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 netdev_features_t features;
244 struct netdev_queue *txq;
245 int ret = NETDEV_TX_BUSY;
247 if (unlikely(!netif_running(dev) ||
248 !netif_carrier_ok(dev)))
251 features = netif_skb_features(skb);
252 if (skb_needs_linearize(skb, features) &&
253 __skb_linearize(skb))
256 txq = skb_get_tx_queue(dev, skb);
260 HARD_TX_LOCK(dev, txq, smp_processor_id());
261 if (!netif_xmit_frozen_or_drv_stopped(txq))
262 ret = netdev_start_xmit(skb, dev, txq, false);
263 HARD_TX_UNLOCK(dev, txq);
267 if (!dev_xmit_complete(ret))
272 atomic_long_inc(&dev->tx_dropped);
274 return NET_XMIT_DROP;
277 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
279 struct net_device *dev;
282 dev = rcu_dereference(po->cached_dev);
290 static void packet_cached_dev_assign(struct packet_sock *po,
291 struct net_device *dev)
293 rcu_assign_pointer(po->cached_dev, dev);
296 static void packet_cached_dev_reset(struct packet_sock *po)
298 RCU_INIT_POINTER(po->cached_dev, NULL);
301 static bool packet_use_direct_xmit(const struct packet_sock *po)
303 return po->xmit == packet_direct_xmit;
306 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
308 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
311 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
313 const struct net_device_ops *ops = dev->netdev_ops;
316 if (ops->ndo_select_queue) {
317 queue_index = ops->ndo_select_queue(dev, skb, NULL,
318 __packet_pick_tx_queue);
319 queue_index = netdev_cap_txqueue(dev, queue_index);
321 queue_index = __packet_pick_tx_queue(dev, skb);
324 skb_set_queue_mapping(skb, queue_index);
327 /* register_prot_hook must be invoked with the po->bind_lock held,
328 * or from a context in which asynchronous accesses to the packet
329 * socket is not possible (packet_create()).
331 static void register_prot_hook(struct sock *sk)
333 struct packet_sock *po = pkt_sk(sk);
337 __fanout_link(sk, po);
339 dev_add_pack(&po->prot_hook);
346 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
347 * held. If the sync parameter is true, we will temporarily drop
348 * the po->bind_lock and do a synchronize_net to make sure no
349 * asynchronous packet processing paths still refer to the elements
350 * of po->prot_hook. If the sync parameter is false, it is the
351 * callers responsibility to take care of this.
353 static void __unregister_prot_hook(struct sock *sk, bool sync)
355 struct packet_sock *po = pkt_sk(sk);
360 __fanout_unlink(sk, po);
362 __dev_remove_pack(&po->prot_hook);
367 spin_unlock(&po->bind_lock);
369 spin_lock(&po->bind_lock);
373 static void unregister_prot_hook(struct sock *sk, bool sync)
375 struct packet_sock *po = pkt_sk(sk);
378 __unregister_prot_hook(sk, sync);
381 static inline struct page * __pure pgv_to_page(void *addr)
383 if (is_vmalloc_addr(addr))
384 return vmalloc_to_page(addr);
385 return virt_to_page(addr);
388 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
390 union tpacket_uhdr h;
393 switch (po->tp_version) {
395 h.h1->tp_status = status;
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
399 h.h2->tp_status = status;
400 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
404 WARN(1, "TPACKET version not supported.\n");
411 static int __packet_get_status(struct packet_sock *po, void *frame)
413 union tpacket_uhdr h;
418 switch (po->tp_version) {
420 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
421 return h.h1->tp_status;
423 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
424 return h.h2->tp_status;
427 WARN(1, "TPACKET version not supported.\n");
433 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
436 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
439 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
440 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
441 return TP_STATUS_TS_RAW_HARDWARE;
443 if (ktime_to_timespec_cond(skb->tstamp, ts))
444 return TP_STATUS_TS_SOFTWARE;
449 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
452 union tpacket_uhdr h;
456 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
460 switch (po->tp_version) {
462 h.h1->tp_sec = ts.tv_sec;
463 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
466 h.h2->tp_sec = ts.tv_sec;
467 h.h2->tp_nsec = ts.tv_nsec;
471 WARN(1, "TPACKET version not supported.\n");
475 /* one flush is safe, as both fields always lie on the same cacheline */
476 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
482 static void *packet_lookup_frame(struct packet_sock *po,
483 struct packet_ring_buffer *rb,
484 unsigned int position,
487 unsigned int pg_vec_pos, frame_offset;
488 union tpacket_uhdr h;
490 pg_vec_pos = position / rb->frames_per_block;
491 frame_offset = position % rb->frames_per_block;
493 h.raw = rb->pg_vec[pg_vec_pos].buffer +
494 (frame_offset * rb->frame_size);
496 if (status != __packet_get_status(po, h.raw))
502 static void *packet_current_frame(struct packet_sock *po,
503 struct packet_ring_buffer *rb,
506 return packet_lookup_frame(po, rb, rb->head, status);
509 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
511 del_timer_sync(&pkc->retire_blk_timer);
514 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
516 struct sk_buff_head *rb_queue)
518 struct tpacket_kbdq_core *pkc;
520 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
521 GET_PBDQC_FROM_RB(&po->rx_ring);
523 spin_lock_bh(&rb_queue->lock);
524 pkc->delete_blk_timer = 1;
525 spin_unlock_bh(&rb_queue->lock);
527 prb_del_retire_blk_timer(pkc);
530 static void prb_init_blk_timer(struct packet_sock *po,
531 struct tpacket_kbdq_core *pkc,
532 void (*func) (unsigned long))
534 init_timer(&pkc->retire_blk_timer);
535 pkc->retire_blk_timer.data = (long)po;
536 pkc->retire_blk_timer.function = func;
537 pkc->retire_blk_timer.expires = jiffies;
540 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
542 struct tpacket_kbdq_core *pkc;
547 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
548 GET_PBDQC_FROM_RB(&po->rx_ring);
549 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
552 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
553 int blk_size_in_bytes)
555 struct net_device *dev;
556 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
557 struct ethtool_cmd ecmd;
562 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
563 if (unlikely(!dev)) {
565 return DEFAULT_PRB_RETIRE_TOV;
567 err = __ethtool_get_settings(dev, &ecmd);
568 speed = ethtool_cmd_speed(&ecmd);
572 * If the link speed is so slow you don't really
573 * need to worry about perf anyways
575 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
576 return DEFAULT_PRB_RETIRE_TOV;
583 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
595 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
596 union tpacket_req_u *req_u)
598 p1->feature_req_word = req_u->req3.tp_feature_req_word;
601 static void init_prb_bdqc(struct packet_sock *po,
602 struct packet_ring_buffer *rb,
604 union tpacket_req_u *req_u, int tx_ring)
606 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
607 struct tpacket_block_desc *pbd;
609 memset(p1, 0x0, sizeof(*p1));
611 p1->knxt_seq_num = 1;
613 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
614 p1->pkblk_start = pg_vec[0].buffer;
615 p1->kblk_size = req_u->req3.tp_block_size;
616 p1->knum_blocks = req_u->req3.tp_block_nr;
617 p1->hdrlen = po->tp_hdrlen;
618 p1->version = po->tp_version;
619 p1->last_kactive_blk_num = 0;
620 po->stats.stats3.tp_freeze_q_cnt = 0;
621 if (req_u->req3.tp_retire_blk_tov)
622 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
624 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
625 req_u->req3.tp_block_size);
626 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
627 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
629 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
630 prb_init_ft_ops(p1, req_u);
631 prb_setup_retire_blk_timer(po, tx_ring);
632 prb_open_block(p1, pbd);
635 /* Do NOT update the last_blk_num first.
636 * Assumes sk_buff_head lock is held.
638 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
640 mod_timer(&pkc->retire_blk_timer,
641 jiffies + pkc->tov_in_jiffies);
642 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
647 * 1) We refresh the timer only when we open a block.
648 * By doing this we don't waste cycles refreshing the timer
649 * on packet-by-packet basis.
651 * With a 1MB block-size, on a 1Gbps line, it will take
652 * i) ~8 ms to fill a block + ii) memcpy etc.
653 * In this cut we are not accounting for the memcpy time.
655 * So, if the user sets the 'tmo' to 10ms then the timer
656 * will never fire while the block is still getting filled
657 * (which is what we want). However, the user could choose
658 * to close a block early and that's fine.
660 * But when the timer does fire, we check whether or not to refresh it.
661 * Since the tmo granularity is in msecs, it is not too expensive
662 * to refresh the timer, lets say every '8' msecs.
663 * Either the user can set the 'tmo' or we can derive it based on
664 * a) line-speed and b) block-size.
665 * prb_calc_retire_blk_tmo() calculates the tmo.
668 static void prb_retire_rx_blk_timer_expired(unsigned long data)
670 struct packet_sock *po = (struct packet_sock *)data;
671 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
673 struct tpacket_block_desc *pbd;
675 spin_lock(&po->sk.sk_receive_queue.lock);
677 frozen = prb_queue_frozen(pkc);
678 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
680 if (unlikely(pkc->delete_blk_timer))
683 /* We only need to plug the race when the block is partially filled.
685 * lock(); increment BLOCK_NUM_PKTS; unlock()
686 * copy_bits() is in progress ...
687 * timer fires on other cpu:
688 * we can't retire the current block because copy_bits
692 if (BLOCK_NUM_PKTS(pbd)) {
693 while (atomic_read(&pkc->blk_fill_in_prog)) {
694 /* Waiting for skb_copy_bits to finish... */
699 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
701 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
702 if (!prb_dispatch_next_block(pkc, po))
707 /* Case 1. Queue was frozen because user-space was
710 if (prb_curr_blk_in_use(pkc, pbd)) {
712 * Ok, user-space is still behind.
713 * So just refresh the timer.
717 /* Case 2. queue was frozen,user-space caught up,
718 * now the link went idle && the timer fired.
719 * We don't have a block to close.So we open this
720 * block and restart the timer.
721 * opening a block thaws the queue,restarts timer
722 * Thawing/timer-refresh is a side effect.
724 prb_open_block(pkc, pbd);
731 _prb_refresh_rx_retire_blk_timer(pkc);
734 spin_unlock(&po->sk.sk_receive_queue.lock);
737 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
738 struct tpacket_block_desc *pbd1, __u32 status)
740 /* Flush everything minus the block header */
742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
747 /* Skip the block header(we know header WILL fit in 4K) */
750 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
751 for (; start < end; start += PAGE_SIZE)
752 flush_dcache_page(pgv_to_page(start));
757 /* Now update the block status. */
759 BLOCK_STATUS(pbd1) = status;
761 /* Flush the block header */
763 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
765 flush_dcache_page(pgv_to_page(start));
775 * 2) Increment active_blk_num
777 * Note:We DONT refresh the timer on purpose.
778 * Because almost always the next block will be opened.
780 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
781 struct tpacket_block_desc *pbd1,
782 struct packet_sock *po, unsigned int stat)
784 __u32 status = TP_STATUS_USER | stat;
786 struct tpacket3_hdr *last_pkt;
787 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
789 if (po->stats.stats3.tp_drops)
790 status |= TP_STATUS_LOSING;
792 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
793 last_pkt->tp_next_offset = 0;
795 /* Get the ts of the last pkt */
796 if (BLOCK_NUM_PKTS(pbd1)) {
797 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
798 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
800 /* Ok, we tmo'd - so get the current time */
803 h1->ts_last_pkt.ts_sec = ts.tv_sec;
804 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
809 /* Flush the block */
810 prb_flush_block(pkc1, pbd1, status);
812 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
815 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
817 pkc->reset_pending_on_curr_blk = 0;
821 * Side effect of opening a block:
823 * 1) prb_queue is thawed.
824 * 2) retire_blk_timer is refreshed.
827 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
828 struct tpacket_block_desc *pbd1)
831 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
835 /* We could have just memset this but we will lose the
836 * flexibility of making the priv area sticky
839 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
840 BLOCK_NUM_PKTS(pbd1) = 0;
841 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
845 h1->ts_first_pkt.ts_sec = ts.tv_sec;
846 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
848 pkc1->pkblk_start = (char *)pbd1;
849 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
851 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
852 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
854 pbd1->version = pkc1->version;
855 pkc1->prev = pkc1->nxt_offset;
856 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
858 prb_thaw_queue(pkc1);
859 _prb_refresh_rx_retire_blk_timer(pkc1);
865 * Queue freeze logic:
866 * 1) Assume tp_block_nr = 8 blocks.
867 * 2) At time 't0', user opens Rx ring.
868 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
869 * 4) user-space is either sleeping or processing block '0'.
870 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
871 * it will close block-7,loop around and try to fill block '0'.
873 * __packet_lookup_frame_in_block
874 * prb_retire_current_block()
875 * prb_dispatch_next_block()
876 * |->(BLOCK_STATUS == USER) evaluates to true
877 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
878 * 6) Now there are two cases:
879 * 6.1) Link goes idle right after the queue is frozen.
880 * But remember, the last open_block() refreshed the timer.
881 * When this timer expires,it will refresh itself so that we can
882 * re-open block-0 in near future.
883 * 6.2) Link is busy and keeps on receiving packets. This is a simple
884 * case and __packet_lookup_frame_in_block will check if block-0
885 * is free and can now be re-used.
887 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
890 pkc->reset_pending_on_curr_blk = 1;
891 po->stats.stats3.tp_freeze_q_cnt++;
894 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
897 * If the next block is free then we will dispatch it
898 * and return a good offset.
899 * Else, we will freeze the queue.
900 * So, caller must check the return value.
902 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
903 struct packet_sock *po)
905 struct tpacket_block_desc *pbd;
909 /* 1. Get current block num */
910 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
912 /* 2. If this block is currently in_use then freeze the queue */
913 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
914 prb_freeze_queue(pkc, po);
920 * open this block and return the offset where the first packet
921 * needs to get stored.
923 prb_open_block(pkc, pbd);
924 return (void *)pkc->nxt_offset;
927 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
928 struct packet_sock *po, unsigned int status)
930 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
932 /* retire/close the current block */
933 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
935 * Plug the case where copy_bits() is in progress on
936 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
937 * have space to copy the pkt in the current block and
938 * called prb_retire_current_block()
940 * We don't need to worry about the TMO case because
941 * the timer-handler already handled this case.
943 if (!(status & TP_STATUS_BLK_TMO)) {
944 while (atomic_read(&pkc->blk_fill_in_prog)) {
945 /* Waiting for skb_copy_bits to finish... */
949 prb_close_block(pkc, pbd, po, status);
954 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
955 struct tpacket_block_desc *pbd)
957 return TP_STATUS_USER & BLOCK_STATUS(pbd);
960 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
962 return pkc->reset_pending_on_curr_blk;
965 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
967 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
968 atomic_dec(&pkc->blk_fill_in_prog);
971 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
972 struct tpacket3_hdr *ppd)
974 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
977 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
978 struct tpacket3_hdr *ppd)
980 ppd->hv1.tp_rxhash = 0;
983 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
984 struct tpacket3_hdr *ppd)
986 if (vlan_tx_tag_present(pkc->skb)) {
987 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
988 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
989 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
991 ppd->hv1.tp_vlan_tci = 0;
992 ppd->hv1.tp_vlan_tpid = 0;
993 ppd->tp_status = TP_STATUS_AVAILABLE;
997 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
998 struct tpacket3_hdr *ppd)
1000 ppd->hv1.tp_padding = 0;
1001 prb_fill_vlan_info(pkc, ppd);
1003 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1004 prb_fill_rxhash(pkc, ppd);
1006 prb_clear_rxhash(pkc, ppd);
1009 static void prb_fill_curr_block(char *curr,
1010 struct tpacket_kbdq_core *pkc,
1011 struct tpacket_block_desc *pbd,
1014 struct tpacket3_hdr *ppd;
1016 ppd = (struct tpacket3_hdr *)curr;
1017 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1019 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1020 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1021 BLOCK_NUM_PKTS(pbd) += 1;
1022 atomic_inc(&pkc->blk_fill_in_prog);
1023 prb_run_all_ft_ops(pkc, ppd);
1026 /* Assumes caller has the sk->rx_queue.lock */
1027 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1028 struct sk_buff *skb,
1033 struct tpacket_kbdq_core *pkc;
1034 struct tpacket_block_desc *pbd;
1037 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1038 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1040 /* Queue is frozen when user space is lagging behind */
1041 if (prb_queue_frozen(pkc)) {
1043 * Check if that last block which caused the queue to freeze,
1044 * is still in_use by user-space.
1046 if (prb_curr_blk_in_use(pkc, pbd)) {
1047 /* Can't record this packet */
1051 * Ok, the block was released by user-space.
1052 * Now let's open that block.
1053 * opening a block also thaws the queue.
1054 * Thawing is a side effect.
1056 prb_open_block(pkc, pbd);
1061 curr = pkc->nxt_offset;
1063 end = (char *)pbd + pkc->kblk_size;
1065 /* first try the current block */
1066 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1067 prb_fill_curr_block(curr, pkc, pbd, len);
1068 return (void *)curr;
1071 /* Ok, close the current block */
1072 prb_retire_current_block(pkc, po, 0);
1074 /* Now, try to dispatch the next block */
1075 curr = (char *)prb_dispatch_next_block(pkc, po);
1077 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1078 prb_fill_curr_block(curr, pkc, pbd, len);
1079 return (void *)curr;
1083 * No free blocks are available.user_space hasn't caught up yet.
1084 * Queue was just frozen and now this packet will get dropped.
1089 static void *packet_current_rx_frame(struct packet_sock *po,
1090 struct sk_buff *skb,
1091 int status, unsigned int len)
1094 switch (po->tp_version) {
1097 curr = packet_lookup_frame(po, &po->rx_ring,
1098 po->rx_ring.head, status);
1101 return __packet_lookup_frame_in_block(po, skb, status, len);
1103 WARN(1, "TPACKET version not supported\n");
1109 static void *prb_lookup_block(struct packet_sock *po,
1110 struct packet_ring_buffer *rb,
1114 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1115 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1117 if (status != BLOCK_STATUS(pbd))
1122 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1125 if (rb->prb_bdqc.kactive_blk_num)
1126 prev = rb->prb_bdqc.kactive_blk_num-1;
1128 prev = rb->prb_bdqc.knum_blocks-1;
1132 /* Assumes caller has held the rx_queue.lock */
1133 static void *__prb_previous_block(struct packet_sock *po,
1134 struct packet_ring_buffer *rb,
1137 unsigned int previous = prb_previous_blk_num(rb);
1138 return prb_lookup_block(po, rb, previous, status);
1141 static void *packet_previous_rx_frame(struct packet_sock *po,
1142 struct packet_ring_buffer *rb,
1145 if (po->tp_version <= TPACKET_V2)
1146 return packet_previous_frame(po, rb, status);
1148 return __prb_previous_block(po, rb, status);
1151 static void packet_increment_rx_head(struct packet_sock *po,
1152 struct packet_ring_buffer *rb)
1154 switch (po->tp_version) {
1157 return packet_increment_head(rb);
1160 WARN(1, "TPACKET version not supported.\n");
1166 static void *packet_previous_frame(struct packet_sock *po,
1167 struct packet_ring_buffer *rb,
1170 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1171 return packet_lookup_frame(po, rb, previous, status);
1174 static void packet_increment_head(struct packet_ring_buffer *buff)
1176 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1179 static void packet_inc_pending(struct packet_ring_buffer *rb)
1181 this_cpu_inc(*rb->pending_refcnt);
1184 static void packet_dec_pending(struct packet_ring_buffer *rb)
1186 this_cpu_dec(*rb->pending_refcnt);
1189 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1191 unsigned int refcnt = 0;
1194 /* We don't use pending refcount in rx_ring. */
1195 if (rb->pending_refcnt == NULL)
1198 for_each_possible_cpu(cpu)
1199 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1204 static int packet_alloc_pending(struct packet_sock *po)
1206 po->rx_ring.pending_refcnt = NULL;
1208 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1209 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1215 static void packet_free_pending(struct packet_sock *po)
1217 free_percpu(po->tx_ring.pending_refcnt);
1220 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1222 struct sock *sk = &po->sk;
1225 if (po->prot_hook.func != tpacket_rcv)
1226 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1229 spin_lock(&sk->sk_receive_queue.lock);
1230 if (po->tp_version == TPACKET_V3)
1231 has_room = prb_lookup_block(po, &po->rx_ring,
1232 po->rx_ring.prb_bdqc.kactive_blk_num,
1235 has_room = packet_lookup_frame(po, &po->rx_ring,
1238 spin_unlock(&sk->sk_receive_queue.lock);
1243 static void packet_sock_destruct(struct sock *sk)
1245 skb_queue_purge(&sk->sk_error_queue);
1247 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1248 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1250 if (!sock_flag(sk, SOCK_DEAD)) {
1251 pr_err("Attempt to release alive packet socket: %p\n", sk);
1255 sk_refcnt_debug_dec(sk);
1258 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1260 int x = atomic_read(&f->rr_cur) + 1;
1268 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1269 struct sk_buff *skb,
1272 return reciprocal_scale(skb_get_hash(skb), num);
1275 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1276 struct sk_buff *skb,
1281 cur = atomic_read(&f->rr_cur);
1282 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1283 fanout_rr_next(f, num))) != cur)
1288 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1289 struct sk_buff *skb,
1292 return smp_processor_id() % num;
1295 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1296 struct sk_buff *skb,
1299 return prandom_u32_max(num);
1302 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1303 struct sk_buff *skb,
1304 unsigned int idx, unsigned int skip,
1309 i = j = min_t(int, f->next[idx], num - 1);
1311 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1323 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1324 struct sk_buff *skb,
1327 return skb_get_queue_mapping(skb) % num;
1330 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1332 return f->flags & (flag >> 8);
1335 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1336 struct packet_type *pt, struct net_device *orig_dev)
1338 struct packet_fanout *f = pt->af_packet_priv;
1339 unsigned int num = f->num_members;
1340 struct packet_sock *po;
1343 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1350 case PACKET_FANOUT_HASH:
1352 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1353 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1357 idx = fanout_demux_hash(f, skb, num);
1359 case PACKET_FANOUT_LB:
1360 idx = fanout_demux_lb(f, skb, num);
1362 case PACKET_FANOUT_CPU:
1363 idx = fanout_demux_cpu(f, skb, num);
1365 case PACKET_FANOUT_RND:
1366 idx = fanout_demux_rnd(f, skb, num);
1368 case PACKET_FANOUT_QM:
1369 idx = fanout_demux_qm(f, skb, num);
1371 case PACKET_FANOUT_ROLLOVER:
1372 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1376 po = pkt_sk(f->arr[idx]);
1377 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1378 unlikely(!packet_rcv_has_room(po, skb))) {
1379 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1380 po = pkt_sk(f->arr[idx]);
1383 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1386 DEFINE_MUTEX(fanout_mutex);
1387 EXPORT_SYMBOL_GPL(fanout_mutex);
1388 static LIST_HEAD(fanout_list);
1390 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1392 struct packet_fanout *f = po->fanout;
1394 spin_lock(&f->lock);
1395 f->arr[f->num_members] = sk;
1398 spin_unlock(&f->lock);
1401 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1403 struct packet_fanout *f = po->fanout;
1406 spin_lock(&f->lock);
1407 for (i = 0; i < f->num_members; i++) {
1408 if (f->arr[i] == sk)
1411 BUG_ON(i >= f->num_members);
1412 f->arr[i] = f->arr[f->num_members - 1];
1414 spin_unlock(&f->lock);
1417 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1419 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1425 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1427 struct packet_sock *po = pkt_sk(sk);
1428 struct packet_fanout *f, *match;
1429 u8 type = type_flags & 0xff;
1430 u8 flags = type_flags >> 8;
1434 case PACKET_FANOUT_ROLLOVER:
1435 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1437 case PACKET_FANOUT_HASH:
1438 case PACKET_FANOUT_LB:
1439 case PACKET_FANOUT_CPU:
1440 case PACKET_FANOUT_RND:
1441 case PACKET_FANOUT_QM:
1453 mutex_lock(&fanout_mutex);
1455 list_for_each_entry(f, &fanout_list, list) {
1457 read_pnet(&f->net) == sock_net(sk)) {
1463 if (match && match->flags != flags)
1467 match = kzalloc(sizeof(*match), GFP_KERNEL);
1470 write_pnet(&match->net, sock_net(sk));
1473 match->flags = flags;
1474 atomic_set(&match->rr_cur, 0);
1475 INIT_LIST_HEAD(&match->list);
1476 spin_lock_init(&match->lock);
1477 atomic_set(&match->sk_ref, 0);
1478 match->prot_hook.type = po->prot_hook.type;
1479 match->prot_hook.dev = po->prot_hook.dev;
1480 match->prot_hook.func = packet_rcv_fanout;
1481 match->prot_hook.af_packet_priv = match;
1482 match->prot_hook.id_match = match_fanout_group;
1483 dev_add_pack(&match->prot_hook);
1484 list_add(&match->list, &fanout_list);
1487 if (match->type == type &&
1488 match->prot_hook.type == po->prot_hook.type &&
1489 match->prot_hook.dev == po->prot_hook.dev) {
1491 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1492 __dev_remove_pack(&po->prot_hook);
1494 atomic_inc(&match->sk_ref);
1495 __fanout_link(sk, po);
1500 mutex_unlock(&fanout_mutex);
1504 static void fanout_release(struct sock *sk)
1506 struct packet_sock *po = pkt_sk(sk);
1507 struct packet_fanout *f;
1513 mutex_lock(&fanout_mutex);
1516 if (atomic_dec_and_test(&f->sk_ref)) {
1518 dev_remove_pack(&f->prot_hook);
1521 mutex_unlock(&fanout_mutex);
1524 static const struct proto_ops packet_ops;
1526 static const struct proto_ops packet_ops_spkt;
1528 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1529 struct packet_type *pt, struct net_device *orig_dev)
1532 struct sockaddr_pkt *spkt;
1535 * When we registered the protocol we saved the socket in the data
1536 * field for just this event.
1539 sk = pt->af_packet_priv;
1542 * Yank back the headers [hope the device set this
1543 * right or kerboom...]
1545 * Incoming packets have ll header pulled,
1548 * For outgoing ones skb->data == skb_mac_header(skb)
1549 * so that this procedure is noop.
1552 if (skb->pkt_type == PACKET_LOOPBACK)
1555 if (!net_eq(dev_net(dev), sock_net(sk)))
1558 skb = skb_share_check(skb, GFP_ATOMIC);
1562 /* drop any routing info */
1565 /* drop conntrack reference */
1568 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1570 skb_push(skb, skb->data - skb_mac_header(skb));
1573 * The SOCK_PACKET socket receives _all_ frames.
1576 spkt->spkt_family = dev->type;
1577 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1578 spkt->spkt_protocol = skb->protocol;
1581 * Charge the memory to the socket. This is done specifically
1582 * to prevent sockets using all the memory up.
1585 if (sock_queue_rcv_skb(sk, skb) == 0)
1596 * Output a raw packet to a device layer. This bypasses all the other
1597 * protocol layers and you must therefore supply it with a complete frame
1600 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1601 struct msghdr *msg, size_t len)
1603 struct sock *sk = sock->sk;
1604 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1605 struct sk_buff *skb = NULL;
1606 struct net_device *dev;
1612 * Get and verify the address.
1616 if (msg->msg_namelen < sizeof(struct sockaddr))
1618 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1619 proto = saddr->spkt_protocol;
1621 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1624 * Find the device first to size check it
1627 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1630 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1636 if (!(dev->flags & IFF_UP))
1640 * You may not queue a frame bigger than the mtu. This is the lowest level
1641 * raw protocol and you must do your own fragmentation at this level.
1644 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1645 if (!netif_supports_nofcs(dev)) {
1646 err = -EPROTONOSUPPORT;
1649 extra_len = 4; /* We're doing our own CRC */
1653 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1657 size_t reserved = LL_RESERVED_SPACE(dev);
1658 int tlen = dev->needed_tailroom;
1659 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1662 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1665 /* FIXME: Save some space for broken drivers that write a hard
1666 * header at transmission time by themselves. PPP is the notable
1667 * one here. This should really be fixed at the driver level.
1669 skb_reserve(skb, reserved);
1670 skb_reset_network_header(skb);
1672 /* Try to align data part correctly */
1677 skb_reset_network_header(skb);
1679 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1685 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1686 /* Earlier code assumed this would be a VLAN pkt,
1687 * double-check this now that we have the actual
1690 struct ethhdr *ehdr;
1691 skb_reset_mac_header(skb);
1692 ehdr = eth_hdr(skb);
1693 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1699 skb->protocol = proto;
1701 skb->priority = sk->sk_priority;
1702 skb->mark = sk->sk_mark;
1704 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1706 if (unlikely(extra_len == 4))
1709 skb_probe_transport_header(skb, 0);
1711 dev_queue_xmit(skb);
1722 static unsigned int run_filter(const struct sk_buff *skb,
1723 const struct sock *sk,
1726 struct sk_filter *filter;
1729 filter = rcu_dereference(sk->sk_filter);
1731 res = SK_RUN_FILTER(filter, skb);
1738 * This function makes lazy skb cloning in hope that most of packets
1739 * are discarded by BPF.
1741 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1742 * and skb->cb are mangled. It works because (and until) packets
1743 * falling here are owned by current CPU. Output packets are cloned
1744 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1745 * sequencially, so that if we return skb to original state on exit,
1746 * we will not harm anyone.
1749 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1750 struct packet_type *pt, struct net_device *orig_dev)
1753 struct sockaddr_ll *sll;
1754 struct packet_sock *po;
1755 u8 *skb_head = skb->data;
1756 int skb_len = skb->len;
1757 unsigned int snaplen, res;
1759 if (skb->pkt_type == PACKET_LOOPBACK)
1762 sk = pt->af_packet_priv;
1765 if (!net_eq(dev_net(dev), sock_net(sk)))
1770 if (dev->header_ops) {
1771 /* The device has an explicit notion of ll header,
1772 * exported to higher levels.
1774 * Otherwise, the device hides details of its frame
1775 * structure, so that corresponding packet head is
1776 * never delivered to user.
1778 if (sk->sk_type != SOCK_DGRAM)
1779 skb_push(skb, skb->data - skb_mac_header(skb));
1780 else if (skb->pkt_type == PACKET_OUTGOING) {
1781 /* Special case: outgoing packets have ll header at head */
1782 skb_pull(skb, skb_network_offset(skb));
1788 res = run_filter(skb, sk, snaplen);
1790 goto drop_n_restore;
1794 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1797 if (skb_shared(skb)) {
1798 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1802 if (skb_head != skb->data) {
1803 skb->data = skb_head;
1810 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1813 sll = &PACKET_SKB_CB(skb)->sa.ll;
1814 sll->sll_family = AF_PACKET;
1815 sll->sll_hatype = dev->type;
1816 sll->sll_protocol = skb->protocol;
1817 sll->sll_pkttype = skb->pkt_type;
1818 if (unlikely(po->origdev))
1819 sll->sll_ifindex = orig_dev->ifindex;
1821 sll->sll_ifindex = dev->ifindex;
1823 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1825 PACKET_SKB_CB(skb)->origlen = skb->len;
1827 if (pskb_trim(skb, snaplen))
1830 skb_set_owner_r(skb, sk);
1834 /* drop conntrack reference */
1837 spin_lock(&sk->sk_receive_queue.lock);
1838 po->stats.stats1.tp_packets++;
1839 skb->dropcount = atomic_read(&sk->sk_drops);
1840 __skb_queue_tail(&sk->sk_receive_queue, skb);
1841 spin_unlock(&sk->sk_receive_queue.lock);
1842 sk->sk_data_ready(sk);
1846 spin_lock(&sk->sk_receive_queue.lock);
1847 po->stats.stats1.tp_drops++;
1848 atomic_inc(&sk->sk_drops);
1849 spin_unlock(&sk->sk_receive_queue.lock);
1852 if (skb_head != skb->data && skb_shared(skb)) {
1853 skb->data = skb_head;
1861 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1862 struct packet_type *pt, struct net_device *orig_dev)
1865 struct packet_sock *po;
1866 struct sockaddr_ll *sll;
1867 union tpacket_uhdr h;
1868 u8 *skb_head = skb->data;
1869 int skb_len = skb->len;
1870 unsigned int snaplen, res;
1871 unsigned long status = TP_STATUS_USER;
1872 unsigned short macoff, netoff, hdrlen;
1873 struct sk_buff *copy_skb = NULL;
1877 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1878 * We may add members to them until current aligned size without forcing
1879 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1881 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1882 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1884 if (skb->pkt_type == PACKET_LOOPBACK)
1887 sk = pt->af_packet_priv;
1890 if (!net_eq(dev_net(dev), sock_net(sk)))
1893 if (dev->header_ops) {
1894 if (sk->sk_type != SOCK_DGRAM)
1895 skb_push(skb, skb->data - skb_mac_header(skb));
1896 else if (skb->pkt_type == PACKET_OUTGOING) {
1897 /* Special case: outgoing packets have ll header at head */
1898 skb_pull(skb, skb_network_offset(skb));
1902 if (skb->ip_summed == CHECKSUM_PARTIAL)
1903 status |= TP_STATUS_CSUMNOTREADY;
1907 res = run_filter(skb, sk, snaplen);
1909 goto drop_n_restore;
1913 if (sk->sk_type == SOCK_DGRAM) {
1914 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1917 unsigned int maclen = skb_network_offset(skb);
1918 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1919 (maclen < 16 ? 16 : maclen)) +
1921 macoff = netoff - maclen;
1923 if (po->tp_version <= TPACKET_V2) {
1924 if (macoff + snaplen > po->rx_ring.frame_size) {
1925 if (po->copy_thresh &&
1926 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1927 if (skb_shared(skb)) {
1928 copy_skb = skb_clone(skb, GFP_ATOMIC);
1930 copy_skb = skb_get(skb);
1931 skb_head = skb->data;
1934 skb_set_owner_r(copy_skb, sk);
1936 snaplen = po->rx_ring.frame_size - macoff;
1937 if ((int)snaplen < 0)
1940 } else if (unlikely(macoff + snaplen >
1941 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1944 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1945 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1946 snaplen, nval, macoff);
1948 if (unlikely((int)snaplen < 0)) {
1950 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1953 spin_lock(&sk->sk_receive_queue.lock);
1954 h.raw = packet_current_rx_frame(po, skb,
1955 TP_STATUS_KERNEL, (macoff+snaplen));
1958 if (po->tp_version <= TPACKET_V2) {
1959 packet_increment_rx_head(po, &po->rx_ring);
1961 * LOSING will be reported till you read the stats,
1962 * because it's COR - Clear On Read.
1963 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1966 if (po->stats.stats1.tp_drops)
1967 status |= TP_STATUS_LOSING;
1969 po->stats.stats1.tp_packets++;
1971 status |= TP_STATUS_COPY;
1972 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1974 spin_unlock(&sk->sk_receive_queue.lock);
1976 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1978 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1979 getnstimeofday(&ts);
1981 status |= ts_status;
1983 switch (po->tp_version) {
1985 h.h1->tp_len = skb->len;
1986 h.h1->tp_snaplen = snaplen;
1987 h.h1->tp_mac = macoff;
1988 h.h1->tp_net = netoff;
1989 h.h1->tp_sec = ts.tv_sec;
1990 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1991 hdrlen = sizeof(*h.h1);
1994 h.h2->tp_len = skb->len;
1995 h.h2->tp_snaplen = snaplen;
1996 h.h2->tp_mac = macoff;
1997 h.h2->tp_net = netoff;
1998 h.h2->tp_sec = ts.tv_sec;
1999 h.h2->tp_nsec = ts.tv_nsec;
2000 if (vlan_tx_tag_present(skb)) {
2001 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
2002 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2003 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2005 h.h2->tp_vlan_tci = 0;
2006 h.h2->tp_vlan_tpid = 0;
2008 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2009 hdrlen = sizeof(*h.h2);
2012 /* tp_nxt_offset,vlan are already populated above.
2013 * So DONT clear those fields here
2015 h.h3->tp_status |= status;
2016 h.h3->tp_len = skb->len;
2017 h.h3->tp_snaplen = snaplen;
2018 h.h3->tp_mac = macoff;
2019 h.h3->tp_net = netoff;
2020 h.h3->tp_sec = ts.tv_sec;
2021 h.h3->tp_nsec = ts.tv_nsec;
2022 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2023 hdrlen = sizeof(*h.h3);
2029 sll = h.raw + TPACKET_ALIGN(hdrlen);
2030 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2031 sll->sll_family = AF_PACKET;
2032 sll->sll_hatype = dev->type;
2033 sll->sll_protocol = skb->protocol;
2034 sll->sll_pkttype = skb->pkt_type;
2035 if (unlikely(po->origdev))
2036 sll->sll_ifindex = orig_dev->ifindex;
2038 sll->sll_ifindex = dev->ifindex;
2042 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2043 if (po->tp_version <= TPACKET_V2) {
2046 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2049 for (start = h.raw; start < end; start += PAGE_SIZE)
2050 flush_dcache_page(pgv_to_page(start));
2055 if (po->tp_version <= TPACKET_V2)
2056 __packet_set_status(po, h.raw, status);
2058 prb_clear_blk_fill_status(&po->rx_ring);
2060 sk->sk_data_ready(sk);
2063 if (skb_head != skb->data && skb_shared(skb)) {
2064 skb->data = skb_head;
2072 po->stats.stats1.tp_drops++;
2073 spin_unlock(&sk->sk_receive_queue.lock);
2075 sk->sk_data_ready(sk);
2076 kfree_skb(copy_skb);
2077 goto drop_n_restore;
2080 static void tpacket_destruct_skb(struct sk_buff *skb)
2082 struct packet_sock *po = pkt_sk(skb->sk);
2084 if (likely(po->tx_ring.pg_vec)) {
2088 ph = skb_shinfo(skb)->destructor_arg;
2089 packet_dec_pending(&po->tx_ring);
2091 ts = __packet_set_timestamp(po, ph, skb);
2092 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2098 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2099 void *frame, struct net_device *dev, int size_max,
2100 __be16 proto, unsigned char *addr, int hlen)
2102 union tpacket_uhdr ph;
2103 int to_write, offset, len, tp_len, nr_frags, len_max;
2104 struct socket *sock = po->sk.sk_socket;
2111 skb->protocol = proto;
2113 skb->priority = po->sk.sk_priority;
2114 skb->mark = po->sk.sk_mark;
2115 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2116 skb_shinfo(skb)->destructor_arg = ph.raw;
2118 switch (po->tp_version) {
2120 tp_len = ph.h2->tp_len;
2123 tp_len = ph.h1->tp_len;
2126 if (unlikely(tp_len > size_max)) {
2127 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2131 skb_reserve(skb, hlen);
2132 skb_reset_network_header(skb);
2134 if (!packet_use_direct_xmit(po))
2135 skb_probe_transport_header(skb, 0);
2136 if (unlikely(po->tp_tx_has_off)) {
2137 int off_min, off_max, off;
2138 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2139 off_max = po->tx_ring.frame_size - tp_len;
2140 if (sock->type == SOCK_DGRAM) {
2141 switch (po->tp_version) {
2143 off = ph.h2->tp_net;
2146 off = ph.h1->tp_net;
2150 switch (po->tp_version) {
2152 off = ph.h2->tp_mac;
2155 off = ph.h1->tp_mac;
2159 if (unlikely((off < off_min) || (off_max < off)))
2161 data = ph.raw + off;
2163 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2167 if (sock->type == SOCK_DGRAM) {
2168 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2170 if (unlikely(err < 0))
2172 } else if (dev->hard_header_len) {
2173 /* net device doesn't like empty head */
2174 if (unlikely(tp_len <= dev->hard_header_len)) {
2175 pr_err("packet size is too short (%d < %d)\n",
2176 tp_len, dev->hard_header_len);
2180 skb_push(skb, dev->hard_header_len);
2181 err = skb_store_bits(skb, 0, data,
2182 dev->hard_header_len);
2186 data += dev->hard_header_len;
2187 to_write -= dev->hard_header_len;
2190 offset = offset_in_page(data);
2191 len_max = PAGE_SIZE - offset;
2192 len = ((to_write > len_max) ? len_max : to_write);
2194 skb->data_len = to_write;
2195 skb->len += to_write;
2196 skb->truesize += to_write;
2197 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2199 while (likely(to_write)) {
2200 nr_frags = skb_shinfo(skb)->nr_frags;
2202 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2203 pr_err("Packet exceed the number of skb frags(%lu)\n",
2208 page = pgv_to_page(data);
2210 flush_dcache_page(page);
2212 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2215 len_max = PAGE_SIZE;
2216 len = ((to_write > len_max) ? len_max : to_write);
2222 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2224 struct sk_buff *skb;
2225 struct net_device *dev;
2227 int err, reserve = 0;
2229 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2230 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2231 int tp_len, size_max;
2232 unsigned char *addr;
2234 int status = TP_STATUS_AVAILABLE;
2237 mutex_lock(&po->pg_vec_lock);
2239 if (likely(saddr == NULL)) {
2240 dev = packet_cached_dev_get(po);
2245 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2247 if (msg->msg_namelen < (saddr->sll_halen
2248 + offsetof(struct sockaddr_ll,
2251 proto = saddr->sll_protocol;
2252 addr = saddr->sll_addr;
2253 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2257 if (unlikely(dev == NULL))
2260 if (unlikely(!(dev->flags & IFF_UP)))
2263 reserve = dev->hard_header_len + VLAN_HLEN;
2264 size_max = po->tx_ring.frame_size
2265 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2267 if (size_max > dev->mtu + reserve)
2268 size_max = dev->mtu + reserve;
2271 ph = packet_current_frame(po, &po->tx_ring,
2272 TP_STATUS_SEND_REQUEST);
2273 if (unlikely(ph == NULL)) {
2274 if (need_wait && need_resched())
2279 status = TP_STATUS_SEND_REQUEST;
2280 hlen = LL_RESERVED_SPACE(dev);
2281 tlen = dev->needed_tailroom;
2282 skb = sock_alloc_send_skb(&po->sk,
2283 hlen + tlen + sizeof(struct sockaddr_ll),
2286 if (unlikely(skb == NULL))
2289 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2291 if (tp_len > dev->mtu + dev->hard_header_len) {
2292 struct ethhdr *ehdr;
2293 /* Earlier code assumed this would be a VLAN pkt,
2294 * double-check this now that we have the actual
2298 skb_reset_mac_header(skb);
2299 ehdr = eth_hdr(skb);
2300 if (ehdr->h_proto != htons(ETH_P_8021Q))
2303 if (unlikely(tp_len < 0)) {
2305 __packet_set_status(po, ph,
2306 TP_STATUS_AVAILABLE);
2307 packet_increment_head(&po->tx_ring);
2311 status = TP_STATUS_WRONG_FORMAT;
2317 packet_pick_tx_queue(dev, skb);
2319 skb->destructor = tpacket_destruct_skb;
2320 __packet_set_status(po, ph, TP_STATUS_SENDING);
2321 packet_inc_pending(&po->tx_ring);
2323 status = TP_STATUS_SEND_REQUEST;
2324 err = po->xmit(skb);
2325 if (unlikely(err > 0)) {
2326 err = net_xmit_errno(err);
2327 if (err && __packet_get_status(po, ph) ==
2328 TP_STATUS_AVAILABLE) {
2329 /* skb was destructed already */
2334 * skb was dropped but not destructed yet;
2335 * let's treat it like congestion or err < 0
2339 packet_increment_head(&po->tx_ring);
2341 } while (likely((ph != NULL) ||
2342 /* Note: packet_read_pending() might be slow if we have
2343 * to call it as it's per_cpu variable, but in fast-path
2344 * we already short-circuit the loop with the first
2345 * condition, and luckily don't have to go that path
2348 (need_wait && packet_read_pending(&po->tx_ring))));
2354 __packet_set_status(po, ph, status);
2359 mutex_unlock(&po->pg_vec_lock);
2363 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2364 size_t reserve, size_t len,
2365 size_t linear, int noblock,
2368 struct sk_buff *skb;
2370 /* Under a page? Don't bother with paged skb. */
2371 if (prepad + len < PAGE_SIZE || !linear)
2374 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2379 skb_reserve(skb, reserve);
2380 skb_put(skb, linear);
2381 skb->data_len = len - linear;
2382 skb->len += len - linear;
2387 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2389 struct sock *sk = sock->sk;
2390 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2391 struct sk_buff *skb;
2392 struct net_device *dev;
2394 unsigned char *addr;
2395 int err, reserve = 0;
2396 struct virtio_net_hdr vnet_hdr = { 0 };
2399 struct packet_sock *po = pkt_sk(sk);
2400 unsigned short gso_type = 0;
2405 * Get and verify the address.
2408 if (likely(saddr == NULL)) {
2409 dev = packet_cached_dev_get(po);
2414 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2416 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2418 proto = saddr->sll_protocol;
2419 addr = saddr->sll_addr;
2420 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2424 if (unlikely(dev == NULL))
2427 if (unlikely(!(dev->flags & IFF_UP)))
2430 if (sock->type == SOCK_RAW)
2431 reserve = dev->hard_header_len;
2432 if (po->has_vnet_hdr) {
2433 vnet_hdr_len = sizeof(vnet_hdr);
2436 if (len < vnet_hdr_len)
2439 len -= vnet_hdr_len;
2441 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2446 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2447 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2449 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2450 vnet_hdr.csum_offset + 2;
2453 if (vnet_hdr.hdr_len > len)
2456 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2457 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2458 case VIRTIO_NET_HDR_GSO_TCPV4:
2459 gso_type = SKB_GSO_TCPV4;
2461 case VIRTIO_NET_HDR_GSO_TCPV6:
2462 gso_type = SKB_GSO_TCPV6;
2464 case VIRTIO_NET_HDR_GSO_UDP:
2465 gso_type = SKB_GSO_UDP;
2471 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2472 gso_type |= SKB_GSO_TCP_ECN;
2474 if (vnet_hdr.gso_size == 0)
2480 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2481 if (!netif_supports_nofcs(dev)) {
2482 err = -EPROTONOSUPPORT;
2485 extra_len = 4; /* We're doing our own CRC */
2489 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2493 hlen = LL_RESERVED_SPACE(dev);
2494 tlen = dev->needed_tailroom;
2495 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2496 msg->msg_flags & MSG_DONTWAIT, &err);
2500 skb_set_network_header(skb, reserve);
2503 if (sock->type == SOCK_DGRAM &&
2504 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2507 /* Returns -EFAULT on error */
2508 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2512 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2514 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2515 /* Earlier code assumed this would be a VLAN pkt,
2516 * double-check this now that we have the actual
2519 struct ethhdr *ehdr;
2520 skb_reset_mac_header(skb);
2521 ehdr = eth_hdr(skb);
2522 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2528 skb->protocol = proto;
2530 skb->priority = sk->sk_priority;
2531 skb->mark = sk->sk_mark;
2533 packet_pick_tx_queue(dev, skb);
2535 if (po->has_vnet_hdr) {
2536 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2537 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2538 vnet_hdr.csum_offset)) {
2544 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2545 skb_shinfo(skb)->gso_type = gso_type;
2547 /* Header must be checked, and gso_segs computed. */
2548 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2549 skb_shinfo(skb)->gso_segs = 0;
2551 len += vnet_hdr_len;
2554 if (!packet_use_direct_xmit(po))
2555 skb_probe_transport_header(skb, reserve);
2556 if (unlikely(extra_len == 4))
2559 err = po->xmit(skb);
2560 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2576 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2577 struct msghdr *msg, size_t len)
2579 struct sock *sk = sock->sk;
2580 struct packet_sock *po = pkt_sk(sk);
2582 if (po->tx_ring.pg_vec)
2583 return tpacket_snd(po, msg);
2585 return packet_snd(sock, msg, len);
2589 * Close a PACKET socket. This is fairly simple. We immediately go
2590 * to 'closed' state and remove our protocol entry in the device list.
2593 static int packet_release(struct socket *sock)
2595 struct sock *sk = sock->sk;
2596 struct packet_sock *po;
2598 union tpacket_req_u req_u;
2606 mutex_lock(&net->packet.sklist_lock);
2607 sk_del_node_init_rcu(sk);
2608 mutex_unlock(&net->packet.sklist_lock);
2611 sock_prot_inuse_add(net, sk->sk_prot, -1);
2614 spin_lock(&po->bind_lock);
2615 unregister_prot_hook(sk, false);
2616 packet_cached_dev_reset(po);
2618 if (po->prot_hook.dev) {
2619 dev_put(po->prot_hook.dev);
2620 po->prot_hook.dev = NULL;
2622 spin_unlock(&po->bind_lock);
2624 packet_flush_mclist(sk);
2626 if (po->rx_ring.pg_vec) {
2627 memset(&req_u, 0, sizeof(req_u));
2628 packet_set_ring(sk, &req_u, 1, 0);
2631 if (po->tx_ring.pg_vec) {
2632 memset(&req_u, 0, sizeof(req_u));
2633 packet_set_ring(sk, &req_u, 1, 1);
2640 * Now the socket is dead. No more input will appear.
2647 skb_queue_purge(&sk->sk_receive_queue);
2648 packet_free_pending(po);
2649 sk_refcnt_debug_release(sk);
2656 * Attach a packet hook.
2659 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2661 struct packet_sock *po = pkt_sk(sk);
2662 const struct net_device *dev_curr;
2674 spin_lock(&po->bind_lock);
2676 proto_curr = po->prot_hook.type;
2677 dev_curr = po->prot_hook.dev;
2679 need_rehook = proto_curr != proto || dev_curr != dev;
2682 unregister_prot_hook(sk, true);
2685 po->prot_hook.type = proto;
2687 if (po->prot_hook.dev)
2688 dev_put(po->prot_hook.dev);
2690 po->prot_hook.dev = dev;
2692 po->ifindex = dev ? dev->ifindex : 0;
2693 packet_cached_dev_assign(po, dev);
2696 if (proto == 0 || !need_rehook)
2699 if (!dev || (dev->flags & IFF_UP)) {
2700 register_prot_hook(sk);
2702 sk->sk_err = ENETDOWN;
2703 if (!sock_flag(sk, SOCK_DEAD))
2704 sk->sk_error_report(sk);
2708 spin_unlock(&po->bind_lock);
2714 * Bind a packet socket to a device
2717 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2720 struct sock *sk = sock->sk;
2722 struct net_device *dev;
2729 if (addr_len != sizeof(struct sockaddr))
2731 strlcpy(name, uaddr->sa_data, sizeof(name));
2733 dev = dev_get_by_name(sock_net(sk), name);
2735 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2739 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2741 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2742 struct sock *sk = sock->sk;
2743 struct net_device *dev = NULL;
2751 if (addr_len < sizeof(struct sockaddr_ll))
2753 if (sll->sll_family != AF_PACKET)
2756 if (sll->sll_ifindex) {
2758 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2762 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2768 static struct proto packet_proto = {
2770 .owner = THIS_MODULE,
2771 .obj_size = sizeof(struct packet_sock),
2775 * Create a packet of type SOCK_PACKET.
2778 static int packet_create(struct net *net, struct socket *sock, int protocol,
2782 struct packet_sock *po;
2783 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2786 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2788 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2789 sock->type != SOCK_PACKET)
2790 return -ESOCKTNOSUPPORT;
2792 sock->state = SS_UNCONNECTED;
2795 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2799 sock->ops = &packet_ops;
2800 if (sock->type == SOCK_PACKET)
2801 sock->ops = &packet_ops_spkt;
2803 sock_init_data(sock, sk);
2806 sk->sk_family = PF_PACKET;
2808 po->xmit = dev_queue_xmit;
2810 err = packet_alloc_pending(po);
2814 packet_cached_dev_reset(po);
2816 sk->sk_destruct = packet_sock_destruct;
2817 sk_refcnt_debug_inc(sk);
2820 * Attach a protocol block
2823 spin_lock_init(&po->bind_lock);
2824 mutex_init(&po->pg_vec_lock);
2825 po->prot_hook.func = packet_rcv;
2827 if (sock->type == SOCK_PACKET)
2828 po->prot_hook.func = packet_rcv_spkt;
2830 po->prot_hook.af_packet_priv = sk;
2833 po->prot_hook.type = proto;
2834 register_prot_hook(sk);
2837 mutex_lock(&net->packet.sklist_lock);
2838 sk_add_node_rcu(sk, &net->packet.sklist);
2839 mutex_unlock(&net->packet.sklist_lock);
2842 sock_prot_inuse_add(net, &packet_proto, 1);
2853 * Pull a packet from our receive queue and hand it to the user.
2854 * If necessary we block.
2857 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2858 struct msghdr *msg, size_t len, int flags)
2860 struct sock *sk = sock->sk;
2861 struct sk_buff *skb;
2863 int vnet_hdr_len = 0;
2866 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2870 /* What error should we return now? EUNATTACH? */
2871 if (pkt_sk(sk)->ifindex < 0)
2875 if (flags & MSG_ERRQUEUE) {
2876 err = sock_recv_errqueue(sk, msg, len,
2877 SOL_PACKET, PACKET_TX_TIMESTAMP);
2882 * Call the generic datagram receiver. This handles all sorts
2883 * of horrible races and re-entrancy so we can forget about it
2884 * in the protocol layers.
2886 * Now it will return ENETDOWN, if device have just gone down,
2887 * but then it will block.
2890 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2893 * An error occurred so return it. Because skb_recv_datagram()
2894 * handles the blocking we don't see and worry about blocking
2901 if (pkt_sk(sk)->has_vnet_hdr) {
2902 struct virtio_net_hdr vnet_hdr = { 0 };
2905 vnet_hdr_len = sizeof(vnet_hdr);
2906 if (len < vnet_hdr_len)
2909 len -= vnet_hdr_len;
2911 if (skb_is_gso(skb)) {
2912 struct skb_shared_info *sinfo = skb_shinfo(skb);
2914 /* This is a hint as to how much should be linear. */
2915 vnet_hdr.hdr_len = skb_headlen(skb);
2916 vnet_hdr.gso_size = sinfo->gso_size;
2917 if (sinfo->gso_type & SKB_GSO_TCPV4)
2918 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2919 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2920 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2921 else if (sinfo->gso_type & SKB_GSO_UDP)
2922 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2923 else if (sinfo->gso_type & SKB_GSO_FCOE)
2927 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2928 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2930 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2932 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2933 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2934 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2935 vnet_hdr.csum_offset = skb->csum_offset;
2936 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2937 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2938 } /* else everything is zero */
2940 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2946 /* You lose any data beyond the buffer you gave. If it worries
2947 * a user program they can ask the device for its MTU
2953 msg->msg_flags |= MSG_TRUNC;
2956 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2960 sock_recv_ts_and_drops(msg, sk, skb);
2962 if (msg->msg_name) {
2963 /* If the address length field is there to be filled
2964 * in, we fill it in now.
2966 if (sock->type == SOCK_PACKET) {
2967 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
2968 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2970 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2971 msg->msg_namelen = sll->sll_halen +
2972 offsetof(struct sockaddr_ll, sll_addr);
2974 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2978 if (pkt_sk(sk)->auxdata) {
2979 struct tpacket_auxdata aux;
2981 aux.tp_status = TP_STATUS_USER;
2982 if (skb->ip_summed == CHECKSUM_PARTIAL)
2983 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2984 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2985 aux.tp_snaplen = skb->len;
2987 aux.tp_net = skb_network_offset(skb);
2988 if (vlan_tx_tag_present(skb)) {
2989 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2990 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2991 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2993 aux.tp_vlan_tci = 0;
2994 aux.tp_vlan_tpid = 0;
2996 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3000 * Free or return the buffer as appropriate. Again this
3001 * hides all the races and re-entrancy issues from us.
3003 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3006 skb_free_datagram(sk, skb);
3011 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3012 int *uaddr_len, int peer)
3014 struct net_device *dev;
3015 struct sock *sk = sock->sk;
3020 uaddr->sa_family = AF_PACKET;
3021 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3023 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3025 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3027 *uaddr_len = sizeof(*uaddr);
3032 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3033 int *uaddr_len, int peer)
3035 struct net_device *dev;
3036 struct sock *sk = sock->sk;
3037 struct packet_sock *po = pkt_sk(sk);
3038 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3043 sll->sll_family = AF_PACKET;
3044 sll->sll_ifindex = po->ifindex;
3045 sll->sll_protocol = po->num;
3046 sll->sll_pkttype = 0;
3048 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3050 sll->sll_hatype = dev->type;
3051 sll->sll_halen = dev->addr_len;
3052 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3054 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3058 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3063 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3067 case PACKET_MR_MULTICAST:
3068 if (i->alen != dev->addr_len)
3071 return dev_mc_add(dev, i->addr);
3073 return dev_mc_del(dev, i->addr);
3075 case PACKET_MR_PROMISC:
3076 return dev_set_promiscuity(dev, what);
3077 case PACKET_MR_ALLMULTI:
3078 return dev_set_allmulti(dev, what);
3079 case PACKET_MR_UNICAST:
3080 if (i->alen != dev->addr_len)
3083 return dev_uc_add(dev, i->addr);
3085 return dev_uc_del(dev, i->addr);
3093 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3095 for ( ; i; i = i->next) {
3096 if (i->ifindex == dev->ifindex)
3097 packet_dev_mc(dev, i, what);
3101 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3103 struct packet_sock *po = pkt_sk(sk);
3104 struct packet_mclist *ml, *i;
3105 struct net_device *dev;
3111 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3116 if (mreq->mr_alen > dev->addr_len)
3120 i = kmalloc(sizeof(*i), GFP_KERNEL);
3125 for (ml = po->mclist; ml; ml = ml->next) {
3126 if (ml->ifindex == mreq->mr_ifindex &&
3127 ml->type == mreq->mr_type &&
3128 ml->alen == mreq->mr_alen &&
3129 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3131 /* Free the new element ... */
3137 i->type = mreq->mr_type;
3138 i->ifindex = mreq->mr_ifindex;
3139 i->alen = mreq->mr_alen;
3140 memcpy(i->addr, mreq->mr_address, i->alen);
3142 i->next = po->mclist;
3144 err = packet_dev_mc(dev, i, 1);
3146 po->mclist = i->next;
3155 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3157 struct packet_mclist *ml, **mlp;
3161 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3162 if (ml->ifindex == mreq->mr_ifindex &&
3163 ml->type == mreq->mr_type &&
3164 ml->alen == mreq->mr_alen &&
3165 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3166 if (--ml->count == 0) {
3167 struct net_device *dev;
3169 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3171 packet_dev_mc(dev, ml, -1);
3179 return -EADDRNOTAVAIL;
3182 static void packet_flush_mclist(struct sock *sk)
3184 struct packet_sock *po = pkt_sk(sk);
3185 struct packet_mclist *ml;
3191 while ((ml = po->mclist) != NULL) {
3192 struct net_device *dev;
3194 po->mclist = ml->next;
3195 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3197 packet_dev_mc(dev, ml, -1);
3204 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3206 struct sock *sk = sock->sk;
3207 struct packet_sock *po = pkt_sk(sk);
3210 if (level != SOL_PACKET)
3211 return -ENOPROTOOPT;
3214 case PACKET_ADD_MEMBERSHIP:
3215 case PACKET_DROP_MEMBERSHIP:
3217 struct packet_mreq_max mreq;
3219 memset(&mreq, 0, sizeof(mreq));
3220 if (len < sizeof(struct packet_mreq))
3222 if (len > sizeof(mreq))
3224 if (copy_from_user(&mreq, optval, len))
3226 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3228 if (optname == PACKET_ADD_MEMBERSHIP)
3229 ret = packet_mc_add(sk, &mreq);
3231 ret = packet_mc_drop(sk, &mreq);
3235 case PACKET_RX_RING:
3236 case PACKET_TX_RING:
3238 union tpacket_req_u req_u;
3241 switch (po->tp_version) {
3244 len = sizeof(req_u.req);
3248 len = sizeof(req_u.req3);
3253 if (pkt_sk(sk)->has_vnet_hdr)
3255 if (copy_from_user(&req_u.req, optval, len))
3257 return packet_set_ring(sk, &req_u, 0,
3258 optname == PACKET_TX_RING);
3260 case PACKET_COPY_THRESH:
3264 if (optlen != sizeof(val))
3266 if (copy_from_user(&val, optval, sizeof(val)))
3269 pkt_sk(sk)->copy_thresh = val;
3272 case PACKET_VERSION:
3276 if (optlen != sizeof(val))
3278 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3280 if (copy_from_user(&val, optval, sizeof(val)))
3286 po->tp_version = val;
3292 case PACKET_RESERVE:
3296 if (optlen != sizeof(val))
3298 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3300 if (copy_from_user(&val, optval, sizeof(val)))
3302 po->tp_reserve = val;
3309 if (optlen != sizeof(val))
3311 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3313 if (copy_from_user(&val, optval, sizeof(val)))
3315 po->tp_loss = !!val;
3318 case PACKET_AUXDATA:
3322 if (optlen < sizeof(val))
3324 if (copy_from_user(&val, optval, sizeof(val)))
3327 po->auxdata = !!val;
3330 case PACKET_ORIGDEV:
3334 if (optlen < sizeof(val))
3336 if (copy_from_user(&val, optval, sizeof(val)))
3339 po->origdev = !!val;
3342 case PACKET_VNET_HDR:
3346 if (sock->type != SOCK_RAW)
3348 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3350 if (optlen < sizeof(val))
3352 if (copy_from_user(&val, optval, sizeof(val)))
3355 po->has_vnet_hdr = !!val;
3358 case PACKET_TIMESTAMP:
3362 if (optlen != sizeof(val))
3364 if (copy_from_user(&val, optval, sizeof(val)))
3367 po->tp_tstamp = val;
3374 if (optlen != sizeof(val))
3376 if (copy_from_user(&val, optval, sizeof(val)))
3379 return fanout_add(sk, val & 0xffff, val >> 16);
3381 case PACKET_TX_HAS_OFF:
3385 if (optlen != sizeof(val))
3387 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3389 if (copy_from_user(&val, optval, sizeof(val)))
3391 po->tp_tx_has_off = !!val;
3394 case PACKET_QDISC_BYPASS:
3398 if (optlen != sizeof(val))
3400 if (copy_from_user(&val, optval, sizeof(val)))
3403 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3407 return -ENOPROTOOPT;
3411 static int packet_getsockopt(struct socket *sock, int level, int optname,
3412 char __user *optval, int __user *optlen)
3415 int val, lv = sizeof(val);
3416 struct sock *sk = sock->sk;
3417 struct packet_sock *po = pkt_sk(sk);
3419 union tpacket_stats_u st;
3421 if (level != SOL_PACKET)
3422 return -ENOPROTOOPT;
3424 if (get_user(len, optlen))
3431 case PACKET_STATISTICS:
3432 spin_lock_bh(&sk->sk_receive_queue.lock);
3433 memcpy(&st, &po->stats, sizeof(st));
3434 memset(&po->stats, 0, sizeof(po->stats));
3435 spin_unlock_bh(&sk->sk_receive_queue.lock);
3437 if (po->tp_version == TPACKET_V3) {
3438 lv = sizeof(struct tpacket_stats_v3);
3439 st.stats3.tp_packets += st.stats3.tp_drops;
3442 lv = sizeof(struct tpacket_stats);
3443 st.stats1.tp_packets += st.stats1.tp_drops;
3448 case PACKET_AUXDATA:
3451 case PACKET_ORIGDEV:
3454 case PACKET_VNET_HDR:
3455 val = po->has_vnet_hdr;
3457 case PACKET_VERSION:
3458 val = po->tp_version;
3461 if (len > sizeof(int))
3463 if (copy_from_user(&val, optval, len))
3467 val = sizeof(struct tpacket_hdr);
3470 val = sizeof(struct tpacket2_hdr);
3473 val = sizeof(struct tpacket3_hdr);
3479 case PACKET_RESERVE:
3480 val = po->tp_reserve;
3485 case PACKET_TIMESTAMP:
3486 val = po->tp_tstamp;
3490 ((u32)po->fanout->id |
3491 ((u32)po->fanout->type << 16) |
3492 ((u32)po->fanout->flags << 24)) :
3495 case PACKET_TX_HAS_OFF:
3496 val = po->tp_tx_has_off;
3498 case PACKET_QDISC_BYPASS:
3499 val = packet_use_direct_xmit(po);
3502 return -ENOPROTOOPT;
3507 if (put_user(len, optlen))
3509 if (copy_to_user(optval, data, len))
3515 static int packet_notifier(struct notifier_block *this,
3516 unsigned long msg, void *ptr)
3519 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3520 struct net *net = dev_net(dev);
3523 sk_for_each_rcu(sk, &net->packet.sklist) {
3524 struct packet_sock *po = pkt_sk(sk);
3527 case NETDEV_UNREGISTER:
3529 packet_dev_mclist(dev, po->mclist, -1);
3533 if (dev->ifindex == po->ifindex) {
3534 spin_lock(&po->bind_lock);
3536 __unregister_prot_hook(sk, false);
3537 sk->sk_err = ENETDOWN;
3538 if (!sock_flag(sk, SOCK_DEAD))
3539 sk->sk_error_report(sk);
3541 if (msg == NETDEV_UNREGISTER) {
3542 packet_cached_dev_reset(po);
3544 if (po->prot_hook.dev)
3545 dev_put(po->prot_hook.dev);
3546 po->prot_hook.dev = NULL;
3548 spin_unlock(&po->bind_lock);
3552 if (dev->ifindex == po->ifindex) {
3553 spin_lock(&po->bind_lock);
3555 register_prot_hook(sk);
3556 spin_unlock(&po->bind_lock);
3566 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3569 struct sock *sk = sock->sk;
3574 int amount = sk_wmem_alloc_get(sk);
3576 return put_user(amount, (int __user *)arg);
3580 struct sk_buff *skb;
3583 spin_lock_bh(&sk->sk_receive_queue.lock);
3584 skb = skb_peek(&sk->sk_receive_queue);
3587 spin_unlock_bh(&sk->sk_receive_queue.lock);
3588 return put_user(amount, (int __user *)arg);
3591 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3593 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3603 case SIOCGIFBRDADDR:
3604 case SIOCSIFBRDADDR:
3605 case SIOCGIFNETMASK:
3606 case SIOCSIFNETMASK:
3607 case SIOCGIFDSTADDR:
3608 case SIOCSIFDSTADDR:
3610 return inet_dgram_ops.ioctl(sock, cmd, arg);
3614 return -ENOIOCTLCMD;
3619 static unsigned int packet_poll(struct file *file, struct socket *sock,
3622 struct sock *sk = sock->sk;
3623 struct packet_sock *po = pkt_sk(sk);
3624 unsigned int mask = datagram_poll(file, sock, wait);
3626 spin_lock_bh(&sk->sk_receive_queue.lock);
3627 if (po->rx_ring.pg_vec) {
3628 if (!packet_previous_rx_frame(po, &po->rx_ring,
3630 mask |= POLLIN | POLLRDNORM;
3632 spin_unlock_bh(&sk->sk_receive_queue.lock);
3633 spin_lock_bh(&sk->sk_write_queue.lock);
3634 if (po->tx_ring.pg_vec) {
3635 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3636 mask |= POLLOUT | POLLWRNORM;
3638 spin_unlock_bh(&sk->sk_write_queue.lock);
3643 /* Dirty? Well, I still did not learn better way to account
3647 static void packet_mm_open(struct vm_area_struct *vma)
3649 struct file *file = vma->vm_file;
3650 struct socket *sock = file->private_data;
3651 struct sock *sk = sock->sk;
3654 atomic_inc(&pkt_sk(sk)->mapped);
3657 static void packet_mm_close(struct vm_area_struct *vma)
3659 struct file *file = vma->vm_file;
3660 struct socket *sock = file->private_data;
3661 struct sock *sk = sock->sk;
3664 atomic_dec(&pkt_sk(sk)->mapped);
3667 static const struct vm_operations_struct packet_mmap_ops = {
3668 .open = packet_mm_open,
3669 .close = packet_mm_close,
3672 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3677 for (i = 0; i < len; i++) {
3678 if (likely(pg_vec[i].buffer)) {
3679 if (is_vmalloc_addr(pg_vec[i].buffer))
3680 vfree(pg_vec[i].buffer);
3682 free_pages((unsigned long)pg_vec[i].buffer,
3684 pg_vec[i].buffer = NULL;
3690 static char *alloc_one_pg_vec_page(unsigned long order)
3693 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3694 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3696 buffer = (char *) __get_free_pages(gfp_flags, order);
3700 /* __get_free_pages failed, fall back to vmalloc */
3701 buffer = vzalloc((1 << order) * PAGE_SIZE);
3705 /* vmalloc failed, lets dig into swap here */
3706 gfp_flags &= ~__GFP_NORETRY;
3707 buffer = (char *) __get_free_pages(gfp_flags, order);
3711 /* complete and utter failure */
3715 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3717 unsigned int block_nr = req->tp_block_nr;
3721 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3722 if (unlikely(!pg_vec))
3725 for (i = 0; i < block_nr; i++) {
3726 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3727 if (unlikely(!pg_vec[i].buffer))
3728 goto out_free_pgvec;
3735 free_pg_vec(pg_vec, order, block_nr);
3740 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3741 int closing, int tx_ring)
3743 struct pgv *pg_vec = NULL;
3744 struct packet_sock *po = pkt_sk(sk);
3745 int was_running, order = 0;
3746 struct packet_ring_buffer *rb;
3747 struct sk_buff_head *rb_queue;
3750 /* Added to avoid minimal code churn */
3751 struct tpacket_req *req = &req_u->req;
3753 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3754 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3755 WARN(1, "Tx-ring is not supported.\n");
3759 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3760 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3764 if (atomic_read(&po->mapped))
3766 if (packet_read_pending(rb))
3770 if (req->tp_block_nr) {
3771 /* Sanity tests and some calculations */
3773 if (unlikely(rb->pg_vec))
3776 switch (po->tp_version) {
3778 po->tp_hdrlen = TPACKET_HDRLEN;
3781 po->tp_hdrlen = TPACKET2_HDRLEN;
3784 po->tp_hdrlen = TPACKET3_HDRLEN;
3789 if (unlikely((int)req->tp_block_size <= 0))
3791 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3793 if (po->tp_version >= TPACKET_V3 &&
3794 (int)(req->tp_block_size -
3795 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3797 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3800 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3803 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3804 if (unlikely(rb->frames_per_block <= 0))
3806 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3811 order = get_order(req->tp_block_size);
3812 pg_vec = alloc_pg_vec(req, order);
3813 if (unlikely(!pg_vec))
3815 switch (po->tp_version) {
3817 /* Transmit path is not supported. We checked
3818 * it above but just being paranoid
3821 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3830 if (unlikely(req->tp_frame_nr))
3836 /* Detach socket from network */
3837 spin_lock(&po->bind_lock);
3838 was_running = po->running;
3842 __unregister_prot_hook(sk, false);
3844 spin_unlock(&po->bind_lock);
3849 mutex_lock(&po->pg_vec_lock);
3850 if (closing || atomic_read(&po->mapped) == 0) {
3852 spin_lock_bh(&rb_queue->lock);
3853 swap(rb->pg_vec, pg_vec);
3854 rb->frame_max = (req->tp_frame_nr - 1);
3856 rb->frame_size = req->tp_frame_size;
3857 spin_unlock_bh(&rb_queue->lock);
3859 swap(rb->pg_vec_order, order);
3860 swap(rb->pg_vec_len, req->tp_block_nr);
3862 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3863 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3864 tpacket_rcv : packet_rcv;
3865 skb_queue_purge(rb_queue);
3866 if (atomic_read(&po->mapped))
3867 pr_err("packet_mmap: vma is busy: %d\n",
3868 atomic_read(&po->mapped));
3870 mutex_unlock(&po->pg_vec_lock);
3872 spin_lock(&po->bind_lock);
3875 register_prot_hook(sk);
3877 spin_unlock(&po->bind_lock);
3878 if (closing && (po->tp_version > TPACKET_V2)) {
3879 /* Because we don't support block-based V3 on tx-ring */
3881 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3886 free_pg_vec(pg_vec, order, req->tp_block_nr);
3891 static int packet_mmap(struct file *file, struct socket *sock,
3892 struct vm_area_struct *vma)
3894 struct sock *sk = sock->sk;
3895 struct packet_sock *po = pkt_sk(sk);
3896 unsigned long size, expected_size;
3897 struct packet_ring_buffer *rb;
3898 unsigned long start;
3905 mutex_lock(&po->pg_vec_lock);
3908 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3910 expected_size += rb->pg_vec_len
3916 if (expected_size == 0)
3919 size = vma->vm_end - vma->vm_start;
3920 if (size != expected_size)
3923 start = vma->vm_start;
3924 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3925 if (rb->pg_vec == NULL)
3928 for (i = 0; i < rb->pg_vec_len; i++) {
3930 void *kaddr = rb->pg_vec[i].buffer;
3933 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3934 page = pgv_to_page(kaddr);
3935 err = vm_insert_page(vma, start, page);
3944 atomic_inc(&po->mapped);
3945 vma->vm_ops = &packet_mmap_ops;
3949 mutex_unlock(&po->pg_vec_lock);
3953 static const struct proto_ops packet_ops_spkt = {
3954 .family = PF_PACKET,
3955 .owner = THIS_MODULE,
3956 .release = packet_release,
3957 .bind = packet_bind_spkt,
3958 .connect = sock_no_connect,
3959 .socketpair = sock_no_socketpair,
3960 .accept = sock_no_accept,
3961 .getname = packet_getname_spkt,
3962 .poll = datagram_poll,
3963 .ioctl = packet_ioctl,
3964 .listen = sock_no_listen,
3965 .shutdown = sock_no_shutdown,
3966 .setsockopt = sock_no_setsockopt,
3967 .getsockopt = sock_no_getsockopt,
3968 .sendmsg = packet_sendmsg_spkt,
3969 .recvmsg = packet_recvmsg,
3970 .mmap = sock_no_mmap,
3971 .sendpage = sock_no_sendpage,
3974 static const struct proto_ops packet_ops = {
3975 .family = PF_PACKET,
3976 .owner = THIS_MODULE,
3977 .release = packet_release,
3978 .bind = packet_bind,
3979 .connect = sock_no_connect,
3980 .socketpair = sock_no_socketpair,
3981 .accept = sock_no_accept,
3982 .getname = packet_getname,
3983 .poll = packet_poll,
3984 .ioctl = packet_ioctl,
3985 .listen = sock_no_listen,
3986 .shutdown = sock_no_shutdown,
3987 .setsockopt = packet_setsockopt,
3988 .getsockopt = packet_getsockopt,
3989 .sendmsg = packet_sendmsg,
3990 .recvmsg = packet_recvmsg,
3991 .mmap = packet_mmap,
3992 .sendpage = sock_no_sendpage,
3995 static const struct net_proto_family packet_family_ops = {
3996 .family = PF_PACKET,
3997 .create = packet_create,
3998 .owner = THIS_MODULE,
4001 static struct notifier_block packet_netdev_notifier = {
4002 .notifier_call = packet_notifier,
4005 #ifdef CONFIG_PROC_FS
4007 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4010 struct net *net = seq_file_net(seq);
4013 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4016 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4018 struct net *net = seq_file_net(seq);
4019 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4022 static void packet_seq_stop(struct seq_file *seq, void *v)
4028 static int packet_seq_show(struct seq_file *seq, void *v)
4030 if (v == SEQ_START_TOKEN)
4031 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4033 struct sock *s = sk_entry(v);
4034 const struct packet_sock *po = pkt_sk(s);
4037 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4039 atomic_read(&s->sk_refcnt),
4044 atomic_read(&s->sk_rmem_alloc),
4045 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4052 static const struct seq_operations packet_seq_ops = {
4053 .start = packet_seq_start,
4054 .next = packet_seq_next,
4055 .stop = packet_seq_stop,
4056 .show = packet_seq_show,
4059 static int packet_seq_open(struct inode *inode, struct file *file)
4061 return seq_open_net(inode, file, &packet_seq_ops,
4062 sizeof(struct seq_net_private));
4065 static const struct file_operations packet_seq_fops = {
4066 .owner = THIS_MODULE,
4067 .open = packet_seq_open,
4069 .llseek = seq_lseek,
4070 .release = seq_release_net,
4075 static int __net_init packet_net_init(struct net *net)
4077 mutex_init(&net->packet.sklist_lock);
4078 INIT_HLIST_HEAD(&net->packet.sklist);
4080 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4086 static void __net_exit packet_net_exit(struct net *net)
4088 remove_proc_entry("packet", net->proc_net);
4091 static struct pernet_operations packet_net_ops = {
4092 .init = packet_net_init,
4093 .exit = packet_net_exit,
4097 static void __exit packet_exit(void)
4099 unregister_netdevice_notifier(&packet_netdev_notifier);
4100 unregister_pernet_subsys(&packet_net_ops);
4101 sock_unregister(PF_PACKET);
4102 proto_unregister(&packet_proto);
4105 static int __init packet_init(void)
4107 int rc = proto_register(&packet_proto, 0);
4112 sock_register(&packet_family_ops);
4113 register_pernet_subsys(&packet_net_ops);
4114 register_netdevice_notifier(&packet_netdev_notifier);
4119 module_init(packet_init);
4120 module_exit(packet_exit);
4121 MODULE_LICENSE("GPL");
4122 MODULE_ALIAS_NETPROTO(PF_PACKET);