2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
218 struct packet_skb_cb {
220 struct sockaddr_pkt pkt;
222 /* Trick: alias skb original length with
223 * ll.sll_family and ll.protocol in order
226 unsigned int origlen;
227 struct sockaddr_ll ll;
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 static int packet_direct_xmit(struct sk_buff *skb)
248 struct net_device *dev = skb->dev;
249 netdev_features_t features;
250 struct netdev_queue *txq;
251 int ret = NETDEV_TX_BUSY;
253 if (unlikely(!netif_running(dev) ||
254 !netif_carrier_ok(dev)))
257 features = netif_skb_features(skb);
258 if (skb_needs_linearize(skb, features) &&
259 __skb_linearize(skb))
262 txq = skb_get_tx_queue(dev, skb);
266 HARD_TX_LOCK(dev, txq, smp_processor_id());
267 if (!netif_xmit_frozen_or_drv_stopped(txq))
268 ret = netdev_start_xmit(skb, dev, txq, false);
269 HARD_TX_UNLOCK(dev, txq);
273 if (!dev_xmit_complete(ret))
278 atomic_long_inc(&dev->tx_dropped);
280 return NET_XMIT_DROP;
283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
285 struct net_device *dev;
288 dev = rcu_dereference(po->cached_dev);
296 static void packet_cached_dev_assign(struct packet_sock *po,
297 struct net_device *dev)
299 rcu_assign_pointer(po->cached_dev, dev);
302 static void packet_cached_dev_reset(struct packet_sock *po)
304 RCU_INIT_POINTER(po->cached_dev, NULL);
307 static bool packet_use_direct_xmit(const struct packet_sock *po)
309 return po->xmit == packet_direct_xmit;
312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
314 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
319 const struct net_device_ops *ops = dev->netdev_ops;
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL,
324 __packet_pick_tx_queue);
325 queue_index = netdev_cap_txqueue(dev, queue_index);
327 queue_index = __packet_pick_tx_queue(dev, skb);
330 skb_set_queue_mapping(skb, queue_index);
333 /* register_prot_hook must be invoked with the po->bind_lock held,
334 * or from a context in which asynchronous accesses to the packet
335 * socket is not possible (packet_create()).
337 static void register_prot_hook(struct sock *sk)
339 struct packet_sock *po = pkt_sk(sk);
343 __fanout_link(sk, po);
345 dev_add_pack(&po->prot_hook);
352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
353 * held. If the sync parameter is true, we will temporarily drop
354 * the po->bind_lock and do a synchronize_net to make sure no
355 * asynchronous packet processing paths still refer to the elements
356 * of po->prot_hook. If the sync parameter is false, it is the
357 * callers responsibility to take care of this.
359 static void __unregister_prot_hook(struct sock *sk, bool sync)
361 struct packet_sock *po = pkt_sk(sk);
366 __fanout_unlink(sk, po);
368 __dev_remove_pack(&po->prot_hook);
373 spin_unlock(&po->bind_lock);
375 spin_lock(&po->bind_lock);
379 static void unregister_prot_hook(struct sock *sk, bool sync)
381 struct packet_sock *po = pkt_sk(sk);
384 __unregister_prot_hook(sk, sync);
387 static inline struct page * __pure pgv_to_page(void *addr)
389 if (is_vmalloc_addr(addr))
390 return vmalloc_to_page(addr);
391 return virt_to_page(addr);
394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
396 union tpacket_uhdr h;
399 switch (po->tp_version) {
401 h.h1->tp_status = status;
402 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 h.h2->tp_status = status;
406 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 WARN(1, "TPACKET version not supported.\n");
417 static int __packet_get_status(struct packet_sock *po, void *frame)
419 union tpacket_uhdr h;
424 switch (po->tp_version) {
426 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
427 return h.h1->tp_status;
429 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
430 return h.h2->tp_status;
433 WARN(1, "TPACKET version not supported.\n");
439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
442 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
445 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
447 return TP_STATUS_TS_RAW_HARDWARE;
449 if (ktime_to_timespec_cond(skb->tstamp, ts))
450 return TP_STATUS_TS_SOFTWARE;
455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
458 union tpacket_uhdr h;
462 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 switch (po->tp_version) {
468 h.h1->tp_sec = ts.tv_sec;
469 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
472 h.h2->tp_sec = ts.tv_sec;
473 h.h2->tp_nsec = ts.tv_nsec;
477 WARN(1, "TPACKET version not supported.\n");
481 /* one flush is safe, as both fields always lie on the same cacheline */
482 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
488 static void *packet_lookup_frame(struct packet_sock *po,
489 struct packet_ring_buffer *rb,
490 unsigned int position,
493 unsigned int pg_vec_pos, frame_offset;
494 union tpacket_uhdr h;
496 pg_vec_pos = position / rb->frames_per_block;
497 frame_offset = position % rb->frames_per_block;
499 h.raw = rb->pg_vec[pg_vec_pos].buffer +
500 (frame_offset * rb->frame_size);
502 if (status != __packet_get_status(po, h.raw))
508 static void *packet_current_frame(struct packet_sock *po,
509 struct packet_ring_buffer *rb,
512 return packet_lookup_frame(po, rb, rb->head, status);
515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
517 del_timer_sync(&pkc->retire_blk_timer);
520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
522 struct sk_buff_head *rb_queue)
524 struct tpacket_kbdq_core *pkc;
526 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
527 GET_PBDQC_FROM_RB(&po->rx_ring);
529 spin_lock_bh(&rb_queue->lock);
530 pkc->delete_blk_timer = 1;
531 spin_unlock_bh(&rb_queue->lock);
533 prb_del_retire_blk_timer(pkc);
536 static void prb_init_blk_timer(struct packet_sock *po,
537 struct tpacket_kbdq_core *pkc,
538 void (*func) (unsigned long))
540 init_timer(&pkc->retire_blk_timer);
541 pkc->retire_blk_timer.data = (long)po;
542 pkc->retire_blk_timer.function = func;
543 pkc->retire_blk_timer.expires = jiffies;
546 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
548 struct tpacket_kbdq_core *pkc;
553 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
554 GET_PBDQC_FROM_RB(&po->rx_ring);
555 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
559 int blk_size_in_bytes)
561 struct net_device *dev;
562 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
563 struct ethtool_cmd ecmd;
568 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
569 if (unlikely(!dev)) {
571 return DEFAULT_PRB_RETIRE_TOV;
573 err = __ethtool_get_settings(dev, &ecmd);
574 speed = ethtool_cmd_speed(&ecmd);
578 * If the link speed is so slow you don't really
579 * need to worry about perf anyways
581 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
582 return DEFAULT_PRB_RETIRE_TOV;
589 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
602 union tpacket_req_u *req_u)
604 p1->feature_req_word = req_u->req3.tp_feature_req_word;
607 static void init_prb_bdqc(struct packet_sock *po,
608 struct packet_ring_buffer *rb,
610 union tpacket_req_u *req_u, int tx_ring)
612 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
613 struct tpacket_block_desc *pbd;
615 memset(p1, 0x0, sizeof(*p1));
617 p1->knxt_seq_num = 1;
619 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
620 p1->pkblk_start = pg_vec[0].buffer;
621 p1->kblk_size = req_u->req3.tp_block_size;
622 p1->knum_blocks = req_u->req3.tp_block_nr;
623 p1->hdrlen = po->tp_hdrlen;
624 p1->version = po->tp_version;
625 p1->last_kactive_blk_num = 0;
626 po->stats.stats3.tp_freeze_q_cnt = 0;
627 if (req_u->req3.tp_retire_blk_tov)
628 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
630 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
631 req_u->req3.tp_block_size);
632 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
633 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
635 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
636 prb_init_ft_ops(p1, req_u);
637 prb_setup_retire_blk_timer(po, tx_ring);
638 prb_open_block(p1, pbd);
641 /* Do NOT update the last_blk_num first.
642 * Assumes sk_buff_head lock is held.
644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
646 mod_timer(&pkc->retire_blk_timer,
647 jiffies + pkc->tov_in_jiffies);
648 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
653 * 1) We refresh the timer only when we open a block.
654 * By doing this we don't waste cycles refreshing the timer
655 * on packet-by-packet basis.
657 * With a 1MB block-size, on a 1Gbps line, it will take
658 * i) ~8 ms to fill a block + ii) memcpy etc.
659 * In this cut we are not accounting for the memcpy time.
661 * So, if the user sets the 'tmo' to 10ms then the timer
662 * will never fire while the block is still getting filled
663 * (which is what we want). However, the user could choose
664 * to close a block early and that's fine.
666 * But when the timer does fire, we check whether or not to refresh it.
667 * Since the tmo granularity is in msecs, it is not too expensive
668 * to refresh the timer, lets say every '8' msecs.
669 * Either the user can set the 'tmo' or we can derive it based on
670 * a) line-speed and b) block-size.
671 * prb_calc_retire_blk_tmo() calculates the tmo.
674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
676 struct packet_sock *po = (struct packet_sock *)data;
677 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
679 struct tpacket_block_desc *pbd;
681 spin_lock(&po->sk.sk_receive_queue.lock);
683 frozen = prb_queue_frozen(pkc);
684 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
686 if (unlikely(pkc->delete_blk_timer))
689 /* We only need to plug the race when the block is partially filled.
691 * lock(); increment BLOCK_NUM_PKTS; unlock()
692 * copy_bits() is in progress ...
693 * timer fires on other cpu:
694 * we can't retire the current block because copy_bits
698 if (BLOCK_NUM_PKTS(pbd)) {
699 while (atomic_read(&pkc->blk_fill_in_prog)) {
700 /* Waiting for skb_copy_bits to finish... */
705 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
707 if (!BLOCK_NUM_PKTS(pbd)) {
708 /* An empty block. Just refresh the timer. */
711 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
712 if (!prb_dispatch_next_block(pkc, po))
717 /* Case 1. Queue was frozen because user-space was
720 if (prb_curr_blk_in_use(pkc, pbd)) {
722 * Ok, user-space is still behind.
723 * So just refresh the timer.
727 /* Case 2. queue was frozen,user-space caught up,
728 * now the link went idle && the timer fired.
729 * We don't have a block to close.So we open this
730 * block and restart the timer.
731 * opening a block thaws the queue,restarts timer
732 * Thawing/timer-refresh is a side effect.
734 prb_open_block(pkc, pbd);
741 _prb_refresh_rx_retire_blk_timer(pkc);
744 spin_unlock(&po->sk.sk_receive_queue.lock);
747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
748 struct tpacket_block_desc *pbd1, __u32 status)
750 /* Flush everything minus the block header */
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
757 /* Skip the block header(we know header WILL fit in 4K) */
760 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
761 for (; start < end; start += PAGE_SIZE)
762 flush_dcache_page(pgv_to_page(start));
767 /* Now update the block status. */
769 BLOCK_STATUS(pbd1) = status;
771 /* Flush the block header */
773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
775 flush_dcache_page(pgv_to_page(start));
785 * 2) Increment active_blk_num
787 * Note:We DONT refresh the timer on purpose.
788 * Because almost always the next block will be opened.
790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
791 struct tpacket_block_desc *pbd1,
792 struct packet_sock *po, unsigned int stat)
794 __u32 status = TP_STATUS_USER | stat;
796 struct tpacket3_hdr *last_pkt;
797 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
798 struct sock *sk = &po->sk;
800 if (po->stats.stats3.tp_drops)
801 status |= TP_STATUS_LOSING;
803 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
804 last_pkt->tp_next_offset = 0;
806 /* Get the ts of the last pkt */
807 if (BLOCK_NUM_PKTS(pbd1)) {
808 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
809 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
811 /* Ok, we tmo'd - so get the current time.
813 * It shouldn't really happen as we don't close empty
814 * blocks. See prb_retire_rx_blk_timer_expired().
818 h1->ts_last_pkt.ts_sec = ts.tv_sec;
819 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
824 /* Flush the block */
825 prb_flush_block(pkc1, pbd1, status);
827 sk->sk_data_ready(sk);
829 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
834 pkc->reset_pending_on_curr_blk = 0;
838 * Side effect of opening a block:
840 * 1) prb_queue is thawed.
841 * 2) retire_blk_timer is refreshed.
844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
845 struct tpacket_block_desc *pbd1)
848 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
852 /* We could have just memset this but we will lose the
853 * flexibility of making the priv area sticky
856 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
857 BLOCK_NUM_PKTS(pbd1) = 0;
858 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
862 h1->ts_first_pkt.ts_sec = ts.tv_sec;
863 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
865 pkc1->pkblk_start = (char *)pbd1;
866 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
868 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
871 pbd1->version = pkc1->version;
872 pkc1->prev = pkc1->nxt_offset;
873 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
875 prb_thaw_queue(pkc1);
876 _prb_refresh_rx_retire_blk_timer(pkc1);
882 * Queue freeze logic:
883 * 1) Assume tp_block_nr = 8 blocks.
884 * 2) At time 't0', user opens Rx ring.
885 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
886 * 4) user-space is either sleeping or processing block '0'.
887 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
888 * it will close block-7,loop around and try to fill block '0'.
890 * __packet_lookup_frame_in_block
891 * prb_retire_current_block()
892 * prb_dispatch_next_block()
893 * |->(BLOCK_STATUS == USER) evaluates to true
894 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
895 * 6) Now there are two cases:
896 * 6.1) Link goes idle right after the queue is frozen.
897 * But remember, the last open_block() refreshed the timer.
898 * When this timer expires,it will refresh itself so that we can
899 * re-open block-0 in near future.
900 * 6.2) Link is busy and keeps on receiving packets. This is a simple
901 * case and __packet_lookup_frame_in_block will check if block-0
902 * is free and can now be re-used.
904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
905 struct packet_sock *po)
907 pkc->reset_pending_on_curr_blk = 1;
908 po->stats.stats3.tp_freeze_q_cnt++;
911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
914 * If the next block is free then we will dispatch it
915 * and return a good offset.
916 * Else, we will freeze the queue.
917 * So, caller must check the return value.
919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
920 struct packet_sock *po)
922 struct tpacket_block_desc *pbd;
926 /* 1. Get current block num */
927 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
929 /* 2. If this block is currently in_use then freeze the queue */
930 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
931 prb_freeze_queue(pkc, po);
937 * open this block and return the offset where the first packet
938 * needs to get stored.
940 prb_open_block(pkc, pbd);
941 return (void *)pkc->nxt_offset;
944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
945 struct packet_sock *po, unsigned int status)
947 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
949 /* retire/close the current block */
950 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
952 * Plug the case where copy_bits() is in progress on
953 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
954 * have space to copy the pkt in the current block and
955 * called prb_retire_current_block()
957 * We don't need to worry about the TMO case because
958 * the timer-handler already handled this case.
960 if (!(status & TP_STATUS_BLK_TMO)) {
961 while (atomic_read(&pkc->blk_fill_in_prog)) {
962 /* Waiting for skb_copy_bits to finish... */
966 prb_close_block(pkc, pbd, po, status);
971 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
972 struct tpacket_block_desc *pbd)
974 return TP_STATUS_USER & BLOCK_STATUS(pbd);
977 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
979 return pkc->reset_pending_on_curr_blk;
982 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
984 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
985 atomic_dec(&pkc->blk_fill_in_prog);
988 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
989 struct tpacket3_hdr *ppd)
991 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
994 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
995 struct tpacket3_hdr *ppd)
997 ppd->hv1.tp_rxhash = 0;
1000 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1001 struct tpacket3_hdr *ppd)
1003 if (skb_vlan_tag_present(pkc->skb)) {
1004 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1005 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1006 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1008 ppd->hv1.tp_vlan_tci = 0;
1009 ppd->hv1.tp_vlan_tpid = 0;
1010 ppd->tp_status = TP_STATUS_AVAILABLE;
1014 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1015 struct tpacket3_hdr *ppd)
1017 ppd->hv1.tp_padding = 0;
1018 prb_fill_vlan_info(pkc, ppd);
1020 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1021 prb_fill_rxhash(pkc, ppd);
1023 prb_clear_rxhash(pkc, ppd);
1026 static void prb_fill_curr_block(char *curr,
1027 struct tpacket_kbdq_core *pkc,
1028 struct tpacket_block_desc *pbd,
1031 struct tpacket3_hdr *ppd;
1033 ppd = (struct tpacket3_hdr *)curr;
1034 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1036 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038 BLOCK_NUM_PKTS(pbd) += 1;
1039 atomic_inc(&pkc->blk_fill_in_prog);
1040 prb_run_all_ft_ops(pkc, ppd);
1043 /* Assumes caller has the sk->rx_queue.lock */
1044 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1045 struct sk_buff *skb,
1050 struct tpacket_kbdq_core *pkc;
1051 struct tpacket_block_desc *pbd;
1054 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1055 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057 /* Queue is frozen when user space is lagging behind */
1058 if (prb_queue_frozen(pkc)) {
1060 * Check if that last block which caused the queue to freeze,
1061 * is still in_use by user-space.
1063 if (prb_curr_blk_in_use(pkc, pbd)) {
1064 /* Can't record this packet */
1068 * Ok, the block was released by user-space.
1069 * Now let's open that block.
1070 * opening a block also thaws the queue.
1071 * Thawing is a side effect.
1073 prb_open_block(pkc, pbd);
1078 curr = pkc->nxt_offset;
1080 end = (char *)pbd + pkc->kblk_size;
1082 /* first try the current block */
1083 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1084 prb_fill_curr_block(curr, pkc, pbd, len);
1085 return (void *)curr;
1088 /* Ok, close the current block */
1089 prb_retire_current_block(pkc, po, 0);
1091 /* Now, try to dispatch the next block */
1092 curr = (char *)prb_dispatch_next_block(pkc, po);
1094 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1095 prb_fill_curr_block(curr, pkc, pbd, len);
1096 return (void *)curr;
1100 * No free blocks are available.user_space hasn't caught up yet.
1101 * Queue was just frozen and now this packet will get dropped.
1106 static void *packet_current_rx_frame(struct packet_sock *po,
1107 struct sk_buff *skb,
1108 int status, unsigned int len)
1111 switch (po->tp_version) {
1114 curr = packet_lookup_frame(po, &po->rx_ring,
1115 po->rx_ring.head, status);
1118 return __packet_lookup_frame_in_block(po, skb, status, len);
1120 WARN(1, "TPACKET version not supported\n");
1126 static void *prb_lookup_block(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1131 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1132 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1134 if (status != BLOCK_STATUS(pbd))
1139 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1142 if (rb->prb_bdqc.kactive_blk_num)
1143 prev = rb->prb_bdqc.kactive_blk_num-1;
1145 prev = rb->prb_bdqc.knum_blocks-1;
1149 /* Assumes caller has held the rx_queue.lock */
1150 static void *__prb_previous_block(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1154 unsigned int previous = prb_previous_blk_num(rb);
1155 return prb_lookup_block(po, rb, previous, status);
1158 static void *packet_previous_rx_frame(struct packet_sock *po,
1159 struct packet_ring_buffer *rb,
1162 if (po->tp_version <= TPACKET_V2)
1163 return packet_previous_frame(po, rb, status);
1165 return __prb_previous_block(po, rb, status);
1168 static void packet_increment_rx_head(struct packet_sock *po,
1169 struct packet_ring_buffer *rb)
1171 switch (po->tp_version) {
1174 return packet_increment_head(rb);
1177 WARN(1, "TPACKET version not supported.\n");
1183 static void *packet_previous_frame(struct packet_sock *po,
1184 struct packet_ring_buffer *rb,
1187 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1188 return packet_lookup_frame(po, rb, previous, status);
1191 static void packet_increment_head(struct packet_ring_buffer *buff)
1193 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1196 static void packet_inc_pending(struct packet_ring_buffer *rb)
1198 this_cpu_inc(*rb->pending_refcnt);
1201 static void packet_dec_pending(struct packet_ring_buffer *rb)
1203 this_cpu_dec(*rb->pending_refcnt);
1206 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1208 unsigned int refcnt = 0;
1211 /* We don't use pending refcount in rx_ring. */
1212 if (rb->pending_refcnt == NULL)
1215 for_each_possible_cpu(cpu)
1216 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1221 static int packet_alloc_pending(struct packet_sock *po)
1223 po->rx_ring.pending_refcnt = NULL;
1225 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1226 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1232 static void packet_free_pending(struct packet_sock *po)
1234 free_percpu(po->tx_ring.pending_refcnt);
1237 #define ROOM_POW_OFF 2
1238 #define ROOM_NONE 0x0
1239 #define ROOM_LOW 0x1
1240 #define ROOM_NORMAL 0x2
1242 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1246 len = po->rx_ring.frame_max + 1;
1247 idx = po->rx_ring.head;
1249 idx += len >> pow_off;
1252 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1255 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1259 len = po->rx_ring.prb_bdqc.knum_blocks;
1260 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1262 idx += len >> pow_off;
1265 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1268 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1270 struct sock *sk = &po->sk;
1271 int ret = ROOM_NONE;
1273 if (po->prot_hook.func != tpacket_rcv) {
1274 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1275 - (skb ? skb->truesize : 0);
1276 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1284 if (po->tp_version == TPACKET_V3) {
1285 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1287 else if (__tpacket_v3_has_room(po, 0))
1290 if (__tpacket_has_room(po, ROOM_POW_OFF))
1292 else if (__tpacket_has_room(po, 0))
1299 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1304 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1305 ret = __packet_rcv_has_room(po, skb);
1306 has_room = ret == ROOM_NORMAL;
1307 if (po->pressure == has_room)
1308 po->pressure = !has_room;
1309 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1314 static void packet_sock_destruct(struct sock *sk)
1316 skb_queue_purge(&sk->sk_error_queue);
1318 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1319 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1321 if (!sock_flag(sk, SOCK_DEAD)) {
1322 pr_err("Attempt to release alive packet socket: %p\n", sk);
1326 sk_refcnt_debug_dec(sk);
1329 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1331 int x = atomic_read(&f->rr_cur) + 1;
1339 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1344 rxhash = skb_get_hash(skb);
1345 for (i = 0; i < ROLLOVER_HLEN; i++)
1346 if (po->rollover->history[i] == rxhash)
1349 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1350 return count > (ROLLOVER_HLEN >> 1);
1353 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1354 struct sk_buff *skb,
1357 return reciprocal_scale(skb_get_hash(skb), num);
1360 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1361 struct sk_buff *skb,
1366 cur = atomic_read(&f->rr_cur);
1367 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1368 fanout_rr_next(f, num))) != cur)
1373 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1374 struct sk_buff *skb,
1377 return smp_processor_id() % num;
1380 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1381 struct sk_buff *skb,
1384 return prandom_u32_max(num);
1387 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1388 struct sk_buff *skb,
1389 unsigned int idx, bool try_self,
1392 struct packet_sock *po, *po_next;
1393 unsigned int i, j, room = ROOM_NONE;
1395 po = pkt_sk(f->arr[idx]);
1398 room = packet_rcv_has_room(po, skb);
1399 if (room == ROOM_NORMAL ||
1400 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1404 i = j = min_t(int, po->rollover->sock, num - 1);
1406 po_next = pkt_sk(f->arr[i]);
1407 if (po_next != po && !po_next->pressure &&
1408 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1410 po->rollover->sock = i;
1411 atomic_long_inc(&po->rollover->num);
1412 if (room == ROOM_LOW)
1413 atomic_long_inc(&po->rollover->num_huge);
1421 atomic_long_inc(&po->rollover->num_failed);
1425 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1426 struct sk_buff *skb,
1429 return skb_get_queue_mapping(skb) % num;
1432 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1434 return f->flags & (flag >> 8);
1437 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1438 struct packet_type *pt, struct net_device *orig_dev)
1440 struct packet_fanout *f = pt->af_packet_priv;
1441 unsigned int num = f->num_members;
1442 struct packet_sock *po;
1445 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1452 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1457 case PACKET_FANOUT_HASH:
1459 idx = fanout_demux_hash(f, skb, num);
1461 case PACKET_FANOUT_LB:
1462 idx = fanout_demux_lb(f, skb, num);
1464 case PACKET_FANOUT_CPU:
1465 idx = fanout_demux_cpu(f, skb, num);
1467 case PACKET_FANOUT_RND:
1468 idx = fanout_demux_rnd(f, skb, num);
1470 case PACKET_FANOUT_QM:
1471 idx = fanout_demux_qm(f, skb, num);
1473 case PACKET_FANOUT_ROLLOVER:
1474 idx = fanout_demux_rollover(f, skb, 0, false, num);
1478 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1479 idx = fanout_demux_rollover(f, skb, idx, true, num);
1481 po = pkt_sk(f->arr[idx]);
1482 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1485 DEFINE_MUTEX(fanout_mutex);
1486 EXPORT_SYMBOL_GPL(fanout_mutex);
1487 static LIST_HEAD(fanout_list);
1489 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1491 struct packet_fanout *f = po->fanout;
1493 spin_lock(&f->lock);
1494 f->arr[f->num_members] = sk;
1497 spin_unlock(&f->lock);
1500 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1502 struct packet_fanout *f = po->fanout;
1505 spin_lock(&f->lock);
1506 for (i = 0; i < f->num_members; i++) {
1507 if (f->arr[i] == sk)
1510 BUG_ON(i >= f->num_members);
1511 f->arr[i] = f->arr[f->num_members - 1];
1513 spin_unlock(&f->lock);
1516 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1518 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1524 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1526 struct packet_sock *po = pkt_sk(sk);
1527 struct packet_fanout *f, *match;
1528 u8 type = type_flags & 0xff;
1529 u8 flags = type_flags >> 8;
1533 case PACKET_FANOUT_ROLLOVER:
1534 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1536 case PACKET_FANOUT_HASH:
1537 case PACKET_FANOUT_LB:
1538 case PACKET_FANOUT_CPU:
1539 case PACKET_FANOUT_RND:
1540 case PACKET_FANOUT_QM:
1552 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) {
1553 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1556 atomic_long_set(&po->rollover->num, 0);
1557 atomic_long_set(&po->rollover->num_huge, 0);
1558 atomic_long_set(&po->rollover->num_failed, 0);
1561 mutex_lock(&fanout_mutex);
1563 list_for_each_entry(f, &fanout_list, list) {
1565 read_pnet(&f->net) == sock_net(sk)) {
1571 if (match && match->flags != flags)
1575 match = kzalloc(sizeof(*match), GFP_KERNEL);
1578 write_pnet(&match->net, sock_net(sk));
1581 match->flags = flags;
1582 atomic_set(&match->rr_cur, 0);
1583 INIT_LIST_HEAD(&match->list);
1584 spin_lock_init(&match->lock);
1585 atomic_set(&match->sk_ref, 0);
1586 match->prot_hook.type = po->prot_hook.type;
1587 match->prot_hook.dev = po->prot_hook.dev;
1588 match->prot_hook.func = packet_rcv_fanout;
1589 match->prot_hook.af_packet_priv = match;
1590 match->prot_hook.id_match = match_fanout_group;
1591 dev_add_pack(&match->prot_hook);
1592 list_add(&match->list, &fanout_list);
1595 if (match->type == type &&
1596 match->prot_hook.type == po->prot_hook.type &&
1597 match->prot_hook.dev == po->prot_hook.dev) {
1599 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1600 __dev_remove_pack(&po->prot_hook);
1602 atomic_inc(&match->sk_ref);
1603 __fanout_link(sk, po);
1608 mutex_unlock(&fanout_mutex);
1610 kfree(po->rollover);
1611 po->rollover = NULL;
1616 static void fanout_release(struct sock *sk)
1618 struct packet_sock *po = pkt_sk(sk);
1619 struct packet_fanout *f;
1625 mutex_lock(&fanout_mutex);
1628 if (atomic_dec_and_test(&f->sk_ref)) {
1630 dev_remove_pack(&f->prot_hook);
1633 mutex_unlock(&fanout_mutex);
1635 kfree(po->rollover);
1638 static const struct proto_ops packet_ops;
1640 static const struct proto_ops packet_ops_spkt;
1642 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1643 struct packet_type *pt, struct net_device *orig_dev)
1646 struct sockaddr_pkt *spkt;
1649 * When we registered the protocol we saved the socket in the data
1650 * field for just this event.
1653 sk = pt->af_packet_priv;
1656 * Yank back the headers [hope the device set this
1657 * right or kerboom...]
1659 * Incoming packets have ll header pulled,
1662 * For outgoing ones skb->data == skb_mac_header(skb)
1663 * so that this procedure is noop.
1666 if (skb->pkt_type == PACKET_LOOPBACK)
1669 if (!net_eq(dev_net(dev), sock_net(sk)))
1672 skb = skb_share_check(skb, GFP_ATOMIC);
1676 /* drop any routing info */
1679 /* drop conntrack reference */
1682 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1684 skb_push(skb, skb->data - skb_mac_header(skb));
1687 * The SOCK_PACKET socket receives _all_ frames.
1690 spkt->spkt_family = dev->type;
1691 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1692 spkt->spkt_protocol = skb->protocol;
1695 * Charge the memory to the socket. This is done specifically
1696 * to prevent sockets using all the memory up.
1699 if (sock_queue_rcv_skb(sk, skb) == 0)
1710 * Output a raw packet to a device layer. This bypasses all the other
1711 * protocol layers and you must therefore supply it with a complete frame
1714 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1717 struct sock *sk = sock->sk;
1718 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1719 struct sk_buff *skb = NULL;
1720 struct net_device *dev;
1726 * Get and verify the address.
1730 if (msg->msg_namelen < sizeof(struct sockaddr))
1732 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1733 proto = saddr->spkt_protocol;
1735 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1738 * Find the device first to size check it
1741 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1744 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1750 if (!(dev->flags & IFF_UP))
1754 * You may not queue a frame bigger than the mtu. This is the lowest level
1755 * raw protocol and you must do your own fragmentation at this level.
1758 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1759 if (!netif_supports_nofcs(dev)) {
1760 err = -EPROTONOSUPPORT;
1763 extra_len = 4; /* We're doing our own CRC */
1767 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1771 size_t reserved = LL_RESERVED_SPACE(dev);
1772 int tlen = dev->needed_tailroom;
1773 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1776 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1779 /* FIXME: Save some space for broken drivers that write a hard
1780 * header at transmission time by themselves. PPP is the notable
1781 * one here. This should really be fixed at the driver level.
1783 skb_reserve(skb, reserved);
1784 skb_reset_network_header(skb);
1786 /* Try to align data part correctly */
1791 skb_reset_network_header(skb);
1793 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1799 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1800 /* Earlier code assumed this would be a VLAN pkt,
1801 * double-check this now that we have the actual
1804 struct ethhdr *ehdr;
1805 skb_reset_mac_header(skb);
1806 ehdr = eth_hdr(skb);
1807 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1813 skb->protocol = proto;
1815 skb->priority = sk->sk_priority;
1816 skb->mark = sk->sk_mark;
1818 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1820 if (unlikely(extra_len == 4))
1823 skb_probe_transport_header(skb, 0);
1825 dev_queue_xmit(skb);
1836 static unsigned int run_filter(const struct sk_buff *skb,
1837 const struct sock *sk,
1840 struct sk_filter *filter;
1843 filter = rcu_dereference(sk->sk_filter);
1845 res = SK_RUN_FILTER(filter, skb);
1852 * This function makes lazy skb cloning in hope that most of packets
1853 * are discarded by BPF.
1855 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1856 * and skb->cb are mangled. It works because (and until) packets
1857 * falling here are owned by current CPU. Output packets are cloned
1858 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1859 * sequencially, so that if we return skb to original state on exit,
1860 * we will not harm anyone.
1863 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1864 struct packet_type *pt, struct net_device *orig_dev)
1867 struct sockaddr_ll *sll;
1868 struct packet_sock *po;
1869 u8 *skb_head = skb->data;
1870 int skb_len = skb->len;
1871 unsigned int snaplen, res;
1873 if (skb->pkt_type == PACKET_LOOPBACK)
1876 sk = pt->af_packet_priv;
1879 if (!net_eq(dev_net(dev), sock_net(sk)))
1884 if (dev->header_ops) {
1885 /* The device has an explicit notion of ll header,
1886 * exported to higher levels.
1888 * Otherwise, the device hides details of its frame
1889 * structure, so that corresponding packet head is
1890 * never delivered to user.
1892 if (sk->sk_type != SOCK_DGRAM)
1893 skb_push(skb, skb->data - skb_mac_header(skb));
1894 else if (skb->pkt_type == PACKET_OUTGOING) {
1895 /* Special case: outgoing packets have ll header at head */
1896 skb_pull(skb, skb_network_offset(skb));
1902 res = run_filter(skb, sk, snaplen);
1904 goto drop_n_restore;
1908 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1911 if (skb_shared(skb)) {
1912 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1916 if (skb_head != skb->data) {
1917 skb->data = skb_head;
1924 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1926 sll = &PACKET_SKB_CB(skb)->sa.ll;
1927 sll->sll_hatype = dev->type;
1928 sll->sll_pkttype = skb->pkt_type;
1929 if (unlikely(po->origdev))
1930 sll->sll_ifindex = orig_dev->ifindex;
1932 sll->sll_ifindex = dev->ifindex;
1934 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1936 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1937 * Use their space for storing the original skb length.
1939 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1941 if (pskb_trim(skb, snaplen))
1944 skb_set_owner_r(skb, sk);
1948 /* drop conntrack reference */
1951 spin_lock(&sk->sk_receive_queue.lock);
1952 po->stats.stats1.tp_packets++;
1953 sock_skb_set_dropcount(sk, skb);
1954 __skb_queue_tail(&sk->sk_receive_queue, skb);
1955 spin_unlock(&sk->sk_receive_queue.lock);
1956 sk->sk_data_ready(sk);
1960 spin_lock(&sk->sk_receive_queue.lock);
1961 po->stats.stats1.tp_drops++;
1962 atomic_inc(&sk->sk_drops);
1963 spin_unlock(&sk->sk_receive_queue.lock);
1966 if (skb_head != skb->data && skb_shared(skb)) {
1967 skb->data = skb_head;
1975 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1976 struct packet_type *pt, struct net_device *orig_dev)
1979 struct packet_sock *po;
1980 struct sockaddr_ll *sll;
1981 union tpacket_uhdr h;
1982 u8 *skb_head = skb->data;
1983 int skb_len = skb->len;
1984 unsigned int snaplen, res;
1985 unsigned long status = TP_STATUS_USER;
1986 unsigned short macoff, netoff, hdrlen;
1987 struct sk_buff *copy_skb = NULL;
1991 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1992 * We may add members to them until current aligned size without forcing
1993 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1995 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1996 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1998 if (skb->pkt_type == PACKET_LOOPBACK)
2001 sk = pt->af_packet_priv;
2004 if (!net_eq(dev_net(dev), sock_net(sk)))
2007 if (dev->header_ops) {
2008 if (sk->sk_type != SOCK_DGRAM)
2009 skb_push(skb, skb->data - skb_mac_header(skb));
2010 else if (skb->pkt_type == PACKET_OUTGOING) {
2011 /* Special case: outgoing packets have ll header at head */
2012 skb_pull(skb, skb_network_offset(skb));
2018 res = run_filter(skb, sk, snaplen);
2020 goto drop_n_restore;
2022 if (skb->ip_summed == CHECKSUM_PARTIAL)
2023 status |= TP_STATUS_CSUMNOTREADY;
2024 else if (skb->pkt_type != PACKET_OUTGOING &&
2025 (skb->ip_summed == CHECKSUM_COMPLETE ||
2026 skb_csum_unnecessary(skb)))
2027 status |= TP_STATUS_CSUM_VALID;
2032 if (sk->sk_type == SOCK_DGRAM) {
2033 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2036 unsigned int maclen = skb_network_offset(skb);
2037 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2038 (maclen < 16 ? 16 : maclen)) +
2040 macoff = netoff - maclen;
2042 if (po->tp_version <= TPACKET_V2) {
2043 if (macoff + snaplen > po->rx_ring.frame_size) {
2044 if (po->copy_thresh &&
2045 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2046 if (skb_shared(skb)) {
2047 copy_skb = skb_clone(skb, GFP_ATOMIC);
2049 copy_skb = skb_get(skb);
2050 skb_head = skb->data;
2053 skb_set_owner_r(copy_skb, sk);
2055 snaplen = po->rx_ring.frame_size - macoff;
2056 if ((int)snaplen < 0)
2059 } else if (unlikely(macoff + snaplen >
2060 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2063 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2064 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2065 snaplen, nval, macoff);
2067 if (unlikely((int)snaplen < 0)) {
2069 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2072 spin_lock(&sk->sk_receive_queue.lock);
2073 h.raw = packet_current_rx_frame(po, skb,
2074 TP_STATUS_KERNEL, (macoff+snaplen));
2077 if (po->tp_version <= TPACKET_V2) {
2078 packet_increment_rx_head(po, &po->rx_ring);
2080 * LOSING will be reported till you read the stats,
2081 * because it's COR - Clear On Read.
2082 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2085 if (po->stats.stats1.tp_drops)
2086 status |= TP_STATUS_LOSING;
2088 po->stats.stats1.tp_packets++;
2090 status |= TP_STATUS_COPY;
2091 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2093 spin_unlock(&sk->sk_receive_queue.lock);
2095 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2097 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2098 getnstimeofday(&ts);
2100 status |= ts_status;
2102 switch (po->tp_version) {
2104 h.h1->tp_len = skb->len;
2105 h.h1->tp_snaplen = snaplen;
2106 h.h1->tp_mac = macoff;
2107 h.h1->tp_net = netoff;
2108 h.h1->tp_sec = ts.tv_sec;
2109 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2110 hdrlen = sizeof(*h.h1);
2113 h.h2->tp_len = skb->len;
2114 h.h2->tp_snaplen = snaplen;
2115 h.h2->tp_mac = macoff;
2116 h.h2->tp_net = netoff;
2117 h.h2->tp_sec = ts.tv_sec;
2118 h.h2->tp_nsec = ts.tv_nsec;
2119 if (skb_vlan_tag_present(skb)) {
2120 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2121 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2122 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2124 h.h2->tp_vlan_tci = 0;
2125 h.h2->tp_vlan_tpid = 0;
2127 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2128 hdrlen = sizeof(*h.h2);
2131 /* tp_nxt_offset,vlan are already populated above.
2132 * So DONT clear those fields here
2134 h.h3->tp_status |= status;
2135 h.h3->tp_len = skb->len;
2136 h.h3->tp_snaplen = snaplen;
2137 h.h3->tp_mac = macoff;
2138 h.h3->tp_net = netoff;
2139 h.h3->tp_sec = ts.tv_sec;
2140 h.h3->tp_nsec = ts.tv_nsec;
2141 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2142 hdrlen = sizeof(*h.h3);
2148 sll = h.raw + TPACKET_ALIGN(hdrlen);
2149 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2150 sll->sll_family = AF_PACKET;
2151 sll->sll_hatype = dev->type;
2152 sll->sll_protocol = skb->protocol;
2153 sll->sll_pkttype = skb->pkt_type;
2154 if (unlikely(po->origdev))
2155 sll->sll_ifindex = orig_dev->ifindex;
2157 sll->sll_ifindex = dev->ifindex;
2161 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2162 if (po->tp_version <= TPACKET_V2) {
2165 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2168 for (start = h.raw; start < end; start += PAGE_SIZE)
2169 flush_dcache_page(pgv_to_page(start));
2174 if (po->tp_version <= TPACKET_V2) {
2175 __packet_set_status(po, h.raw, status);
2176 sk->sk_data_ready(sk);
2178 prb_clear_blk_fill_status(&po->rx_ring);
2182 if (skb_head != skb->data && skb_shared(skb)) {
2183 skb->data = skb_head;
2191 po->stats.stats1.tp_drops++;
2192 spin_unlock(&sk->sk_receive_queue.lock);
2194 sk->sk_data_ready(sk);
2195 kfree_skb(copy_skb);
2196 goto drop_n_restore;
2199 static void tpacket_destruct_skb(struct sk_buff *skb)
2201 struct packet_sock *po = pkt_sk(skb->sk);
2203 if (likely(po->tx_ring.pg_vec)) {
2207 ph = skb_shinfo(skb)->destructor_arg;
2208 packet_dec_pending(&po->tx_ring);
2210 ts = __packet_set_timestamp(po, ph, skb);
2211 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2217 static bool ll_header_truncated(const struct net_device *dev, int len)
2219 /* net device doesn't like empty head */
2220 if (unlikely(len <= dev->hard_header_len)) {
2221 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2222 current->comm, len, dev->hard_header_len);
2229 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2230 void *frame, struct net_device *dev, int size_max,
2231 __be16 proto, unsigned char *addr, int hlen)
2233 union tpacket_uhdr ph;
2234 int to_write, offset, len, tp_len, nr_frags, len_max;
2235 struct socket *sock = po->sk.sk_socket;
2242 skb->protocol = proto;
2244 skb->priority = po->sk.sk_priority;
2245 skb->mark = po->sk.sk_mark;
2246 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2247 skb_shinfo(skb)->destructor_arg = ph.raw;
2249 switch (po->tp_version) {
2251 tp_len = ph.h2->tp_len;
2254 tp_len = ph.h1->tp_len;
2257 if (unlikely(tp_len > size_max)) {
2258 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2262 skb_reserve(skb, hlen);
2263 skb_reset_network_header(skb);
2265 if (!packet_use_direct_xmit(po))
2266 skb_probe_transport_header(skb, 0);
2267 if (unlikely(po->tp_tx_has_off)) {
2268 int off_min, off_max, off;
2269 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2270 off_max = po->tx_ring.frame_size - tp_len;
2271 if (sock->type == SOCK_DGRAM) {
2272 switch (po->tp_version) {
2274 off = ph.h2->tp_net;
2277 off = ph.h1->tp_net;
2281 switch (po->tp_version) {
2283 off = ph.h2->tp_mac;
2286 off = ph.h1->tp_mac;
2290 if (unlikely((off < off_min) || (off_max < off)))
2292 data = ph.raw + off;
2294 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2298 if (sock->type == SOCK_DGRAM) {
2299 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2301 if (unlikely(err < 0))
2303 } else if (dev->hard_header_len) {
2304 if (ll_header_truncated(dev, tp_len))
2307 skb_push(skb, dev->hard_header_len);
2308 err = skb_store_bits(skb, 0, data,
2309 dev->hard_header_len);
2313 data += dev->hard_header_len;
2314 to_write -= dev->hard_header_len;
2317 offset = offset_in_page(data);
2318 len_max = PAGE_SIZE - offset;
2319 len = ((to_write > len_max) ? len_max : to_write);
2321 skb->data_len = to_write;
2322 skb->len += to_write;
2323 skb->truesize += to_write;
2324 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2326 while (likely(to_write)) {
2327 nr_frags = skb_shinfo(skb)->nr_frags;
2329 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2330 pr_err("Packet exceed the number of skb frags(%lu)\n",
2335 page = pgv_to_page(data);
2337 flush_dcache_page(page);
2339 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2342 len_max = PAGE_SIZE;
2343 len = ((to_write > len_max) ? len_max : to_write);
2349 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2351 struct sk_buff *skb;
2352 struct net_device *dev;
2354 int err, reserve = 0;
2356 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2357 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2358 int tp_len, size_max;
2359 unsigned char *addr;
2361 int status = TP_STATUS_AVAILABLE;
2364 mutex_lock(&po->pg_vec_lock);
2366 if (likely(saddr == NULL)) {
2367 dev = packet_cached_dev_get(po);
2372 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2374 if (msg->msg_namelen < (saddr->sll_halen
2375 + offsetof(struct sockaddr_ll,
2378 proto = saddr->sll_protocol;
2379 addr = saddr->sll_addr;
2380 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2384 if (unlikely(dev == NULL))
2387 if (unlikely(!(dev->flags & IFF_UP)))
2390 reserve = dev->hard_header_len + VLAN_HLEN;
2391 size_max = po->tx_ring.frame_size
2392 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2394 if (size_max > dev->mtu + reserve)
2395 size_max = dev->mtu + reserve;
2398 ph = packet_current_frame(po, &po->tx_ring,
2399 TP_STATUS_SEND_REQUEST);
2400 if (unlikely(ph == NULL)) {
2401 if (need_wait && need_resched())
2406 status = TP_STATUS_SEND_REQUEST;
2407 hlen = LL_RESERVED_SPACE(dev);
2408 tlen = dev->needed_tailroom;
2409 skb = sock_alloc_send_skb(&po->sk,
2410 hlen + tlen + sizeof(struct sockaddr_ll),
2413 if (unlikely(skb == NULL)) {
2414 /* we assume the socket was initially writeable ... */
2415 if (likely(len_sum > 0))
2419 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2421 if (tp_len > dev->mtu + dev->hard_header_len) {
2422 struct ethhdr *ehdr;
2423 /* Earlier code assumed this would be a VLAN pkt,
2424 * double-check this now that we have the actual
2428 skb_reset_mac_header(skb);
2429 ehdr = eth_hdr(skb);
2430 if (ehdr->h_proto != htons(ETH_P_8021Q))
2433 if (unlikely(tp_len < 0)) {
2435 __packet_set_status(po, ph,
2436 TP_STATUS_AVAILABLE);
2437 packet_increment_head(&po->tx_ring);
2441 status = TP_STATUS_WRONG_FORMAT;
2447 packet_pick_tx_queue(dev, skb);
2449 skb->destructor = tpacket_destruct_skb;
2450 __packet_set_status(po, ph, TP_STATUS_SENDING);
2451 packet_inc_pending(&po->tx_ring);
2453 status = TP_STATUS_SEND_REQUEST;
2454 err = po->xmit(skb);
2455 if (unlikely(err > 0)) {
2456 err = net_xmit_errno(err);
2457 if (err && __packet_get_status(po, ph) ==
2458 TP_STATUS_AVAILABLE) {
2459 /* skb was destructed already */
2464 * skb was dropped but not destructed yet;
2465 * let's treat it like congestion or err < 0
2469 packet_increment_head(&po->tx_ring);
2471 } while (likely((ph != NULL) ||
2472 /* Note: packet_read_pending() might be slow if we have
2473 * to call it as it's per_cpu variable, but in fast-path
2474 * we already short-circuit the loop with the first
2475 * condition, and luckily don't have to go that path
2478 (need_wait && packet_read_pending(&po->tx_ring))));
2484 __packet_set_status(po, ph, status);
2489 mutex_unlock(&po->pg_vec_lock);
2493 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2494 size_t reserve, size_t len,
2495 size_t linear, int noblock,
2498 struct sk_buff *skb;
2500 /* Under a page? Don't bother with paged skb. */
2501 if (prepad + len < PAGE_SIZE || !linear)
2504 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2509 skb_reserve(skb, reserve);
2510 skb_put(skb, linear);
2511 skb->data_len = len - linear;
2512 skb->len += len - linear;
2517 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2519 struct sock *sk = sock->sk;
2520 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2521 struct sk_buff *skb;
2522 struct net_device *dev;
2524 unsigned char *addr;
2525 int err, reserve = 0;
2526 struct virtio_net_hdr vnet_hdr = { 0 };
2529 struct packet_sock *po = pkt_sk(sk);
2530 unsigned short gso_type = 0;
2536 * Get and verify the address.
2539 if (likely(saddr == NULL)) {
2540 dev = packet_cached_dev_get(po);
2545 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2547 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2549 proto = saddr->sll_protocol;
2550 addr = saddr->sll_addr;
2551 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2555 if (unlikely(dev == NULL))
2558 if (unlikely(!(dev->flags & IFF_UP)))
2561 if (sock->type == SOCK_RAW)
2562 reserve = dev->hard_header_len;
2563 if (po->has_vnet_hdr) {
2564 vnet_hdr_len = sizeof(vnet_hdr);
2567 if (len < vnet_hdr_len)
2570 len -= vnet_hdr_len;
2573 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2574 if (n != vnet_hdr_len)
2577 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2578 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2579 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2580 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2581 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2582 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2583 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2586 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2589 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2590 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2591 case VIRTIO_NET_HDR_GSO_TCPV4:
2592 gso_type = SKB_GSO_TCPV4;
2594 case VIRTIO_NET_HDR_GSO_TCPV6:
2595 gso_type = SKB_GSO_TCPV6;
2597 case VIRTIO_NET_HDR_GSO_UDP:
2598 gso_type = SKB_GSO_UDP;
2604 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2605 gso_type |= SKB_GSO_TCP_ECN;
2607 if (vnet_hdr.gso_size == 0)
2613 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2614 if (!netif_supports_nofcs(dev)) {
2615 err = -EPROTONOSUPPORT;
2618 extra_len = 4; /* We're doing our own CRC */
2622 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2626 hlen = LL_RESERVED_SPACE(dev);
2627 tlen = dev->needed_tailroom;
2628 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2629 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2630 msg->msg_flags & MSG_DONTWAIT, &err);
2634 skb_set_network_header(skb, reserve);
2637 if (sock->type == SOCK_DGRAM) {
2638 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2639 if (unlikely(offset < 0))
2642 if (ll_header_truncated(dev, len))
2646 /* Returns -EFAULT on error */
2647 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2651 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2653 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2654 /* Earlier code assumed this would be a VLAN pkt,
2655 * double-check this now that we have the actual
2658 struct ethhdr *ehdr;
2659 skb_reset_mac_header(skb);
2660 ehdr = eth_hdr(skb);
2661 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2667 skb->protocol = proto;
2669 skb->priority = sk->sk_priority;
2670 skb->mark = sk->sk_mark;
2672 packet_pick_tx_queue(dev, skb);
2674 if (po->has_vnet_hdr) {
2675 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2676 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2677 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2678 if (!skb_partial_csum_set(skb, s, o)) {
2684 skb_shinfo(skb)->gso_size =
2685 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2686 skb_shinfo(skb)->gso_type = gso_type;
2688 /* Header must be checked, and gso_segs computed. */
2689 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2690 skb_shinfo(skb)->gso_segs = 0;
2692 len += vnet_hdr_len;
2695 if (!packet_use_direct_xmit(po))
2696 skb_probe_transport_header(skb, reserve);
2697 if (unlikely(extra_len == 4))
2700 err = po->xmit(skb);
2701 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2717 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2719 struct sock *sk = sock->sk;
2720 struct packet_sock *po = pkt_sk(sk);
2722 if (po->tx_ring.pg_vec)
2723 return tpacket_snd(po, msg);
2725 return packet_snd(sock, msg, len);
2729 * Close a PACKET socket. This is fairly simple. We immediately go
2730 * to 'closed' state and remove our protocol entry in the device list.
2733 static int packet_release(struct socket *sock)
2735 struct sock *sk = sock->sk;
2736 struct packet_sock *po;
2738 union tpacket_req_u req_u;
2746 mutex_lock(&net->packet.sklist_lock);
2747 sk_del_node_init_rcu(sk);
2748 mutex_unlock(&net->packet.sklist_lock);
2751 sock_prot_inuse_add(net, sk->sk_prot, -1);
2754 spin_lock(&po->bind_lock);
2755 unregister_prot_hook(sk, false);
2756 packet_cached_dev_reset(po);
2758 if (po->prot_hook.dev) {
2759 dev_put(po->prot_hook.dev);
2760 po->prot_hook.dev = NULL;
2762 spin_unlock(&po->bind_lock);
2764 packet_flush_mclist(sk);
2766 if (po->rx_ring.pg_vec) {
2767 memset(&req_u, 0, sizeof(req_u));
2768 packet_set_ring(sk, &req_u, 1, 0);
2771 if (po->tx_ring.pg_vec) {
2772 memset(&req_u, 0, sizeof(req_u));
2773 packet_set_ring(sk, &req_u, 1, 1);
2780 * Now the socket is dead. No more input will appear.
2787 skb_queue_purge(&sk->sk_receive_queue);
2788 packet_free_pending(po);
2789 sk_refcnt_debug_release(sk);
2796 * Attach a packet hook.
2799 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2801 struct packet_sock *po = pkt_sk(sk);
2802 const struct net_device *dev_curr;
2814 spin_lock(&po->bind_lock);
2816 proto_curr = po->prot_hook.type;
2817 dev_curr = po->prot_hook.dev;
2819 need_rehook = proto_curr != proto || dev_curr != dev;
2822 unregister_prot_hook(sk, true);
2825 po->prot_hook.type = proto;
2827 if (po->prot_hook.dev)
2828 dev_put(po->prot_hook.dev);
2830 po->prot_hook.dev = dev;
2832 po->ifindex = dev ? dev->ifindex : 0;
2833 packet_cached_dev_assign(po, dev);
2836 if (proto == 0 || !need_rehook)
2839 if (!dev || (dev->flags & IFF_UP)) {
2840 register_prot_hook(sk);
2842 sk->sk_err = ENETDOWN;
2843 if (!sock_flag(sk, SOCK_DEAD))
2844 sk->sk_error_report(sk);
2848 spin_unlock(&po->bind_lock);
2854 * Bind a packet socket to a device
2857 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2860 struct sock *sk = sock->sk;
2862 struct net_device *dev;
2869 if (addr_len != sizeof(struct sockaddr))
2871 strlcpy(name, uaddr->sa_data, sizeof(name));
2873 dev = dev_get_by_name(sock_net(sk), name);
2875 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2879 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2881 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2882 struct sock *sk = sock->sk;
2883 struct net_device *dev = NULL;
2891 if (addr_len < sizeof(struct sockaddr_ll))
2893 if (sll->sll_family != AF_PACKET)
2896 if (sll->sll_ifindex) {
2898 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2902 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2908 static struct proto packet_proto = {
2910 .owner = THIS_MODULE,
2911 .obj_size = sizeof(struct packet_sock),
2915 * Create a packet of type SOCK_PACKET.
2918 static int packet_create(struct net *net, struct socket *sock, int protocol,
2922 struct packet_sock *po;
2923 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2926 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2928 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2929 sock->type != SOCK_PACKET)
2930 return -ESOCKTNOSUPPORT;
2932 sock->state = SS_UNCONNECTED;
2935 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2939 sock->ops = &packet_ops;
2940 if (sock->type == SOCK_PACKET)
2941 sock->ops = &packet_ops_spkt;
2943 sock_init_data(sock, sk);
2946 sk->sk_family = PF_PACKET;
2948 po->xmit = dev_queue_xmit;
2950 err = packet_alloc_pending(po);
2954 packet_cached_dev_reset(po);
2956 sk->sk_destruct = packet_sock_destruct;
2957 sk_refcnt_debug_inc(sk);
2960 * Attach a protocol block
2963 spin_lock_init(&po->bind_lock);
2964 mutex_init(&po->pg_vec_lock);
2965 po->rollover = NULL;
2966 po->prot_hook.func = packet_rcv;
2968 if (sock->type == SOCK_PACKET)
2969 po->prot_hook.func = packet_rcv_spkt;
2971 po->prot_hook.af_packet_priv = sk;
2974 po->prot_hook.type = proto;
2975 register_prot_hook(sk);
2978 mutex_lock(&net->packet.sklist_lock);
2979 sk_add_node_rcu(sk, &net->packet.sklist);
2980 mutex_unlock(&net->packet.sklist_lock);
2983 sock_prot_inuse_add(net, &packet_proto, 1);
2994 * Pull a packet from our receive queue and hand it to the user.
2995 * If necessary we block.
2998 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3001 struct sock *sk = sock->sk;
3002 struct sk_buff *skb;
3004 int vnet_hdr_len = 0;
3005 unsigned int origlen = 0;
3008 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3012 /* What error should we return now? EUNATTACH? */
3013 if (pkt_sk(sk)->ifindex < 0)
3017 if (flags & MSG_ERRQUEUE) {
3018 err = sock_recv_errqueue(sk, msg, len,
3019 SOL_PACKET, PACKET_TX_TIMESTAMP);
3024 * Call the generic datagram receiver. This handles all sorts
3025 * of horrible races and re-entrancy so we can forget about it
3026 * in the protocol layers.
3028 * Now it will return ENETDOWN, if device have just gone down,
3029 * but then it will block.
3032 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3035 * An error occurred so return it. Because skb_recv_datagram()
3036 * handles the blocking we don't see and worry about blocking
3043 if (pkt_sk(sk)->pressure)
3044 packet_rcv_has_room(pkt_sk(sk), NULL);
3046 if (pkt_sk(sk)->has_vnet_hdr) {
3047 struct virtio_net_hdr vnet_hdr = { 0 };
3050 vnet_hdr_len = sizeof(vnet_hdr);
3051 if (len < vnet_hdr_len)
3054 len -= vnet_hdr_len;
3056 if (skb_is_gso(skb)) {
3057 struct skb_shared_info *sinfo = skb_shinfo(skb);
3059 /* This is a hint as to how much should be linear. */
3061 __cpu_to_virtio16(false, skb_headlen(skb));
3063 __cpu_to_virtio16(false, sinfo->gso_size);
3064 if (sinfo->gso_type & SKB_GSO_TCPV4)
3065 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3066 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3067 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3068 else if (sinfo->gso_type & SKB_GSO_UDP)
3069 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3070 else if (sinfo->gso_type & SKB_GSO_FCOE)
3074 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3075 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3077 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3079 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3080 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3081 vnet_hdr.csum_start = __cpu_to_virtio16(false,
3082 skb_checksum_start_offset(skb));
3083 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3085 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3086 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3087 } /* else everything is zero */
3089 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3094 /* You lose any data beyond the buffer you gave. If it worries
3095 * a user program they can ask the device for its MTU
3101 msg->msg_flags |= MSG_TRUNC;
3104 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3108 if (sock->type != SOCK_PACKET) {
3109 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3111 /* Original length was stored in sockaddr_ll fields */
3112 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3113 sll->sll_family = AF_PACKET;
3114 sll->sll_protocol = skb->protocol;
3117 sock_recv_ts_and_drops(msg, sk, skb);
3119 if (msg->msg_name) {
3120 /* If the address length field is there to be filled
3121 * in, we fill it in now.
3123 if (sock->type == SOCK_PACKET) {
3124 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3125 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3127 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3129 msg->msg_namelen = sll->sll_halen +
3130 offsetof(struct sockaddr_ll, sll_addr);
3132 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3136 if (pkt_sk(sk)->auxdata) {
3137 struct tpacket_auxdata aux;
3139 aux.tp_status = TP_STATUS_USER;
3140 if (skb->ip_summed == CHECKSUM_PARTIAL)
3141 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3142 else if (skb->pkt_type != PACKET_OUTGOING &&
3143 (skb->ip_summed == CHECKSUM_COMPLETE ||
3144 skb_csum_unnecessary(skb)))
3145 aux.tp_status |= TP_STATUS_CSUM_VALID;
3147 aux.tp_len = origlen;
3148 aux.tp_snaplen = skb->len;
3150 aux.tp_net = skb_network_offset(skb);
3151 if (skb_vlan_tag_present(skb)) {
3152 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3153 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3154 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3156 aux.tp_vlan_tci = 0;
3157 aux.tp_vlan_tpid = 0;
3159 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3163 * Free or return the buffer as appropriate. Again this
3164 * hides all the races and re-entrancy issues from us.
3166 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3169 skb_free_datagram(sk, skb);
3174 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3175 int *uaddr_len, int peer)
3177 struct net_device *dev;
3178 struct sock *sk = sock->sk;
3183 uaddr->sa_family = AF_PACKET;
3184 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3186 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3188 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3190 *uaddr_len = sizeof(*uaddr);
3195 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3196 int *uaddr_len, int peer)
3198 struct net_device *dev;
3199 struct sock *sk = sock->sk;
3200 struct packet_sock *po = pkt_sk(sk);
3201 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3206 sll->sll_family = AF_PACKET;
3207 sll->sll_ifindex = po->ifindex;
3208 sll->sll_protocol = po->num;
3209 sll->sll_pkttype = 0;
3211 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3213 sll->sll_hatype = dev->type;
3214 sll->sll_halen = dev->addr_len;
3215 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3217 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3221 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3226 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3230 case PACKET_MR_MULTICAST:
3231 if (i->alen != dev->addr_len)
3234 return dev_mc_add(dev, i->addr);
3236 return dev_mc_del(dev, i->addr);
3238 case PACKET_MR_PROMISC:
3239 return dev_set_promiscuity(dev, what);
3240 case PACKET_MR_ALLMULTI:
3241 return dev_set_allmulti(dev, what);
3242 case PACKET_MR_UNICAST:
3243 if (i->alen != dev->addr_len)
3246 return dev_uc_add(dev, i->addr);
3248 return dev_uc_del(dev, i->addr);
3256 static void packet_dev_mclist_delete(struct net_device *dev,
3257 struct packet_mclist **mlp)
3259 struct packet_mclist *ml;
3261 while ((ml = *mlp) != NULL) {
3262 if (ml->ifindex == dev->ifindex) {
3263 packet_dev_mc(dev, ml, -1);
3271 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3273 struct packet_sock *po = pkt_sk(sk);
3274 struct packet_mclist *ml, *i;
3275 struct net_device *dev;
3281 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3286 if (mreq->mr_alen > dev->addr_len)
3290 i = kmalloc(sizeof(*i), GFP_KERNEL);
3295 for (ml = po->mclist; ml; ml = ml->next) {
3296 if (ml->ifindex == mreq->mr_ifindex &&
3297 ml->type == mreq->mr_type &&
3298 ml->alen == mreq->mr_alen &&
3299 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3301 /* Free the new element ... */
3307 i->type = mreq->mr_type;
3308 i->ifindex = mreq->mr_ifindex;
3309 i->alen = mreq->mr_alen;
3310 memcpy(i->addr, mreq->mr_address, i->alen);
3312 i->next = po->mclist;
3314 err = packet_dev_mc(dev, i, 1);
3316 po->mclist = i->next;
3325 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3327 struct packet_mclist *ml, **mlp;
3331 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3332 if (ml->ifindex == mreq->mr_ifindex &&
3333 ml->type == mreq->mr_type &&
3334 ml->alen == mreq->mr_alen &&
3335 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3336 if (--ml->count == 0) {
3337 struct net_device *dev;
3339 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3341 packet_dev_mc(dev, ml, -1);
3351 static void packet_flush_mclist(struct sock *sk)
3353 struct packet_sock *po = pkt_sk(sk);
3354 struct packet_mclist *ml;
3360 while ((ml = po->mclist) != NULL) {
3361 struct net_device *dev;
3363 po->mclist = ml->next;
3364 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3366 packet_dev_mc(dev, ml, -1);
3373 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3375 struct sock *sk = sock->sk;
3376 struct packet_sock *po = pkt_sk(sk);
3379 if (level != SOL_PACKET)
3380 return -ENOPROTOOPT;
3383 case PACKET_ADD_MEMBERSHIP:
3384 case PACKET_DROP_MEMBERSHIP:
3386 struct packet_mreq_max mreq;
3388 memset(&mreq, 0, sizeof(mreq));
3389 if (len < sizeof(struct packet_mreq))
3391 if (len > sizeof(mreq))
3393 if (copy_from_user(&mreq, optval, len))
3395 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3397 if (optname == PACKET_ADD_MEMBERSHIP)
3398 ret = packet_mc_add(sk, &mreq);
3400 ret = packet_mc_drop(sk, &mreq);
3404 case PACKET_RX_RING:
3405 case PACKET_TX_RING:
3407 union tpacket_req_u req_u;
3410 switch (po->tp_version) {
3413 len = sizeof(req_u.req);
3417 len = sizeof(req_u.req3);
3422 if (pkt_sk(sk)->has_vnet_hdr)
3424 if (copy_from_user(&req_u.req, optval, len))
3426 return packet_set_ring(sk, &req_u, 0,
3427 optname == PACKET_TX_RING);
3429 case PACKET_COPY_THRESH:
3433 if (optlen != sizeof(val))
3435 if (copy_from_user(&val, optval, sizeof(val)))
3438 pkt_sk(sk)->copy_thresh = val;
3441 case PACKET_VERSION:
3445 if (optlen != sizeof(val))
3447 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3449 if (copy_from_user(&val, optval, sizeof(val)))
3455 po->tp_version = val;
3461 case PACKET_RESERVE:
3465 if (optlen != sizeof(val))
3467 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3469 if (copy_from_user(&val, optval, sizeof(val)))
3471 po->tp_reserve = val;
3478 if (optlen != sizeof(val))
3480 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3482 if (copy_from_user(&val, optval, sizeof(val)))
3484 po->tp_loss = !!val;
3487 case PACKET_AUXDATA:
3491 if (optlen < sizeof(val))
3493 if (copy_from_user(&val, optval, sizeof(val)))
3496 po->auxdata = !!val;
3499 case PACKET_ORIGDEV:
3503 if (optlen < sizeof(val))
3505 if (copy_from_user(&val, optval, sizeof(val)))
3508 po->origdev = !!val;
3511 case PACKET_VNET_HDR:
3515 if (sock->type != SOCK_RAW)
3517 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3519 if (optlen < sizeof(val))
3521 if (copy_from_user(&val, optval, sizeof(val)))
3524 po->has_vnet_hdr = !!val;
3527 case PACKET_TIMESTAMP:
3531 if (optlen != sizeof(val))
3533 if (copy_from_user(&val, optval, sizeof(val)))
3536 po->tp_tstamp = val;
3543 if (optlen != sizeof(val))
3545 if (copy_from_user(&val, optval, sizeof(val)))
3548 return fanout_add(sk, val & 0xffff, val >> 16);
3550 case PACKET_TX_HAS_OFF:
3554 if (optlen != sizeof(val))
3556 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3558 if (copy_from_user(&val, optval, sizeof(val)))
3560 po->tp_tx_has_off = !!val;
3563 case PACKET_QDISC_BYPASS:
3567 if (optlen != sizeof(val))
3569 if (copy_from_user(&val, optval, sizeof(val)))
3572 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3576 return -ENOPROTOOPT;
3580 static int packet_getsockopt(struct socket *sock, int level, int optname,
3581 char __user *optval, int __user *optlen)
3584 int val, lv = sizeof(val);
3585 struct sock *sk = sock->sk;
3586 struct packet_sock *po = pkt_sk(sk);
3588 union tpacket_stats_u st;
3589 struct tpacket_rollover_stats rstats;
3591 if (level != SOL_PACKET)
3592 return -ENOPROTOOPT;
3594 if (get_user(len, optlen))
3601 case PACKET_STATISTICS:
3602 spin_lock_bh(&sk->sk_receive_queue.lock);
3603 memcpy(&st, &po->stats, sizeof(st));
3604 memset(&po->stats, 0, sizeof(po->stats));
3605 spin_unlock_bh(&sk->sk_receive_queue.lock);
3607 if (po->tp_version == TPACKET_V3) {
3608 lv = sizeof(struct tpacket_stats_v3);
3609 st.stats3.tp_packets += st.stats3.tp_drops;
3612 lv = sizeof(struct tpacket_stats);
3613 st.stats1.tp_packets += st.stats1.tp_drops;
3618 case PACKET_AUXDATA:
3621 case PACKET_ORIGDEV:
3624 case PACKET_VNET_HDR:
3625 val = po->has_vnet_hdr;
3627 case PACKET_VERSION:
3628 val = po->tp_version;
3631 if (len > sizeof(int))
3633 if (copy_from_user(&val, optval, len))
3637 val = sizeof(struct tpacket_hdr);
3640 val = sizeof(struct tpacket2_hdr);
3643 val = sizeof(struct tpacket3_hdr);
3649 case PACKET_RESERVE:
3650 val = po->tp_reserve;
3655 case PACKET_TIMESTAMP:
3656 val = po->tp_tstamp;
3660 ((u32)po->fanout->id |
3661 ((u32)po->fanout->type << 16) |
3662 ((u32)po->fanout->flags << 24)) :
3665 case PACKET_ROLLOVER_STATS:
3668 rstats.tp_all = atomic_long_read(&po->rollover->num);
3669 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3670 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3672 lv = sizeof(rstats);
3674 case PACKET_TX_HAS_OFF:
3675 val = po->tp_tx_has_off;
3677 case PACKET_QDISC_BYPASS:
3678 val = packet_use_direct_xmit(po);
3681 return -ENOPROTOOPT;
3686 if (put_user(len, optlen))
3688 if (copy_to_user(optval, data, len))
3694 static int packet_notifier(struct notifier_block *this,
3695 unsigned long msg, void *ptr)
3698 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3699 struct net *net = dev_net(dev);
3702 sk_for_each_rcu(sk, &net->packet.sklist) {
3703 struct packet_sock *po = pkt_sk(sk);
3706 case NETDEV_UNREGISTER:
3708 packet_dev_mclist_delete(dev, &po->mclist);
3712 if (dev->ifindex == po->ifindex) {
3713 spin_lock(&po->bind_lock);
3715 __unregister_prot_hook(sk, false);
3716 sk->sk_err = ENETDOWN;
3717 if (!sock_flag(sk, SOCK_DEAD))
3718 sk->sk_error_report(sk);
3720 if (msg == NETDEV_UNREGISTER) {
3721 packet_cached_dev_reset(po);
3723 if (po->prot_hook.dev)
3724 dev_put(po->prot_hook.dev);
3725 po->prot_hook.dev = NULL;
3727 spin_unlock(&po->bind_lock);
3731 if (dev->ifindex == po->ifindex) {
3732 spin_lock(&po->bind_lock);
3734 register_prot_hook(sk);
3735 spin_unlock(&po->bind_lock);
3745 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3748 struct sock *sk = sock->sk;
3753 int amount = sk_wmem_alloc_get(sk);
3755 return put_user(amount, (int __user *)arg);
3759 struct sk_buff *skb;
3762 spin_lock_bh(&sk->sk_receive_queue.lock);
3763 skb = skb_peek(&sk->sk_receive_queue);
3766 spin_unlock_bh(&sk->sk_receive_queue.lock);
3767 return put_user(amount, (int __user *)arg);
3770 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3772 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3782 case SIOCGIFBRDADDR:
3783 case SIOCSIFBRDADDR:
3784 case SIOCGIFNETMASK:
3785 case SIOCSIFNETMASK:
3786 case SIOCGIFDSTADDR:
3787 case SIOCSIFDSTADDR:
3789 return inet_dgram_ops.ioctl(sock, cmd, arg);
3793 return -ENOIOCTLCMD;
3798 static unsigned int packet_poll(struct file *file, struct socket *sock,
3801 struct sock *sk = sock->sk;
3802 struct packet_sock *po = pkt_sk(sk);
3803 unsigned int mask = datagram_poll(file, sock, wait);
3805 spin_lock_bh(&sk->sk_receive_queue.lock);
3806 if (po->rx_ring.pg_vec) {
3807 if (!packet_previous_rx_frame(po, &po->rx_ring,
3809 mask |= POLLIN | POLLRDNORM;
3811 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3813 spin_unlock_bh(&sk->sk_receive_queue.lock);
3814 spin_lock_bh(&sk->sk_write_queue.lock);
3815 if (po->tx_ring.pg_vec) {
3816 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3817 mask |= POLLOUT | POLLWRNORM;
3819 spin_unlock_bh(&sk->sk_write_queue.lock);
3824 /* Dirty? Well, I still did not learn better way to account
3828 static void packet_mm_open(struct vm_area_struct *vma)
3830 struct file *file = vma->vm_file;
3831 struct socket *sock = file->private_data;
3832 struct sock *sk = sock->sk;
3835 atomic_inc(&pkt_sk(sk)->mapped);
3838 static void packet_mm_close(struct vm_area_struct *vma)
3840 struct file *file = vma->vm_file;
3841 struct socket *sock = file->private_data;
3842 struct sock *sk = sock->sk;
3845 atomic_dec(&pkt_sk(sk)->mapped);
3848 static const struct vm_operations_struct packet_mmap_ops = {
3849 .open = packet_mm_open,
3850 .close = packet_mm_close,
3853 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3858 for (i = 0; i < len; i++) {
3859 if (likely(pg_vec[i].buffer)) {
3860 if (is_vmalloc_addr(pg_vec[i].buffer))
3861 vfree(pg_vec[i].buffer);
3863 free_pages((unsigned long)pg_vec[i].buffer,
3865 pg_vec[i].buffer = NULL;
3871 static char *alloc_one_pg_vec_page(unsigned long order)
3874 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3875 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3877 buffer = (char *) __get_free_pages(gfp_flags, order);
3881 /* __get_free_pages failed, fall back to vmalloc */
3882 buffer = vzalloc((1 << order) * PAGE_SIZE);
3886 /* vmalloc failed, lets dig into swap here */
3887 gfp_flags &= ~__GFP_NORETRY;
3888 buffer = (char *) __get_free_pages(gfp_flags, order);
3892 /* complete and utter failure */
3896 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3898 unsigned int block_nr = req->tp_block_nr;
3902 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3903 if (unlikely(!pg_vec))
3906 for (i = 0; i < block_nr; i++) {
3907 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3908 if (unlikely(!pg_vec[i].buffer))
3909 goto out_free_pgvec;
3916 free_pg_vec(pg_vec, order, block_nr);
3921 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3922 int closing, int tx_ring)
3924 struct pgv *pg_vec = NULL;
3925 struct packet_sock *po = pkt_sk(sk);
3926 int was_running, order = 0;
3927 struct packet_ring_buffer *rb;
3928 struct sk_buff_head *rb_queue;
3931 /* Added to avoid minimal code churn */
3932 struct tpacket_req *req = &req_u->req;
3934 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3935 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3936 WARN(1, "Tx-ring is not supported.\n");
3940 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3941 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3945 if (atomic_read(&po->mapped))
3947 if (packet_read_pending(rb))
3951 if (req->tp_block_nr) {
3952 /* Sanity tests and some calculations */
3954 if (unlikely(rb->pg_vec))
3957 switch (po->tp_version) {
3959 po->tp_hdrlen = TPACKET_HDRLEN;
3962 po->tp_hdrlen = TPACKET2_HDRLEN;
3965 po->tp_hdrlen = TPACKET3_HDRLEN;
3970 if (unlikely((int)req->tp_block_size <= 0))
3972 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3974 if (po->tp_version >= TPACKET_V3 &&
3975 (int)(req->tp_block_size -
3976 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3978 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3981 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3984 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3985 if (unlikely(rb->frames_per_block <= 0))
3987 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3992 order = get_order(req->tp_block_size);
3993 pg_vec = alloc_pg_vec(req, order);
3994 if (unlikely(!pg_vec))
3996 switch (po->tp_version) {
3998 /* Transmit path is not supported. We checked
3999 * it above but just being paranoid
4002 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
4011 if (unlikely(req->tp_frame_nr))
4017 /* Detach socket from network */
4018 spin_lock(&po->bind_lock);
4019 was_running = po->running;
4023 __unregister_prot_hook(sk, false);
4025 spin_unlock(&po->bind_lock);
4030 mutex_lock(&po->pg_vec_lock);
4031 if (closing || atomic_read(&po->mapped) == 0) {
4033 spin_lock_bh(&rb_queue->lock);
4034 swap(rb->pg_vec, pg_vec);
4035 rb->frame_max = (req->tp_frame_nr - 1);
4037 rb->frame_size = req->tp_frame_size;
4038 spin_unlock_bh(&rb_queue->lock);
4040 swap(rb->pg_vec_order, order);
4041 swap(rb->pg_vec_len, req->tp_block_nr);
4043 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4044 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4045 tpacket_rcv : packet_rcv;
4046 skb_queue_purge(rb_queue);
4047 if (atomic_read(&po->mapped))
4048 pr_err("packet_mmap: vma is busy: %d\n",
4049 atomic_read(&po->mapped));
4051 mutex_unlock(&po->pg_vec_lock);
4053 spin_lock(&po->bind_lock);
4056 register_prot_hook(sk);
4058 spin_unlock(&po->bind_lock);
4059 if (closing && (po->tp_version > TPACKET_V2)) {
4060 /* Because we don't support block-based V3 on tx-ring */
4062 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4067 free_pg_vec(pg_vec, order, req->tp_block_nr);
4072 static int packet_mmap(struct file *file, struct socket *sock,
4073 struct vm_area_struct *vma)
4075 struct sock *sk = sock->sk;
4076 struct packet_sock *po = pkt_sk(sk);
4077 unsigned long size, expected_size;
4078 struct packet_ring_buffer *rb;
4079 unsigned long start;
4086 mutex_lock(&po->pg_vec_lock);
4089 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4091 expected_size += rb->pg_vec_len
4097 if (expected_size == 0)
4100 size = vma->vm_end - vma->vm_start;
4101 if (size != expected_size)
4104 start = vma->vm_start;
4105 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4106 if (rb->pg_vec == NULL)
4109 for (i = 0; i < rb->pg_vec_len; i++) {
4111 void *kaddr = rb->pg_vec[i].buffer;
4114 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4115 page = pgv_to_page(kaddr);
4116 err = vm_insert_page(vma, start, page);
4125 atomic_inc(&po->mapped);
4126 vma->vm_ops = &packet_mmap_ops;
4130 mutex_unlock(&po->pg_vec_lock);
4134 static const struct proto_ops packet_ops_spkt = {
4135 .family = PF_PACKET,
4136 .owner = THIS_MODULE,
4137 .release = packet_release,
4138 .bind = packet_bind_spkt,
4139 .connect = sock_no_connect,
4140 .socketpair = sock_no_socketpair,
4141 .accept = sock_no_accept,
4142 .getname = packet_getname_spkt,
4143 .poll = datagram_poll,
4144 .ioctl = packet_ioctl,
4145 .listen = sock_no_listen,
4146 .shutdown = sock_no_shutdown,
4147 .setsockopt = sock_no_setsockopt,
4148 .getsockopt = sock_no_getsockopt,
4149 .sendmsg = packet_sendmsg_spkt,
4150 .recvmsg = packet_recvmsg,
4151 .mmap = sock_no_mmap,
4152 .sendpage = sock_no_sendpage,
4155 static const struct proto_ops packet_ops = {
4156 .family = PF_PACKET,
4157 .owner = THIS_MODULE,
4158 .release = packet_release,
4159 .bind = packet_bind,
4160 .connect = sock_no_connect,
4161 .socketpair = sock_no_socketpair,
4162 .accept = sock_no_accept,
4163 .getname = packet_getname,
4164 .poll = packet_poll,
4165 .ioctl = packet_ioctl,
4166 .listen = sock_no_listen,
4167 .shutdown = sock_no_shutdown,
4168 .setsockopt = packet_setsockopt,
4169 .getsockopt = packet_getsockopt,
4170 .sendmsg = packet_sendmsg,
4171 .recvmsg = packet_recvmsg,
4172 .mmap = packet_mmap,
4173 .sendpage = sock_no_sendpage,
4176 static const struct net_proto_family packet_family_ops = {
4177 .family = PF_PACKET,
4178 .create = packet_create,
4179 .owner = THIS_MODULE,
4182 static struct notifier_block packet_netdev_notifier = {
4183 .notifier_call = packet_notifier,
4186 #ifdef CONFIG_PROC_FS
4188 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4191 struct net *net = seq_file_net(seq);
4194 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4197 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4199 struct net *net = seq_file_net(seq);
4200 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4203 static void packet_seq_stop(struct seq_file *seq, void *v)
4209 static int packet_seq_show(struct seq_file *seq, void *v)
4211 if (v == SEQ_START_TOKEN)
4212 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4214 struct sock *s = sk_entry(v);
4215 const struct packet_sock *po = pkt_sk(s);
4218 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4220 atomic_read(&s->sk_refcnt),
4225 atomic_read(&s->sk_rmem_alloc),
4226 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4233 static const struct seq_operations packet_seq_ops = {
4234 .start = packet_seq_start,
4235 .next = packet_seq_next,
4236 .stop = packet_seq_stop,
4237 .show = packet_seq_show,
4240 static int packet_seq_open(struct inode *inode, struct file *file)
4242 return seq_open_net(inode, file, &packet_seq_ops,
4243 sizeof(struct seq_net_private));
4246 static const struct file_operations packet_seq_fops = {
4247 .owner = THIS_MODULE,
4248 .open = packet_seq_open,
4250 .llseek = seq_lseek,
4251 .release = seq_release_net,
4256 static int __net_init packet_net_init(struct net *net)
4258 mutex_init(&net->packet.sklist_lock);
4259 INIT_HLIST_HEAD(&net->packet.sklist);
4261 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4267 static void __net_exit packet_net_exit(struct net *net)
4269 remove_proc_entry("packet", net->proc_net);
4272 static struct pernet_operations packet_net_ops = {
4273 .init = packet_net_init,
4274 .exit = packet_net_exit,
4278 static void __exit packet_exit(void)
4280 unregister_netdevice_notifier(&packet_netdev_notifier);
4281 unregister_pernet_subsys(&packet_net_ops);
4282 sock_unregister(PF_PACKET);
4283 proto_unregister(&packet_proto);
4286 static int __init packet_init(void)
4288 int rc = proto_register(&packet_proto, 0);
4293 sock_register(&packet_family_ops);
4294 register_pernet_subsys(&packet_net_ops);
4295 register_netdevice_notifier(&packet_netdev_notifier);
4300 module_init(packet_init);
4301 module_exit(packet_exit);
4302 MODULE_LICENSE("GPL");
4303 MODULE_ALIAS_NETPROTO(PF_PACKET);