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 {
219 unsigned int origlen;
221 struct sockaddr_pkt pkt;
222 struct sockaddr_ll ll;
226 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
228 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
229 #define GET_PBLOCK_DESC(x, bid) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
233 #define GET_NEXT_PRB_BLK_NUM(x) \
234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 ((x)->kactive_blk_num+1) : 0)
237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
240 static int packet_direct_xmit(struct sk_buff *skb)
242 struct net_device *dev = skb->dev;
243 const struct net_device_ops *ops = dev->netdev_ops;
244 netdev_features_t features;
245 struct netdev_queue *txq;
246 int ret = NETDEV_TX_BUSY;
249 if (unlikely(!netif_running(dev) ||
250 !netif_carrier_ok(dev)))
253 features = netif_skb_features(skb);
254 if (skb_needs_linearize(skb, features) &&
255 __skb_linearize(skb))
258 queue_map = skb_get_queue_mapping(skb);
259 txq = netdev_get_tx_queue(dev, queue_map);
263 HARD_TX_LOCK(dev, txq, smp_processor_id());
264 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
265 ret = ops->ndo_start_xmit(skb, dev);
266 if (ret == NETDEV_TX_OK)
267 txq_trans_update(txq);
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 __pure struct page *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 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
447 return TP_STATUS_TS_SYS_HARDWARE;
448 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
453 if (ktime_to_timespec_cond(skb->tstamp, ts))
454 return TP_STATUS_TS_SOFTWARE;
459 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
462 union tpacket_uhdr h;
466 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
470 switch (po->tp_version) {
472 h.h1->tp_sec = ts.tv_sec;
473 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
476 h.h2->tp_sec = ts.tv_sec;
477 h.h2->tp_nsec = ts.tv_nsec;
481 WARN(1, "TPACKET version not supported.\n");
485 /* one flush is safe, as both fields always lie on the same cacheline */
486 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
492 static void *packet_lookup_frame(struct packet_sock *po,
493 struct packet_ring_buffer *rb,
494 unsigned int position,
497 unsigned int pg_vec_pos, frame_offset;
498 union tpacket_uhdr h;
500 pg_vec_pos = position / rb->frames_per_block;
501 frame_offset = position % rb->frames_per_block;
503 h.raw = rb->pg_vec[pg_vec_pos].buffer +
504 (frame_offset * rb->frame_size);
506 if (status != __packet_get_status(po, h.raw))
512 static void *packet_current_frame(struct packet_sock *po,
513 struct packet_ring_buffer *rb,
516 return packet_lookup_frame(po, rb, rb->head, status);
519 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
521 del_timer_sync(&pkc->retire_blk_timer);
524 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
526 struct sk_buff_head *rb_queue)
528 struct tpacket_kbdq_core *pkc;
530 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
531 GET_PBDQC_FROM_RB(&po->rx_ring);
533 spin_lock_bh(&rb_queue->lock);
534 pkc->delete_blk_timer = 1;
535 spin_unlock_bh(&rb_queue->lock);
537 prb_del_retire_blk_timer(pkc);
540 static void prb_init_blk_timer(struct packet_sock *po,
541 struct tpacket_kbdq_core *pkc,
542 void (*func) (unsigned long))
544 init_timer(&pkc->retire_blk_timer);
545 pkc->retire_blk_timer.data = (long)po;
546 pkc->retire_blk_timer.function = func;
547 pkc->retire_blk_timer.expires = jiffies;
550 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
552 struct tpacket_kbdq_core *pkc;
557 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
558 GET_PBDQC_FROM_RB(&po->rx_ring);
559 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
562 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
563 int blk_size_in_bytes)
565 struct net_device *dev;
566 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
567 struct ethtool_cmd ecmd;
572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
573 if (unlikely(!dev)) {
575 return DEFAULT_PRB_RETIRE_TOV;
577 err = __ethtool_get_settings(dev, &ecmd);
578 speed = ethtool_cmd_speed(&ecmd);
582 * If the link speed is so slow you don't really
583 * need to worry about perf anyways
585 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
586 return DEFAULT_PRB_RETIRE_TOV;
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
614 union tpacket_req_u *req_u, int tx_ring)
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
619 memset(p1, 0x0, sizeof(*p1));
621 p1->knxt_seq_num = 1;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
639 prb_init_ft_ops(p1, req_u);
640 prb_setup_retire_blk_timer(po, tx_ring);
641 prb_open_block(p1, pbd);
644 /* Do NOT update the last_blk_num first.
645 * Assumes sk_buff_head lock is held.
647 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
649 mod_timer(&pkc->retire_blk_timer,
650 jiffies + pkc->tov_in_jiffies);
651 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
656 * 1) We refresh the timer only when we open a block.
657 * By doing this we don't waste cycles refreshing the timer
658 * on packet-by-packet basis.
660 * With a 1MB block-size, on a 1Gbps line, it will take
661 * i) ~8 ms to fill a block + ii) memcpy etc.
662 * In this cut we are not accounting for the memcpy time.
664 * So, if the user sets the 'tmo' to 10ms then the timer
665 * will never fire while the block is still getting filled
666 * (which is what we want). However, the user could choose
667 * to close a block early and that's fine.
669 * But when the timer does fire, we check whether or not to refresh it.
670 * Since the tmo granularity is in msecs, it is not too expensive
671 * to refresh the timer, lets say every '8' msecs.
672 * Either the user can set the 'tmo' or we can derive it based on
673 * a) line-speed and b) block-size.
674 * prb_calc_retire_blk_tmo() calculates the tmo.
677 static void prb_retire_rx_blk_timer_expired(unsigned long data)
679 struct packet_sock *po = (struct packet_sock *)data;
680 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
682 struct tpacket_block_desc *pbd;
684 spin_lock(&po->sk.sk_receive_queue.lock);
686 frozen = prb_queue_frozen(pkc);
687 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
689 if (unlikely(pkc->delete_blk_timer))
692 /* We only need to plug the race when the block is partially filled.
694 * lock(); increment BLOCK_NUM_PKTS; unlock()
695 * copy_bits() is in progress ...
696 * timer fires on other cpu:
697 * we can't retire the current block because copy_bits
701 if (BLOCK_NUM_PKTS(pbd)) {
702 while (atomic_read(&pkc->blk_fill_in_prog)) {
703 /* Waiting for skb_copy_bits to finish... */
708 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
710 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
711 if (!prb_dispatch_next_block(pkc, po))
716 /* Case 1. Queue was frozen because user-space was
719 if (prb_curr_blk_in_use(pkc, pbd)) {
721 * Ok, user-space is still behind.
722 * So just refresh the timer.
726 /* Case 2. queue was frozen,user-space caught up,
727 * now the link went idle && the timer fired.
728 * We don't have a block to close.So we open this
729 * block and restart the timer.
730 * opening a block thaws the queue,restarts timer
731 * Thawing/timer-refresh is a side effect.
733 prb_open_block(pkc, pbd);
740 _prb_refresh_rx_retire_blk_timer(pkc);
743 spin_unlock(&po->sk.sk_receive_queue.lock);
746 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
747 struct tpacket_block_desc *pbd1, __u32 status)
749 /* Flush everything minus the block header */
751 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
756 /* Skip the block header(we know header WILL fit in 4K) */
759 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
760 for (; start < end; start += PAGE_SIZE)
761 flush_dcache_page(pgv_to_page(start));
766 /* Now update the block status. */
768 BLOCK_STATUS(pbd1) = status;
770 /* Flush the block header */
772 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
774 flush_dcache_page(pgv_to_page(start));
784 * 2) Increment active_blk_num
786 * Note:We DONT refresh the timer on purpose.
787 * Because almost always the next block will be opened.
789 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
790 struct tpacket_block_desc *pbd1,
791 struct packet_sock *po, unsigned int stat)
793 __u32 status = TP_STATUS_USER | stat;
795 struct tpacket3_hdr *last_pkt;
796 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
798 if (po->stats.stats3.tp_drops)
799 status |= TP_STATUS_LOSING;
801 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
802 last_pkt->tp_next_offset = 0;
804 /* Get the ts of the last pkt */
805 if (BLOCK_NUM_PKTS(pbd1)) {
806 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
807 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
809 /* Ok, we tmo'd - so get the current time */
812 h1->ts_last_pkt.ts_sec = ts.tv_sec;
813 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
818 /* Flush the block */
819 prb_flush_block(pkc1, pbd1, status);
821 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
824 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
826 pkc->reset_pending_on_curr_blk = 0;
830 * Side effect of opening a block:
832 * 1) prb_queue is thawed.
833 * 2) retire_blk_timer is refreshed.
836 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
837 struct tpacket_block_desc *pbd1)
840 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
844 /* We could have just memset this but we will lose the
845 * flexibility of making the priv area sticky
848 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
849 BLOCK_NUM_PKTS(pbd1) = 0;
850 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
854 h1->ts_first_pkt.ts_sec = ts.tv_sec;
855 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
857 pkc1->pkblk_start = (char *)pbd1;
858 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
860 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
861 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
863 pbd1->version = pkc1->version;
864 pkc1->prev = pkc1->nxt_offset;
865 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
867 prb_thaw_queue(pkc1);
868 _prb_refresh_rx_retire_blk_timer(pkc1);
874 * Queue freeze logic:
875 * 1) Assume tp_block_nr = 8 blocks.
876 * 2) At time 't0', user opens Rx ring.
877 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
878 * 4) user-space is either sleeping or processing block '0'.
879 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
880 * it will close block-7,loop around and try to fill block '0'.
882 * __packet_lookup_frame_in_block
883 * prb_retire_current_block()
884 * prb_dispatch_next_block()
885 * |->(BLOCK_STATUS == USER) evaluates to true
886 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
887 * 6) Now there are two cases:
888 * 6.1) Link goes idle right after the queue is frozen.
889 * But remember, the last open_block() refreshed the timer.
890 * When this timer expires,it will refresh itself so that we can
891 * re-open block-0 in near future.
892 * 6.2) Link is busy and keeps on receiving packets. This is a simple
893 * case and __packet_lookup_frame_in_block will check if block-0
894 * is free and can now be re-used.
896 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
897 struct packet_sock *po)
899 pkc->reset_pending_on_curr_blk = 1;
900 po->stats.stats3.tp_freeze_q_cnt++;
903 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
906 * If the next block is free then we will dispatch it
907 * and return a good offset.
908 * Else, we will freeze the queue.
909 * So, caller must check the return value.
911 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
912 struct packet_sock *po)
914 struct tpacket_block_desc *pbd;
918 /* 1. Get current block num */
919 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
921 /* 2. If this block is currently in_use then freeze the queue */
922 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
923 prb_freeze_queue(pkc, po);
929 * open this block and return the offset where the first packet
930 * needs to get stored.
932 prb_open_block(pkc, pbd);
933 return (void *)pkc->nxt_offset;
936 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
937 struct packet_sock *po, unsigned int status)
939 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
941 /* retire/close the current block */
942 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
944 * Plug the case where copy_bits() is in progress on
945 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
946 * have space to copy the pkt in the current block and
947 * called prb_retire_current_block()
949 * We don't need to worry about the TMO case because
950 * the timer-handler already handled this case.
952 if (!(status & TP_STATUS_BLK_TMO)) {
953 while (atomic_read(&pkc->blk_fill_in_prog)) {
954 /* Waiting for skb_copy_bits to finish... */
958 prb_close_block(pkc, pbd, po, status);
963 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
964 struct tpacket_block_desc *pbd)
966 return TP_STATUS_USER & BLOCK_STATUS(pbd);
969 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
971 return pkc->reset_pending_on_curr_blk;
974 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
976 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
977 atomic_dec(&pkc->blk_fill_in_prog);
980 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
981 struct tpacket3_hdr *ppd)
983 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
986 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
987 struct tpacket3_hdr *ppd)
989 ppd->hv1.tp_rxhash = 0;
992 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
993 struct tpacket3_hdr *ppd)
995 if (vlan_tx_tag_present(pkc->skb)) {
996 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
997 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
998 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1000 ppd->hv1.tp_vlan_tci = 0;
1001 ppd->hv1.tp_vlan_tpid = 0;
1002 ppd->tp_status = TP_STATUS_AVAILABLE;
1006 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1007 struct tpacket3_hdr *ppd)
1009 ppd->hv1.tp_padding = 0;
1010 prb_fill_vlan_info(pkc, ppd);
1012 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1013 prb_fill_rxhash(pkc, ppd);
1015 prb_clear_rxhash(pkc, ppd);
1018 static void prb_fill_curr_block(char *curr,
1019 struct tpacket_kbdq_core *pkc,
1020 struct tpacket_block_desc *pbd,
1023 struct tpacket3_hdr *ppd;
1025 ppd = (struct tpacket3_hdr *)curr;
1026 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1028 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1029 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1030 BLOCK_NUM_PKTS(pbd) += 1;
1031 atomic_inc(&pkc->blk_fill_in_prog);
1032 prb_run_all_ft_ops(pkc, ppd);
1035 /* Assumes caller has the sk->rx_queue.lock */
1036 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1037 struct sk_buff *skb,
1042 struct tpacket_kbdq_core *pkc;
1043 struct tpacket_block_desc *pbd;
1046 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1047 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1049 /* Queue is frozen when user space is lagging behind */
1050 if (prb_queue_frozen(pkc)) {
1052 * Check if that last block which caused the queue to freeze,
1053 * is still in_use by user-space.
1055 if (prb_curr_blk_in_use(pkc, pbd)) {
1056 /* Can't record this packet */
1060 * Ok, the block was released by user-space.
1061 * Now let's open that block.
1062 * opening a block also thaws the queue.
1063 * Thawing is a side effect.
1065 prb_open_block(pkc, pbd);
1070 curr = pkc->nxt_offset;
1072 end = (char *)pbd + pkc->kblk_size;
1074 /* first try the current block */
1075 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1076 prb_fill_curr_block(curr, pkc, pbd, len);
1077 return (void *)curr;
1080 /* Ok, close the current block */
1081 prb_retire_current_block(pkc, po, 0);
1083 /* Now, try to dispatch the next block */
1084 curr = (char *)prb_dispatch_next_block(pkc, po);
1086 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1087 prb_fill_curr_block(curr, pkc, pbd, len);
1088 return (void *)curr;
1092 * No free blocks are available.user_space hasn't caught up yet.
1093 * Queue was just frozen and now this packet will get dropped.
1098 static void *packet_current_rx_frame(struct packet_sock *po,
1099 struct sk_buff *skb,
1100 int status, unsigned int len)
1103 switch (po->tp_version) {
1106 curr = packet_lookup_frame(po, &po->rx_ring,
1107 po->rx_ring.head, status);
1110 return __packet_lookup_frame_in_block(po, skb, status, len);
1112 WARN(1, "TPACKET version not supported\n");
1118 static void *prb_lookup_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1123 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1124 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1126 if (status != BLOCK_STATUS(pbd))
1131 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1134 if (rb->prb_bdqc.kactive_blk_num)
1135 prev = rb->prb_bdqc.kactive_blk_num-1;
1137 prev = rb->prb_bdqc.knum_blocks-1;
1141 /* Assumes caller has held the rx_queue.lock */
1142 static void *__prb_previous_block(struct packet_sock *po,
1143 struct packet_ring_buffer *rb,
1146 unsigned int previous = prb_previous_blk_num(rb);
1147 return prb_lookup_block(po, rb, previous, status);
1150 static void *packet_previous_rx_frame(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1154 if (po->tp_version <= TPACKET_V2)
1155 return packet_previous_frame(po, rb, status);
1157 return __prb_previous_block(po, rb, status);
1160 static void packet_increment_rx_head(struct packet_sock *po,
1161 struct packet_ring_buffer *rb)
1163 switch (po->tp_version) {
1166 return packet_increment_head(rb);
1169 WARN(1, "TPACKET version not supported.\n");
1175 static void *packet_previous_frame(struct packet_sock *po,
1176 struct packet_ring_buffer *rb,
1179 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1180 return packet_lookup_frame(po, rb, previous, status);
1183 static void packet_increment_head(struct packet_ring_buffer *buff)
1185 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1188 static void packet_inc_pending(struct packet_ring_buffer *rb)
1190 this_cpu_inc(*rb->pending_refcnt);
1193 static void packet_dec_pending(struct packet_ring_buffer *rb)
1195 this_cpu_dec(*rb->pending_refcnt);
1198 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1200 unsigned int refcnt = 0;
1203 /* We don't use pending refcount in rx_ring. */
1204 if (rb->pending_refcnt == NULL)
1207 for_each_possible_cpu(cpu)
1208 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1213 static int packet_alloc_pending(struct packet_sock *po)
1215 po->rx_ring.pending_refcnt = NULL;
1217 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1218 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1224 static void packet_free_pending(struct packet_sock *po)
1226 free_percpu(po->tx_ring.pending_refcnt);
1229 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1231 struct sock *sk = &po->sk;
1234 if (po->prot_hook.func != tpacket_rcv)
1235 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1238 spin_lock(&sk->sk_receive_queue.lock);
1239 if (po->tp_version == TPACKET_V3)
1240 has_room = prb_lookup_block(po, &po->rx_ring,
1241 po->rx_ring.prb_bdqc.kactive_blk_num,
1244 has_room = packet_lookup_frame(po, &po->rx_ring,
1247 spin_unlock(&sk->sk_receive_queue.lock);
1252 static void packet_sock_destruct(struct sock *sk)
1254 skb_queue_purge(&sk->sk_error_queue);
1256 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1257 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1259 if (!sock_flag(sk, SOCK_DEAD)) {
1260 pr_err("Attempt to release alive packet socket: %p\n", sk);
1264 sk_refcnt_debug_dec(sk);
1267 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1269 int x = atomic_read(&f->rr_cur) + 1;
1277 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1278 struct sk_buff *skb,
1281 return reciprocal_scale(skb_get_hash(skb), num);
1284 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1285 struct sk_buff *skb,
1290 cur = atomic_read(&f->rr_cur);
1291 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1292 fanout_rr_next(f, num))) != cur)
1297 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1298 struct sk_buff *skb,
1301 return smp_processor_id() % num;
1304 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1305 struct sk_buff *skb,
1308 return prandom_u32_max(num);
1311 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1312 struct sk_buff *skb,
1313 unsigned int idx, unsigned int skip,
1318 i = j = min_t(int, f->next[idx], num - 1);
1320 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1332 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1333 struct sk_buff *skb,
1336 return skb_get_queue_mapping(skb) % num;
1339 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1341 return f->flags & (flag >> 8);
1344 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1345 struct packet_type *pt, struct net_device *orig_dev)
1347 struct packet_fanout *f = pt->af_packet_priv;
1348 unsigned int num = f->num_members;
1349 struct packet_sock *po;
1352 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1359 case PACKET_FANOUT_HASH:
1361 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1362 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1366 idx = fanout_demux_hash(f, skb, num);
1368 case PACKET_FANOUT_LB:
1369 idx = fanout_demux_lb(f, skb, num);
1371 case PACKET_FANOUT_CPU:
1372 idx = fanout_demux_cpu(f, skb, num);
1374 case PACKET_FANOUT_RND:
1375 idx = fanout_demux_rnd(f, skb, num);
1377 case PACKET_FANOUT_QM:
1378 idx = fanout_demux_qm(f, skb, num);
1380 case PACKET_FANOUT_ROLLOVER:
1381 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1385 po = pkt_sk(f->arr[idx]);
1386 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1387 unlikely(!packet_rcv_has_room(po, skb))) {
1388 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1389 po = pkt_sk(f->arr[idx]);
1392 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1395 DEFINE_MUTEX(fanout_mutex);
1396 EXPORT_SYMBOL_GPL(fanout_mutex);
1397 static LIST_HEAD(fanout_list);
1399 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1401 struct packet_fanout *f = po->fanout;
1403 spin_lock(&f->lock);
1404 f->arr[f->num_members] = sk;
1407 spin_unlock(&f->lock);
1410 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1412 struct packet_fanout *f = po->fanout;
1415 spin_lock(&f->lock);
1416 for (i = 0; i < f->num_members; i++) {
1417 if (f->arr[i] == sk)
1420 BUG_ON(i >= f->num_members);
1421 f->arr[i] = f->arr[f->num_members - 1];
1423 spin_unlock(&f->lock);
1426 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1428 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1434 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1436 struct packet_sock *po = pkt_sk(sk);
1437 struct packet_fanout *f, *match;
1438 u8 type = type_flags & 0xff;
1439 u8 flags = type_flags >> 8;
1443 case PACKET_FANOUT_ROLLOVER:
1444 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1446 case PACKET_FANOUT_HASH:
1447 case PACKET_FANOUT_LB:
1448 case PACKET_FANOUT_CPU:
1449 case PACKET_FANOUT_RND:
1450 case PACKET_FANOUT_QM:
1462 mutex_lock(&fanout_mutex);
1464 list_for_each_entry(f, &fanout_list, list) {
1466 read_pnet(&f->net) == sock_net(sk)) {
1472 if (match && match->flags != flags)
1476 match = kzalloc(sizeof(*match), GFP_KERNEL);
1479 write_pnet(&match->net, sock_net(sk));
1482 match->flags = flags;
1483 atomic_set(&match->rr_cur, 0);
1484 INIT_LIST_HEAD(&match->list);
1485 spin_lock_init(&match->lock);
1486 atomic_set(&match->sk_ref, 0);
1487 match->prot_hook.type = po->prot_hook.type;
1488 match->prot_hook.dev = po->prot_hook.dev;
1489 match->prot_hook.func = packet_rcv_fanout;
1490 match->prot_hook.af_packet_priv = match;
1491 match->prot_hook.id_match = match_fanout_group;
1492 dev_add_pack(&match->prot_hook);
1493 list_add(&match->list, &fanout_list);
1496 if (match->type == type &&
1497 match->prot_hook.type == po->prot_hook.type &&
1498 match->prot_hook.dev == po->prot_hook.dev) {
1500 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1501 __dev_remove_pack(&po->prot_hook);
1503 atomic_inc(&match->sk_ref);
1504 __fanout_link(sk, po);
1509 mutex_unlock(&fanout_mutex);
1513 static void fanout_release(struct sock *sk)
1515 struct packet_sock *po = pkt_sk(sk);
1516 struct packet_fanout *f;
1522 mutex_lock(&fanout_mutex);
1525 if (atomic_dec_and_test(&f->sk_ref)) {
1527 dev_remove_pack(&f->prot_hook);
1530 mutex_unlock(&fanout_mutex);
1533 static const struct proto_ops packet_ops;
1535 static const struct proto_ops packet_ops_spkt;
1537 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1538 struct packet_type *pt, struct net_device *orig_dev)
1541 struct sockaddr_pkt *spkt;
1544 * When we registered the protocol we saved the socket in the data
1545 * field for just this event.
1548 sk = pt->af_packet_priv;
1551 * Yank back the headers [hope the device set this
1552 * right or kerboom...]
1554 * Incoming packets have ll header pulled,
1557 * For outgoing ones skb->data == skb_mac_header(skb)
1558 * so that this procedure is noop.
1561 if (skb->pkt_type == PACKET_LOOPBACK)
1564 if (!net_eq(dev_net(dev), sock_net(sk)))
1567 skb = skb_share_check(skb, GFP_ATOMIC);
1571 /* drop any routing info */
1574 /* drop conntrack reference */
1577 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1579 skb_push(skb, skb->data - skb_mac_header(skb));
1582 * The SOCK_PACKET socket receives _all_ frames.
1585 spkt->spkt_family = dev->type;
1586 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1587 spkt->spkt_protocol = skb->protocol;
1590 * Charge the memory to the socket. This is done specifically
1591 * to prevent sockets using all the memory up.
1594 if (sock_queue_rcv_skb(sk, skb) == 0)
1605 * Output a raw packet to a device layer. This bypasses all the other
1606 * protocol layers and you must therefore supply it with a complete frame
1609 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1610 struct msghdr *msg, size_t len)
1612 struct sock *sk = sock->sk;
1613 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1614 struct sk_buff *skb = NULL;
1615 struct net_device *dev;
1621 * Get and verify the address.
1625 if (msg->msg_namelen < sizeof(struct sockaddr))
1627 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1628 proto = saddr->spkt_protocol;
1630 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1633 * Find the device first to size check it
1636 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1639 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1645 if (!(dev->flags & IFF_UP))
1649 * You may not queue a frame bigger than the mtu. This is the lowest level
1650 * raw protocol and you must do your own fragmentation at this level.
1653 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1654 if (!netif_supports_nofcs(dev)) {
1655 err = -EPROTONOSUPPORT;
1658 extra_len = 4; /* We're doing our own CRC */
1662 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1666 size_t reserved = LL_RESERVED_SPACE(dev);
1667 int tlen = dev->needed_tailroom;
1668 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1671 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1674 /* FIXME: Save some space for broken drivers that write a hard
1675 * header at transmission time by themselves. PPP is the notable
1676 * one here. This should really be fixed at the driver level.
1678 skb_reserve(skb, reserved);
1679 skb_reset_network_header(skb);
1681 /* Try to align data part correctly */
1686 skb_reset_network_header(skb);
1688 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1694 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1695 /* Earlier code assumed this would be a VLAN pkt,
1696 * double-check this now that we have the actual
1699 struct ethhdr *ehdr;
1700 skb_reset_mac_header(skb);
1701 ehdr = eth_hdr(skb);
1702 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1708 skb->protocol = proto;
1710 skb->priority = sk->sk_priority;
1711 skb->mark = sk->sk_mark;
1713 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1715 if (unlikely(extra_len == 4))
1718 skb_probe_transport_header(skb, 0);
1720 dev_queue_xmit(skb);
1731 static unsigned int run_filter(const struct sk_buff *skb,
1732 const struct sock *sk,
1735 struct sk_filter *filter;
1738 filter = rcu_dereference(sk->sk_filter);
1740 res = SK_RUN_FILTER(filter, skb);
1747 * This function makes lazy skb cloning in hope that most of packets
1748 * are discarded by BPF.
1750 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1751 * and skb->cb are mangled. It works because (and until) packets
1752 * falling here are owned by current CPU. Output packets are cloned
1753 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1754 * sequencially, so that if we return skb to original state on exit,
1755 * we will not harm anyone.
1758 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1759 struct packet_type *pt, struct net_device *orig_dev)
1762 struct sockaddr_ll *sll;
1763 struct packet_sock *po;
1764 u8 *skb_head = skb->data;
1765 int skb_len = skb->len;
1766 unsigned int snaplen, res;
1768 if (skb->pkt_type == PACKET_LOOPBACK)
1771 sk = pt->af_packet_priv;
1774 if (!net_eq(dev_net(dev), sock_net(sk)))
1779 if (dev->header_ops) {
1780 /* The device has an explicit notion of ll header,
1781 * exported to higher levels.
1783 * Otherwise, the device hides details of its frame
1784 * structure, so that corresponding packet head is
1785 * never delivered to user.
1787 if (sk->sk_type != SOCK_DGRAM)
1788 skb_push(skb, skb->data - skb_mac_header(skb));
1789 else if (skb->pkt_type == PACKET_OUTGOING) {
1790 /* Special case: outgoing packets have ll header at head */
1791 skb_pull(skb, skb_network_offset(skb));
1797 res = run_filter(skb, sk, snaplen);
1799 goto drop_n_restore;
1803 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1806 if (skb_shared(skb)) {
1807 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1811 if (skb_head != skb->data) {
1812 skb->data = skb_head;
1819 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1822 sll = &PACKET_SKB_CB(skb)->sa.ll;
1823 sll->sll_family = AF_PACKET;
1824 sll->sll_hatype = dev->type;
1825 sll->sll_protocol = skb->protocol;
1826 sll->sll_pkttype = skb->pkt_type;
1827 if (unlikely(po->origdev))
1828 sll->sll_ifindex = orig_dev->ifindex;
1830 sll->sll_ifindex = dev->ifindex;
1832 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1834 PACKET_SKB_CB(skb)->origlen = skb->len;
1836 if (pskb_trim(skb, snaplen))
1839 skb_set_owner_r(skb, sk);
1843 /* drop conntrack reference */
1846 spin_lock(&sk->sk_receive_queue.lock);
1847 po->stats.stats1.tp_packets++;
1848 skb->dropcount = atomic_read(&sk->sk_drops);
1849 __skb_queue_tail(&sk->sk_receive_queue, skb);
1850 spin_unlock(&sk->sk_receive_queue.lock);
1851 sk->sk_data_ready(sk);
1855 spin_lock(&sk->sk_receive_queue.lock);
1856 po->stats.stats1.tp_drops++;
1857 atomic_inc(&sk->sk_drops);
1858 spin_unlock(&sk->sk_receive_queue.lock);
1861 if (skb_head != skb->data && skb_shared(skb)) {
1862 skb->data = skb_head;
1870 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1871 struct packet_type *pt, struct net_device *orig_dev)
1874 struct packet_sock *po;
1875 struct sockaddr_ll *sll;
1876 union tpacket_uhdr h;
1877 u8 *skb_head = skb->data;
1878 int skb_len = skb->len;
1879 unsigned int snaplen, res;
1880 unsigned long status = TP_STATUS_USER;
1881 unsigned short macoff, netoff, hdrlen;
1882 struct sk_buff *copy_skb = NULL;
1886 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1887 * We may add members to them until current aligned size without forcing
1888 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1890 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1891 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1893 if (skb->pkt_type == PACKET_LOOPBACK)
1896 sk = pt->af_packet_priv;
1899 if (!net_eq(dev_net(dev), sock_net(sk)))
1902 if (dev->header_ops) {
1903 if (sk->sk_type != SOCK_DGRAM)
1904 skb_push(skb, skb->data - skb_mac_header(skb));
1905 else if (skb->pkt_type == PACKET_OUTGOING) {
1906 /* Special case: outgoing packets have ll header at head */
1907 skb_pull(skb, skb_network_offset(skb));
1911 if (skb->ip_summed == CHECKSUM_PARTIAL)
1912 status |= TP_STATUS_CSUMNOTREADY;
1916 res = run_filter(skb, sk, snaplen);
1918 goto drop_n_restore;
1922 if (sk->sk_type == SOCK_DGRAM) {
1923 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1926 unsigned int maclen = skb_network_offset(skb);
1927 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1928 (maclen < 16 ? 16 : maclen)) +
1930 macoff = netoff - maclen;
1932 if (po->tp_version <= TPACKET_V2) {
1933 if (macoff + snaplen > po->rx_ring.frame_size) {
1934 if (po->copy_thresh &&
1935 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1936 if (skb_shared(skb)) {
1937 copy_skb = skb_clone(skb, GFP_ATOMIC);
1939 copy_skb = skb_get(skb);
1940 skb_head = skb->data;
1943 skb_set_owner_r(copy_skb, sk);
1945 snaplen = po->rx_ring.frame_size - macoff;
1946 if ((int)snaplen < 0)
1950 spin_lock(&sk->sk_receive_queue.lock);
1951 h.raw = packet_current_rx_frame(po, skb,
1952 TP_STATUS_KERNEL, (macoff+snaplen));
1955 if (po->tp_version <= TPACKET_V2) {
1956 packet_increment_rx_head(po, &po->rx_ring);
1958 * LOSING will be reported till you read the stats,
1959 * because it's COR - Clear On Read.
1960 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1963 if (po->stats.stats1.tp_drops)
1964 status |= TP_STATUS_LOSING;
1966 po->stats.stats1.tp_packets++;
1968 status |= TP_STATUS_COPY;
1969 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1971 spin_unlock(&sk->sk_receive_queue.lock);
1973 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1975 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1976 getnstimeofday(&ts);
1978 status |= ts_status;
1980 switch (po->tp_version) {
1982 h.h1->tp_len = skb->len;
1983 h.h1->tp_snaplen = snaplen;
1984 h.h1->tp_mac = macoff;
1985 h.h1->tp_net = netoff;
1986 h.h1->tp_sec = ts.tv_sec;
1987 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1988 hdrlen = sizeof(*h.h1);
1991 h.h2->tp_len = skb->len;
1992 h.h2->tp_snaplen = snaplen;
1993 h.h2->tp_mac = macoff;
1994 h.h2->tp_net = netoff;
1995 h.h2->tp_sec = ts.tv_sec;
1996 h.h2->tp_nsec = ts.tv_nsec;
1997 if (vlan_tx_tag_present(skb)) {
1998 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1999 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2000 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2002 h.h2->tp_vlan_tci = 0;
2003 h.h2->tp_vlan_tpid = 0;
2005 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2006 hdrlen = sizeof(*h.h2);
2009 /* tp_nxt_offset,vlan are already populated above.
2010 * So DONT clear those fields here
2012 h.h3->tp_status |= status;
2013 h.h3->tp_len = skb->len;
2014 h.h3->tp_snaplen = snaplen;
2015 h.h3->tp_mac = macoff;
2016 h.h3->tp_net = netoff;
2017 h.h3->tp_sec = ts.tv_sec;
2018 h.h3->tp_nsec = ts.tv_nsec;
2019 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2020 hdrlen = sizeof(*h.h3);
2026 sll = h.raw + TPACKET_ALIGN(hdrlen);
2027 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2028 sll->sll_family = AF_PACKET;
2029 sll->sll_hatype = dev->type;
2030 sll->sll_protocol = skb->protocol;
2031 sll->sll_pkttype = skb->pkt_type;
2032 if (unlikely(po->origdev))
2033 sll->sll_ifindex = orig_dev->ifindex;
2035 sll->sll_ifindex = dev->ifindex;
2039 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2040 if (po->tp_version <= TPACKET_V2) {
2043 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2046 for (start = h.raw; start < end; start += PAGE_SIZE)
2047 flush_dcache_page(pgv_to_page(start));
2052 if (po->tp_version <= TPACKET_V2)
2053 __packet_set_status(po, h.raw, status);
2055 prb_clear_blk_fill_status(&po->rx_ring);
2057 sk->sk_data_ready(sk);
2060 if (skb_head != skb->data && skb_shared(skb)) {
2061 skb->data = skb_head;
2069 po->stats.stats1.tp_drops++;
2070 spin_unlock(&sk->sk_receive_queue.lock);
2072 sk->sk_data_ready(sk);
2073 kfree_skb(copy_skb);
2074 goto drop_n_restore;
2077 static void tpacket_destruct_skb(struct sk_buff *skb)
2079 struct packet_sock *po = pkt_sk(skb->sk);
2081 if (likely(po->tx_ring.pg_vec)) {
2085 ph = skb_shinfo(skb)->destructor_arg;
2086 packet_dec_pending(&po->tx_ring);
2088 ts = __packet_set_timestamp(po, ph, skb);
2089 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2095 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2096 void *frame, struct net_device *dev, int size_max,
2097 __be16 proto, unsigned char *addr, int hlen)
2099 union tpacket_uhdr ph;
2100 int to_write, offset, len, tp_len, nr_frags, len_max;
2101 struct socket *sock = po->sk.sk_socket;
2108 skb->protocol = proto;
2110 skb->priority = po->sk.sk_priority;
2111 skb->mark = po->sk.sk_mark;
2112 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2113 skb_shinfo(skb)->destructor_arg = ph.raw;
2115 switch (po->tp_version) {
2117 tp_len = ph.h2->tp_len;
2120 tp_len = ph.h1->tp_len;
2123 if (unlikely(tp_len > size_max)) {
2124 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2128 skb_reserve(skb, hlen);
2129 skb_reset_network_header(skb);
2131 if (!packet_use_direct_xmit(po))
2132 skb_probe_transport_header(skb, 0);
2133 if (unlikely(po->tp_tx_has_off)) {
2134 int off_min, off_max, off;
2135 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2136 off_max = po->tx_ring.frame_size - tp_len;
2137 if (sock->type == SOCK_DGRAM) {
2138 switch (po->tp_version) {
2140 off = ph.h2->tp_net;
2143 off = ph.h1->tp_net;
2147 switch (po->tp_version) {
2149 off = ph.h2->tp_mac;
2152 off = ph.h1->tp_mac;
2156 if (unlikely((off < off_min) || (off_max < off)))
2158 data = ph.raw + off;
2160 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2164 if (sock->type == SOCK_DGRAM) {
2165 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2167 if (unlikely(err < 0))
2169 } else if (dev->hard_header_len) {
2170 /* net device doesn't like empty head */
2171 if (unlikely(tp_len <= dev->hard_header_len)) {
2172 pr_err("packet size is too short (%d < %d)\n",
2173 tp_len, dev->hard_header_len);
2177 skb_push(skb, dev->hard_header_len);
2178 err = skb_store_bits(skb, 0, data,
2179 dev->hard_header_len);
2183 data += dev->hard_header_len;
2184 to_write -= dev->hard_header_len;
2187 offset = offset_in_page(data);
2188 len_max = PAGE_SIZE - offset;
2189 len = ((to_write > len_max) ? len_max : to_write);
2191 skb->data_len = to_write;
2192 skb->len += to_write;
2193 skb->truesize += to_write;
2194 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2196 while (likely(to_write)) {
2197 nr_frags = skb_shinfo(skb)->nr_frags;
2199 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2200 pr_err("Packet exceed the number of skb frags(%lu)\n",
2205 page = pgv_to_page(data);
2207 flush_dcache_page(page);
2209 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2212 len_max = PAGE_SIZE;
2213 len = ((to_write > len_max) ? len_max : to_write);
2219 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2221 struct sk_buff *skb;
2222 struct net_device *dev;
2224 int err, reserve = 0;
2226 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2227 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2228 int tp_len, size_max;
2229 unsigned char *addr;
2231 int status = TP_STATUS_AVAILABLE;
2234 mutex_lock(&po->pg_vec_lock);
2236 if (likely(saddr == NULL)) {
2237 dev = packet_cached_dev_get(po);
2242 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2244 if (msg->msg_namelen < (saddr->sll_halen
2245 + offsetof(struct sockaddr_ll,
2248 proto = saddr->sll_protocol;
2249 addr = saddr->sll_addr;
2250 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2254 if (unlikely(dev == NULL))
2257 if (unlikely(!(dev->flags & IFF_UP)))
2260 reserve = dev->hard_header_len + VLAN_HLEN;
2261 size_max = po->tx_ring.frame_size
2262 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2264 if (size_max > dev->mtu + reserve)
2265 size_max = dev->mtu + reserve;
2268 ph = packet_current_frame(po, &po->tx_ring,
2269 TP_STATUS_SEND_REQUEST);
2270 if (unlikely(ph == NULL)) {
2271 if (need_wait && need_resched())
2276 status = TP_STATUS_SEND_REQUEST;
2277 hlen = LL_RESERVED_SPACE(dev);
2278 tlen = dev->needed_tailroom;
2279 skb = sock_alloc_send_skb(&po->sk,
2280 hlen + tlen + sizeof(struct sockaddr_ll),
2283 if (unlikely(skb == NULL))
2286 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2288 if (tp_len > dev->mtu + dev->hard_header_len) {
2289 struct ethhdr *ehdr;
2290 /* Earlier code assumed this would be a VLAN pkt,
2291 * double-check this now that we have the actual
2295 skb_reset_mac_header(skb);
2296 ehdr = eth_hdr(skb);
2297 if (ehdr->h_proto != htons(ETH_P_8021Q))
2300 if (unlikely(tp_len < 0)) {
2302 __packet_set_status(po, ph,
2303 TP_STATUS_AVAILABLE);
2304 packet_increment_head(&po->tx_ring);
2308 status = TP_STATUS_WRONG_FORMAT;
2314 packet_pick_tx_queue(dev, skb);
2316 skb->destructor = tpacket_destruct_skb;
2317 __packet_set_status(po, ph, TP_STATUS_SENDING);
2318 packet_inc_pending(&po->tx_ring);
2320 status = TP_STATUS_SEND_REQUEST;
2321 err = po->xmit(skb);
2322 if (unlikely(err > 0)) {
2323 err = net_xmit_errno(err);
2324 if (err && __packet_get_status(po, ph) ==
2325 TP_STATUS_AVAILABLE) {
2326 /* skb was destructed already */
2331 * skb was dropped but not destructed yet;
2332 * let's treat it like congestion or err < 0
2336 packet_increment_head(&po->tx_ring);
2338 } while (likely((ph != NULL) ||
2339 /* Note: packet_read_pending() might be slow if we have
2340 * to call it as it's per_cpu variable, but in fast-path
2341 * we already short-circuit the loop with the first
2342 * condition, and luckily don't have to go that path
2345 (need_wait && packet_read_pending(&po->tx_ring))));
2351 __packet_set_status(po, ph, status);
2356 mutex_unlock(&po->pg_vec_lock);
2360 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2361 size_t reserve, size_t len,
2362 size_t linear, int noblock,
2365 struct sk_buff *skb;
2367 /* Under a page? Don't bother with paged skb. */
2368 if (prepad + len < PAGE_SIZE || !linear)
2371 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2376 skb_reserve(skb, reserve);
2377 skb_put(skb, linear);
2378 skb->data_len = len - linear;
2379 skb->len += len - linear;
2384 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2386 struct sock *sk = sock->sk;
2387 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2388 struct sk_buff *skb;
2389 struct net_device *dev;
2391 unsigned char *addr;
2392 int err, reserve = 0;
2393 struct virtio_net_hdr vnet_hdr = { 0 };
2396 struct packet_sock *po = pkt_sk(sk);
2397 unsigned short gso_type = 0;
2402 * Get and verify the address.
2405 if (likely(saddr == NULL)) {
2406 dev = packet_cached_dev_get(po);
2411 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2413 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2415 proto = saddr->sll_protocol;
2416 addr = saddr->sll_addr;
2417 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2421 if (unlikely(dev == NULL))
2424 if (unlikely(!(dev->flags & IFF_UP)))
2427 if (sock->type == SOCK_RAW)
2428 reserve = dev->hard_header_len;
2429 if (po->has_vnet_hdr) {
2430 vnet_hdr_len = sizeof(vnet_hdr);
2433 if (len < vnet_hdr_len)
2436 len -= vnet_hdr_len;
2438 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2443 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2444 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2446 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2447 vnet_hdr.csum_offset + 2;
2450 if (vnet_hdr.hdr_len > len)
2453 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2454 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2455 case VIRTIO_NET_HDR_GSO_TCPV4:
2456 gso_type = SKB_GSO_TCPV4;
2458 case VIRTIO_NET_HDR_GSO_TCPV6:
2459 gso_type = SKB_GSO_TCPV6;
2461 case VIRTIO_NET_HDR_GSO_UDP:
2462 gso_type = SKB_GSO_UDP;
2468 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2469 gso_type |= SKB_GSO_TCP_ECN;
2471 if (vnet_hdr.gso_size == 0)
2477 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2478 if (!netif_supports_nofcs(dev)) {
2479 err = -EPROTONOSUPPORT;
2482 extra_len = 4; /* We're doing our own CRC */
2486 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2490 hlen = LL_RESERVED_SPACE(dev);
2491 tlen = dev->needed_tailroom;
2492 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2493 msg->msg_flags & MSG_DONTWAIT, &err);
2497 skb_set_network_header(skb, reserve);
2500 if (sock->type == SOCK_DGRAM &&
2501 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2504 /* Returns -EFAULT on error */
2505 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2509 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2511 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2512 /* Earlier code assumed this would be a VLAN pkt,
2513 * double-check this now that we have the actual
2516 struct ethhdr *ehdr;
2517 skb_reset_mac_header(skb);
2518 ehdr = eth_hdr(skb);
2519 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2525 skb->protocol = proto;
2527 skb->priority = sk->sk_priority;
2528 skb->mark = sk->sk_mark;
2530 packet_pick_tx_queue(dev, skb);
2532 if (po->has_vnet_hdr) {
2533 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2534 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2535 vnet_hdr.csum_offset)) {
2541 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2542 skb_shinfo(skb)->gso_type = gso_type;
2544 /* Header must be checked, and gso_segs computed. */
2545 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2546 skb_shinfo(skb)->gso_segs = 0;
2548 len += vnet_hdr_len;
2551 if (!packet_use_direct_xmit(po))
2552 skb_probe_transport_header(skb, reserve);
2553 if (unlikely(extra_len == 4))
2556 err = po->xmit(skb);
2557 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2573 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2574 struct msghdr *msg, size_t len)
2576 struct sock *sk = sock->sk;
2577 struct packet_sock *po = pkt_sk(sk);
2579 if (po->tx_ring.pg_vec)
2580 return tpacket_snd(po, msg);
2582 return packet_snd(sock, msg, len);
2586 * Close a PACKET socket. This is fairly simple. We immediately go
2587 * to 'closed' state and remove our protocol entry in the device list.
2590 static int packet_release(struct socket *sock)
2592 struct sock *sk = sock->sk;
2593 struct packet_sock *po;
2595 union tpacket_req_u req_u;
2603 mutex_lock(&net->packet.sklist_lock);
2604 sk_del_node_init_rcu(sk);
2605 mutex_unlock(&net->packet.sklist_lock);
2608 sock_prot_inuse_add(net, sk->sk_prot, -1);
2611 spin_lock(&po->bind_lock);
2612 unregister_prot_hook(sk, false);
2613 packet_cached_dev_reset(po);
2615 if (po->prot_hook.dev) {
2616 dev_put(po->prot_hook.dev);
2617 po->prot_hook.dev = NULL;
2619 spin_unlock(&po->bind_lock);
2621 packet_flush_mclist(sk);
2623 if (po->rx_ring.pg_vec) {
2624 memset(&req_u, 0, sizeof(req_u));
2625 packet_set_ring(sk, &req_u, 1, 0);
2628 if (po->tx_ring.pg_vec) {
2629 memset(&req_u, 0, sizeof(req_u));
2630 packet_set_ring(sk, &req_u, 1, 1);
2637 * Now the socket is dead. No more input will appear.
2644 skb_queue_purge(&sk->sk_receive_queue);
2645 packet_free_pending(po);
2646 sk_refcnt_debug_release(sk);
2653 * Attach a packet hook.
2656 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2658 struct packet_sock *po = pkt_sk(sk);
2659 const struct net_device *dev_curr;
2671 spin_lock(&po->bind_lock);
2673 proto_curr = po->prot_hook.type;
2674 dev_curr = po->prot_hook.dev;
2676 need_rehook = proto_curr != proto || dev_curr != dev;
2679 unregister_prot_hook(sk, true);
2682 po->prot_hook.type = proto;
2684 if (po->prot_hook.dev)
2685 dev_put(po->prot_hook.dev);
2687 po->prot_hook.dev = dev;
2689 po->ifindex = dev ? dev->ifindex : 0;
2690 packet_cached_dev_assign(po, dev);
2693 if (proto == 0 || !need_rehook)
2696 if (!dev || (dev->flags & IFF_UP)) {
2697 register_prot_hook(sk);
2699 sk->sk_err = ENETDOWN;
2700 if (!sock_flag(sk, SOCK_DEAD))
2701 sk->sk_error_report(sk);
2705 spin_unlock(&po->bind_lock);
2711 * Bind a packet socket to a device
2714 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2717 struct sock *sk = sock->sk;
2719 struct net_device *dev;
2726 if (addr_len != sizeof(struct sockaddr))
2728 strlcpy(name, uaddr->sa_data, sizeof(name));
2730 dev = dev_get_by_name(sock_net(sk), name);
2732 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2736 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2738 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2739 struct sock *sk = sock->sk;
2740 struct net_device *dev = NULL;
2748 if (addr_len < sizeof(struct sockaddr_ll))
2750 if (sll->sll_family != AF_PACKET)
2753 if (sll->sll_ifindex) {
2755 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2759 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2765 static struct proto packet_proto = {
2767 .owner = THIS_MODULE,
2768 .obj_size = sizeof(struct packet_sock),
2772 * Create a packet of type SOCK_PACKET.
2775 static int packet_create(struct net *net, struct socket *sock, int protocol,
2779 struct packet_sock *po;
2780 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2783 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2785 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2786 sock->type != SOCK_PACKET)
2787 return -ESOCKTNOSUPPORT;
2789 sock->state = SS_UNCONNECTED;
2792 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2796 sock->ops = &packet_ops;
2797 if (sock->type == SOCK_PACKET)
2798 sock->ops = &packet_ops_spkt;
2800 sock_init_data(sock, sk);
2803 sk->sk_family = PF_PACKET;
2805 po->xmit = dev_queue_xmit;
2807 err = packet_alloc_pending(po);
2811 packet_cached_dev_reset(po);
2813 sk->sk_destruct = packet_sock_destruct;
2814 sk_refcnt_debug_inc(sk);
2817 * Attach a protocol block
2820 spin_lock_init(&po->bind_lock);
2821 mutex_init(&po->pg_vec_lock);
2822 po->prot_hook.func = packet_rcv;
2824 if (sock->type == SOCK_PACKET)
2825 po->prot_hook.func = packet_rcv_spkt;
2827 po->prot_hook.af_packet_priv = sk;
2830 po->prot_hook.type = proto;
2831 register_prot_hook(sk);
2834 mutex_lock(&net->packet.sklist_lock);
2835 sk_add_node_rcu(sk, &net->packet.sklist);
2836 mutex_unlock(&net->packet.sklist_lock);
2839 sock_prot_inuse_add(net, &packet_proto, 1);
2850 * Pull a packet from our receive queue and hand it to the user.
2851 * If necessary we block.
2854 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2855 struct msghdr *msg, size_t len, int flags)
2857 struct sock *sk = sock->sk;
2858 struct sk_buff *skb;
2860 int vnet_hdr_len = 0;
2863 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2867 /* What error should we return now? EUNATTACH? */
2868 if (pkt_sk(sk)->ifindex < 0)
2872 if (flags & MSG_ERRQUEUE) {
2873 err = sock_recv_errqueue(sk, msg, len,
2874 SOL_PACKET, PACKET_TX_TIMESTAMP);
2879 * Call the generic datagram receiver. This handles all sorts
2880 * of horrible races and re-entrancy so we can forget about it
2881 * in the protocol layers.
2883 * Now it will return ENETDOWN, if device have just gone down,
2884 * but then it will block.
2887 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2890 * An error occurred so return it. Because skb_recv_datagram()
2891 * handles the blocking we don't see and worry about blocking
2898 if (pkt_sk(sk)->has_vnet_hdr) {
2899 struct virtio_net_hdr vnet_hdr = { 0 };
2902 vnet_hdr_len = sizeof(vnet_hdr);
2903 if (len < vnet_hdr_len)
2906 len -= vnet_hdr_len;
2908 if (skb_is_gso(skb)) {
2909 struct skb_shared_info *sinfo = skb_shinfo(skb);
2911 /* This is a hint as to how much should be linear. */
2912 vnet_hdr.hdr_len = skb_headlen(skb);
2913 vnet_hdr.gso_size = sinfo->gso_size;
2914 if (sinfo->gso_type & SKB_GSO_TCPV4)
2915 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2916 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2917 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2918 else if (sinfo->gso_type & SKB_GSO_UDP)
2919 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2920 else if (sinfo->gso_type & SKB_GSO_FCOE)
2924 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2925 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2927 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2929 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2930 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2931 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2932 vnet_hdr.csum_offset = skb->csum_offset;
2933 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2934 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2935 } /* else everything is zero */
2937 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2943 /* You lose any data beyond the buffer you gave. If it worries
2944 * a user program they can ask the device for its MTU
2950 msg->msg_flags |= MSG_TRUNC;
2953 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2957 sock_recv_ts_and_drops(msg, sk, skb);
2959 if (msg->msg_name) {
2960 /* If the address length field is there to be filled
2961 * in, we fill it in now.
2963 if (sock->type == SOCK_PACKET) {
2964 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
2965 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2967 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2968 msg->msg_namelen = sll->sll_halen +
2969 offsetof(struct sockaddr_ll, sll_addr);
2971 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2975 if (pkt_sk(sk)->auxdata) {
2976 struct tpacket_auxdata aux;
2978 aux.tp_status = TP_STATUS_USER;
2979 if (skb->ip_summed == CHECKSUM_PARTIAL)
2980 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2981 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2982 aux.tp_snaplen = skb->len;
2984 aux.tp_net = skb_network_offset(skb);
2985 if (vlan_tx_tag_present(skb)) {
2986 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2987 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2988 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2990 aux.tp_vlan_tci = 0;
2991 aux.tp_vlan_tpid = 0;
2993 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2997 * Free or return the buffer as appropriate. Again this
2998 * hides all the races and re-entrancy issues from us.
3000 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3003 skb_free_datagram(sk, skb);
3008 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3009 int *uaddr_len, int peer)
3011 struct net_device *dev;
3012 struct sock *sk = sock->sk;
3017 uaddr->sa_family = AF_PACKET;
3018 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3020 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3022 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3024 *uaddr_len = sizeof(*uaddr);
3029 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3030 int *uaddr_len, int peer)
3032 struct net_device *dev;
3033 struct sock *sk = sock->sk;
3034 struct packet_sock *po = pkt_sk(sk);
3035 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3040 sll->sll_family = AF_PACKET;
3041 sll->sll_ifindex = po->ifindex;
3042 sll->sll_protocol = po->num;
3043 sll->sll_pkttype = 0;
3045 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3047 sll->sll_hatype = dev->type;
3048 sll->sll_halen = dev->addr_len;
3049 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3051 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3055 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3060 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3064 case PACKET_MR_MULTICAST:
3065 if (i->alen != dev->addr_len)
3068 return dev_mc_add(dev, i->addr);
3070 return dev_mc_del(dev, i->addr);
3072 case PACKET_MR_PROMISC:
3073 return dev_set_promiscuity(dev, what);
3075 case PACKET_MR_ALLMULTI:
3076 return dev_set_allmulti(dev, what);
3078 case PACKET_MR_UNICAST:
3079 if (i->alen != dev->addr_len)
3082 return dev_uc_add(dev, i->addr);
3084 return dev_uc_del(dev, i->addr);
3092 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3094 for ( ; i; i = i->next) {
3095 if (i->ifindex == dev->ifindex)
3096 packet_dev_mc(dev, i, what);
3100 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3102 struct packet_sock *po = pkt_sk(sk);
3103 struct packet_mclist *ml, *i;
3104 struct net_device *dev;
3110 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3115 if (mreq->mr_alen > dev->addr_len)
3119 i = kmalloc(sizeof(*i), GFP_KERNEL);
3124 for (ml = po->mclist; ml; ml = ml->next) {
3125 if (ml->ifindex == mreq->mr_ifindex &&
3126 ml->type == mreq->mr_type &&
3127 ml->alen == mreq->mr_alen &&
3128 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3130 /* Free the new element ... */
3136 i->type = mreq->mr_type;
3137 i->ifindex = mreq->mr_ifindex;
3138 i->alen = mreq->mr_alen;
3139 memcpy(i->addr, mreq->mr_address, i->alen);
3141 i->next = po->mclist;
3143 err = packet_dev_mc(dev, i, 1);
3145 po->mclist = i->next;
3154 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3156 struct packet_mclist *ml, **mlp;
3160 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3161 if (ml->ifindex == mreq->mr_ifindex &&
3162 ml->type == mreq->mr_type &&
3163 ml->alen == mreq->mr_alen &&
3164 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3165 if (--ml->count == 0) {
3166 struct net_device *dev;
3168 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3170 packet_dev_mc(dev, ml, -1);
3178 return -EADDRNOTAVAIL;
3181 static void packet_flush_mclist(struct sock *sk)
3183 struct packet_sock *po = pkt_sk(sk);
3184 struct packet_mclist *ml;
3190 while ((ml = po->mclist) != NULL) {
3191 struct net_device *dev;
3193 po->mclist = ml->next;
3194 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3196 packet_dev_mc(dev, ml, -1);
3203 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3205 struct sock *sk = sock->sk;
3206 struct packet_sock *po = pkt_sk(sk);
3209 if (level != SOL_PACKET)
3210 return -ENOPROTOOPT;
3213 case PACKET_ADD_MEMBERSHIP:
3214 case PACKET_DROP_MEMBERSHIP:
3216 struct packet_mreq_max mreq;
3218 memset(&mreq, 0, sizeof(mreq));
3219 if (len < sizeof(struct packet_mreq))
3221 if (len > sizeof(mreq))
3223 if (copy_from_user(&mreq, optval, len))
3225 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3227 if (optname == PACKET_ADD_MEMBERSHIP)
3228 ret = packet_mc_add(sk, &mreq);
3230 ret = packet_mc_drop(sk, &mreq);
3234 case PACKET_RX_RING:
3235 case PACKET_TX_RING:
3237 union tpacket_req_u req_u;
3240 switch (po->tp_version) {
3243 len = sizeof(req_u.req);
3247 len = sizeof(req_u.req3);
3252 if (pkt_sk(sk)->has_vnet_hdr)
3254 if (copy_from_user(&req_u.req, optval, len))
3256 return packet_set_ring(sk, &req_u, 0,
3257 optname == PACKET_TX_RING);
3259 case PACKET_COPY_THRESH:
3263 if (optlen != sizeof(val))
3265 if (copy_from_user(&val, optval, sizeof(val)))
3268 pkt_sk(sk)->copy_thresh = val;
3271 case PACKET_VERSION:
3275 if (optlen != sizeof(val))
3277 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3279 if (copy_from_user(&val, optval, sizeof(val)))
3285 po->tp_version = val;
3291 case PACKET_RESERVE:
3295 if (optlen != sizeof(val))
3297 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3299 if (copy_from_user(&val, optval, sizeof(val)))
3301 po->tp_reserve = val;
3308 if (optlen != sizeof(val))
3310 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3312 if (copy_from_user(&val, optval, sizeof(val)))
3314 po->tp_loss = !!val;
3317 case PACKET_AUXDATA:
3321 if (optlen < sizeof(val))
3323 if (copy_from_user(&val, optval, sizeof(val)))
3326 po->auxdata = !!val;
3329 case PACKET_ORIGDEV:
3333 if (optlen < sizeof(val))
3335 if (copy_from_user(&val, optval, sizeof(val)))
3338 po->origdev = !!val;
3341 case PACKET_VNET_HDR:
3345 if (sock->type != SOCK_RAW)
3347 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3349 if (optlen < sizeof(val))
3351 if (copy_from_user(&val, optval, sizeof(val)))
3354 po->has_vnet_hdr = !!val;
3357 case PACKET_TIMESTAMP:
3361 if (optlen != sizeof(val))
3363 if (copy_from_user(&val, optval, sizeof(val)))
3366 po->tp_tstamp = val;
3373 if (optlen != sizeof(val))
3375 if (copy_from_user(&val, optval, sizeof(val)))
3378 return fanout_add(sk, val & 0xffff, val >> 16);
3380 case PACKET_TX_HAS_OFF:
3384 if (optlen != sizeof(val))
3386 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3388 if (copy_from_user(&val, optval, sizeof(val)))
3390 po->tp_tx_has_off = !!val;
3393 case PACKET_QDISC_BYPASS:
3397 if (optlen != sizeof(val))
3399 if (copy_from_user(&val, optval, sizeof(val)))
3402 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3406 return -ENOPROTOOPT;
3410 static int packet_getsockopt(struct socket *sock, int level, int optname,
3411 char __user *optval, int __user *optlen)
3414 int val, lv = sizeof(val);
3415 struct sock *sk = sock->sk;
3416 struct packet_sock *po = pkt_sk(sk);
3418 union tpacket_stats_u st;
3420 if (level != SOL_PACKET)
3421 return -ENOPROTOOPT;
3423 if (get_user(len, optlen))
3430 case PACKET_STATISTICS:
3431 spin_lock_bh(&sk->sk_receive_queue.lock);
3432 memcpy(&st, &po->stats, sizeof(st));
3433 memset(&po->stats, 0, sizeof(po->stats));
3434 spin_unlock_bh(&sk->sk_receive_queue.lock);
3436 if (po->tp_version == TPACKET_V3) {
3437 lv = sizeof(struct tpacket_stats_v3);
3438 st.stats3.tp_packets += st.stats3.tp_drops;
3441 lv = sizeof(struct tpacket_stats);
3442 st.stats1.tp_packets += st.stats1.tp_drops;
3447 case PACKET_AUXDATA:
3450 case PACKET_ORIGDEV:
3453 case PACKET_VNET_HDR:
3454 val = po->has_vnet_hdr;
3456 case PACKET_VERSION:
3457 val = po->tp_version;
3460 if (len > sizeof(int))
3462 if (copy_from_user(&val, optval, len))
3466 val = sizeof(struct tpacket_hdr);
3469 val = sizeof(struct tpacket2_hdr);
3472 val = sizeof(struct tpacket3_hdr);
3478 case PACKET_RESERVE:
3479 val = po->tp_reserve;
3484 case PACKET_TIMESTAMP:
3485 val = po->tp_tstamp;
3489 ((u32)po->fanout->id |
3490 ((u32)po->fanout->type << 16) |
3491 ((u32)po->fanout->flags << 24)) :
3494 case PACKET_TX_HAS_OFF:
3495 val = po->tp_tx_has_off;
3497 case PACKET_QDISC_BYPASS:
3498 val = packet_use_direct_xmit(po);
3501 return -ENOPROTOOPT;
3506 if (put_user(len, optlen))
3508 if (copy_to_user(optval, data, len))
3514 static int packet_notifier(struct notifier_block *this,
3515 unsigned long msg, void *ptr)
3518 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3519 struct net *net = dev_net(dev);
3522 sk_for_each_rcu(sk, &net->packet.sklist) {
3523 struct packet_sock *po = pkt_sk(sk);
3526 case NETDEV_UNREGISTER:
3528 packet_dev_mclist(dev, po->mclist, -1);
3532 if (dev->ifindex == po->ifindex) {
3533 spin_lock(&po->bind_lock);
3535 __unregister_prot_hook(sk, false);
3536 sk->sk_err = ENETDOWN;
3537 if (!sock_flag(sk, SOCK_DEAD))
3538 sk->sk_error_report(sk);
3540 if (msg == NETDEV_UNREGISTER) {
3541 packet_cached_dev_reset(po);
3543 if (po->prot_hook.dev)
3544 dev_put(po->prot_hook.dev);
3545 po->prot_hook.dev = NULL;
3547 spin_unlock(&po->bind_lock);
3551 if (dev->ifindex == po->ifindex) {
3552 spin_lock(&po->bind_lock);
3554 register_prot_hook(sk);
3555 spin_unlock(&po->bind_lock);
3565 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3568 struct sock *sk = sock->sk;
3573 int amount = sk_wmem_alloc_get(sk);
3575 return put_user(amount, (int __user *)arg);
3579 struct sk_buff *skb;
3582 spin_lock_bh(&sk->sk_receive_queue.lock);
3583 skb = skb_peek(&sk->sk_receive_queue);
3586 spin_unlock_bh(&sk->sk_receive_queue.lock);
3587 return put_user(amount, (int __user *)arg);
3590 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3592 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3602 case SIOCGIFBRDADDR:
3603 case SIOCSIFBRDADDR:
3604 case SIOCGIFNETMASK:
3605 case SIOCSIFNETMASK:
3606 case SIOCGIFDSTADDR:
3607 case SIOCSIFDSTADDR:
3609 return inet_dgram_ops.ioctl(sock, cmd, arg);
3613 return -ENOIOCTLCMD;
3618 static unsigned int packet_poll(struct file *file, struct socket *sock,
3621 struct sock *sk = sock->sk;
3622 struct packet_sock *po = pkt_sk(sk);
3623 unsigned int mask = datagram_poll(file, sock, wait);
3625 spin_lock_bh(&sk->sk_receive_queue.lock);
3626 if (po->rx_ring.pg_vec) {
3627 if (!packet_previous_rx_frame(po, &po->rx_ring,
3629 mask |= POLLIN | POLLRDNORM;
3631 spin_unlock_bh(&sk->sk_receive_queue.lock);
3632 spin_lock_bh(&sk->sk_write_queue.lock);
3633 if (po->tx_ring.pg_vec) {
3634 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3635 mask |= POLLOUT | POLLWRNORM;
3637 spin_unlock_bh(&sk->sk_write_queue.lock);
3642 /* Dirty? Well, I still did not learn better way to account
3646 static void packet_mm_open(struct vm_area_struct *vma)
3648 struct file *file = vma->vm_file;
3649 struct socket *sock = file->private_data;
3650 struct sock *sk = sock->sk;
3653 atomic_inc(&pkt_sk(sk)->mapped);
3656 static void packet_mm_close(struct vm_area_struct *vma)
3658 struct file *file = vma->vm_file;
3659 struct socket *sock = file->private_data;
3660 struct sock *sk = sock->sk;
3663 atomic_dec(&pkt_sk(sk)->mapped);
3666 static const struct vm_operations_struct packet_mmap_ops = {
3667 .open = packet_mm_open,
3668 .close = packet_mm_close,
3671 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3676 for (i = 0; i < len; i++) {
3677 if (likely(pg_vec[i].buffer)) {
3678 if (is_vmalloc_addr(pg_vec[i].buffer))
3679 vfree(pg_vec[i].buffer);
3681 free_pages((unsigned long)pg_vec[i].buffer,
3683 pg_vec[i].buffer = NULL;
3689 static char *alloc_one_pg_vec_page(unsigned long order)
3692 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3693 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3695 buffer = (char *) __get_free_pages(gfp_flags, order);
3699 /* __get_free_pages failed, fall back to vmalloc */
3700 buffer = vzalloc((1 << order) * PAGE_SIZE);
3704 /* vmalloc failed, lets dig into swap here */
3705 gfp_flags &= ~__GFP_NORETRY;
3706 buffer = (char *) __get_free_pages(gfp_flags, order);
3710 /* complete and utter failure */
3714 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3716 unsigned int block_nr = req->tp_block_nr;
3720 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3721 if (unlikely(!pg_vec))
3724 for (i = 0; i < block_nr; i++) {
3725 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3726 if (unlikely(!pg_vec[i].buffer))
3727 goto out_free_pgvec;
3734 free_pg_vec(pg_vec, order, block_nr);
3739 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3740 int closing, int tx_ring)
3742 struct pgv *pg_vec = NULL;
3743 struct packet_sock *po = pkt_sk(sk);
3744 int was_running, order = 0;
3745 struct packet_ring_buffer *rb;
3746 struct sk_buff_head *rb_queue;
3749 /* Added to avoid minimal code churn */
3750 struct tpacket_req *req = &req_u->req;
3752 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3753 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3754 WARN(1, "Tx-ring is not supported.\n");
3758 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3759 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3763 if (atomic_read(&po->mapped))
3765 if (packet_read_pending(rb))
3769 if (req->tp_block_nr) {
3770 /* Sanity tests and some calculations */
3772 if (unlikely(rb->pg_vec))
3775 switch (po->tp_version) {
3777 po->tp_hdrlen = TPACKET_HDRLEN;
3780 po->tp_hdrlen = TPACKET2_HDRLEN;
3783 po->tp_hdrlen = TPACKET3_HDRLEN;
3788 if (unlikely((int)req->tp_block_size <= 0))
3790 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3792 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3795 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3798 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3799 if (unlikely(rb->frames_per_block <= 0))
3801 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3806 order = get_order(req->tp_block_size);
3807 pg_vec = alloc_pg_vec(req, order);
3808 if (unlikely(!pg_vec))
3810 switch (po->tp_version) {
3812 /* Transmit path is not supported. We checked
3813 * it above but just being paranoid
3816 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3825 if (unlikely(req->tp_frame_nr))
3831 /* Detach socket from network */
3832 spin_lock(&po->bind_lock);
3833 was_running = po->running;
3837 __unregister_prot_hook(sk, false);
3839 spin_unlock(&po->bind_lock);
3844 mutex_lock(&po->pg_vec_lock);
3845 if (closing || atomic_read(&po->mapped) == 0) {
3847 spin_lock_bh(&rb_queue->lock);
3848 swap(rb->pg_vec, pg_vec);
3849 rb->frame_max = (req->tp_frame_nr - 1);
3851 rb->frame_size = req->tp_frame_size;
3852 spin_unlock_bh(&rb_queue->lock);
3854 swap(rb->pg_vec_order, order);
3855 swap(rb->pg_vec_len, req->tp_block_nr);
3857 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3858 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3859 tpacket_rcv : packet_rcv;
3860 skb_queue_purge(rb_queue);
3861 if (atomic_read(&po->mapped))
3862 pr_err("packet_mmap: vma is busy: %d\n",
3863 atomic_read(&po->mapped));
3865 mutex_unlock(&po->pg_vec_lock);
3867 spin_lock(&po->bind_lock);
3870 register_prot_hook(sk);
3872 spin_unlock(&po->bind_lock);
3873 if (closing && (po->tp_version > TPACKET_V2)) {
3874 /* Because we don't support block-based V3 on tx-ring */
3876 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3881 free_pg_vec(pg_vec, order, req->tp_block_nr);
3886 static int packet_mmap(struct file *file, struct socket *sock,
3887 struct vm_area_struct *vma)
3889 struct sock *sk = sock->sk;
3890 struct packet_sock *po = pkt_sk(sk);
3891 unsigned long size, expected_size;
3892 struct packet_ring_buffer *rb;
3893 unsigned long start;
3900 mutex_lock(&po->pg_vec_lock);
3903 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3905 expected_size += rb->pg_vec_len
3911 if (expected_size == 0)
3914 size = vma->vm_end - vma->vm_start;
3915 if (size != expected_size)
3918 start = vma->vm_start;
3919 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3920 if (rb->pg_vec == NULL)
3923 for (i = 0; i < rb->pg_vec_len; i++) {
3925 void *kaddr = rb->pg_vec[i].buffer;
3928 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3929 page = pgv_to_page(kaddr);
3930 err = vm_insert_page(vma, start, page);
3939 atomic_inc(&po->mapped);
3940 vma->vm_ops = &packet_mmap_ops;
3944 mutex_unlock(&po->pg_vec_lock);
3948 static const struct proto_ops packet_ops_spkt = {
3949 .family = PF_PACKET,
3950 .owner = THIS_MODULE,
3951 .release = packet_release,
3952 .bind = packet_bind_spkt,
3953 .connect = sock_no_connect,
3954 .socketpair = sock_no_socketpair,
3955 .accept = sock_no_accept,
3956 .getname = packet_getname_spkt,
3957 .poll = datagram_poll,
3958 .ioctl = packet_ioctl,
3959 .listen = sock_no_listen,
3960 .shutdown = sock_no_shutdown,
3961 .setsockopt = sock_no_setsockopt,
3962 .getsockopt = sock_no_getsockopt,
3963 .sendmsg = packet_sendmsg_spkt,
3964 .recvmsg = packet_recvmsg,
3965 .mmap = sock_no_mmap,
3966 .sendpage = sock_no_sendpage,
3969 static const struct proto_ops packet_ops = {
3970 .family = PF_PACKET,
3971 .owner = THIS_MODULE,
3972 .release = packet_release,
3973 .bind = packet_bind,
3974 .connect = sock_no_connect,
3975 .socketpair = sock_no_socketpair,
3976 .accept = sock_no_accept,
3977 .getname = packet_getname,
3978 .poll = packet_poll,
3979 .ioctl = packet_ioctl,
3980 .listen = sock_no_listen,
3981 .shutdown = sock_no_shutdown,
3982 .setsockopt = packet_setsockopt,
3983 .getsockopt = packet_getsockopt,
3984 .sendmsg = packet_sendmsg,
3985 .recvmsg = packet_recvmsg,
3986 .mmap = packet_mmap,
3987 .sendpage = sock_no_sendpage,
3990 static const struct net_proto_family packet_family_ops = {
3991 .family = PF_PACKET,
3992 .create = packet_create,
3993 .owner = THIS_MODULE,
3996 static struct notifier_block packet_netdev_notifier = {
3997 .notifier_call = packet_notifier,
4000 #ifdef CONFIG_PROC_FS
4002 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4005 struct net *net = seq_file_net(seq);
4008 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4011 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4013 struct net *net = seq_file_net(seq);
4014 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4017 static void packet_seq_stop(struct seq_file *seq, void *v)
4023 static int packet_seq_show(struct seq_file *seq, void *v)
4025 if (v == SEQ_START_TOKEN)
4026 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4028 struct sock *s = sk_entry(v);
4029 const struct packet_sock *po = pkt_sk(s);
4032 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4034 atomic_read(&s->sk_refcnt),
4039 atomic_read(&s->sk_rmem_alloc),
4040 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4047 static const struct seq_operations packet_seq_ops = {
4048 .start = packet_seq_start,
4049 .next = packet_seq_next,
4050 .stop = packet_seq_stop,
4051 .show = packet_seq_show,
4054 static int packet_seq_open(struct inode *inode, struct file *file)
4056 return seq_open_net(inode, file, &packet_seq_ops,
4057 sizeof(struct seq_net_private));
4060 static const struct file_operations packet_seq_fops = {
4061 .owner = THIS_MODULE,
4062 .open = packet_seq_open,
4064 .llseek = seq_lseek,
4065 .release = seq_release_net,
4070 static int __net_init packet_net_init(struct net *net)
4072 mutex_init(&net->packet.sklist_lock);
4073 INIT_HLIST_HEAD(&net->packet.sklist);
4075 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4081 static void __net_exit packet_net_exit(struct net *net)
4083 remove_proc_entry("packet", net->proc_net);
4086 static struct pernet_operations packet_net_ops = {
4087 .init = packet_net_init,
4088 .exit = packet_net_exit,
4092 static void __exit packet_exit(void)
4094 unregister_netdevice_notifier(&packet_netdev_notifier);
4095 unregister_pernet_subsys(&packet_net_ops);
4096 sock_unregister(PF_PACKET);
4097 proto_unregister(&packet_proto);
4100 static int __init packet_init(void)
4102 int rc = proto_register(&packet_proto, 0);
4107 sock_register(&packet_family_ops);
4108 register_pernet_subsys(&packet_net_ops);
4109 register_netdevice_notifier(&packet_netdev_notifier);
4114 module_init(packet_init);
4115 module_exit(packet_exit);
4116 MODULE_LICENSE("GPL");
4117 MODULE_ALIAS_NETPROTO(PF_PACKET);