layers and you must therefore supply it with a complete frame
1693 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1696 struct sock *sk = sock->sk;
1697 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1698 struct sk_buff *skb = NULL;
1699 struct net_device *dev;
1705 * Get and verify the address.
1709 if (msg->msg_namelen < sizeof(struct sockaddr))
1711 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1712 proto = saddr->spkt_protocol;
1714 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1717 * Find the device first to size check it
1720 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1723 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1729 if (!(dev->flags & IFF_UP))
1733 * You may not queue a frame bigger than the mtu. This is the lowest level
1734 * raw protocol and you must do your own fragmentation at this level.
1737 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1738 if (!netif_supports_nofcs(dev)) {
1739 err = -EPROTONOSUPPORT;
1742 extra_len = 4; /* We're doing our own CRC */
1746 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1750 size_t reserved = LL_RESERVED_SPACE(dev);
1751 int tlen = dev->needed_tailroom;
1752 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1755 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1758 /* FIXME: Save some space for broken drivers that write a hard
1759 * header at transmission time by themselves. PPP is the notable
1760 * one here. This should really be fixed at the driver level.
1762 skb_reserve(skb, reserved);
1763 skb_reset_network_header(skb);
1765 /* Try to align data part correctly */
1770 skb_reset_network_header(skb);
1772 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1778 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1779 /* Earlier code assumed this would be a VLAN pkt,
1780 * double-check this now that we have the actual
1783 struct ethhdr *ehdr;
1784 skb_reset_mac_header(skb);
1785 ehdr = eth_hdr(skb);
1786 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1792 skb->protocol = proto;
1794 skb->priority = sk->sk_priority;
1795 skb->mark = sk->sk_mark;
1797 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1799 if (unlikely(extra_len == 4))
1802 skb_probe_transport_header(skb, 0);
1804 dev_queue_xmit(skb);
1815 static unsigned int run_filter(const struct sk_buff *skb,
1816 const struct sock *sk,
1819 struct sk_filter *filter;
1822 filter = rcu_dereference(sk->sk_filter);
1824 res = SK_RUN_FILTER(filter, skb);
1831 * This function makes lazy skb cloning in hope that most of packets
1832 * are discarded by BPF.
1834 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1835 * and skb->cb are mangled. It works because (and until) packets
1836 * falling here are owned by current CPU. Output packets are cloned
1837 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1838 * sequencially, so that if we return skb to original state on exit,
1839 * we will not harm anyone.
1842 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1843 struct packet_type *pt, struct net_device *orig_dev)
1846 struct sockaddr_ll *sll;
1847 struct packet_sock *po;
1848 u8 *skb_head = skb->data;
1849 int skb_len = skb->len;
1850 unsigned int snaplen, res;
1852 if (skb->pkt_type == PACKET_LOOPBACK)
1855 sk = pt->af_packet_priv;
1858 if (!net_eq(dev_net(dev), sock_net(sk)))
1863 if (dev->header_ops) {
1864 /* The device has an explicit notion of ll header,
1865 * exported to higher levels.
1867 * Otherwise, the device hides details of its frame
1868 * structure, so that corresponding packet head is
1869 * never delivered to user.
1871 if (sk->sk_type != SOCK_DGRAM)
1872 skb_push(skb, skb->data - skb_mac_header(skb));
1873 else if (skb->pkt_type == PACKET_OUTGOING) {
1874 /* Special case: outgoing packets have ll header at head */
1875 skb_pull(skb, skb_network_offset(skb));
1881 res = run_filter(skb, sk, snaplen);
1883 goto drop_n_restore;
1887 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1890 if (skb_shared(skb)) {
1891 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1895 if (skb_head != skb->data) {
1896 skb->data = skb_head;
1903 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1905 sll = &PACKET_SKB_CB(skb)->sa.ll;
1906 sll->sll_hatype = dev->type;
1907 sll->sll_pkttype = skb->pkt_type;
1908 if (unlikely(po->origdev))
1909 sll->sll_ifindex = orig_dev->ifindex;
1911 sll->sll_ifindex = dev->ifindex;
1913 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1915 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1916 * Use their space for storing the original skb length.
1918 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1920 if (pskb_trim(skb, snaplen))
1923 skb_set_owner_r(skb, sk);
1927 /* drop conntrack reference */
1930 spin_lock(&sk->sk_receive_queue.lock);
1931 po->stats.stats1.tp_packets++;
1932 sock_skb_set_dropcount(sk, skb);
1933 __skb_queue_tail(&sk->sk_receive_queue, skb);
1934 spin_unlock(&sk->sk_receive_queue.lock);
1935 sk->sk_data_ready(sk);
1939 spin_lock(&sk->sk_receive_queue.lock);
1940 po->stats.stats1.tp_drops++;
1941 atomic_inc(&sk->sk_drops);
1942 spin_unlock(&sk->sk_receive_queue.lock);
1945 if (skb_head != skb->data && skb_shared(skb)) {
1946 skb->data = skb_head;
1954 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1955 struct packet_type *pt, struct net_device *orig_dev)
1958 struct packet_sock *po;
1959 struct sockaddr_ll *sll;
1960 union tpacket_uhdr h;
1961 u8 *skb_head = skb->data;
1962 int skb_len = skb->len;
1963 unsigned int snaplen, res;
1964 unsigned long status = TP_STATUS_USER;
1965 unsigned short macoff, netoff, hdrlen;
1966 struct sk_buff *copy_skb = NULL;
1970 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1971 * We may add members to them until current aligned size without forcing
1972 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1974 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1975 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1977 if (skb->pkt_type == PACKET_LOOPBACK)
1980 sk = pt->af_packet_priv;
1983 if (!net_eq(dev_net(dev), sock_net(sk)))
1986 if (dev->header_ops) {
1987 if (sk->sk_type != SOCK_DGRAM)
1988 skb_push(skb, skb->data - skb_mac_header(skb));
1989 else if (skb->pkt_type == PACKET_OUTGOING) {
1990 /* Special case: outgoing packets have ll header at head */
1991 skb_pull(skb, skb_network_offset(skb));
1997 res = run_filter(skb, sk, snaplen);
1999 goto drop_n_restore;
2001 if (skb->ip_summed == CHECKSUM_PARTIAL)
2002 status |= TP_STATUS_CSUMNOTREADY;
2003 else if (skb->pkt_type != PACKET_OUTGOING &&
2004 (skb->ip_summed == CHECKSUM_COMPLETE ||
2005 skb_csum_unnecessary(skb)))
2006 status |= TP_STATUS_CSUM_VALID;
2011 if (sk->sk_type == SOCK_DGRAM) {
2012 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2015 unsigned int maclen = skb_network_offset(skb);
2016 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2017 (maclen < 16 ? 16 : maclen)) +
2019 macoff = netoff - maclen;
2021 if (po->tp_version <= TPACKET_V2) {
2022 if (macoff + snaplen > po->rx_ring.frame_size) {
2023 if (po->copy_thresh &&
2024 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2025 if (skb_shared(skb)) {
2026 copy_skb = skb_clone(skb, GFP_ATOMIC);
2028 copy_skb = skb_get(skb);
2029 skb_head = skb->data;
2032 skb_set_owner_r(copy_skb, sk);
2034 snaplen = po->rx_ring.frame_size - macoff;
2035 if ((int)snaplen < 0)
2038 } else if (unlikely(macoff + snaplen >
2039 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2042 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2043 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2044 snaplen, nval, macoff);
2046 if (unlikely((int)snaplen < 0)) {
2048 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2051 spin_lock(&sk->sk_receive_queue.lock);
2052 h.raw = packet_current_rx_frame(po, skb,
2053 TP_STATUS_KERNEL, (macoff+snaplen));
2056 if (po->tp_version <= TPACKET_V2) {
2057 packet_increment_rx_head(po, &po->rx_ring);
2059 * LOSING will be reported till you read the stats,
2060 * because it's COR - Clear On Read.
2061 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2064 if (po->stats.stats1.tp_drops)
2065 status |= TP_STATUS_LOSING;
2067 po->stats.stats1.tp_packets++;
2069 status |= TP_STATUS_COPY;
2070 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2072 spin_unlock(&sk->sk_receive_queue.lock);
2074 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2076 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2077 getnstimeofday(&ts);
2079 status |= ts_status;
2081 switch (po->tp_version) {
2083 h.h1->tp_len = skb->len;
2084 h.h1->tp_snaplen = snaplen;
2085 h.h1->tp_mac = macoff;
2086 h.h1->tp_net = netoff;
2087 h.h1->tp_sec = ts.tv_sec;
2088 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2089 hdrlen = sizeof(*h.h1);
2092 h.h2->tp_len = skb->len;
2093 h.h2->tp_snaplen = snaplen;
2094 h.h2->tp_mac = macoff;
2095 h.h2->tp_net = netoff;
2096 h.h2->tp_sec = ts.tv_sec;
2097 h.h2->tp_nsec = ts.tv_nsec;
2098 if (skb_vlan_tag_present(skb)) {
2099 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2100 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2101 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2103 h.h2->tp_vlan_tci = 0;
2104 h.h2->tp_vlan_tpid = 0;
2106 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2107 hdrlen = sizeof(*h.h2);
2110 /* tp_nxt_offset,vlan are already populated above.
2111 * So DONT clear those fields here
2113 h.h3->tp_status |= status;
2114 h.h3->tp_len = skb->len;
2115 h.h3->tp_snaplen = snaplen;
2116 h.h3->tp_mac = macoff;
2117 h.h3->tp_net = netoff;
2118 h.h3->tp_sec = ts.tv_sec;
2119 h.h3->tp_nsec = ts.tv_nsec;
2120 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2121 hdrlen = sizeof(*h.h3);
2127 sll = h.raw + TPACKET_ALIGN(hdrlen);
2128 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2129 sll->sll_family = AF_PACKET;
2130 sll->sll_hatype = dev->type;
2131 sll->sll_protocol = skb->protocol;
2132 sll->sll_pkttype = skb->pkt_type;
2133 if (unlikely(po->origdev))
2134 sll->sll_ifindex = orig_dev->ifindex;
2136 sll->sll_ifindex = dev->ifindex;
2140 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2141 if (po->tp_version <= TPACKET_V2) {
2144 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2147 for (start = h.raw; start < end; start += PAGE_SIZE)
2148 flush_dcache_page(pgv_to_page(start));
2153 if (po->tp_version <= TPACKET_V2) {
2154 __packet_set_status(po, h.raw, status);
2155 sk->sk_data_ready(sk);
2157 prb_clear_blk_fill_status(&po->rx_ring);
2161 if (skb_head != skb->data && skb_shared(skb)) {
2162 skb->data = skb_head;
2170 po->stats.stats1.tp_drops++;
2171 spin_unlock(&sk->sk_receive_queue.lock);
2173 sk->sk_data_ready(sk);
2174 kfree_skb(copy_skb);
2175 goto drop_n_restore;
2178 static void tpacket_destruct_skb(struct sk_buff *skb)
2180 struct packet_sock *po = pkt_sk(skb->sk);
2182 if (likely(po->tx_ring.pg_vec)) {
2186 ph = skb_shinfo(skb)->destructor_arg;
2187 packet_dec_pending(&po->tx_ring);
2189 ts = __packet_set_timestamp(po, ph, skb);
2190 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2196 static bool ll_header_truncated(const struct net_device *dev, int len)
2198 /* net device doesn't like empty head */
2199 if (unlikely(len <= dev->hard_header_len)) {
2200 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2201 current->comm, len, dev->hard_header_len);
2208 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2209 void *frame, struct net_device *dev, int size_max,
2210 __be16 proto, unsigned char *addr, int hlen)
2212 union tpacket_uhdr ph;
2213 int to_write, offset, len, tp_len, nr_frags, len_max;
2214 struct socket *sock = po->sk.sk_socket;
2221 skb->protocol = proto;
2223 skb->priority = po->sk.sk_priority;
2224 skb->mark = po->sk.sk_mark;
2225 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2226 skb_shinfo(skb)->destructor_arg = ph.raw;
2228 switch (po->tp_version) {
2230 tp_len = ph.h2->tp_len;
2233 tp_len = ph.h1->tp_len;
2236 if (unlikely(tp_len > size_max)) {
2237 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2241 skb_reserve(skb, hlen);
2242 skb_reset_network_header(skb);
2244 if (!packet_use_direct_xmit(po))
2245 skb_probe_transport_header(skb, 0);
2246 if (unlikely(po->tp_tx_has_off)) {
2247 int off_min, off_max, off;
2248 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2249 off_max = po->tx_ring.frame_size - tp_len;
2250 if (sock->type == SOCK_DGRAM) {
2251 switch (po->tp_version) {
2253 off = ph.h2->tp_net;
2256 off = ph.h1->tp_net;
2260 switch (po->tp_version) {
2262 off = ph.h2->tp_mac;
2265 off = ph.h1->tp_mac;
2269 if (unlikely((off < off_min) || (off_max < off)))
2271 data = ph.raw + off;
2273 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2277 if (sock->type == SOCK_DGRAM) {
2278 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2280 if (unlikely(err < 0))
2282 } else if (dev->hard_header_len) {
2283 if (ll_header_truncated(dev, tp_len))
2286 skb_push(skb, dev->hard_header_len);
2287 err = skb_store_bits(skb, 0, data,
2288 dev->hard_header_len);
2292 data += dev->hard_header_len;
2293 to_write -= dev->hard_header_len;
2296 offset = offset_in_page(data);
2297 len_max = PAGE_SIZE - offset;
2298 len = ((to_write > len_max) ? len_max : to_write);
2300 skb->data_len = to_write;
2301 skb->len += to_write;
2302 skb->truesize += to_write;
2303 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2305 while (likely(to_write)) {
2306 nr_frags = skb_shinfo(skb)->nr_frags;
2308 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2309 pr_err("Packet exceed the number of skb frags(%lu)\n",
2314 page = pgv_to_page(data);
2316 flush_dcache_page(page);
2318 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2321 len_max = PAGE_SIZE;
2322 len = ((to_write > len_max) ? len_max : to_write);
2328 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2330 struct sk_buff *skb;
2331 struct net_device *dev;
2333 int err, reserve = 0;
2335 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2336 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2337 int tp_len, size_max;
2338 unsigned char *addr;
2340 int status = TP_STATUS_AVAILABLE;
2343 mutex_lock(&po->pg_vec_lock);
2345 if (likely(saddr == NULL)) {
2346 dev = packet_cached_dev_get(po);
2351 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2353 if (msg->msg_namelen < (saddr->sll_halen
2354 + offsetof(struct sockaddr_ll,
2357 proto = saddr->sll_protocol;
2358 addr = saddr->sll_addr;
2359 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2363 if (unlikely(dev == NULL))
2366 if (unlikely(!(dev->flags & IFF_UP)))
2369 reserve = dev->hard_header_len + VLAN_HLEN;
2370 size_max = po->tx_ring.frame_size
2371 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2373 if (size_max > dev->mtu + reserve)
2374 size_max = dev->mtu + reserve;
2377 ph = packet_current_frame(po, &po->tx_ring,
2378 TP_STATUS_SEND_REQUEST);
2379 if (unlikely(ph == NULL)) {
2380 if (need_wait && need_resched())
2385 status = TP_STATUS_SEND_REQUEST;
2386 hlen = LL_RESERVED_SPACE(dev);
2387 tlen = dev->needed_tailroom;
2388 skb = sock_alloc_send_skb(&po->sk,
2389 hlen + tlen + sizeof(struct sockaddr_ll),
2392 if (unlikely(skb == NULL)) {
2393 /* we assume the socket was initially writeable ... */
2394 if (likely(len_sum > 0))
2398 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2400 if (tp_len > dev->mtu + dev->hard_header_len) {
2401 struct ethhdr *ehdr;
2402 /* Earlier code assumed this would be a VLAN pkt,
2403 * double-check this now that we have the actual
2407 skb_reset_mac_header(skb);
2408 ehdr = eth_hdr(skb);
2409 if (ehdr->h_proto != htons(ETH_P_8021Q))
2412 if (unlikely(tp_len < 0)) {
2414 __packet_set_status(po, ph,
2415 TP_STATUS_AVAILABLE);
2416 packet_increment_head(&po->tx_ring);
2420 status = TP_STATUS_WRONG_FORMAT;
2426 packet_pick_tx_queue(dev, skb);
2428 skb->destructor = tpacket_destruct_skb;
2429 __packet_set_status(po, ph, TP_STATUS_SENDING);
2430 packet_inc_pending(&po->tx_ring);
2432 status = TP_STATUS_SEND_REQUEST;
2433 err = po->xmit(skb);
2434 if (unlikely(err > 0)) {
2435 err = net_xmit_errno(err);
2436 if (err && __packet_get_status(po, ph) ==
2437 TP_STATUS_AVAILABLE) {
2438 /* skb was destructed already */
2443 * skb was dropped but not destructed yet;
2444 * let's treat it like congestion or err < 0
2448 packet_increment_head(&po->tx_ring);
2450 } while (likely((ph != NULL) ||
2451 /* Note: packet_read_pending() might be slow if we have
2452 * to call it as it's per_cpu variable, but in fast-path
2453 * we already short-circuit the loop with the first
2454 * condition, and luckily don't have to go that path
2457 (need_wait && packet_read_pending(&po->tx_ring))));
2463 __packet_set_status(po, ph, status);
2468 mutex_unlock(&po->pg_vec_lock);
2472 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2473 size_t reserve, size_t len,
2474 size_t linear, int noblock,
2477 struct sk_buff *skb;
2479 /* Under a page? Don't bother with paged skb. */
2480 if (prepad + len < PAGE_SIZE || !linear)
2483 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2488 skb_reserve(skb, reserve);
2489 skb_put(skb, linear);
2490 skb->data_len = len - linear;
2491 skb->len += len - linear;
2496 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2498 struct sock *sk = sock->sk;
2499 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2500 struct sk_buff *skb;
2501 struct net_device *dev;
2503 unsigned char *addr;
2504 int err, reserve = 0;
2505 struct virtio_net_hdr vnet_hdr = { 0 };
2508 struct packet_sock *po = pkt_sk(sk);
2509 unsigned short gso_type = 0;
2515 * Get and verify the address.
2518 if (likely(saddr == NULL)) {
2519 dev = packet_cached_dev_get(po);
2524 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2526 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2528 proto = saddr->sll_protocol;
2529 addr = saddr->sll_addr;
2530 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2534 if (unlikely(dev == NULL))
2537 if (unlikely(!(dev->flags & IFF_UP)))
2540 if (sock->type == SOCK_RAW)
2541 reserve = dev->hard_header_len;
2542 if (po->has_vnet_hdr) {
2543 vnet_hdr_len = sizeof(vnet_hdr);
2546 if (len < vnet_hdr_len)
2549 len -= vnet_hdr_len;
2552 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2553 if (n != vnet_hdr_len)
2556 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2557 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2558 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2559 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2560 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2561 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2562 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2565 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2568 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2569 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2570 case VIRTIO_NET_HDR_GSO_TCPV4:
2571 gso_type = SKB_GSO_TCPV4;
2573 case VIRTIO_NET_HDR_GSO_TCPV6:
2574 gso_type = SKB_GSO_TCPV6;
2576 case VIRTIO_NET_HDR_GSO_UDP:
2577 gso_type = SKB_GSO_UDP;
2583 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2584 gso_type |= SKB_GSO_TCP_ECN;
2586 if (vnet_hdr.gso_size == 0)
2592 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2593 if (!netif_supports_nofcs(dev)) {
2594 err = -EPROTONOSUPPORT;
2597 extra_len = 4; /* We're doing our own CRC */
2601 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2605 hlen = LL_RESERVED_SPACE(dev);
2606 tlen = dev->needed_tailroom;
2607 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2608 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2609 msg->msg_flags & MSG_DONTWAIT, &err);
2613 skb_set_network_header(skb, reserve);
2616 if (sock->type == SOCK_DGRAM) {
2617 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2618 if (unlikely(offset < 0))
2621 if (ll_header_truncated(dev, len))
2625 /* Returns -EFAULT on error */
2626 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2630 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2632 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2633 /* Earlier code assumed this would be a VLAN pkt,
2634 * double-check this now that we have the actual
2637 struct ethhdr *ehdr;
2638 skb_reset_mac_header(skb);
2639 ehdr = eth_hdr(skb);
2640 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2646 skb->protocol = proto;
2648 skb->priority = sk->sk_priority;
2649 skb->mark = sk->sk_mark;
2651 packet_pick_tx_queue(dev, skb);
2653 if (po->has_vnet_hdr) {
2654 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2655 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2656 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2657 if (!skb_partial_csum_set(skb, s, o)) {
2663 skb_shinfo(skb)->gso_size =
2664 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2665 skb_shinfo(skb)->gso_type = gso_type;
2667 /* Header must be checked, and gso_segs computed. */
2668 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2669 skb_shinfo(skb)->gso_segs = 0;
2671 len += vnet_hdr_len;
2674 if (!packet_use_direct_xmit(po))
2675 skb_probe_transport_header(skb, reserve);
2676 if (unlikely(extra_len == 4))
2679 err = po->xmit(skb);
2680 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2696 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2698 struct sock *sk = sock->sk;
2699 struct packet_sock *po = pkt_sk(sk);
2701 if (po->tx_ring.pg_vec)
2702 return tpacket_snd(po, msg);
2704 return packet_snd(sock, msg, len);
2708 * Close a PACKET socket. This is fairly simple. We immediately go
2709 * to 'closed' state and remove our protocol entry in the device list.
2712 static int packet_release(struct socket *sock)
2714 struct sock *sk = sock->sk;
2715 struct packet_sock *po;
2717 union tpacket_req_u req_u;
2725 mutex_lock(&net->packet.sklist_lock);
2726 sk_del_node_init_rcu(sk);
2727 mutex_unlock(&net->packet.sklist_lock);
2730 sock_prot_inuse_add(net, sk->sk_prot, -1);
2733 spin_lock(&po->bind_lock);
2734 unregister_prot_hook(sk, false);
2735 packet_cached_dev_reset(po);
2737 if (po->prot_hook.dev) {
2738 dev_put(po->prot_hook.dev);
2739 po->prot_hook.dev = NULL;
2741 spin_unlock(&po->bind_lock);
2743 packet_flush_mclist(sk);
2745 if (po->rx_ring.pg_vec) {
2746 memset(&req_u, 0, sizeof(req_u));
2747 packet_set_ring(sk, &req_u, 1, 0);
2750 if (po->tx_ring.pg_vec) {
2751 memset(&req_u, 0, sizeof(req_u));
2752 packet_set_ring(sk, &req_u, 1, 1);
2759 * Now the socket is dead. No more input will appear.
2766 skb_queue_purge(&sk->sk_receive_queue);
2767 packet_free_pending(po);
2768 sk_refcnt_debug_release(sk);
2775 * Attach a packet hook.
2778 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2780 struct packet_sock *po = pkt_sk(sk);
2781 const struct net_device *dev_curr;
2793 spin_lock(&po->bind_lock);
2795 proto_curr = po->prot_hook.type;
2796 dev_curr = po->prot_hook.dev;
2798 need_rehook = proto_curr != proto || dev_curr != dev;
2801 unregister_prot_hook(sk, true);
2804 po->prot_hook.type = proto;
2806 if (po->prot_hook.dev)
2807 dev_put(po->prot_hook.dev);
2809 po->prot_hook.dev = dev;
2811 po->ifindex = dev ? dev->ifindex : 0;
2812 packet_cached_dev_assign(po, dev);
2815 if (proto == 0 || !need_rehook)
2818 if (!dev || (dev->flags & IFF_UP)) {
2819 register_prot_hook(sk);
2821 sk->sk_err = ENETDOWN;
2822 if (!sock_flag(sk, SOCK_DEAD))
2823 sk->sk_error_report(sk);
2827 spin_unlock(&po->bind_lock);
2833 * Bind a packet socket to a device
2836 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2839 struct sock *sk = sock->sk;
2841 struct net_device *dev;
2848 if (addr_len != sizeof(struct sockaddr))
2850 strlcpy(name, uaddr->sa_data, sizeof(name));
2852 dev = dev_get_by_name(sock_net(sk), name);
2854 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2858 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2860 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2861 struct sock *sk = sock->sk;
2862 struct net_device *dev = NULL;
2870 if (addr_len < sizeof(struct sockaddr_ll))
2872 if (sll->sll_family != AF_PACKET)
2875 if (sll->sll_ifindex) {
2877 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2881 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2887 static struct proto packet_proto = {
2889 .owner = THIS_MODULE,
2890 .obj_size = sizeof(struct packet_sock),
2894 * Create a packet of type SOCK_PACKET.
2897 static int packet_create(struct net *net, struct socket *sock, int protocol,
2901 struct packet_sock *po;
2902 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2905 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2907 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2908 sock->type != SOCK_PACKET)
2909 return -ESOCKTNOSUPPORT;
2911 sock->state = SS_UNCONNECTED;
2914 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2918 sock->ops = &packet_ops;
2919 if (sock->type == SOCK_PACKET)
2920 sock->ops = &packet_ops_spkt;
2922 sock_init_data(sock, sk);
2925 sk->sk_family = PF_PACKET;
2927 po->xmit = dev_queue_xmit;
2929 err = packet_alloc_pending(po);
2933 packet_cached_dev_reset(po);
2935 sk->sk_destruct = packet_sock_destruct;
2936 sk_refcnt_debug_inc(sk);
2939 * Attach a protocol block
2942 spin_lock_init(&po->bind_lock);
2943 mutex_init(&po->pg_vec_lock);
2944 po->rollover = NULL;
2945 po->prot_hook.func = packet_rcv;
2947 if (sock->type == SOCK_PACKET)
2948 po->prot_hook.func = packet_rcv_spkt;
2950 po->prot_hook.af_packet_priv = sk;
2953 po->prot_hook.type = proto;
2954 register_prot_hook(sk);
2957 mutex_lock(&net->packet.sklist_lock);
2958 sk_add_node_rcu(sk, &net->packet.sklist);
2959 mutex_unlock(&net->packet.sklist_lock);
2962 sock_prot_inuse_add(net, &packet_proto, 1);
2973 * Pull a packet from our receive queue and hand it to the user.
2974 * If necessary we block.
2977 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2980 struct sock *sk = sock->sk;
2981 struct sk_buff *skb;
2983 int vnet_hdr_len = 0;
2984 unsigned int origlen = 0;
2987 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2991 /* What error should we return now? EUNATTACH? */
2992 if (pkt_sk(sk)->ifindex < 0)
2996 if (flags & MSG_ERRQUEUE) {
2997 err = sock_recv_errqueue(sk, msg, len,
2998 SOL_PACKET, PACKET_TX_TIMESTAMP);
3003 * Call the generic datagram receiver. This handles all sorts
3004 * of horrible races and re-entrancy so we can forget about it
3005 * in the protocol layers.
3007 * Now it will return ENETDOWN, if device have just gone down,
3008 * but then it will block.
3011 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3014 * An error occurred so return it. Because skb_recv_datagram()
3015 * handles the blocking we don't see and worry about blocking
3022 if (pkt_sk(sk)->pressure)
3023 packet_rcv_has_room(pkt_sk(sk), NULL);
3025 if (pkt_sk(sk)->has_vnet_hdr) {
3026 struct virtio_net_hdr vnet_hdr = { 0 };
3029 vnet_hdr_len = sizeof(vnet_hdr);
3030 if (len < vnet_hdr_len)
3033 len -= vnet_hdr_len;
3035 if (skb_is_gso(skb)) {
3036 struct skb_shared_info *sinfo = skb_shinfo(skb);
3038 /* This is a hint as to how much should be linear. */
3040 __cpu_to_virtio16(false, skb_headlen(skb));
3042 __cpu_to_virtio16(false, sinfo->gso_size);
3043 if (sinfo->gso_type & SKB_GSO_TCPV4)
3044 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3045 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3046 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3047 else if (sinfo->gso_type & SKB_GSO_UDP)
3048 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3049 else if (sinfo->gso_type & SKB_GSO_FCOE)
3053 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3054 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3056 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3058 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3059 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3060 vnet_hdr.csum_start = __cpu_to_virtio16(false,
3061 skb_checksum_start_offset(skb));
3062 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3064 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3065 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3066 } /* else everything is zero */
3068 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3073 /* You lose any data beyond the buffer you gave. If it worries
3074 * a user program they can ask the device for its MTU
3080 msg->msg_flags |= MSG_TRUNC;
3083 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3087 if (sock->type != SOCK_PACKET) {
3088 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3090 /* Original length was stored in sockaddr_ll fields */
3091 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3092 sll->sll_family = AF_PACKET;
3093 sll->sll_protocol = skb->protocol;
3096 sock_recv_ts_and_drops(msg, sk, skb);
3098 if (msg->msg_name) {
3099 /* If the address length field is there to be filled
3100 * in, we fill it in now.
3102 if (sock->type == SOCK_PACKET) {
3103 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3104 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3106 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3108 msg->msg_namelen = sll->sll_halen +
3109 offsetof(struct sockaddr_ll, sll_addr);
3111 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3115 if (pkt_sk(sk)->auxdata) {
3116 struct tpacket_auxdata aux;
3118 aux.tp_status = TP_STATUS_USER;
3119 if (skb->ip_summed == CHECKSUM_PARTIAL)
3120 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3121 else if (skb->pkt_type != PACKET_OUTGOING &&
3122 (skb->ip_summed == CHECKSUM_COMPLETE ||
3123 skb_csum_unnecessary(skb)))
3124 aux.tp_status |= TP_STATUS_CSUM_VALID;
3126 aux.tp_len = origlen;
3127 aux.tp_snaplen = skb->len;
3129 aux.tp_net = skb_network_offset(skb);
3130 if (skb_vlan_tag_present(skb)) {
3131 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3132 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3133 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3135 aux.tp_vlan_tci = 0;
3136 aux.tp_vlan_tpid = 0;
3138 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3142 * Free or return the buffer as appropriate. Again this
3143 * hides all the races and re-entrancy issues from us.
3145 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3148 skb_free_datagram(sk, skb);
3153 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3154 int *uaddr_len, int peer)
3156 struct net_device *dev;
3157 struct sock *sk = sock->sk;
3162 uaddr->sa_family = AF_PACKET;
3163 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3165 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3167 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3169 *uaddr_len = sizeof(*uaddr);
3174 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3175 int *uaddr_len, int peer)
3177 struct net_device *dev;
3178 struct sock *sk = sock->sk;
3179 struct packet_sock *po = pkt_sk(sk);
3180 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3185 sll->sll_family = AF_PACKET;
3186 sll->sll_ifindex = po->ifindex;
3187 sll->sll_protocol = po->num;
3188 sll->sll_pkttype = 0;
3190 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3192 sll->sll_hatype = dev->type;
3193 sll->sll_halen = dev->addr_len;
3194 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3196 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3200 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3205 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3209 case PACKET_MR_MULTICAST:
3210 if (i->alen != dev->addr_len)
3213 return dev_mc_add(dev, i->addr);
3215 return dev_mc_del(dev, i->addr);
3217 case PACKET_MR_PROMISC:
3218 return dev_set_promiscuity(dev, what);
3219 case PACKET_MR_ALLMULTI:
3220 return dev_set_allmulti(dev, what);
3221 case PACKET_MR_UNICAST:
3222 if (i->alen != dev->addr_len)
3225 return dev_uc_add(dev, i->addr);
3227 return dev_uc_del(dev, i->addr);
3235 static void packet_dev_mclist_delete(struct net_device *dev,
3236 struct packet_mclist **mlp)
3238 struct packet_mclist *ml;
3240 while ((ml = *mlp) != NULL) {
3241 if (ml->ifindex == dev->ifindex) {
3242 packet_dev_mc(dev, ml, -1);
3250 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3252 struct packet_sock *po = pkt_sk(sk);
3253 struct packet_mclist *ml, *i;
3254 struct net_device *dev;
3260 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3265 if (mreq->mr_alen > dev->addr_len)
3269 i = kmalloc(sizeof(*i), GFP_KERNEL);
3274 for (ml = po->mclist; ml; ml = ml->next) {
3275 if (ml->ifindex == mreq->mr_ifindex &&
3276 ml->type == mreq->mr_type &&
3277 ml->alen == mreq->mr_alen &&
3278 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3280 /* Free the new element ... */
3286 i->type = mreq->mr_type;
3287 i->ifindex = mreq->mr_ifindex;
3288 i->alen = mreq->mr_alen;
3289 memcpy(i->addr, mreq->mr_address, i->alen);
3291 i->next = po->mclist;
3293 err = packet_dev_mc(dev, i, 1);
3295 po->mclist = i->next;
3304 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3306 struct packet_mclist *ml, **mlp;
3310 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3311 if (ml->ifindex == mreq->mr_ifindex &&
3312 ml->type == mreq->mr_type &&
3313 ml->alen == mreq->mr_alen &&
3314 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3315 if (--ml->count == 0) {
3316 struct net_device *dev;
3318 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3320 packet_dev_mc(dev, ml, -1);
3330 static void packet_flush_mclist(struct sock *sk)
3332 struct packet_sock *po = pkt_sk(sk);
3333 struct packet_mclist *ml;
3339 while ((ml = po->mclist) != NULL) {
3340 struct net_device *dev;
3342 po->mclist = ml->next;
3343 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3345 packet_dev_mc(dev, ml, -1);
3352 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3354 struct sock *sk = sock->sk;
3355 struct packet_sock *po = pkt_sk(sk);
3358 if (level != SOL_PACKET)
3359 return -ENOPROTOOPT;
3362 case PACKET_ADD_MEMBERSHIP:
3363 case PACKET_DROP_MEMBERSHIP:
3365 struct packet_mreq_max mreq;
3367 memset(&mreq, 0, sizeof(mreq));
3368 if (len < sizeof(struct packet_mreq))
3370 if (len > sizeof(mreq))
3372 if (copy_from_user(&mreq, optval, len))
3374 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3376 if (optname == PACKET_ADD_MEMBERSHIP)
3377 ret = packet_mc_add(sk, &mreq);
3379 ret = packet_mc_drop(sk, &mreq);
3383 case PACKET_RX_RING:
3384 case PACKET_TX_RING:
3386 union tpacket_req_u req_u;
3389 switch (po->tp_version) {
3392 len = sizeof(req_u.req);
3396 len = sizeof(req_u.req3);
3401 if (pkt_sk(sk)->has_vnet_hdr)
3403 if (copy_from_user(&req_u.req, optval, len))
3405 return packet_set_ring(sk, &req_u, 0,
3406 optname == PACKET_TX_RING);
3408 case PACKET_COPY_THRESH:
3412 if (optlen != sizeof(val))
3414 if (copy_from_user(&val, optval, sizeof(val)))
3417 pkt_sk(sk)->copy_thresh = val;
3420 case PACKET_VERSION:
3424 if (optlen != sizeof(val))
3426 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3428 if (copy_from_user(&val, optval, sizeof(val)))
3434 po->tp_version = val;
3440 case PACKET_RESERVE:
3444 if (optlen != sizeof(val))
3446 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3448 if (copy_from_user(&val, optval, sizeof(val)))
3450 po->tp_reserve = val;
3457 if (optlen != sizeof(val))
3459 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3461 if (copy_from_user(&val, optval, sizeof(val)))
3463 po->tp_loss = !!val;
3466 case PACKET_AUXDATA:
3470 if (optlen < sizeof(val))
3472 if (copy_from_user(&val, optval, sizeof(val)))
3475 po->auxdata = !!val;
3478 case PACKET_ORIGDEV:
3482 if (optlen < sizeof(val))
3484 if (copy_from_user(&val, optval, sizeof(val)))
3487 po->origdev = !!val;
3490 case PACKET_VNET_HDR:
3494 if (sock->type != SOCK_RAW)
3496 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3498 if (optlen < sizeof(val))
3500 if (copy_from_user(&val, optval, sizeof(val)))
3503 po->has_vnet_hdr = !!val;
3506 case PACKET_TIMESTAMP:
3510 if (optlen != sizeof(val))
3512 if (copy_from_user(&val, optval, sizeof(val)))
3515 po->tp_tstamp = val;
3522 if (optlen != sizeof(val))
3524 if (copy_from_user(&val, optval, sizeof(val)))
3527 return fanout_add(sk, val & 0xffff, val >> 16);
3529 case PACKET_TX_HAS_OFF:
3533 if (optlen != sizeof(val))
3535 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3537 if (copy_from_user(&val, optval, sizeof(val)))
3539 po->tp_tx_has_off = !!val;
3542 case PACKET_QDISC_BYPASS:
3546 if (optlen != sizeof(val))
3548 if (copy_from_user(&val, optval, sizeof(val)))
3551 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3555 return -ENOPROTOOPT;
3559 static int packet_getsockopt(struct socket *sock, int level, int optname,
3560 char __user *optval, int __user *optlen)
3563 int val, lv = sizeof(val);
3564 struct sock *sk = sock->sk;
3565 struct packet_sock *po = pkt_sk(sk);
3567 union tpacket_stats_u st;
3569 if (level != SOL_PACKET)
3570 return -ENOPROTOOPT;
3572 if (get_user(len, optlen))
3579 case PACKET_STATISTICS:
3580 spin_lock_bh(&sk->sk_receive_queue.lock);
3581 memcpy(&st, &po->stats, sizeof(st));
3582 memset(&po->stats, 0, sizeof(po->stats));
3583 spin_unlock_bh(&sk->sk_receive_queue.lock);
3585 if (po->tp_version == TPACKET_V3) {
3586 lv = sizeof(struct tpacket_stats_v3);
3587 st.stats3.tp_packets += st.stats3.tp_drops;
3590 lv = sizeof(struct tpacket_stats);
3591 st.stats1.tp_packets += st.stats1.tp_drops;
3596 case PACKET_AUXDATA:
3599 case PACKET_ORIGDEV:
3602 case PACKET_VNET_HDR:
3603 val = po->has_vnet_hdr;
3605 case PACKET_VERSION:
3606 val = po->tp_version;
3609 if (len > sizeof(int))
3611 if (copy_from_user(&val, optval, len))
3615 val = sizeof(struct tpacket_hdr);
3618 val = sizeof(struct tpacket2_hdr);
3621 val = sizeof(struct tpacket3_hdr);
3627 case PACKET_RESERVE:
3628 val = po->tp_reserve;
3633 case PACKET_TIMESTAMP:
3634 val = po->tp_tstamp;
3638 ((u32)po->fanout->id |
3639 ((u32)po->fanout->type << 16) |
3640 ((u32)po->fanout->flags << 24)) :
3643 case PACKET_TX_HAS_OFF:
3644 val = po->tp_tx_has_off;
3646 case PACKET_QDISC_BYPASS:
3647 val = packet_use_direct_xmit(po);
3650 return -ENOPROTOOPT;
3655 if (put_user(len, optlen))
3657 if (copy_to_user(optval, data, len))
3663 static int packet_notifier(struct notifier_block *this,
3664 unsigned long msg, void *ptr)
3667 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3668 struct net *net = dev_net(dev);
3671 sk_for_each_rcu(sk, &net->packet.sklist) {
3672 struct packet_sock *po = pkt_sk(sk);
3675 case NETDEV_UNREGISTER:
3677 packet_dev_mclist_delete(dev, &po->mclist);
3681 if (dev->ifindex == po->ifindex) {
3682 spin_lock(&po->bind_lock);
3684 __unregister_prot_hook(sk, false);
3685 sk->sk_err = ENETDOWN;
3686 if (!sock_flag(sk, SOCK_DEAD))
3687 sk->sk_error_report(sk);
3689 if (msg == NETDEV_UNREGISTER) {
3690 packet_cached_dev_reset(po);
3692 if (po->prot_hook.dev)
3693 dev_put(po->prot_hook.dev);
3694 po->prot_hook.dev = NULL;
3696 spin_unlock(&po->bind_lock);
3700 if (dev->ifindex == po->ifindex) {
3701 spin_lock(&po->bind_lock);
3703 register_prot_hook(sk);
3704 spin_unlock(&po->bind_lock);
3714 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3717 struct sock *sk = sock->sk;
3722 int amount = sk_wmem_alloc_get(sk);
3724 return put_user(amount, (int __user *)arg);
3728 struct sk_buff *skb;
3731 spin_lock_bh(&sk->sk_receive_queue.lock);
3732 skb = skb_peek(&sk->sk_receive_queue);
3735 spin_unlock_bh(&sk->sk_receive_queue.lock);
3736 return put_user(amount, (int __user *)arg);
3739 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3741 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3751 case SIOCGIFBRDADDR:
3752 case SIOCSIFBRDADDR:
3753 case SIOCGIFNETMASK:
3754 case SIOCSIFNETMASK:
3755 case SIOCGIFDSTADDR:
3756 case SIOCSIFDSTADDR:
3758 return inet_dgram_ops.ioctl(sock, cmd, arg);
3762 return -ENOIOCTLCMD;
3767 static unsigned int packet_poll(struct file *file, struct socket *sock,
3770 struct sock *sk = sock->sk;
3771 struct packet_sock *po = pkt_sk(sk);
3772 unsigned int mask = datagram_poll(file, sock, wait);
3774 spin_lock_bh(&sk->sk_receive_queue.lock);
3775 if (po->rx_ring.pg_vec) {
3776 if (!packet_previous_rx_frame(po, &po->rx_ring,
3778 mask |= POLLIN | POLLRDNORM;
3780 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3781 xchg(&po->pressure, 0);
3782 spin_unlock_bh(&sk->sk_receive_queue.lock);
3783 spin_lock_bh(&sk->sk_write_queue.lock);
3784 if (po->tx_ring.pg_vec) {
3785 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3786 mask |= POLLOUT | POLLWRNORM;
3788 spin_unlock_bh(&sk->sk_write_queue.lock);
3793 /* Dirty? Well, I still did not learn better way to account
3797 static void packet_mm_open(struct vm_area_struct *vma)
3799 struct file *file = vma->vm_file;
3800 struct socket *sock = file->private_data;
3801 struct sock *sk = sock->sk;
3804 atomic_inc(&pkt_sk(sk)->mapped);
3807 static void packet_mm_close(struct vm_area_struct *vma)
3809 struct file *file = vma->vm_file;
3810 struct socket *sock = file->private_data;
3811 struct sock *sk = sock->sk;
3814 atomic_dec(&pkt_sk(sk)->mapped);
3817 static const struct vm_operations_struct packet_mmap_ops = {
3818 .open = packet_mm_open,
3819 .close = packet_mm_close,
3822 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3827 for (i = 0; i < len; i++) {
3828 if (likely(pg_vec[i].buffer)) {
3829 if (is_vmalloc_addr(pg_vec[i].buffer))
3830 vfree(pg_vec[i].buffer);
3832 free_pages((unsigned long)pg_vec[i].buffer,
3834 pg_vec[i].buffer = NULL;
3840 static char *alloc_one_pg_vec_page(unsigned long order)
3843 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3844 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3846 buffer = (char *) __get_free_pages(gfp_flags, order);
3850 /* __get_free_pages failed, fall back to vmalloc */
3851 buffer = vzalloc((1 << order) * PAGE_SIZE);
3855 /* vmalloc failed, lets dig into swap here */
3856 gfp_flags &= ~__GFP_NORETRY;
3857 buffer = (char *) __get_free_pages(gfp_flags, order);
3861 /* complete and utter failure */
3865 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3867 unsigned int block_nr = req->tp_block_nr;
3871 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3872 if (unlikely(!pg_vec))
3875 for (i = 0; i < block_nr; i++) {
3876 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3877 if (unlikely(!pg_vec[i].buffer))
3878 goto out_free_pgvec;
3885 free_pg_vec(pg_vec, order, block_nr);
3890 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3891 int closing, int tx_ring)
3893 struct pgv *pg_vec = NULL;
3894 struct packet_sock *po = pkt_sk(sk);
3895 int was_running, order = 0;
3896 struct packet_ring_buffer *rb;
3897 struct sk_buff_head *rb_queue;
3900 /* Added to avoid minimal code churn */
3901 struct tpacket_req *req = &req_u->req;
3903 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3904 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3905 WARN(1, "Tx-ring is not supported.\n");
3909 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3910 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3914 if (atomic_read(&po->mapped))
3916 if (packet_read_pending(rb))
3920 if (req->tp_block_nr) {
3921 /* Sanity tests and some calculations */
3923 if (unlikely(rb->pg_vec))
3926 switch (po->tp_version) {
3928 po->tp_hdrlen = TPACKET_HDRLEN;
3931 po->tp_hdrlen = TPACKET2_HDRLEN;
3934 po->tp_hdrlen = TPACKET3_HDRLEN;
3939 if (unlikely((int)req->tp_block_size <= 0))
3941 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3943 if (po->tp_version >= TPACKET_V3 &&
3944 (int)(req->tp_block_size -
3945 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3947 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3950 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3953 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3954 if (unlikely(rb->frames_per_block <= 0))
3956 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3961 order = get_order(req->tp_block_size);
3962 pg_vec = alloc_pg_vec(req, order);
3963 if (unlikely(!pg_vec))
3965 switch (po->tp_version) {
3967 /* Transmit path is not supported. We checked
3968 * it above but just being paranoid
3971 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3980 if (unlikely(req->tp_frame_nr))
3986 /* Detach socket from network */
3987 spin_lock(&po->bind_lock);
3988 was_running = po->running;
3992 __unregister_prot_hook(sk, false);
3994 spin_unlock(&po->bind_lock);
3999 mutex_lock(&po->pg_vec_lock);
4000 if (closing || atomic_read(&po->mapped) == 0) {
4002 spin_lock_bh(&rb_queue->lock);
4003 swap(rb->pg_vec, pg_vec);
4004 rb->frame_max = (req->tp_frame_nr - 1);
4006 rb->frame_size = req->tp_frame_size;
4007 spin_unlock_bh(&rb_queue->lock);
4009 swap(rb->pg_vec_order, order);
4010 swap(rb->pg_vec_len, req->tp_block_nr);
4012 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4013 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4014 tpacket_rcv : packet_rcv;
4015 skb_queue_purge(rb_queue);
4016 if (atomic_read(&po->mapped))
4017 pr_err("packet_mmap: vma is busy: %d\n",
4018 atomic_read(&po->mapped));
4020 mutex_unlock(&po->pg_vec_lock);
4022 spin_lock(&po->bind_lock);
4025 register_prot_hook(sk);
4027 spin_unlock(&po->bind_lock);
4028 if (closing && (po->tp_version > TPACKET_V2)) {
4029 /* Because we don't support block-based V3 on tx-ring */
4031 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4036 free_pg_vec(pg_vec, order, req->tp_block_nr);
4041 static int packet_mmap(struct file *file, struct socket *sock,
4042 struct vm_area_struct *vma)
4044 struct sock *sk = sock->sk;
4045 struct packet_sock *po = pkt_sk(sk);
4046 unsigned long size, expected_size;
4047 struct packet_ring_buffer *rb;
4048 unsigned long start;
4055 mutex_lock(&po->pg_vec_lock);
4058 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4060 expected_size += rb->pg_vec_len
4066 if (expected_size == 0)
4069 size = vma->vm_end - vma->vm_start;
4070 if (size != expected_size)
4073 start = vma->vm_start;
4074 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4075 if (rb->pg_vec == NULL)
4078 for (i = 0; i < rb->pg_vec_len; i++) {
4080 void *kaddr = rb->pg_vec[i].buffer;
4083 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4084 page = pgv_to_page(kaddr);
4085 err = vm_insert_page(vma, start, page);
4094 atomic_inc(&po->mapped);
4095 vma->vm_ops = &packet_mmap_ops;
4099 mutex_unlock(&po->pg_vec_lock);
4103 static const struct proto_ops packet_ops_spkt = {
4104 .family = PF_PACKET,
4105 .owner = THIS_MODULE,
4106 .release = packet_release,
4107 .bind = packet_bind_spkt,
4108 .connect = sock_no_connect,
4109 .socketpair = sock_no_socketpair,
4110 .accept = sock_no_accept,
4111 .getname = packet_getname_spkt,
4112 .poll = datagram_poll,
4113 .ioctl = packet_ioctl,
4114 .listen = sock_no_listen,
4115 .shutdown = sock_no_shutdown,
4116 .setsockopt = sock_no_setsockopt,
4117 .getsockopt = sock_no_getsockopt,
4118 .sendmsg = packet_sendmsg_spkt,
4119 .recvmsg = packet_recvmsg,
4120 .mmap = sock_no_mmap,
4121 .sendpage = sock_no_sendpage,
4124 static const struct proto_ops packet_ops = {
4125 .family = PF_PACKET,
4126 .owner = THIS_MODULE,
4127 .release = packet_release,
4128 .bind = packet_bind,
4129 .connect = sock_no_connect,
4130 .socketpair = sock_no_socketpair,
4131 .accept = sock_no_accept,
4132 .getname = packet_getname,
4133 .poll = packet_poll,
4134 .ioctl = packet_ioctl,
4135 .listen = sock_no_listen,
4136 .shutdown = sock_no_shutdown,
4137 .setsockopt = packet_setsockopt,
4138 .getsockopt = packet_getsockopt,
4139 .sendmsg = packet_sendmsg,
4140 .recvmsg = packet_recvmsg,
4141 .mmap = packet_mmap,
4142 .sendpage = sock_no_sendpage,
4145 static const struct net_proto_family packet_family_ops = {
4146 .family = PF_PACKET,
4147 .create = packet_create,
4148 .owner = THIS_MODULE,
4151 static struct notifier_block packet_netdev_notifier = {
4152 .notifier_call = packet_notifier,
4155 #ifdef CONFIG_PROC_FS
4157 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4160 struct net *net = seq_file_net(seq);
4163 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4166 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4168 struct net *net = seq_file_net(seq);
4169 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4172 static void packet_seq_stop(struct seq_file *seq, void *v)
4178 static int packet_seq_show(struct seq_file *seq, void *v)
4180 if (v == SEQ_START_TOKEN)
4181 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4183 struct sock *s = sk_entry(v);
4184 const struct packet_sock *po = pkt_sk(s);
4187 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4189 atomic_read(&s->sk_refcnt),
4194 atomic_read(&s->sk_rmem_alloc),
4195 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4202 static const struct seq_operations packet_seq_ops = {
4203 .start = packet_seq_start,
4204 .next = packet_seq_next,
4205 .stop = packet_seq_stop,
4206 .show = packet_seq_show,
4209 static int packet_seq_open(struct inode *inode, struct file *file)
4211 return seq_open_net(inode, file, &packet_seq_ops,
4212 sizeof(struct seq_net_private));
4215 static const struct file_operations packet_seq_fops = {
4216 .owner = THIS_MODULE,
4217 .open = packet_seq_open,
4219 .llseek = seq_lseek,
4220 .release = seq_release_net,
4225 static int __net_init packet_net_init(struct net *net)
4227 mutex_init(&net->packet.sklist_lock);
4228 INIT_HLIST_HEAD(&net->packet.sklist);
4230 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4236 static void __net_exit packet_net_exit(struct net *net)
4238 remove_proc_entry("packet", net->proc_net);
4241 static struct pernet_operations packet_net_ops = {
4242 .init = packet_net_init,
4243 .exit = packet_net_exit,
4247 static void __exit packet_exit(void)
4249 unregister_netdevice_notifier(&packet_netdev_notifier);
4250 unregister_pernet_subsys(&packet_net_ops);
4251 sock_unregister(PF_PACKET);
4252 proto_unregister(&packet_proto);
4255 static int __init packet_init(void)
4257 int rc = proto_register(&packet_proto, 0);
4262 sock_register(&packet_family_ops);
4263 register_pernet_subsys(&packet_net_ops);
4264 register_netdevice_notifier(&packet_netdev_notifier);
4269 module_init(packet_init);
4270 module_exit(packet_exit);
4271 MODULE_LICENSE("GPL");
4272 MODULE_ALIAS_NETPROTO(PF_PACKET);