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/delay.h>
67 #include <linux/kmod.h>
68 #include <linux/slab.h>
69 #include <linux/vmalloc.h>
70 #include <net/net_namespace.h>
72 #include <net/protocol.h>
73 #include <linux/skbuff.h>
75 #include <linux/errno.h>
76 #include <linux/timer.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
80 #include <asm/cacheflush.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
92 #include <linux/percpu.h>
94 #include <net/inet_common.h>
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
113 Incoming, dev->hard_header!=NULL
114 mac_header -> ll header
117 Outgoing, dev->hard_header!=NULL
118 mac_header -> ll header
121 Incoming, dev->hard_header==NULL
122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
124 assymetry between rx and tx paths.
127 Outgoing, dev->hard_header==NULL
128 mac_header -> data. ll header is still not built!
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
138 dev->hard_header != NULL
139 mac_header -> ll header
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
150 /* Private packet socket structures. */
152 /* identical to struct packet_mreq except it has
153 * a longer address field.
155 struct packet_mreq_max {
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 #define PGV_FROM_VMALLOC 1
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
199 struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206 struct tpacket_block_desc *);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209 static void prb_init_blk_timer(struct packet_sock *,
210 struct tpacket_kbdq_core *,
211 void (*func) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 struct tpacket3_hdr *);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 struct tpacket3_hdr *);
217 static void packet_flush_mclist(struct sock *sk);
219 struct packet_skb_cb {
221 struct sockaddr_pkt pkt;
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
227 unsigned int origlen;
228 struct sockaddr_ll ll;
233 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
235 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
236 #define GET_PBLOCK_DESC(x, bid) \
237 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
240 #define GET_NEXT_PRB_BLK_NUM(x) \
241 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
242 ((x)->kactive_blk_num+1) : 0)
244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
245 static void __fanout_link(struct sock *sk, struct packet_sock *po);
247 static int packet_direct_xmit(struct sk_buff *skb)
249 struct net_device *dev = skb->dev;
250 netdev_features_t features;
251 struct netdev_queue *txq;
252 int ret = NETDEV_TX_BUSY;
254 if (unlikely(!netif_running(dev) ||
255 !netif_carrier_ok(dev)))
258 features = netif_skb_features(skb);
259 if (skb_needs_linearize(skb, features) &&
260 __skb_linearize(skb))
263 txq = skb_get_tx_queue(dev, skb);
267 HARD_TX_LOCK(dev, txq, smp_processor_id());
268 if (!netif_xmit_frozen_or_drv_stopped(txq))
269 ret = netdev_start_xmit(skb, dev, txq, false);
270 HARD_TX_UNLOCK(dev, txq);
274 if (!dev_xmit_complete(ret))
279 atomic_long_inc(&dev->tx_dropped);
281 return NET_XMIT_DROP;
284 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
286 struct net_device *dev;
289 dev = rcu_dereference(po->cached_dev);
297 static void packet_cached_dev_assign(struct packet_sock *po,
298 struct net_device *dev)
300 rcu_assign_pointer(po->cached_dev, dev);
303 static void packet_cached_dev_reset(struct packet_sock *po)
305 RCU_INIT_POINTER(po->cached_dev, NULL);
308 static bool packet_use_direct_xmit(const struct packet_sock *po)
310 return po->xmit == packet_direct_xmit;
313 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
315 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
318 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
320 const struct net_device_ops *ops = dev->netdev_ops;
323 if (ops->ndo_select_queue) {
324 queue_index = ops->ndo_select_queue(dev, skb, NULL,
325 __packet_pick_tx_queue);
326 queue_index = netdev_cap_txqueue(dev, queue_index);
328 queue_index = __packet_pick_tx_queue(dev, skb);
331 skb_set_queue_mapping(skb, queue_index);
334 /* register_prot_hook must be invoked with the po->bind_lock held,
335 * or from a context in which asynchronous accesses to the packet
336 * socket is not possible (packet_create()).
338 static void register_prot_hook(struct sock *sk)
340 struct packet_sock *po = pkt_sk(sk);
344 __fanout_link(sk, po);
346 dev_add_pack(&po->prot_hook);
353 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
354 * held. If the sync parameter is true, we will temporarily drop
355 * the po->bind_lock and do a synchronize_net to make sure no
356 * asynchronous packet processing paths still refer to the elements
357 * of po->prot_hook. If the sync parameter is false, it is the
358 * callers responsibility to take care of this.
360 static void __unregister_prot_hook(struct sock *sk, bool sync)
362 struct packet_sock *po = pkt_sk(sk);
367 __fanout_unlink(sk, po);
369 __dev_remove_pack(&po->prot_hook);
374 spin_unlock(&po->bind_lock);
376 spin_lock(&po->bind_lock);
380 static void unregister_prot_hook(struct sock *sk, bool sync)
382 struct packet_sock *po = pkt_sk(sk);
385 __unregister_prot_hook(sk, sync);
388 static inline struct page * __pure pgv_to_page(void *addr)
390 if (is_vmalloc_addr(addr))
391 return vmalloc_to_page(addr);
392 return virt_to_page(addr);
395 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
397 union tpacket_uhdr h;
400 switch (po->tp_version) {
402 h.h1->tp_status = status;
403 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
406 h.h2->tp_status = status;
407 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
411 WARN(1, "TPACKET version not supported.\n");
418 static int __packet_get_status(struct packet_sock *po, void *frame)
420 union tpacket_uhdr h;
425 switch (po->tp_version) {
427 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
428 return h.h1->tp_status;
430 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
431 return h.h2->tp_status;
434 WARN(1, "TPACKET version not supported.\n");
440 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
443 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
446 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
447 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
448 return TP_STATUS_TS_RAW_HARDWARE;
450 if (ktime_to_timespec_cond(skb->tstamp, ts))
451 return TP_STATUS_TS_SOFTWARE;
456 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
459 union tpacket_uhdr h;
463 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
467 switch (po->tp_version) {
469 h.h1->tp_sec = ts.tv_sec;
470 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
473 h.h2->tp_sec = ts.tv_sec;
474 h.h2->tp_nsec = ts.tv_nsec;
478 WARN(1, "TPACKET version not supported.\n");
482 /* one flush is safe, as both fields always lie on the same cacheline */
483 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
489 static void *packet_lookup_frame(struct packet_sock *po,
490 struct packet_ring_buffer *rb,
491 unsigned int position,
494 unsigned int pg_vec_pos, frame_offset;
495 union tpacket_uhdr h;
497 pg_vec_pos = position / rb->frames_per_block;
498 frame_offset = position % rb->frames_per_block;
500 h.raw = rb->pg_vec[pg_vec_pos].buffer +
501 (frame_offset * rb->frame_size);
503 if (status != __packet_get_status(po, h.raw))
509 static void *packet_current_frame(struct packet_sock *po,
510 struct packet_ring_buffer *rb,
513 return packet_lookup_frame(po, rb, rb->head, status);
516 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
518 del_timer_sync(&pkc->retire_blk_timer);
521 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
523 struct sk_buff_head *rb_queue)
525 struct tpacket_kbdq_core *pkc;
527 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
528 GET_PBDQC_FROM_RB(&po->rx_ring);
530 spin_lock_bh(&rb_queue->lock);
531 pkc->delete_blk_timer = 1;
532 spin_unlock_bh(&rb_queue->lock);
534 prb_del_retire_blk_timer(pkc);
537 static void prb_init_blk_timer(struct packet_sock *po,
538 struct tpacket_kbdq_core *pkc,
539 void (*func) (unsigned long))
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
547 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
549 struct tpacket_kbdq_core *pkc;
554 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
555 GET_PBDQC_FROM_RB(&po->rx_ring);
556 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
559 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
560 int blk_size_in_bytes)
562 struct net_device *dev;
563 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
564 struct ethtool_cmd ecmd;
569 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
570 if (unlikely(!dev)) {
572 return DEFAULT_PRB_RETIRE_TOV;
574 err = __ethtool_get_settings(dev, &ecmd);
575 speed = ethtool_cmd_speed(&ecmd);
579 * If the link speed is so slow you don't really
580 * need to worry about perf anyways
582 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
583 return DEFAULT_PRB_RETIRE_TOV;
590 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
602 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
603 union tpacket_req_u *req_u)
605 p1->feature_req_word = req_u->req3.tp_feature_req_word;
608 static void init_prb_bdqc(struct packet_sock *po,
609 struct packet_ring_buffer *rb,
611 union tpacket_req_u *req_u, int tx_ring)
613 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
614 struct tpacket_block_desc *pbd;
616 memset(p1, 0x0, sizeof(*p1));
618 p1->knxt_seq_num = 1;
620 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
621 p1->pkblk_start = pg_vec[0].buffer;
622 p1->kblk_size = req_u->req3.tp_block_size;
623 p1->knum_blocks = req_u->req3.tp_block_nr;
624 p1->hdrlen = po->tp_hdrlen;
625 p1->version = po->tp_version;
626 p1->last_kactive_blk_num = 0;
627 po->stats.stats3.tp_freeze_q_cnt = 0;
628 if (req_u->req3.tp_retire_blk_tov)
629 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
631 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
632 req_u->req3.tp_block_size);
633 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
634 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
636 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
637 prb_init_ft_ops(p1, req_u);
638 prb_setup_retire_blk_timer(po, tx_ring);
639 prb_open_block(p1, pbd);
642 /* Do NOT update the last_blk_num first.
643 * Assumes sk_buff_head lock is held.
645 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
647 mod_timer(&pkc->retire_blk_timer,
648 jiffies + pkc->tov_in_jiffies);
649 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
654 * 1) We refresh the timer only when we open a block.
655 * By doing this we don't waste cycles refreshing the timer
656 * on packet-by-packet basis.
658 * With a 1MB block-size, on a 1Gbps line, it will take
659 * i) ~8 ms to fill a block + ii) memcpy etc.
660 * In this cut we are not accounting for the memcpy time.
662 * So, if the user sets the 'tmo' to 10ms then the timer
663 * will never fire while the block is still getting filled
664 * (which is what we want). However, the user could choose
665 * to close a block early and that's fine.
667 * But when the timer does fire, we check whether or not to refresh it.
668 * Since the tmo granularity is in msecs, it is not too expensive
669 * to refresh the timer, lets say every '8' msecs.
670 * Either the user can set the 'tmo' or we can derive it based on
671 * a) line-speed and b) block-size.
672 * prb_calc_retire_blk_tmo() calculates the tmo.
675 static void prb_retire_rx_blk_timer_expired(unsigned long data)
677 struct packet_sock *po = (struct packet_sock *)data;
678 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
680 struct tpacket_block_desc *pbd;
682 spin_lock(&po->sk.sk_receive_queue.lock);
684 frozen = prb_queue_frozen(pkc);
685 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
687 if (unlikely(pkc->delete_blk_timer))
690 /* We only need to plug the race when the block is partially filled.
692 * lock(); increment BLOCK_NUM_PKTS; unlock()
693 * copy_bits() is in progress ...
694 * timer fires on other cpu:
695 * we can't retire the current block because copy_bits
699 if (BLOCK_NUM_PKTS(pbd)) {
700 while (atomic_read(&pkc->blk_fill_in_prog)) {
701 /* Waiting for skb_copy_bits to finish... */
706 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
708 if (!BLOCK_NUM_PKTS(pbd)) {
709 /* An empty block. Just refresh the timer. */
712 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
713 if (!prb_dispatch_next_block(pkc, po))
718 /* Case 1. Queue was frozen because user-space was
721 if (prb_curr_blk_in_use(pkc, pbd)) {
723 * Ok, user-space is still behind.
724 * So just refresh the timer.
728 /* Case 2. queue was frozen,user-space caught up,
729 * now the link went idle && the timer fired.
730 * We don't have a block to close.So we open this
731 * block and restart the timer.
732 * opening a block thaws the queue,restarts timer
733 * Thawing/timer-refresh is a side effect.
735 prb_open_block(pkc, pbd);
742 _prb_refresh_rx_retire_blk_timer(pkc);
745 spin_unlock(&po->sk.sk_receive_queue.lock);
748 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
749 struct tpacket_block_desc *pbd1, __u32 status)
751 /* Flush everything minus the block header */
753 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
758 /* Skip the block header(we know header WILL fit in 4K) */
761 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
762 for (; start < end; start += PAGE_SIZE)
763 flush_dcache_page(pgv_to_page(start));
768 /* Now update the block status. */
770 BLOCK_STATUS(pbd1) = status;
772 /* Flush the block header */
774 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
776 flush_dcache_page(pgv_to_page(start));
786 * 2) Increment active_blk_num
788 * Note:We DONT refresh the timer on purpose.
789 * Because almost always the next block will be opened.
791 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
792 struct tpacket_block_desc *pbd1,
793 struct packet_sock *po, unsigned int stat)
795 __u32 status = TP_STATUS_USER | stat;
797 struct tpacket3_hdr *last_pkt;
798 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
799 struct sock *sk = &po->sk;
801 if (po->stats.stats3.tp_drops)
802 status |= TP_STATUS_LOSING;
804 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
805 last_pkt->tp_next_offset = 0;
807 /* Get the ts of the last pkt */
808 if (BLOCK_NUM_PKTS(pbd1)) {
809 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
810 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
812 /* Ok, we tmo'd - so get the current time.
814 * It shouldn't really happen as we don't close empty
815 * blocks. See prb_retire_rx_blk_timer_expired().
819 h1->ts_last_pkt.ts_sec = ts.tv_sec;
820 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
825 /* Flush the block */
826 prb_flush_block(pkc1, pbd1, status);
828 sk->sk_data_ready(sk);
830 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
833 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
835 pkc->reset_pending_on_curr_blk = 0;
839 * Side effect of opening a block:
841 * 1) prb_queue is thawed.
842 * 2) retire_blk_timer is refreshed.
845 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
846 struct tpacket_block_desc *pbd1)
849 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
853 /* We could have just memset this but we will lose the
854 * flexibility of making the priv area sticky
857 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
858 BLOCK_NUM_PKTS(pbd1) = 0;
859 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863 h1->ts_first_pkt.ts_sec = ts.tv_sec;
864 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
866 pkc1->pkblk_start = (char *)pbd1;
867 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
870 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
872 pbd1->version = pkc1->version;
873 pkc1->prev = pkc1->nxt_offset;
874 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
876 prb_thaw_queue(pkc1);
877 _prb_refresh_rx_retire_blk_timer(pkc1);
883 * Queue freeze logic:
884 * 1) Assume tp_block_nr = 8 blocks.
885 * 2) At time 't0', user opens Rx ring.
886 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
887 * 4) user-space is either sleeping or processing block '0'.
888 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
889 * it will close block-7,loop around and try to fill block '0'.
891 * __packet_lookup_frame_in_block
892 * prb_retire_current_block()
893 * prb_dispatch_next_block()
894 * |->(BLOCK_STATUS == USER) evaluates to true
895 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
896 * 6) Now there are two cases:
897 * 6.1) Link goes idle right after the queue is frozen.
898 * But remember, the last open_block() refreshed the timer.
899 * When this timer expires,it will refresh itself so that we can
900 * re-open block-0 in near future.
901 * 6.2) Link is busy and keeps on receiving packets. This is a simple
902 * case and __packet_lookup_frame_in_block will check if block-0
903 * is free and can now be re-used.
905 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
906 struct packet_sock *po)
908 pkc->reset_pending_on_curr_blk = 1;
909 po->stats.stats3.tp_freeze_q_cnt++;
912 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
915 * If the next block is free then we will dispatch it
916 * and return a good offset.
917 * Else, we will freeze the queue.
918 * So, caller must check the return value.
920 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
921 struct packet_sock *po)
923 struct tpacket_block_desc *pbd;
927 /* 1. Get current block num */
928 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
930 /* 2. If this block is currently in_use then freeze the queue */
931 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
932 prb_freeze_queue(pkc, po);
938 * open this block and return the offset where the first packet
939 * needs to get stored.
941 prb_open_block(pkc, pbd);
942 return (void *)pkc->nxt_offset;
945 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
946 struct packet_sock *po, unsigned int status)
948 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
950 /* retire/close the current block */
951 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
953 * Plug the case where copy_bits() is in progress on
954 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
955 * have space to copy the pkt in the current block and
956 * called prb_retire_current_block()
958 * We don't need to worry about the TMO case because
959 * the timer-handler already handled this case.
961 if (!(status & TP_STATUS_BLK_TMO)) {
962 while (atomic_read(&pkc->blk_fill_in_prog)) {
963 /* Waiting for skb_copy_bits to finish... */
967 prb_close_block(pkc, pbd, po, status);
972 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
973 struct tpacket_block_desc *pbd)
975 return TP_STATUS_USER & BLOCK_STATUS(pbd);
978 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
980 return pkc->reset_pending_on_curr_blk;
983 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
985 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
986 atomic_dec(&pkc->blk_fill_in_prog);
989 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
990 struct tpacket3_hdr *ppd)
992 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
995 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
996 struct tpacket3_hdr *ppd)
998 ppd->hv1.tp_rxhash = 0;
1001 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1002 struct tpacket3_hdr *ppd)
1004 if (skb_vlan_tag_present(pkc->skb)) {
1005 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1006 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1007 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1009 ppd->hv1.tp_vlan_tci = 0;
1010 ppd->hv1.tp_vlan_tpid = 0;
1011 ppd->tp_status = TP_STATUS_AVAILABLE;
1015 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1016 struct tpacket3_hdr *ppd)
1018 ppd->hv1.tp_padding = 0;
1019 prb_fill_vlan_info(pkc, ppd);
1021 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1022 prb_fill_rxhash(pkc, ppd);
1024 prb_clear_rxhash(pkc, ppd);
1027 static void prb_fill_curr_block(char *curr,
1028 struct tpacket_kbdq_core *pkc,
1029 struct tpacket_block_desc *pbd,
1032 struct tpacket3_hdr *ppd;
1034 ppd = (struct tpacket3_hdr *)curr;
1035 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1037 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1039 BLOCK_NUM_PKTS(pbd) += 1;
1040 atomic_inc(&pkc->blk_fill_in_prog);
1041 prb_run_all_ft_ops(pkc, ppd);
1044 /* Assumes caller has the sk->rx_queue.lock */
1045 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1046 struct sk_buff *skb,
1051 struct tpacket_kbdq_core *pkc;
1052 struct tpacket_block_desc *pbd;
1055 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1056 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058 /* Queue is frozen when user space is lagging behind */
1059 if (prb_queue_frozen(pkc)) {
1061 * Check if that last block which caused the queue to freeze,
1062 * is still in_use by user-space.
1064 if (prb_curr_blk_in_use(pkc, pbd)) {
1065 /* Can't record this packet */
1069 * Ok, the block was released by user-space.
1070 * Now let's open that block.
1071 * opening a block also thaws the queue.
1072 * Thawing is a side effect.
1074 prb_open_block(pkc, pbd);
1079 curr = pkc->nxt_offset;
1081 end = (char *)pbd + pkc->kblk_size;
1083 /* first try the current block */
1084 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1085 prb_fill_curr_block(curr, pkc, pbd, len);
1086 return (void *)curr;
1089 /* Ok, close the current block */
1090 prb_retire_current_block(pkc, po, 0);
1092 /* Now, try to dispatch the next block */
1093 curr = (char *)prb_dispatch_next_block(pkc, po);
1095 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1096 prb_fill_curr_block(curr, pkc, pbd, len);
1097 return (void *)curr;
1101 * No free blocks are available.user_space hasn't caught up yet.
1102 * Queue was just frozen and now this packet will get dropped.
1107 static void *packet_current_rx_frame(struct packet_sock *po,
1108 struct sk_buff *skb,
1109 int status, unsigned int len)
1112 switch (po->tp_version) {
1115 curr = packet_lookup_frame(po, &po->rx_ring,
1116 po->rx_ring.head, status);
1119 return __packet_lookup_frame_in_block(po, skb, status, len);
1121 WARN(1, "TPACKET version not supported\n");
1127 static void *prb_lookup_block(struct packet_sock *po,
1128 struct packet_ring_buffer *rb,
1132 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1133 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1135 if (status != BLOCK_STATUS(pbd))
1140 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1143 if (rb->prb_bdqc.kactive_blk_num)
1144 prev = rb->prb_bdqc.kactive_blk_num-1;
1146 prev = rb->prb_bdqc.knum_blocks-1;
1150 /* Assumes caller has held the rx_queue.lock */
1151 static void *__prb_previous_block(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1155 unsigned int previous = prb_previous_blk_num(rb);
1156 return prb_lookup_block(po, rb, previous, status);
1159 static void *packet_previous_rx_frame(struct packet_sock *po,
1160 struct packet_ring_buffer *rb,
1163 if (po->tp_version <= TPACKET_V2)
1164 return packet_previous_frame(po, rb, status);
1166 return __prb_previous_block(po, rb, status);
1169 static void packet_increment_rx_head(struct packet_sock *po,
1170 struct packet_ring_buffer *rb)
1172 switch (po->tp_version) {
1175 return packet_increment_head(rb);
1178 WARN(1, "TPACKET version not supported.\n");
1184 static void *packet_previous_frame(struct packet_sock *po,
1185 struct packet_ring_buffer *rb,
1188 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1189 return packet_lookup_frame(po, rb, previous, status);
1192 static void packet_increment_head(struct packet_ring_buffer *buff)
1194 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1197 static void packet_inc_pending(struct packet_ring_buffer *rb)
1199 this_cpu_inc(*rb->pending_refcnt);
1202 static void packet_dec_pending(struct packet_ring_buffer *rb)
1204 this_cpu_dec(*rb->pending_refcnt);
1207 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1209 unsigned int refcnt = 0;
1212 /* We don't use pending refcount in rx_ring. */
1213 if (rb->pending_refcnt == NULL)
1216 for_each_possible_cpu(cpu)
1217 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1222 static int packet_alloc_pending(struct packet_sock *po)
1224 po->rx_ring.pending_refcnt = NULL;
1226 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1227 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1233 static void packet_free_pending(struct packet_sock *po)
1235 free_percpu(po->tx_ring.pending_refcnt);
1238 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1240 struct sock *sk = &po->sk;
1243 if (po->prot_hook.func != tpacket_rcv)
1244 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1247 spin_lock(&sk->sk_receive_queue.lock);
1248 if (po->tp_version == TPACKET_V3)
1249 has_room = prb_lookup_block(po, &po->rx_ring,
1250 po->rx_ring.prb_bdqc.kactive_blk_num,
1253 has_room = packet_lookup_frame(po, &po->rx_ring,
1256 spin_unlock(&sk->sk_receive_queue.lock);
1261 static void packet_sock_destruct(struct sock *sk)
1263 skb_queue_purge(&sk->sk_error_queue);
1265 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1266 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1268 if (!sock_flag(sk, SOCK_DEAD)) {
1269 pr_err("Attempt to release alive packet socket: %p\n", sk);
1273 sk_refcnt_debug_dec(sk);
1276 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1277 struct sk_buff *skb,
1280 return reciprocal_scale(skb_get_hash(skb), num);
1283 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1284 struct sk_buff *skb,
1287 unsigned int val = atomic_inc_return(&f->rr_cur);
1292 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1293 struct sk_buff *skb,
1296 return smp_processor_id() % num;
1299 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1300 struct sk_buff *skb,
1303 return prandom_u32_max(num);
1306 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1307 struct sk_buff *skb,
1308 unsigned int idx, unsigned int skip,
1313 i = j = min_t(int, f->next[idx], num - 1);
1315 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1327 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1328 struct sk_buff *skb,
1331 return skb_get_queue_mapping(skb) % num;
1334 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1336 return f->flags & (flag >> 8);
1339 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1340 struct packet_type *pt, struct net_device *orig_dev)
1342 struct packet_fanout *f = pt->af_packet_priv;
1343 unsigned int num = READ_ONCE(f->num_members);
1344 struct packet_sock *po;
1347 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1353 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1354 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1359 case PACKET_FANOUT_HASH:
1361 idx = fanout_demux_hash(f, skb, num);
1363 case PACKET_FANOUT_LB:
1364 idx = fanout_demux_lb(f, skb, num);
1366 case PACKET_FANOUT_CPU:
1367 idx = fanout_demux_cpu(f, skb, num);
1369 case PACKET_FANOUT_RND:
1370 idx = fanout_demux_rnd(f, skb, num);
1372 case PACKET_FANOUT_QM:
1373 idx = fanout_demux_qm(f, skb, num);
1375 case PACKET_FANOUT_ROLLOVER:
1376 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1380 po = pkt_sk(f->arr[idx]);
1381 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1382 unlikely(!packet_rcv_has_room(po, skb))) {
1383 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1384 po = pkt_sk(f->arr[idx]);
1387 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1390 DEFINE_MUTEX(fanout_mutex);
1391 EXPORT_SYMBOL_GPL(fanout_mutex);
1392 static LIST_HEAD(fanout_list);
1394 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1396 struct packet_fanout *f = po->fanout;
1398 spin_lock(&f->lock);
1399 f->arr[f->num_members] = sk;
1402 spin_unlock(&f->lock);
1405 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1407 struct packet_fanout *f = po->fanout;
1410 spin_lock(&f->lock);
1411 for (i = 0; i < f->num_members; i++) {
1412 if (f->arr[i] == sk)
1415 BUG_ON(i >= f->num_members);
1416 f->arr[i] = f->arr[f->num_members - 1];
1418 spin_unlock(&f->lock);
1421 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1423 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1429 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1431 struct packet_sock *po = pkt_sk(sk);
1432 struct packet_fanout *f, *match;
1433 u8 type = type_flags & 0xff;
1434 u8 flags = type_flags >> 8;
1438 case PACKET_FANOUT_ROLLOVER:
1439 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1441 case PACKET_FANOUT_HASH:
1442 case PACKET_FANOUT_LB:
1443 case PACKET_FANOUT_CPU:
1444 case PACKET_FANOUT_RND:
1445 case PACKET_FANOUT_QM:
1457 mutex_lock(&fanout_mutex);
1459 list_for_each_entry(f, &fanout_list, list) {
1461 read_pnet(&f->net) == sock_net(sk)) {
1467 if (match && match->flags != flags)
1471 match = kzalloc(sizeof(*match), GFP_KERNEL);
1474 write_pnet(&match->net, sock_net(sk));
1477 match->flags = flags;
1478 atomic_set(&match->rr_cur, 0);
1479 INIT_LIST_HEAD(&match->list);
1480 spin_lock_init(&match->lock);
1481 atomic_set(&match->sk_ref, 0);
1482 match->prot_hook.type = po->prot_hook.type;
1483 match->prot_hook.dev = po->prot_hook.dev;
1484 match->prot_hook.func = packet_rcv_fanout;
1485 match->prot_hook.af_packet_priv = match;
1486 match->prot_hook.id_match = match_fanout_group;
1487 dev_add_pack(&match->prot_hook);
1488 list_add(&match->list, &fanout_list);
1491 if (match->type == type &&
1492 match->prot_hook.type == po->prot_hook.type &&
1493 match->prot_hook.dev == po->prot_hook.dev) {
1495 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1496 __dev_remove_pack(&po->prot_hook);
1498 atomic_inc(&match->sk_ref);
1499 __fanout_link(sk, po);
1504 mutex_unlock(&fanout_mutex);
1508 static void fanout_release(struct sock *sk)
1510 struct packet_sock *po = pkt_sk(sk);
1511 struct packet_fanout *f;
1517 mutex_lock(&fanout_mutex);
1520 if (atomic_dec_and_test(&f->sk_ref)) {
1522 dev_remove_pack(&f->prot_hook);
1525 mutex_unlock(&fanout_mutex);
1528 static const struct proto_ops packet_ops;
1530 static const struct proto_ops packet_ops_spkt;
1532 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1533 struct packet_type *pt, struct net_device *orig_dev)
1536 struct sockaddr_pkt *spkt;
1539 * When we registered the protocol we saved the socket in the data
1540 * field for just this event.
1543 sk = pt->af_packet_priv;
1546 * Yank back the headers [hope the device set this
1547 * right or kerboom...]
1549 * Incoming packets have ll header pulled,
1552 * For outgoing ones skb->data == skb_mac_header(skb)
1553 * so that this procedure is noop.
1556 if (skb->pkt_type == PACKET_LOOPBACK)
1559 if (!net_eq(dev_net(dev), sock_net(sk)))
1562 skb = skb_share_check(skb, GFP_ATOMIC);
1566 /* drop any routing info */
1569 /* drop conntrack reference */
1572 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1574 skb_push(skb, skb->data - skb_mac_header(skb));
1577 * The SOCK_PACKET socket receives _all_ frames.
1580 spkt->spkt_family = dev->type;
1581 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1582 spkt->spkt_protocol = skb->protocol;
1585 * Charge the memory to the socket. This is done specifically
1586 * to prevent sockets using all the memory up.
1589 if (sock_queue_rcv_skb(sk, skb) == 0)
1600 * Output a raw packet to a device layer. This bypasses all the other
1601 * protocol layers and you must therefore supply it with a complete frame
1604 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1607 struct sock *sk = sock->sk;
1608 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1609 struct sk_buff *skb = NULL;
1610 struct net_device *dev;
1616 * Get and verify the address.
1620 if (msg->msg_namelen < sizeof(struct sockaddr))
1622 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1623 proto = saddr->spkt_protocol;
1625 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1628 * Find the device first to size check it
1631 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1634 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1640 if (!(dev->flags & IFF_UP))
1644 * You may not queue a frame bigger than the mtu. This is the lowest level
1645 * raw protocol and you must do your own fragmentation at this level.
1648 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1649 if (!netif_supports_nofcs(dev)) {
1650 err = -EPROTONOSUPPORT;
1653 extra_len = 4; /* We're doing our own CRC */
1657 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1661 size_t reserved = LL_RESERVED_SPACE(dev);
1662 int tlen = dev->needed_tailroom;
1663 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1666 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1669 /* FIXME: Save some space for broken drivers that write a hard
1670 * header at transmission time by themselves. PPP is the notable
1671 * one here. This should really be fixed at the driver level.
1673 skb_reserve(skb, reserved);
1674 skb_reset_network_header(skb);
1676 /* Try to align data part correctly */
1681 skb_reset_network_header(skb);
1683 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1689 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1690 /* Earlier code assumed this would be a VLAN pkt,
1691 * double-check this now that we have the actual
1694 struct ethhdr *ehdr;
1695 skb_reset_mac_header(skb);
1696 ehdr = eth_hdr(skb);
1697 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1703 skb->protocol = proto;
1705 skb->priority = sk->sk_priority;
1706 skb->mark = sk->sk_mark;
1708 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1710 if (unlikely(extra_len == 4))
1713 skb_probe_transport_header(skb, 0);
1715 dev_queue_xmit(skb);
1726 static unsigned int run_filter(const struct sk_buff *skb,
1727 const struct sock *sk,
1730 struct sk_filter *filter;
1733 filter = rcu_dereference(sk->sk_filter);
1735 res = SK_RUN_FILTER(filter, skb);
1742 * This function makes lazy skb cloning in hope that most of packets
1743 * are discarded by BPF.
1745 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1746 * and skb->cb are mangled. It works because (and until) packets
1747 * falling here are owned by current CPU. Output packets are cloned
1748 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1749 * sequencially, so that if we return skb to original state on exit,
1750 * we will not harm anyone.
1753 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1754 struct packet_type *pt, struct net_device *orig_dev)
1757 struct sockaddr_ll *sll;
1758 struct packet_sock *po;
1759 u8 *skb_head = skb->data;
1760 int skb_len = skb->len;
1761 unsigned int snaplen, res;
1763 if (skb->pkt_type == PACKET_LOOPBACK)
1766 sk = pt->af_packet_priv;
1769 if (!net_eq(dev_net(dev), sock_net(sk)))
1774 if (dev->header_ops) {
1775 /* The device has an explicit notion of ll header,
1776 * exported to higher levels.
1778 * Otherwise, the device hides details of its frame
1779 * structure, so that corresponding packet head is
1780 * never delivered to user.
1782 if (sk->sk_type != SOCK_DGRAM)
1783 skb_push(skb, skb->data - skb_mac_header(skb));
1784 else if (skb->pkt_type == PACKET_OUTGOING) {
1785 /* Special case: outgoing packets have ll header at head */
1786 skb_pull(skb, skb_network_offset(skb));
1792 res = run_filter(skb, sk, snaplen);
1794 goto drop_n_restore;
1798 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1801 if (skb_shared(skb)) {
1802 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1806 if (skb_head != skb->data) {
1807 skb->data = skb_head;
1814 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1816 sll = &PACKET_SKB_CB(skb)->sa.ll;
1817 sll->sll_hatype = dev->type;
1818 sll->sll_pkttype = skb->pkt_type;
1819 if (unlikely(po->origdev))
1820 sll->sll_ifindex = orig_dev->ifindex;
1822 sll->sll_ifindex = dev->ifindex;
1824 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1826 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1827 * Use their space for storing the original skb length.
1829 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1831 if (pskb_trim(skb, snaplen))
1834 skb_set_owner_r(skb, sk);
1838 /* drop conntrack reference */
1841 spin_lock(&sk->sk_receive_queue.lock);
1842 po->stats.stats1.tp_packets++;
1843 sock_skb_set_dropcount(sk, skb);
1844 __skb_queue_tail(&sk->sk_receive_queue, skb);
1845 spin_unlock(&sk->sk_receive_queue.lock);
1846 sk->sk_data_ready(sk);
1850 spin_lock(&sk->sk_receive_queue.lock);
1851 po->stats.stats1.tp_drops++;
1852 atomic_inc(&sk->sk_drops);
1853 spin_unlock(&sk->sk_receive_queue.lock);
1856 if (skb_head != skb->data && skb_shared(skb)) {
1857 skb->data = skb_head;
1865 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1866 struct packet_type *pt, struct net_device *orig_dev)
1869 struct packet_sock *po;
1870 struct sockaddr_ll *sll;
1871 union tpacket_uhdr h;
1872 u8 *skb_head = skb->data;
1873 int skb_len = skb->len;
1874 unsigned int snaplen, res;
1875 unsigned long status = TP_STATUS_USER;
1876 unsigned short macoff, netoff, hdrlen;
1877 struct sk_buff *copy_skb = NULL;
1881 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1882 * We may add members to them until current aligned size without forcing
1883 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1885 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1886 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1888 if (skb->pkt_type == PACKET_LOOPBACK)
1891 sk = pt->af_packet_priv;
1894 if (!net_eq(dev_net(dev), sock_net(sk)))
1897 if (dev->header_ops) {
1898 if (sk->sk_type != SOCK_DGRAM)
1899 skb_push(skb, skb->data - skb_mac_header(skb));
1900 else if (skb->pkt_type == PACKET_OUTGOING) {
1901 /* Special case: outgoing packets have ll header at head */
1902 skb_pull(skb, skb_network_offset(skb));
1908 res = run_filter(skb, sk, snaplen);
1910 goto drop_n_restore;
1912 if (skb->ip_summed == CHECKSUM_PARTIAL)
1913 status |= TP_STATUS_CSUMNOTREADY;
1914 else if (skb->pkt_type != PACKET_OUTGOING &&
1915 (skb->ip_summed == CHECKSUM_COMPLETE ||
1916 skb_csum_unnecessary(skb)))
1917 status |= TP_STATUS_CSUM_VALID;
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)
1949 } else if (unlikely(macoff + snaplen >
1950 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1953 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1954 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1955 snaplen, nval, macoff);
1957 if (unlikely((int)snaplen < 0)) {
1959 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1962 spin_lock(&sk->sk_receive_queue.lock);
1963 h.raw = packet_current_rx_frame(po, skb,
1964 TP_STATUS_KERNEL, (macoff+snaplen));
1967 if (po->tp_version <= TPACKET_V2) {
1968 packet_increment_rx_head(po, &po->rx_ring);
1970 * LOSING will be reported till you read the stats,
1971 * because it's COR - Clear On Read.
1972 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1975 if (po->stats.stats1.tp_drops)
1976 status |= TP_STATUS_LOSING;
1978 po->stats.stats1.tp_packets++;
1980 status |= TP_STATUS_COPY;
1981 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1983 spin_unlock(&sk->sk_receive_queue.lock);
1985 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1987 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1988 getnstimeofday(&ts);
1990 status |= ts_status;
1992 switch (po->tp_version) {
1994 h.h1->tp_len = skb->len;
1995 h.h1->tp_snaplen = snaplen;
1996 h.h1->tp_mac = macoff;
1997 h.h1->tp_net = netoff;
1998 h.h1->tp_sec = ts.tv_sec;
1999 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2000 hdrlen = sizeof(*h.h1);
2003 h.h2->tp_len = skb->len;
2004 h.h2->tp_snaplen = snaplen;
2005 h.h2->tp_mac = macoff;
2006 h.h2->tp_net = netoff;
2007 h.h2->tp_sec = ts.tv_sec;
2008 h.h2->tp_nsec = ts.tv_nsec;
2009 if (skb_vlan_tag_present(skb)) {
2010 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2011 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2012 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2014 h.h2->tp_vlan_tci = 0;
2015 h.h2->tp_vlan_tpid = 0;
2017 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2018 hdrlen = sizeof(*h.h2);
2021 /* tp_nxt_offset,vlan are already populated above.
2022 * So DONT clear those fields here
2024 h.h3->tp_status |= status;
2025 h.h3->tp_len = skb->len;
2026 h.h3->tp_snaplen = snaplen;
2027 h.h3->tp_mac = macoff;
2028 h.h3->tp_net = netoff;
2029 h.h3->tp_sec = ts.tv_sec;
2030 h.h3->tp_nsec = ts.tv_nsec;
2031 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2032 hdrlen = sizeof(*h.h3);
2038 sll = h.raw + TPACKET_ALIGN(hdrlen);
2039 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2040 sll->sll_family = AF_PACKET;
2041 sll->sll_hatype = dev->type;
2042 sll->sll_protocol = skb->protocol;
2043 sll->sll_pkttype = skb->pkt_type;
2044 if (unlikely(po->origdev))
2045 sll->sll_ifindex = orig_dev->ifindex;
2047 sll->sll_ifindex = dev->ifindex;
2051 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2052 if (po->tp_version <= TPACKET_V2) {
2055 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2058 for (start = h.raw; start < end; start += PAGE_SIZE)
2059 flush_dcache_page(pgv_to_page(start));
2064 if (po->tp_version <= TPACKET_V2) {
2065 __packet_set_status(po, h.raw, status);
2066 sk->sk_data_ready(sk);
2068 prb_clear_blk_fill_status(&po->rx_ring);
2072 if (skb_head != skb->data && skb_shared(skb)) {
2073 skb->data = skb_head;
2081 po->stats.stats1.tp_drops++;
2082 spin_unlock(&sk->sk_receive_queue.lock);
2084 sk->sk_data_ready(sk);
2085 kfree_skb(copy_skb);
2086 goto drop_n_restore;
2089 static void tpacket_destruct_skb(struct sk_buff *skb)
2091 struct packet_sock *po = pkt_sk(skb->sk);
2093 if (likely(po->tx_ring.pg_vec)) {
2097 ph = skb_shinfo(skb)->destructor_arg;
2098 packet_dec_pending(&po->tx_ring);
2100 ts = __packet_set_timestamp(po, ph, skb);
2101 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2107 static bool ll_header_truncated(const struct net_device *dev, int len)
2109 /* net device doesn't like empty head */
2110 if (unlikely(len <= dev->hard_header_len)) {
2111 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2112 current->comm, len, dev->hard_header_len);
2119 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2120 void *frame, struct net_device *dev, int size_max,
2121 __be16 proto, unsigned char *addr, int hlen)
2123 union tpacket_uhdr ph;
2124 int to_write, offset, len, tp_len, nr_frags, len_max;
2125 struct socket *sock = po->sk.sk_socket;
2132 skb->protocol = proto;
2134 skb->priority = po->sk.sk_priority;
2135 skb->mark = po->sk.sk_mark;
2136 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2137 skb_shinfo(skb)->destructor_arg = ph.raw;
2139 switch (po->tp_version) {
2141 tp_len = ph.h2->tp_len;
2144 tp_len = ph.h1->tp_len;
2147 if (unlikely(tp_len > size_max)) {
2148 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2152 skb_reserve(skb, hlen);
2153 skb_reset_network_header(skb);
2155 if (!packet_use_direct_xmit(po))
2156 skb_probe_transport_header(skb, 0);
2157 if (unlikely(po->tp_tx_has_off)) {
2158 int off_min, off_max, off;
2159 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2160 off_max = po->tx_ring.frame_size - tp_len;
2161 if (sock->type == SOCK_DGRAM) {
2162 switch (po->tp_version) {
2164 off = ph.h2->tp_net;
2167 off = ph.h1->tp_net;
2171 switch (po->tp_version) {
2173 off = ph.h2->tp_mac;
2176 off = ph.h1->tp_mac;
2180 if (unlikely((off < off_min) || (off_max < off)))
2182 data = ph.raw + off;
2184 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2188 if (sock->type == SOCK_DGRAM) {
2189 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2191 if (unlikely(err < 0))
2193 } else if (dev->hard_header_len) {
2194 if (ll_header_truncated(dev, tp_len))
2197 skb_push(skb, dev->hard_header_len);
2198 err = skb_store_bits(skb, 0, data,
2199 dev->hard_header_len);
2203 data += dev->hard_header_len;
2204 to_write -= dev->hard_header_len;
2207 offset = offset_in_page(data);
2208 len_max = PAGE_SIZE - offset;
2209 len = ((to_write > len_max) ? len_max : to_write);
2211 skb->data_len = to_write;
2212 skb->len += to_write;
2213 skb->truesize += to_write;
2214 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2216 while (likely(to_write)) {
2217 nr_frags = skb_shinfo(skb)->nr_frags;
2219 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2220 pr_err("Packet exceed the number of skb frags(%lu)\n",
2225 page = pgv_to_page(data);
2227 flush_dcache_page(page);
2229 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2232 len_max = PAGE_SIZE;
2233 len = ((to_write > len_max) ? len_max : to_write);
2239 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2241 struct sk_buff *skb;
2242 struct net_device *dev;
2244 int err, reserve = 0;
2246 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2247 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2248 int tp_len, size_max;
2249 unsigned char *addr;
2251 int status = TP_STATUS_AVAILABLE;
2254 mutex_lock(&po->pg_vec_lock);
2256 if (likely(saddr == NULL)) {
2257 dev = packet_cached_dev_get(po);
2262 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2264 if (msg->msg_namelen < (saddr->sll_halen
2265 + offsetof(struct sockaddr_ll,
2268 proto = saddr->sll_protocol;
2269 addr = saddr->sll_addr;
2270 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2274 if (unlikely(dev == NULL))
2277 if (unlikely(!(dev->flags & IFF_UP)))
2280 reserve = dev->hard_header_len + VLAN_HLEN;
2281 size_max = po->tx_ring.frame_size
2282 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2284 if (size_max > dev->mtu + reserve)
2285 size_max = dev->mtu + reserve;
2288 ph = packet_current_frame(po, &po->tx_ring,
2289 TP_STATUS_SEND_REQUEST);
2290 if (unlikely(ph == NULL)) {
2291 if (need_wait && need_resched())
2296 status = TP_STATUS_SEND_REQUEST;
2297 hlen = LL_RESERVED_SPACE(dev);
2298 tlen = dev->needed_tailroom;
2299 skb = sock_alloc_send_skb(&po->sk,
2300 hlen + tlen + sizeof(struct sockaddr_ll),
2303 if (unlikely(skb == NULL)) {
2304 /* we assume the socket was initially writeable ... */
2305 if (likely(len_sum > 0))
2309 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2311 if (tp_len > dev->mtu + dev->hard_header_len) {
2312 struct ethhdr *ehdr;
2313 /* Earlier code assumed this would be a VLAN pkt,
2314 * double-check this now that we have the actual
2318 skb_reset_mac_header(skb);
2319 ehdr = eth_hdr(skb);
2320 if (ehdr->h_proto != htons(ETH_P_8021Q))
2323 if (unlikely(tp_len < 0)) {
2325 __packet_set_status(po, ph,
2326 TP_STATUS_AVAILABLE);
2327 packet_increment_head(&po->tx_ring);
2331 status = TP_STATUS_WRONG_FORMAT;
2337 packet_pick_tx_queue(dev, skb);
2339 skb->destructor = tpacket_destruct_skb;
2340 __packet_set_status(po, ph, TP_STATUS_SENDING);
2341 packet_inc_pending(&po->tx_ring);
2343 status = TP_STATUS_SEND_REQUEST;
2344 err = po->xmit(skb);
2345 if (unlikely(err > 0)) {
2346 err = net_xmit_errno(err);
2347 if (err && __packet_get_status(po, ph) ==
2348 TP_STATUS_AVAILABLE) {
2349 /* skb was destructed already */
2354 * skb was dropped but not destructed yet;
2355 * let's treat it like congestion or err < 0
2359 packet_increment_head(&po->tx_ring);
2361 } while (likely((ph != NULL) ||
2362 /* Note: packet_read_pending() might be slow if we have
2363 * to call it as it's per_cpu variable, but in fast-path
2364 * we already short-circuit the loop with the first
2365 * condition, and luckily don't have to go that path
2368 (need_wait && packet_read_pending(&po->tx_ring))));
2374 __packet_set_status(po, ph, status);
2379 mutex_unlock(&po->pg_vec_lock);
2383 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2384 size_t reserve, size_t len,
2385 size_t linear, int noblock,
2388 struct sk_buff *skb;
2390 /* Under a page? Don't bother with paged skb. */
2391 if (prepad + len < PAGE_SIZE || !linear)
2394 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2399 skb_reserve(skb, reserve);
2400 skb_put(skb, linear);
2401 skb->data_len = len - linear;
2402 skb->len += len - linear;
2407 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2409 struct sock *sk = sock->sk;
2410 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2411 struct sk_buff *skb;
2412 struct net_device *dev;
2414 unsigned char *addr;
2415 int err, reserve = 0;
2416 struct virtio_net_hdr vnet_hdr = { 0 };
2419 struct packet_sock *po = pkt_sk(sk);
2420 unsigned short gso_type = 0;
2426 * Get and verify the address.
2429 if (likely(saddr == NULL)) {
2430 dev = packet_cached_dev_get(po);
2435 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2437 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2439 proto = saddr->sll_protocol;
2440 addr = saddr->sll_addr;
2441 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2445 if (unlikely(dev == NULL))
2448 if (unlikely(!(dev->flags & IFF_UP)))
2451 if (sock->type == SOCK_RAW)
2452 reserve = dev->hard_header_len;
2453 if (po->has_vnet_hdr) {
2454 vnet_hdr_len = sizeof(vnet_hdr);
2457 if (len < vnet_hdr_len)
2460 len -= vnet_hdr_len;
2463 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2464 if (n != vnet_hdr_len)
2467 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2468 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2469 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2470 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2471 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2472 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2473 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2476 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2479 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2480 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2481 case VIRTIO_NET_HDR_GSO_TCPV4:
2482 gso_type = SKB_GSO_TCPV4;
2484 case VIRTIO_NET_HDR_GSO_TCPV6:
2485 gso_type = SKB_GSO_TCPV6;
2487 case VIRTIO_NET_HDR_GSO_UDP:
2488 gso_type = SKB_GSO_UDP;
2494 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2495 gso_type |= SKB_GSO_TCP_ECN;
2497 if (vnet_hdr.gso_size == 0)
2503 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2504 if (!netif_supports_nofcs(dev)) {
2505 err = -EPROTONOSUPPORT;
2508 extra_len = 4; /* We're doing our own CRC */
2512 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2516 hlen = LL_RESERVED_SPACE(dev);
2517 tlen = dev->needed_tailroom;
2518 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2519 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2520 msg->msg_flags & MSG_DONTWAIT, &err);
2524 skb_set_network_header(skb, reserve);
2527 if (sock->type == SOCK_DGRAM) {
2528 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2529 if (unlikely(offset < 0))
2532 if (ll_header_truncated(dev, len))
2536 /* Returns -EFAULT on error */
2537 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2541 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2543 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2544 /* Earlier code assumed this would be a VLAN pkt,
2545 * double-check this now that we have the actual
2548 struct ethhdr *ehdr;
2549 skb_reset_mac_header(skb);
2550 ehdr = eth_hdr(skb);
2551 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2557 skb->protocol = proto;
2559 skb->priority = sk->sk_priority;
2560 skb->mark = sk->sk_mark;
2562 packet_pick_tx_queue(dev, skb);
2564 if (po->has_vnet_hdr) {
2565 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2566 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2567 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2568 if (!skb_partial_csum_set(skb, s, o)) {
2574 skb_shinfo(skb)->gso_size =
2575 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2576 skb_shinfo(skb)->gso_type = gso_type;
2578 /* Header must be checked, and gso_segs computed. */
2579 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2580 skb_shinfo(skb)->gso_segs = 0;
2582 len += vnet_hdr_len;
2585 if (!packet_use_direct_xmit(po))
2586 skb_probe_transport_header(skb, reserve);
2587 if (unlikely(extra_len == 4))
2590 err = po->xmit(skb);
2591 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2607 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2609 struct sock *sk = sock->sk;
2610 struct packet_sock *po = pkt_sk(sk);
2612 if (po->tx_ring.pg_vec)
2613 return tpacket_snd(po, msg);
2615 return packet_snd(sock, msg, len);
2619 * Close a PACKET socket. This is fairly simple. We immediately go
2620 * to 'closed' state and remove our protocol entry in the device list.
2623 static int packet_release(struct socket *sock)
2625 struct sock *sk = sock->sk;
2626 struct packet_sock *po;
2628 union tpacket_req_u req_u;
2636 mutex_lock(&net->packet.sklist_lock);
2637 sk_del_node_init_rcu(sk);
2638 mutex_unlock(&net->packet.sklist_lock);
2641 sock_prot_inuse_add(net, sk->sk_prot, -1);
2644 spin_lock(&po->bind_lock);
2645 unregister_prot_hook(sk, false);
2646 packet_cached_dev_reset(po);
2648 if (po->prot_hook.dev) {
2649 dev_put(po->prot_hook.dev);
2650 po->prot_hook.dev = NULL;
2652 spin_unlock(&po->bind_lock);
2654 packet_flush_mclist(sk);
2656 if (po->rx_ring.pg_vec) {
2657 memset(&req_u, 0, sizeof(req_u));
2658 packet_set_ring(sk, &req_u, 1, 0);
2661 if (po->tx_ring.pg_vec) {
2662 memset(&req_u, 0, sizeof(req_u));
2663 packet_set_ring(sk, &req_u, 1, 1);
2670 * Now the socket is dead. No more input will appear.
2677 skb_queue_purge(&sk->sk_receive_queue);
2678 packet_free_pending(po);
2679 sk_refcnt_debug_release(sk);
2686 * Attach a packet hook.
2689 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2691 struct packet_sock *po = pkt_sk(sk);
2692 const struct net_device *dev_curr;
2704 spin_lock(&po->bind_lock);
2706 proto_curr = po->prot_hook.type;
2707 dev_curr = po->prot_hook.dev;
2709 need_rehook = proto_curr != proto || dev_curr != dev;
2712 unregister_prot_hook(sk, true);
2715 po->prot_hook.type = proto;
2717 if (po->prot_hook.dev)
2718 dev_put(po->prot_hook.dev);
2720 po->prot_hook.dev = dev;
2722 po->ifindex = dev ? dev->ifindex : 0;
2723 packet_cached_dev_assign(po, dev);
2726 if (proto == 0 || !need_rehook)
2729 if (!dev || (dev->flags & IFF_UP)) {
2730 register_prot_hook(sk);
2732 sk->sk_err = ENETDOWN;
2733 if (!sock_flag(sk, SOCK_DEAD))
2734 sk->sk_error_report(sk);
2738 spin_unlock(&po->bind_lock);
2744 * Bind a packet socket to a device
2747 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2750 struct sock *sk = sock->sk;
2752 struct net_device *dev;
2759 if (addr_len != sizeof(struct sockaddr))
2761 strlcpy(name, uaddr->sa_data, sizeof(name));
2763 dev = dev_get_by_name(sock_net(sk), name);
2765 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2769 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2771 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2772 struct sock *sk = sock->sk;
2773 struct net_device *dev = NULL;
2781 if (addr_len < sizeof(struct sockaddr_ll))
2783 if (sll->sll_family != AF_PACKET)
2786 if (sll->sll_ifindex) {
2788 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2792 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2798 static struct proto packet_proto = {
2800 .owner = THIS_MODULE,
2801 .obj_size = sizeof(struct packet_sock),
2805 * Create a packet of type SOCK_PACKET.
2808 static int packet_create(struct net *net, struct socket *sock, int protocol,
2812 struct packet_sock *po;
2813 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2816 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2818 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2819 sock->type != SOCK_PACKET)
2820 return -ESOCKTNOSUPPORT;
2822 sock->state = SS_UNCONNECTED;
2825 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2829 sock->ops = &packet_ops;
2830 if (sock->type == SOCK_PACKET)
2831 sock->ops = &packet_ops_spkt;
2833 sock_init_data(sock, sk);
2836 sk->sk_family = PF_PACKET;
2838 po->xmit = dev_queue_xmit;
2840 err = packet_alloc_pending(po);
2844 packet_cached_dev_reset(po);
2846 sk->sk_destruct = packet_sock_destruct;
2847 sk_refcnt_debug_inc(sk);
2850 * Attach a protocol block
2853 spin_lock_init(&po->bind_lock);
2854 mutex_init(&po->pg_vec_lock);
2855 po->prot_hook.func = packet_rcv;
2857 if (sock->type == SOCK_PACKET)
2858 po->prot_hook.func = packet_rcv_spkt;
2860 po->prot_hook.af_packet_priv = sk;
2863 po->prot_hook.type = proto;
2864 register_prot_hook(sk);
2867 mutex_lock(&net->packet.sklist_lock);
2868 sk_add_node_rcu(sk, &net->packet.sklist);
2869 mutex_unlock(&net->packet.sklist_lock);
2872 sock_prot_inuse_add(net, &packet_proto, 1);
2883 * Pull a packet from our receive queue and hand it to the user.
2884 * If necessary we block.
2887 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2890 struct sock *sk = sock->sk;
2891 struct sk_buff *skb;
2893 int vnet_hdr_len = 0;
2894 unsigned int origlen = 0;
2897 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2901 /* What error should we return now? EUNATTACH? */
2902 if (pkt_sk(sk)->ifindex < 0)
2906 if (flags & MSG_ERRQUEUE) {
2907 err = sock_recv_errqueue(sk, msg, len,
2908 SOL_PACKET, PACKET_TX_TIMESTAMP);
2913 * Call the generic datagram receiver. This handles all sorts
2914 * of horrible races and re-entrancy so we can forget about it
2915 * in the protocol layers.
2917 * Now it will return ENETDOWN, if device have just gone down,
2918 * but then it will block.
2921 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2924 * An error occurred so return it. Because skb_recv_datagram()
2925 * handles the blocking we don't see and worry about blocking
2932 if (pkt_sk(sk)->has_vnet_hdr) {
2933 struct virtio_net_hdr vnet_hdr = { 0 };
2936 vnet_hdr_len = sizeof(vnet_hdr);
2937 if (len < vnet_hdr_len)
2940 len -= vnet_hdr_len;
2942 if (skb_is_gso(skb)) {
2943 struct skb_shared_info *sinfo = skb_shinfo(skb);
2945 /* This is a hint as to how much should be linear. */
2947 __cpu_to_virtio16(false, skb_headlen(skb));
2949 __cpu_to_virtio16(false, sinfo->gso_size);
2950 if (sinfo->gso_type & SKB_GSO_TCPV4)
2951 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2952 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2953 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2954 else if (sinfo->gso_type & SKB_GSO_UDP)
2955 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2956 else if (sinfo->gso_type & SKB_GSO_FCOE)
2960 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2961 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2963 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2965 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2966 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2967 vnet_hdr.csum_start = __cpu_to_virtio16(false,
2968 skb_checksum_start_offset(skb));
2969 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2971 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2972 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2973 } /* else everything is zero */
2975 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
2980 /* You lose any data beyond the buffer you gave. If it worries
2981 * a user program they can ask the device for its MTU
2987 msg->msg_flags |= MSG_TRUNC;
2990 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2994 if (sock->type != SOCK_PACKET) {
2995 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2997 /* Original length was stored in sockaddr_ll fields */
2998 origlen = PACKET_SKB_CB(skb)->sa.origlen;
2999 sll->sll_family = AF_PACKET;
3000 sll->sll_protocol = skb->protocol;
3003 sock_recv_ts_and_drops(msg, sk, skb);
3005 if (msg->msg_name) {
3006 /* If the address length field is there to be filled
3007 * in, we fill it in now.
3009 if (sock->type == SOCK_PACKET) {
3010 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3011 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3013 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3015 msg->msg_namelen = sll->sll_halen +
3016 offsetof(struct sockaddr_ll, sll_addr);
3018 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3022 if (pkt_sk(sk)->auxdata) {
3023 struct tpacket_auxdata aux;
3025 aux.tp_status = TP_STATUS_USER;
3026 if (skb->ip_summed == CHECKSUM_PARTIAL)
3027 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3028 else if (skb->pkt_type != PACKET_OUTGOING &&
3029 (skb->ip_summed == CHECKSUM_COMPLETE ||
3030 skb_csum_unnecessary(skb)))
3031 aux.tp_status |= TP_STATUS_CSUM_VALID;
3033 aux.tp_len = origlen;
3034 aux.tp_snaplen = skb->len;
3036 aux.tp_net = skb_network_offset(skb);
3037 if (skb_vlan_tag_present(skb)) {
3038 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3039 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3040 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3042 aux.tp_vlan_tci = 0;
3043 aux.tp_vlan_tpid = 0;
3045 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3049 * Free or return the buffer as appropriate. Again this
3050 * hides all the races and re-entrancy issues from us.
3052 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3055 skb_free_datagram(sk, skb);
3060 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3061 int *uaddr_len, int peer)
3063 struct net_device *dev;
3064 struct sock *sk = sock->sk;
3069 uaddr->sa_family = AF_PACKET;
3070 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3072 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3074 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3076 *uaddr_len = sizeof(*uaddr);
3081 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3082 int *uaddr_len, int peer)
3084 struct net_device *dev;
3085 struct sock *sk = sock->sk;
3086 struct packet_sock *po = pkt_sk(sk);
3087 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3092 sll->sll_family = AF_PACKET;
3093 sll->sll_ifindex = po->ifindex;
3094 sll->sll_protocol = po->num;
3095 sll->sll_pkttype = 0;
3097 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3099 sll->sll_hatype = dev->type;
3100 sll->sll_halen = dev->addr_len;
3101 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3103 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3107 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3112 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3116 case PACKET_MR_MULTICAST:
3117 if (i->alen != dev->addr_len)
3120 return dev_mc_add(dev, i->addr);
3122 return dev_mc_del(dev, i->addr);
3124 case PACKET_MR_PROMISC:
3125 return dev_set_promiscuity(dev, what);
3126 case PACKET_MR_ALLMULTI:
3127 return dev_set_allmulti(dev, what);
3128 case PACKET_MR_UNICAST:
3129 if (i->alen != dev->addr_len)
3132 return dev_uc_add(dev, i->addr);
3134 return dev_uc_del(dev, i->addr);
3142 static void packet_dev_mclist_delete(struct net_device *dev,
3143 struct packet_mclist **mlp)
3145 struct packet_mclist *ml;
3147 while ((ml = *mlp) != NULL) {
3148 if (ml->ifindex == dev->ifindex) {
3149 packet_dev_mc(dev, ml, -1);
3157 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3159 struct packet_sock *po = pkt_sk(sk);
3160 struct packet_mclist *ml, *i;
3161 struct net_device *dev;
3167 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3172 if (mreq->mr_alen > dev->addr_len)
3176 i = kmalloc(sizeof(*i), GFP_KERNEL);
3181 for (ml = po->mclist; ml; ml = ml->next) {
3182 if (ml->ifindex == mreq->mr_ifindex &&
3183 ml->type == mreq->mr_type &&
3184 ml->alen == mreq->mr_alen &&
3185 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3187 /* Free the new element ... */
3193 i->type = mreq->mr_type;
3194 i->ifindex = mreq->mr_ifindex;
3195 i->alen = mreq->mr_alen;
3196 memcpy(i->addr, mreq->mr_address, i->alen);
3198 i->next = po->mclist;
3200 err = packet_dev_mc(dev, i, 1);
3202 po->mclist = i->next;
3211 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3213 struct packet_mclist *ml, **mlp;
3217 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3218 if (ml->ifindex == mreq->mr_ifindex &&
3219 ml->type == mreq->mr_type &&
3220 ml->alen == mreq->mr_alen &&
3221 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3222 if (--ml->count == 0) {
3223 struct net_device *dev;
3225 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3227 packet_dev_mc(dev, ml, -1);
3237 static void packet_flush_mclist(struct sock *sk)
3239 struct packet_sock *po = pkt_sk(sk);
3240 struct packet_mclist *ml;
3246 while ((ml = po->mclist) != NULL) {
3247 struct net_device *dev;
3249 po->mclist = ml->next;
3250 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3252 packet_dev_mc(dev, ml, -1);
3259 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3261 struct sock *sk = sock->sk;
3262 struct packet_sock *po = pkt_sk(sk);
3265 if (level != SOL_PACKET)
3266 return -ENOPROTOOPT;
3269 case PACKET_ADD_MEMBERSHIP:
3270 case PACKET_DROP_MEMBERSHIP:
3272 struct packet_mreq_max mreq;
3274 memset(&mreq, 0, sizeof(mreq));
3275 if (len < sizeof(struct packet_mreq))
3277 if (len > sizeof(mreq))
3279 if (copy_from_user(&mreq, optval, len))
3281 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3283 if (optname == PACKET_ADD_MEMBERSHIP)
3284 ret = packet_mc_add(sk, &mreq);
3286 ret = packet_mc_drop(sk, &mreq);
3290 case PACKET_RX_RING:
3291 case PACKET_TX_RING:
3293 union tpacket_req_u req_u;
3296 switch (po->tp_version) {
3299 len = sizeof(req_u.req);
3303 len = sizeof(req_u.req3);
3308 if (pkt_sk(sk)->has_vnet_hdr)
3310 if (copy_from_user(&req_u.req, optval, len))
3312 return packet_set_ring(sk, &req_u, 0,
3313 optname == PACKET_TX_RING);
3315 case PACKET_COPY_THRESH:
3319 if (optlen != sizeof(val))
3321 if (copy_from_user(&val, optval, sizeof(val)))
3324 pkt_sk(sk)->copy_thresh = val;
3327 case PACKET_VERSION:
3331 if (optlen != sizeof(val))
3333 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3335 if (copy_from_user(&val, optval, sizeof(val)))
3341 po->tp_version = val;
3347 case PACKET_RESERVE:
3351 if (optlen != sizeof(val))
3353 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3355 if (copy_from_user(&val, optval, sizeof(val)))
3357 po->tp_reserve = val;
3364 if (optlen != sizeof(val))
3366 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3368 if (copy_from_user(&val, optval, sizeof(val)))
3370 po->tp_loss = !!val;
3373 case PACKET_AUXDATA:
3377 if (optlen < sizeof(val))
3379 if (copy_from_user(&val, optval, sizeof(val)))
3382 po->auxdata = !!val;
3385 case PACKET_ORIGDEV:
3389 if (optlen < sizeof(val))
3391 if (copy_from_user(&val, optval, sizeof(val)))
3394 po->origdev = !!val;
3397 case PACKET_VNET_HDR:
3401 if (sock->type != SOCK_RAW)
3403 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3405 if (optlen < sizeof(val))
3407 if (copy_from_user(&val, optval, sizeof(val)))
3410 po->has_vnet_hdr = !!val;
3413 case PACKET_TIMESTAMP:
3417 if (optlen != sizeof(val))
3419 if (copy_from_user(&val, optval, sizeof(val)))
3422 po->tp_tstamp = val;
3429 if (optlen != sizeof(val))
3431 if (copy_from_user(&val, optval, sizeof(val)))
3434 return fanout_add(sk, val & 0xffff, val >> 16);
3436 case PACKET_TX_HAS_OFF:
3440 if (optlen != sizeof(val))
3442 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3444 if (copy_from_user(&val, optval, sizeof(val)))
3446 po->tp_tx_has_off = !!val;
3449 case PACKET_QDISC_BYPASS:
3453 if (optlen != sizeof(val))
3455 if (copy_from_user(&val, optval, sizeof(val)))
3458 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3462 return -ENOPROTOOPT;
3466 static int packet_getsockopt(struct socket *sock, int level, int optname,
3467 char __user *optval, int __user *optlen)
3470 int val, lv = sizeof(val);
3471 struct sock *sk = sock->sk;
3472 struct packet_sock *po = pkt_sk(sk);
3474 union tpacket_stats_u st;
3476 if (level != SOL_PACKET)
3477 return -ENOPROTOOPT;
3479 if (get_user(len, optlen))
3486 case PACKET_STATISTICS:
3487 spin_lock_bh(&sk->sk_receive_queue.lock);
3488 memcpy(&st, &po->stats, sizeof(st));
3489 memset(&po->stats, 0, sizeof(po->stats));
3490 spin_unlock_bh(&sk->sk_receive_queue.lock);
3492 if (po->tp_version == TPACKET_V3) {
3493 lv = sizeof(struct tpacket_stats_v3);
3494 st.stats3.tp_packets += st.stats3.tp_drops;
3497 lv = sizeof(struct tpacket_stats);
3498 st.stats1.tp_packets += st.stats1.tp_drops;
3503 case PACKET_AUXDATA:
3506 case PACKET_ORIGDEV:
3509 case PACKET_VNET_HDR:
3510 val = po->has_vnet_hdr;
3512 case PACKET_VERSION:
3513 val = po->tp_version;
3516 if (len > sizeof(int))
3518 if (copy_from_user(&val, optval, len))
3522 val = sizeof(struct tpacket_hdr);
3525 val = sizeof(struct tpacket2_hdr);
3528 val = sizeof(struct tpacket3_hdr);
3534 case PACKET_RESERVE:
3535 val = po->tp_reserve;
3540 case PACKET_TIMESTAMP:
3541 val = po->tp_tstamp;
3545 ((u32)po->fanout->id |
3546 ((u32)po->fanout->type << 16) |
3547 ((u32)po->fanout->flags << 24)) :
3550 case PACKET_TX_HAS_OFF:
3551 val = po->tp_tx_has_off;
3553 case PACKET_QDISC_BYPASS:
3554 val = packet_use_direct_xmit(po);
3557 return -ENOPROTOOPT;
3562 if (put_user(len, optlen))
3564 if (copy_to_user(optval, data, len))
3570 static int packet_notifier(struct notifier_block *this,
3571 unsigned long msg, void *ptr)
3574 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3575 struct net *net = dev_net(dev);
3578 sk_for_each_rcu(sk, &net->packet.sklist) {
3579 struct packet_sock *po = pkt_sk(sk);
3582 case NETDEV_UNREGISTER:
3584 packet_dev_mclist_delete(dev, &po->mclist);
3588 if (dev->ifindex == po->ifindex) {
3589 spin_lock(&po->bind_lock);
3591 __unregister_prot_hook(sk, false);
3592 sk->sk_err = ENETDOWN;
3593 if (!sock_flag(sk, SOCK_DEAD))
3594 sk->sk_error_report(sk);
3596 if (msg == NETDEV_UNREGISTER) {
3597 packet_cached_dev_reset(po);
3599 if (po->prot_hook.dev)
3600 dev_put(po->prot_hook.dev);
3601 po->prot_hook.dev = NULL;
3603 spin_unlock(&po->bind_lock);
3607 if (dev->ifindex == po->ifindex) {
3608 spin_lock(&po->bind_lock);
3610 register_prot_hook(sk);
3611 spin_unlock(&po->bind_lock);
3621 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3624 struct sock *sk = sock->sk;
3629 int amount = sk_wmem_alloc_get(sk);
3631 return put_user(amount, (int __user *)arg);
3635 struct sk_buff *skb;
3638 spin_lock_bh(&sk->sk_receive_queue.lock);
3639 skb = skb_peek(&sk->sk_receive_queue);
3642 spin_unlock_bh(&sk->sk_receive_queue.lock);
3643 return put_user(amount, (int __user *)arg);
3646 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3648 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3658 case SIOCGIFBRDADDR:
3659 case SIOCSIFBRDADDR:
3660 case SIOCGIFNETMASK:
3661 case SIOCSIFNETMASK:
3662 case SIOCGIFDSTADDR:
3663 case SIOCSIFDSTADDR:
3665 return inet_dgram_ops.ioctl(sock, cmd, arg);
3669 return -ENOIOCTLCMD;
3674 static unsigned int packet_poll(struct file *file, struct socket *sock,
3677 struct sock *sk = sock->sk;
3678 struct packet_sock *po = pkt_sk(sk);
3679 unsigned int mask = datagram_poll(file, sock, wait);
3681 spin_lock_bh(&sk->sk_receive_queue.lock);
3682 if (po->rx_ring.pg_vec) {
3683 if (!packet_previous_rx_frame(po, &po->rx_ring,
3685 mask |= POLLIN | POLLRDNORM;
3687 spin_unlock_bh(&sk->sk_receive_queue.lock);
3688 spin_lock_bh(&sk->sk_write_queue.lock);
3689 if (po->tx_ring.pg_vec) {
3690 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3691 mask |= POLLOUT | POLLWRNORM;
3693 spin_unlock_bh(&sk->sk_write_queue.lock);
3698 /* Dirty? Well, I still did not learn better way to account
3702 static void packet_mm_open(struct vm_area_struct *vma)
3704 struct file *file = vma->vm_file;
3705 struct socket *sock = file->private_data;
3706 struct sock *sk = sock->sk;
3709 atomic_inc(&pkt_sk(sk)->mapped);
3712 static void packet_mm_close(struct vm_area_struct *vma)
3714 struct file *file = vma->vm_file;
3715 struct socket *sock = file->private_data;
3716 struct sock *sk = sock->sk;
3719 atomic_dec(&pkt_sk(sk)->mapped);
3722 static const struct vm_operations_struct packet_mmap_ops = {
3723 .open = packet_mm_open,
3724 .close = packet_mm_close,
3727 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3732 for (i = 0; i < len; i++) {
3733 if (likely(pg_vec[i].buffer)) {
3734 if (is_vmalloc_addr(pg_vec[i].buffer))
3735 vfree(pg_vec[i].buffer);
3737 free_pages((unsigned long)pg_vec[i].buffer,
3739 pg_vec[i].buffer = NULL;
3745 static char *alloc_one_pg_vec_page(unsigned long order)
3748 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3749 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3751 buffer = (char *) __get_free_pages(gfp_flags, order);
3755 /* __get_free_pages failed, fall back to vmalloc */
3756 buffer = vzalloc((1 << order) * PAGE_SIZE);
3760 /* vmalloc failed, lets dig into swap here */
3761 gfp_flags &= ~__GFP_NORETRY;
3762 buffer = (char *) __get_free_pages(gfp_flags, order);
3766 /* complete and utter failure */
3770 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3772 unsigned int block_nr = req->tp_block_nr;
3776 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3777 if (unlikely(!pg_vec))
3780 for (i = 0; i < block_nr; i++) {
3781 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3782 if (unlikely(!pg_vec[i].buffer))
3783 goto out_free_pgvec;
3790 free_pg_vec(pg_vec, order, block_nr);
3795 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3796 int closing, int tx_ring)
3798 struct pgv *pg_vec = NULL;
3799 struct packet_sock *po = pkt_sk(sk);
3800 int was_running, order = 0;
3801 struct packet_ring_buffer *rb;
3802 struct sk_buff_head *rb_queue;
3805 /* Added to avoid minimal code churn */
3806 struct tpacket_req *req = &req_u->req;
3808 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3809 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3810 WARN(1, "Tx-ring is not supported.\n");
3814 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3815 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3819 if (atomic_read(&po->mapped))
3821 if (packet_read_pending(rb))
3825 if (req->tp_block_nr) {
3826 /* Sanity tests and some calculations */
3828 if (unlikely(rb->pg_vec))
3831 switch (po->tp_version) {
3833 po->tp_hdrlen = TPACKET_HDRLEN;
3836 po->tp_hdrlen = TPACKET2_HDRLEN;
3839 po->tp_hdrlen = TPACKET3_HDRLEN;
3844 if (unlikely((int)req->tp_block_size <= 0))
3846 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3848 if (po->tp_version >= TPACKET_V3 &&
3849 (int)(req->tp_block_size -
3850 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3852 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3855 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3858 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3859 if (unlikely(rb->frames_per_block <= 0))
3861 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3866 order = get_order(req->tp_block_size);
3867 pg_vec = alloc_pg_vec(req, order);
3868 if (unlikely(!pg_vec))
3870 switch (po->tp_version) {
3872 /* Transmit path is not supported. We checked
3873 * it above but just being paranoid
3876 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3885 if (unlikely(req->tp_frame_nr))
3891 /* Detach socket from network */
3892 spin_lock(&po->bind_lock);
3893 was_running = po->running;
3897 __unregister_prot_hook(sk, false);
3899 spin_unlock(&po->bind_lock);
3904 mutex_lock(&po->pg_vec_lock);
3905 if (closing || atomic_read(&po->mapped) == 0) {
3907 spin_lock_bh(&rb_queue->lock);
3908 swap(rb->pg_vec, pg_vec);
3909 rb->frame_max = (req->tp_frame_nr - 1);
3911 rb->frame_size = req->tp_frame_size;
3912 spin_unlock_bh(&rb_queue->lock);
3914 swap(rb->pg_vec_order, order);
3915 swap(rb->pg_vec_len, req->tp_block_nr);
3917 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3918 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3919 tpacket_rcv : packet_rcv;
3920 skb_queue_purge(rb_queue);
3921 if (atomic_read(&po->mapped))
3922 pr_err("packet_mmap: vma is busy: %d\n",
3923 atomic_read(&po->mapped));
3925 mutex_unlock(&po->pg_vec_lock);
3927 spin_lock(&po->bind_lock);
3930 register_prot_hook(sk);
3932 spin_unlock(&po->bind_lock);
3933 if (closing && (po->tp_version > TPACKET_V2)) {
3934 /* Because we don't support block-based V3 on tx-ring */
3936 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3941 free_pg_vec(pg_vec, order, req->tp_block_nr);
3946 static int packet_mmap(struct file *file, struct socket *sock,
3947 struct vm_area_struct *vma)
3949 struct sock *sk = sock->sk;
3950 struct packet_sock *po = pkt_sk(sk);
3951 unsigned long size, expected_size;
3952 struct packet_ring_buffer *rb;
3953 unsigned long start;
3960 mutex_lock(&po->pg_vec_lock);
3963 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3965 expected_size += rb->pg_vec_len
3971 if (expected_size == 0)
3974 size = vma->vm_end - vma->vm_start;
3975 if (size != expected_size)
3978 start = vma->vm_start;
3979 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3980 if (rb->pg_vec == NULL)
3983 for (i = 0; i < rb->pg_vec_len; i++) {
3985 void *kaddr = rb->pg_vec[i].buffer;
3988 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3989 page = pgv_to_page(kaddr);
3990 err = vm_insert_page(vma, start, page);
3999 atomic_inc(&po->mapped);
4000 vma->vm_ops = &packet_mmap_ops;
4004 mutex_unlock(&po->pg_vec_lock);
4008 static const struct proto_ops packet_ops_spkt = {
4009 .family = PF_PACKET,
4010 .owner = THIS_MODULE,
4011 .release = packet_release,
4012 .bind = packet_bind_spkt,
4013 .connect = sock_no_connect,
4014 .socketpair = sock_no_socketpair,
4015 .accept = sock_no_accept,
4016 .getname = packet_getname_spkt,
4017 .poll = datagram_poll,
4018 .ioctl = packet_ioctl,
4019 .listen = sock_no_listen,
4020 .shutdown = sock_no_shutdown,
4021 .setsockopt = sock_no_setsockopt,
4022 .getsockopt = sock_no_getsockopt,
4023 .sendmsg = packet_sendmsg_spkt,
4024 .recvmsg = packet_recvmsg,
4025 .mmap = sock_no_mmap,
4026 .sendpage = sock_no_sendpage,
4029 static const struct proto_ops packet_ops = {
4030 .family = PF_PACKET,
4031 .owner = THIS_MODULE,
4032 .release = packet_release,
4033 .bind = packet_bind,
4034 .connect = sock_no_connect,
4035 .socketpair = sock_no_socketpair,
4036 .accept = sock_no_accept,
4037 .getname = packet_getname,
4038 .poll = packet_poll,
4039 .ioctl = packet_ioctl,
4040 .listen = sock_no_listen,
4041 .shutdown = sock_no_shutdown,
4042 .setsockopt = packet_setsockopt,
4043 .getsockopt = packet_getsockopt,
4044 .sendmsg = packet_sendmsg,
4045 .recvmsg = packet_recvmsg,
4046 .mmap = packet_mmap,
4047 .sendpage = sock_no_sendpage,
4050 static const struct net_proto_family packet_family_ops = {
4051 .family = PF_PACKET,
4052 .create = packet_create,
4053 .owner = THIS_MODULE,
4056 static struct notifier_block packet_netdev_notifier = {
4057 .notifier_call = packet_notifier,
4060 #ifdef CONFIG_PROC_FS
4062 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4065 struct net *net = seq_file_net(seq);
4068 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4071 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4073 struct net *net = seq_file_net(seq);
4074 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4077 static void packet_seq_stop(struct seq_file *seq, void *v)
4083 static int packet_seq_show(struct seq_file *seq, void *v)
4085 if (v == SEQ_START_TOKEN)
4086 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4088 struct sock *s = sk_entry(v);
4089 const struct packet_sock *po = pkt_sk(s);
4092 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4094 atomic_read(&s->sk_refcnt),
4099 atomic_read(&s->sk_rmem_alloc),
4100 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4107 static const struct seq_operations packet_seq_ops = {
4108 .start = packet_seq_start,
4109 .next = packet_seq_next,
4110 .stop = packet_seq_stop,
4111 .show = packet_seq_show,
4114 static int packet_seq_open(struct inode *inode, struct file *file)
4116 return seq_open_net(inode, file, &packet_seq_ops,
4117 sizeof(struct seq_net_private));
4120 static const struct file_operations packet_seq_fops = {
4121 .owner = THIS_MODULE,
4122 .open = packet_seq_open,
4124 .llseek = seq_lseek,
4125 .release = seq_release_net,
4130 static int __net_init packet_net_init(struct net *net)
4132 mutex_init(&net->packet.sklist_lock);
4133 INIT_HLIST_HEAD(&net->packet.sklist);
4135 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4141 static void __net_exit packet_net_exit(struct net *net)
4143 remove_proc_entry("packet", net->proc_net);
4146 static struct pernet_operations packet_net_ops = {
4147 .init = packet_net_init,
4148 .exit = packet_net_exit,
4152 static void __exit packet_exit(void)
4154 unregister_netdevice_notifier(&packet_netdev_notifier);
4155 unregister_pernet_subsys(&packet_net_ops);
4156 sock_unregister(PF_PACKET);
4157 proto_unregister(&packet_proto);
4160 static int __init packet_init(void)
4162 int rc = proto_register(&packet_proto, 0);
4167 sock_register(&packet_family_ops);
4168 register_pernet_subsys(&packet_net_ops);
4169 register_netdevice_notifier(&packet_netdev_notifier);
4174 module_init(packet_init);
4175 module_exit(packet_exit);
4176 MODULE_LICENSE("GPL");
4177 MODULE_ALIAS_NETPROTO(PF_PACKET);