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>
95 #include <linux/bpf.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 vio_le() virtio_legacy_is_little_endian()
235 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
237 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
249 static int packet_direct_xmit(struct sk_buff *skb)
251 struct net_device *dev = skb->dev;
252 struct sk_buff *orig_skb = skb;
253 struct netdev_queue *txq;
254 int ret = NETDEV_TX_BUSY;
256 if (unlikely(!netif_running(dev) ||
257 !netif_carrier_ok(dev)))
260 skb = validate_xmit_skb_list(skb, dev);
264 txq = skb_get_tx_queue(dev, skb);
268 HARD_TX_LOCK(dev, txq, smp_processor_id());
269 if (!netif_xmit_frozen_or_drv_stopped(txq))
270 ret = netdev_start_xmit(skb, dev, txq, false);
271 HARD_TX_UNLOCK(dev, txq);
275 if (!dev_xmit_complete(ret))
280 atomic_long_inc(&dev->tx_dropped);
282 return NET_XMIT_DROP;
285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 struct net_device *dev;
290 dev = rcu_dereference(po->cached_dev);
298 static void packet_cached_dev_assign(struct packet_sock *po,
299 struct net_device *dev)
301 rcu_assign_pointer(po->cached_dev, dev);
304 static void packet_cached_dev_reset(struct packet_sock *po)
306 RCU_INIT_POINTER(po->cached_dev, NULL);
309 static bool packet_use_direct_xmit(const struct packet_sock *po)
311 return po->xmit == packet_direct_xmit;
314 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
316 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
319 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321 const struct net_device_ops *ops = dev->netdev_ops;
324 if (ops->ndo_select_queue) {
325 queue_index = ops->ndo_select_queue(dev, skb, NULL,
326 __packet_pick_tx_queue);
327 queue_index = netdev_cap_txqueue(dev, queue_index);
329 queue_index = __packet_pick_tx_queue(dev, skb);
332 skb_set_queue_mapping(skb, queue_index);
335 /* register_prot_hook must be invoked with the po->bind_lock held,
336 * or from a context in which asynchronous accesses to the packet
337 * socket is not possible (packet_create()).
339 static void register_prot_hook(struct sock *sk)
341 struct packet_sock *po = pkt_sk(sk);
345 __fanout_link(sk, po);
347 dev_add_pack(&po->prot_hook);
354 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
355 * held. If the sync parameter is true, we will temporarily drop
356 * the po->bind_lock and do a synchronize_net to make sure no
357 * asynchronous packet processing paths still refer to the elements
358 * of po->prot_hook. If the sync parameter is false, it is the
359 * callers responsibility to take care of this.
361 static void __unregister_prot_hook(struct sock *sk, bool sync)
363 struct packet_sock *po = pkt_sk(sk);
368 __fanout_unlink(sk, po);
370 __dev_remove_pack(&po->prot_hook);
375 spin_unlock(&po->bind_lock);
377 spin_lock(&po->bind_lock);
381 static void unregister_prot_hook(struct sock *sk, bool sync)
383 struct packet_sock *po = pkt_sk(sk);
386 __unregister_prot_hook(sk, sync);
389 static inline struct page * __pure pgv_to_page(void *addr)
391 if (is_vmalloc_addr(addr))
392 return vmalloc_to_page(addr);
393 return virt_to_page(addr);
396 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
398 union tpacket_uhdr h;
401 switch (po->tp_version) {
403 h.h1->tp_status = status;
404 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
407 h.h2->tp_status = status;
408 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
412 WARN(1, "TPACKET version not supported.\n");
419 static int __packet_get_status(struct packet_sock *po, void *frame)
421 union tpacket_uhdr h;
426 switch (po->tp_version) {
428 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
429 return h.h1->tp_status;
431 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
432 return h.h2->tp_status;
435 WARN(1, "TPACKET version not supported.\n");
441 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
444 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
447 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
448 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
449 return TP_STATUS_TS_RAW_HARDWARE;
451 if (ktime_to_timespec_cond(skb->tstamp, ts))
452 return TP_STATUS_TS_SOFTWARE;
457 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
460 union tpacket_uhdr h;
464 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
468 switch (po->tp_version) {
470 h.h1->tp_sec = ts.tv_sec;
471 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
474 h.h2->tp_sec = ts.tv_sec;
475 h.h2->tp_nsec = ts.tv_nsec;
479 WARN(1, "TPACKET version not supported.\n");
483 /* one flush is safe, as both fields always lie on the same cacheline */
484 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
490 static void *packet_lookup_frame(struct packet_sock *po,
491 struct packet_ring_buffer *rb,
492 unsigned int position,
495 unsigned int pg_vec_pos, frame_offset;
496 union tpacket_uhdr h;
498 pg_vec_pos = position / rb->frames_per_block;
499 frame_offset = position % rb->frames_per_block;
501 h.raw = rb->pg_vec[pg_vec_pos].buffer +
502 (frame_offset * rb->frame_size);
504 if (status != __packet_get_status(po, h.raw))
510 static void *packet_current_frame(struct packet_sock *po,
511 struct packet_ring_buffer *rb,
514 return packet_lookup_frame(po, rb, rb->head, status);
517 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
519 del_timer_sync(&pkc->retire_blk_timer);
522 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 = 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)
548 struct tpacket_kbdq_core *pkc;
550 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
551 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
554 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
555 int blk_size_in_bytes)
557 struct net_device *dev;
558 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
559 struct ethtool_cmd ecmd;
564 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
565 if (unlikely(!dev)) {
567 return DEFAULT_PRB_RETIRE_TOV;
569 err = __ethtool_get_settings(dev, &ecmd);
570 speed = ethtool_cmd_speed(&ecmd);
574 * If the link speed is so slow you don't really
575 * need to worry about perf anyways
577 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
578 return DEFAULT_PRB_RETIRE_TOV;
585 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598 union tpacket_req_u *req_u)
600 p1->feature_req_word = req_u->req3.tp_feature_req_word;
603 static void init_prb_bdqc(struct packet_sock *po,
604 struct packet_ring_buffer *rb,
606 union tpacket_req_u *req_u)
608 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609 struct tpacket_block_desc *pbd;
611 memset(p1, 0x0, sizeof(*p1));
613 p1->knxt_seq_num = 1;
615 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616 p1->pkblk_start = pg_vec[0].buffer;
617 p1->kblk_size = req_u->req3.tp_block_size;
618 p1->knum_blocks = req_u->req3.tp_block_nr;
619 p1->hdrlen = po->tp_hdrlen;
620 p1->version = po->tp_version;
621 p1->last_kactive_blk_num = 0;
622 po->stats.stats3.tp_freeze_q_cnt = 0;
623 if (req_u->req3.tp_retire_blk_tov)
624 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627 req_u->req3.tp_block_size);
628 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632 prb_init_ft_ops(p1, req_u);
633 prb_setup_retire_blk_timer(po);
634 prb_open_block(p1, pbd);
637 /* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
642 mod_timer(&pkc->retire_blk_timer,
643 jiffies + pkc->tov_in_jiffies);
644 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
672 struct packet_sock *po = (struct packet_sock *)data;
673 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
675 struct tpacket_block_desc *pbd;
677 spin_lock(&po->sk.sk_receive_queue.lock);
679 frozen = prb_queue_frozen(pkc);
680 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682 if (unlikely(pkc->delete_blk_timer))
685 /* We only need to plug the race when the block is partially filled.
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
694 if (BLOCK_NUM_PKTS(pbd)) {
695 while (atomic_read(&pkc->blk_fill_in_prog)) {
696 /* Waiting for skb_copy_bits to finish... */
701 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 if (!BLOCK_NUM_PKTS(pbd)) {
704 /* An empty block. Just refresh the timer. */
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
713 /* Case 1. Queue was frozen because user-space was
716 if (prb_curr_blk_in_use(pkc, pbd)) {
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
730 prb_open_block(pkc, pbd);
737 _prb_refresh_rx_retire_blk_timer(pkc);
740 spin_unlock(&po->sk.sk_receive_queue.lock);
743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744 struct tpacket_block_desc *pbd1, __u32 status)
746 /* Flush everything minus the block header */
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753 /* Skip the block header(we know header WILL fit in 4K) */
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
763 /* Now update the block status. */
765 BLOCK_STATUS(pbd1) = status;
767 /* Flush the block header */
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 flush_dcache_page(pgv_to_page(start));
781 * 2) Increment active_blk_num
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
788 struct packet_sock *po, unsigned int stat)
790 __u32 status = TP_STATUS_USER | stat;
792 struct tpacket3_hdr *last_pkt;
793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794 struct sock *sk = &po->sk;
796 if (po->stats.stats3.tp_drops)
797 status |= TP_STATUS_LOSING;
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
807 /* Ok, we tmo'd - so get the current time.
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
814 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
820 /* Flush the block */
821 prb_flush_block(pkc1, pbd1, status);
823 sk->sk_data_ready(sk);
825 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
830 pkc->reset_pending_on_curr_blk = 0;
834 * Side effect of opening a block:
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841 struct tpacket_block_desc *pbd1)
844 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
852 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853 BLOCK_NUM_PKTS(pbd1) = 0;
854 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
858 h1->ts_first_pkt.ts_sec = ts.tv_sec;
859 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 pkc1->pkblk_start = (char *)pbd1;
862 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 pbd1->version = pkc1->version;
868 pkc1->prev = pkc1->nxt_offset;
869 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 prb_thaw_queue(pkc1);
872 _prb_refresh_rx_retire_blk_timer(pkc1);
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po)
903 pkc->reset_pending_on_curr_blk = 1;
904 po->stats.stats3.tp_freeze_q_cnt++;
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 struct packet_sock *po)
918 struct tpacket_block_desc *pbd;
922 /* 1. Get current block num */
923 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 prb_freeze_queue(pkc, po);
933 * open this block and return the offset where the first packet
934 * needs to get stored.
936 prb_open_block(pkc, pbd);
937 return (void *)pkc->nxt_offset;
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 struct packet_sock *po, unsigned int status)
943 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
956 if (!(status & TP_STATUS_BLK_TMO)) {
957 while (atomic_read(&pkc->blk_fill_in_prog)) {
958 /* Waiting for skb_copy_bits to finish... */
962 prb_close_block(pkc, pbd, po, status);
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968 struct tpacket_block_desc *pbd)
970 return TP_STATUS_USER & BLOCK_STATUS(pbd);
973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
975 return pkc->reset_pending_on_curr_blk;
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
981 atomic_dec(&pkc->blk_fill_in_prog);
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 struct tpacket3_hdr *ppd)
987 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 struct tpacket3_hdr *ppd)
993 ppd->hv1.tp_rxhash = 0;
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 struct tpacket3_hdr *ppd)
999 if (skb_vlan_tag_present(pkc->skb)) {
1000 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004 ppd->hv1.tp_vlan_tci = 0;
1005 ppd->hv1.tp_vlan_tpid = 0;
1006 ppd->tp_status = TP_STATUS_AVAILABLE;
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1013 ppd->hv1.tp_padding = 0;
1014 prb_fill_vlan_info(pkc, ppd);
1016 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017 prb_fill_rxhash(pkc, ppd);
1019 prb_clear_rxhash(pkc, ppd);
1022 static void prb_fill_curr_block(char *curr,
1023 struct tpacket_kbdq_core *pkc,
1024 struct tpacket_block_desc *pbd,
1027 struct tpacket3_hdr *ppd;
1029 ppd = (struct tpacket3_hdr *)curr;
1030 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_NUM_PKTS(pbd) += 1;
1035 atomic_inc(&pkc->blk_fill_in_prog);
1036 prb_run_all_ft_ops(pkc, ppd);
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041 struct sk_buff *skb,
1046 struct tpacket_kbdq_core *pkc;
1047 struct tpacket_block_desc *pbd;
1050 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc)) {
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1059 if (prb_curr_blk_in_use(pkc, pbd)) {
1060 /* Can't record this packet */
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1069 prb_open_block(pkc, pbd);
1074 curr = pkc->nxt_offset;
1076 end = (char *)pbd + pkc->kblk_size;
1078 /* first try the current block */
1079 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080 prb_fill_curr_block(curr, pkc, pbd, len);
1081 return (void *)curr;
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc, po, 0);
1087 /* Now, try to dispatch the next block */
1088 curr = (char *)prb_dispatch_next_block(pkc, po);
1090 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103 struct sk_buff *skb,
1104 int status, unsigned int len)
1107 switch (po->tp_version) {
1110 curr = packet_lookup_frame(po, &po->rx_ring,
1111 po->rx_ring.head, status);
1114 return __packet_lookup_frame_in_block(po, skb, status, len);
1116 WARN(1, "TPACKET version not supported\n");
1122 static void *prb_lookup_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1127 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1128 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130 if (status != BLOCK_STATUS(pbd))
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1138 if (rb->prb_bdqc.kactive_blk_num)
1139 prev = rb->prb_bdqc.kactive_blk_num-1;
1141 prev = rb->prb_bdqc.knum_blocks-1;
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1150 unsigned int previous = prb_previous_blk_num(rb);
1151 return prb_lookup_block(po, rb, previous, status);
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 if (po->tp_version <= TPACKET_V2)
1159 return packet_previous_frame(po, rb, status);
1161 return __prb_previous_block(po, rb, status);
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165 struct packet_ring_buffer *rb)
1167 switch (po->tp_version) {
1170 return packet_increment_head(rb);
1173 WARN(1, "TPACKET version not supported.\n");
1179 static void *packet_previous_frame(struct packet_sock *po,
1180 struct packet_ring_buffer *rb,
1183 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184 return packet_lookup_frame(po, rb, previous, status);
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 this_cpu_inc(*rb->pending_refcnt);
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 this_cpu_dec(*rb->pending_refcnt);
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 unsigned int refcnt = 0;
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb->pending_refcnt == NULL)
1211 for_each_possible_cpu(cpu)
1212 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217 static int packet_alloc_pending(struct packet_sock *po)
1219 po->rx_ring.pending_refcnt = NULL;
1221 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1228 static void packet_free_pending(struct packet_sock *po)
1230 free_percpu(po->tx_ring.pending_refcnt);
1233 #define ROOM_POW_OFF 2
1234 #define ROOM_NONE 0x0
1235 #define ROOM_LOW 0x1
1236 #define ROOM_NORMAL 0x2
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1242 len = po->rx_ring.frame_max + 1;
1243 idx = po->rx_ring.head;
1245 idx += len >> pow_off;
1248 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1255 len = po->rx_ring.prb_bdqc.knum_blocks;
1256 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 idx += len >> pow_off;
1261 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 struct sock *sk = &po->sk;
1267 int ret = ROOM_NONE;
1269 if (po->prot_hook.func != tpacket_rcv) {
1270 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271 - (skb ? skb->truesize : 0);
1272 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280 if (po->tp_version == TPACKET_V3) {
1281 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 else if (__tpacket_v3_has_room(po, 0))
1286 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_has_room(po, 0))
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1300 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301 ret = __packet_rcv_has_room(po, skb);
1302 has_room = ret == ROOM_NORMAL;
1303 if (po->pressure == has_room)
1304 po->pressure = !has_room;
1305 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1310 static void packet_sock_destruct(struct sock *sk)
1312 skb_queue_purge(&sk->sk_error_queue);
1314 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317 if (!sock_flag(sk, SOCK_DEAD)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk);
1322 sk_refcnt_debug_dec(sk);
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1330 rxhash = skb_get_hash(skb);
1331 for (i = 0; i < ROLLOVER_HLEN; i++)
1332 if (po->rollover->history[i] == rxhash)
1335 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336 return count > (ROLLOVER_HLEN >> 1);
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340 struct sk_buff *skb,
1343 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 unsigned int val = atomic_inc_return(&f->rr_cur);
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356 struct sk_buff *skb,
1359 return smp_processor_id() % num;
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return prandom_u32_max(num);
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370 struct sk_buff *skb,
1371 unsigned int idx, bool try_self,
1374 struct packet_sock *po, *po_next, *po_skip = NULL;
1375 unsigned int i, j, room = ROOM_NONE;
1377 po = pkt_sk(f->arr[idx]);
1380 room = packet_rcv_has_room(po, skb);
1381 if (room == ROOM_NORMAL ||
1382 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1387 i = j = min_t(int, po->rollover->sock, num - 1);
1389 po_next = pkt_sk(f->arr[i]);
1390 if (po_next != po_skip && !po_next->pressure &&
1391 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1393 po->rollover->sock = i;
1394 atomic_long_inc(&po->rollover->num);
1395 if (room == ROOM_LOW)
1396 atomic_long_inc(&po->rollover->num_huge);
1404 atomic_long_inc(&po->rollover->num_failed);
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409 struct sk_buff *skb,
1412 return skb_get_queue_mapping(skb) % num;
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416 struct sk_buff *skb,
1419 struct bpf_prog *prog;
1420 unsigned int ret = 0;
1423 prog = rcu_dereference(f->bpf_prog);
1425 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433 return f->flags & (flag >> 8);
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437 struct packet_type *pt, struct net_device *orig_dev)
1439 struct packet_fanout *f = pt->af_packet_priv;
1440 unsigned int num = READ_ONCE(f->num_members);
1441 struct net *net = read_pnet(&f->net);
1442 struct packet_sock *po;
1445 if (!net_eq(dev_net(dev), net) || !num) {
1450 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1456 case PACKET_FANOUT_HASH:
1458 idx = fanout_demux_hash(f, skb, num);
1460 case PACKET_FANOUT_LB:
1461 idx = fanout_demux_lb(f, skb, num);
1463 case PACKET_FANOUT_CPU:
1464 idx = fanout_demux_cpu(f, skb, num);
1466 case PACKET_FANOUT_RND:
1467 idx = fanout_demux_rnd(f, skb, num);
1469 case PACKET_FANOUT_QM:
1470 idx = fanout_demux_qm(f, skb, num);
1472 case PACKET_FANOUT_ROLLOVER:
1473 idx = fanout_demux_rollover(f, skb, 0, false, num);
1475 case PACKET_FANOUT_CBPF:
1476 case PACKET_FANOUT_EBPF:
1477 idx = fanout_demux_bpf(f, skb, num);
1481 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482 idx = fanout_demux_rollover(f, skb, idx, true, num);
1484 po = pkt_sk(f->arr[idx]);
1485 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494 struct packet_fanout *f = po->fanout;
1496 spin_lock(&f->lock);
1497 f->arr[f->num_members] = sk;
1500 spin_unlock(&f->lock);
1503 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505 struct packet_fanout *f = po->fanout;
1508 spin_lock(&f->lock);
1509 for (i = 0; i < f->num_members; i++) {
1510 if (f->arr[i] == sk)
1513 BUG_ON(i >= f->num_members);
1514 f->arr[i] = f->arr[f->num_members - 1];
1516 spin_unlock(&f->lock);
1519 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1521 if (sk->sk_family != PF_PACKET)
1524 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1527 static void fanout_init_data(struct packet_fanout *f)
1530 case PACKET_FANOUT_LB:
1531 atomic_set(&f->rr_cur, 0);
1533 case PACKET_FANOUT_CBPF:
1534 case PACKET_FANOUT_EBPF:
1535 RCU_INIT_POINTER(f->bpf_prog, NULL);
1540 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542 struct bpf_prog *old;
1544 spin_lock(&f->lock);
1545 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546 rcu_assign_pointer(f->bpf_prog, new);
1547 spin_unlock(&f->lock);
1551 bpf_prog_destroy(old);
1555 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1558 struct bpf_prog *new;
1559 struct sock_fprog fprog;
1562 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564 if (len != sizeof(fprog))
1566 if (copy_from_user(&fprog, data, len))
1569 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1573 __fanout_set_data_bpf(po->fanout, new);
1577 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1580 struct bpf_prog *new;
1583 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585 if (len != sizeof(fd))
1587 if (copy_from_user(&fd, data, len))
1590 new = bpf_prog_get(fd);
1592 return PTR_ERR(new);
1593 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1598 __fanout_set_data_bpf(po->fanout, new);
1602 static int fanout_set_data(struct packet_sock *po, char __user *data,
1605 switch (po->fanout->type) {
1606 case PACKET_FANOUT_CBPF:
1607 return fanout_set_data_cbpf(po, data, len);
1608 case PACKET_FANOUT_EBPF:
1609 return fanout_set_data_ebpf(po, data, len);
1615 static void fanout_release_data(struct packet_fanout *f)
1618 case PACKET_FANOUT_CBPF:
1619 case PACKET_FANOUT_EBPF:
1620 __fanout_set_data_bpf(f, NULL);
1624 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1626 struct packet_rollover *rollover = NULL;
1627 struct packet_sock *po = pkt_sk(sk);
1628 struct packet_fanout *f, *match;
1629 u8 type = type_flags & 0xff;
1630 u8 flags = type_flags >> 8;
1634 case PACKET_FANOUT_ROLLOVER:
1635 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1637 case PACKET_FANOUT_HASH:
1638 case PACKET_FANOUT_LB:
1639 case PACKET_FANOUT_CPU:
1640 case PACKET_FANOUT_RND:
1641 case PACKET_FANOUT_QM:
1642 case PACKET_FANOUT_CBPF:
1643 case PACKET_FANOUT_EBPF:
1649 mutex_lock(&fanout_mutex);
1659 if (type == PACKET_FANOUT_ROLLOVER ||
1660 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1662 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1665 atomic_long_set(&rollover->num, 0);
1666 atomic_long_set(&rollover->num_huge, 0);
1667 atomic_long_set(&rollover->num_failed, 0);
1668 po->rollover = rollover;
1672 list_for_each_entry(f, &fanout_list, list) {
1674 read_pnet(&f->net) == sock_net(sk)) {
1680 if (match && match->flags != flags)
1684 match = kzalloc(sizeof(*match), GFP_KERNEL);
1687 write_pnet(&match->net, sock_net(sk));
1690 match->flags = flags;
1691 INIT_LIST_HEAD(&match->list);
1692 spin_lock_init(&match->lock);
1693 atomic_set(&match->sk_ref, 0);
1694 fanout_init_data(match);
1695 match->prot_hook.type = po->prot_hook.type;
1696 match->prot_hook.dev = po->prot_hook.dev;
1697 match->prot_hook.func = packet_rcv_fanout;
1698 match->prot_hook.af_packet_priv = match;
1699 match->prot_hook.id_match = match_fanout_group;
1700 dev_add_pack(&match->prot_hook);
1701 list_add(&match->list, &fanout_list);
1704 if (match->type == type &&
1705 match->prot_hook.type == po->prot_hook.type &&
1706 match->prot_hook.dev == po->prot_hook.dev) {
1708 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1709 __dev_remove_pack(&po->prot_hook);
1711 atomic_inc(&match->sk_ref);
1712 __fanout_link(sk, po);
1717 if (err && rollover) {
1719 po->rollover = NULL;
1721 mutex_unlock(&fanout_mutex);
1725 static void fanout_release(struct sock *sk)
1727 struct packet_sock *po = pkt_sk(sk);
1728 struct packet_fanout *f;
1730 mutex_lock(&fanout_mutex);
1735 if (atomic_dec_and_test(&f->sk_ref)) {
1737 dev_remove_pack(&f->prot_hook);
1738 fanout_release_data(f);
1743 kfree_rcu(po->rollover, rcu);
1745 mutex_unlock(&fanout_mutex);
1748 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1749 struct sk_buff *skb)
1751 /* Earlier code assumed this would be a VLAN pkt, double-check
1752 * this now that we have the actual packet in hand. We can only
1753 * do this check on Ethernet devices.
1755 if (unlikely(dev->type != ARPHRD_ETHER))
1758 skb_reset_mac_header(skb);
1759 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1762 static const struct proto_ops packet_ops;
1764 static const struct proto_ops packet_ops_spkt;
1766 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1767 struct packet_type *pt, struct net_device *orig_dev)
1770 struct sockaddr_pkt *spkt;
1773 * When we registered the protocol we saved the socket in the data
1774 * field for just this event.
1777 sk = pt->af_packet_priv;
1780 * Yank back the headers [hope the device set this
1781 * right or kerboom...]
1783 * Incoming packets have ll header pulled,
1786 * For outgoing ones skb->data == skb_mac_header(skb)
1787 * so that this procedure is noop.
1790 if (skb->pkt_type == PACKET_LOOPBACK)
1793 if (!net_eq(dev_net(dev), sock_net(sk)))
1796 skb = skb_share_check(skb, GFP_ATOMIC);
1800 /* drop any routing info */
1803 /* drop conntrack reference */
1806 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1808 skb_push(skb, skb->data - skb_mac_header(skb));
1811 * The SOCK_PACKET socket receives _all_ frames.
1814 spkt->spkt_family = dev->type;
1815 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1816 spkt->spkt_protocol = skb->protocol;
1819 * Charge the memory to the socket. This is done specifically
1820 * to prevent sockets using all the memory up.
1823 if (sock_queue_rcv_skb(sk, skb) == 0)
1834 * Output a raw packet to a device layer. This bypasses all the other
1835 * protocol layers and you must therefore supply it with a complete frame
1838 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1841 struct sock *sk = sock->sk;
1842 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1843 struct sk_buff *skb = NULL;
1844 struct net_device *dev;
1850 * Get and verify the address.
1854 if (msg->msg_namelen < sizeof(struct sockaddr))
1856 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1857 proto = saddr->spkt_protocol;
1859 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1862 * Find the device first to size check it
1865 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1868 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1874 if (!(dev->flags & IFF_UP))
1878 * You may not queue a frame bigger than the mtu. This is the lowest level
1879 * raw protocol and you must do your own fragmentation at this level.
1882 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1883 if (!netif_supports_nofcs(dev)) {
1884 err = -EPROTONOSUPPORT;
1887 extra_len = 4; /* We're doing our own CRC */
1891 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1895 size_t reserved = LL_RESERVED_SPACE(dev);
1896 int tlen = dev->needed_tailroom;
1897 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1900 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1903 /* FIXME: Save some space for broken drivers that write a hard
1904 * header at transmission time by themselves. PPP is the notable
1905 * one here. This should really be fixed at the driver level.
1907 skb_reserve(skb, reserved);
1908 skb_reset_network_header(skb);
1910 /* Try to align data part correctly */
1915 skb_reset_network_header(skb);
1917 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1923 if (!dev_validate_header(dev, skb->data, len)) {
1927 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1928 !packet_extra_vlan_len_allowed(dev, skb)) {
1933 skb->protocol = proto;
1935 skb->priority = sk->sk_priority;
1936 skb->mark = sk->sk_mark;
1938 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1940 if (unlikely(extra_len == 4))
1943 skb_probe_transport_header(skb, 0);
1945 dev_queue_xmit(skb);
1956 static unsigned int run_filter(struct sk_buff *skb,
1957 const struct sock *sk,
1960 struct sk_filter *filter;
1963 filter = rcu_dereference(sk->sk_filter);
1965 res = bpf_prog_run_clear_cb(filter->prog, skb);
1972 * This function makes lazy skb cloning in hope that most of packets
1973 * are discarded by BPF.
1975 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1976 * and skb->cb are mangled. It works because (and until) packets
1977 * falling here are owned by current CPU. Output packets are cloned
1978 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1979 * sequencially, so that if we return skb to original state on exit,
1980 * we will not harm anyone.
1983 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1984 struct packet_type *pt, struct net_device *orig_dev)
1987 struct sockaddr_ll *sll;
1988 struct packet_sock *po;
1989 u8 *skb_head = skb->data;
1990 int skb_len = skb->len;
1991 unsigned int snaplen, res;
1993 if (skb->pkt_type == PACKET_LOOPBACK)
1996 sk = pt->af_packet_priv;
1999 if (!net_eq(dev_net(dev), sock_net(sk)))
2004 if (dev->header_ops) {
2005 /* The device has an explicit notion of ll header,
2006 * exported to higher levels.
2008 * Otherwise, the device hides details of its frame
2009 * structure, so that corresponding packet head is
2010 * never delivered to user.
2012 if (sk->sk_type != SOCK_DGRAM)
2013 skb_push(skb, skb->data - skb_mac_header(skb));
2014 else if (skb->pkt_type == PACKET_OUTGOING) {
2015 /* Special case: outgoing packets have ll header at head */
2016 skb_pull(skb, skb_network_offset(skb));
2022 res = run_filter(skb, sk, snaplen);
2024 goto drop_n_restore;
2028 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2031 if (skb_shared(skb)) {
2032 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2036 if (skb_head != skb->data) {
2037 skb->data = skb_head;
2044 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2046 sll = &PACKET_SKB_CB(skb)->sa.ll;
2047 sll->sll_hatype = dev->type;
2048 sll->sll_pkttype = skb->pkt_type;
2049 if (unlikely(po->origdev))
2050 sll->sll_ifindex = orig_dev->ifindex;
2052 sll->sll_ifindex = dev->ifindex;
2054 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2056 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2057 * Use their space for storing the original skb length.
2059 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2061 if (pskb_trim(skb, snaplen))
2064 skb_set_owner_r(skb, sk);
2068 /* drop conntrack reference */
2071 spin_lock(&sk->sk_receive_queue.lock);
2072 po->stats.stats1.tp_packets++;
2073 sock_skb_set_dropcount(sk, skb);
2074 __skb_queue_tail(&sk->sk_receive_queue, skb);
2075 spin_unlock(&sk->sk_receive_queue.lock);
2076 sk->sk_data_ready(sk);
2080 spin_lock(&sk->sk_receive_queue.lock);
2081 po->stats.stats1.tp_drops++;
2082 atomic_inc(&sk->sk_drops);
2083 spin_unlock(&sk->sk_receive_queue.lock);
2086 if (skb_head != skb->data && skb_shared(skb)) {
2087 skb->data = skb_head;
2095 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2096 struct packet_type *pt, struct net_device *orig_dev)
2099 struct packet_sock *po;
2100 struct sockaddr_ll *sll;
2101 union tpacket_uhdr h;
2102 u8 *skb_head = skb->data;
2103 int skb_len = skb->len;
2104 unsigned int snaplen, res;
2105 unsigned long status = TP_STATUS_USER;
2106 unsigned short macoff, netoff, hdrlen;
2107 struct sk_buff *copy_skb = NULL;
2111 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2112 * We may add members to them until current aligned size without forcing
2113 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2115 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2116 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2118 if (skb->pkt_type == PACKET_LOOPBACK)
2121 sk = pt->af_packet_priv;
2124 if (!net_eq(dev_net(dev), sock_net(sk)))
2127 if (dev->header_ops) {
2128 if (sk->sk_type != SOCK_DGRAM)
2129 skb_push(skb, skb->data - skb_mac_header(skb));
2130 else if (skb->pkt_type == PACKET_OUTGOING) {
2131 /* Special case: outgoing packets have ll header at head */
2132 skb_pull(skb, skb_network_offset(skb));
2138 res = run_filter(skb, sk, snaplen);
2140 goto drop_n_restore;
2142 if (skb->ip_summed == CHECKSUM_PARTIAL)
2143 status |= TP_STATUS_CSUMNOTREADY;
2144 else if (skb->pkt_type != PACKET_OUTGOING &&
2145 (skb->ip_summed == CHECKSUM_COMPLETE ||
2146 skb_csum_unnecessary(skb)))
2147 status |= TP_STATUS_CSUM_VALID;
2152 if (sk->sk_type == SOCK_DGRAM) {
2153 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2156 unsigned int maclen = skb_network_offset(skb);
2157 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2158 (maclen < 16 ? 16 : maclen)) +
2160 macoff = netoff - maclen;
2162 if (po->tp_version <= TPACKET_V2) {
2163 if (macoff + snaplen > po->rx_ring.frame_size) {
2164 if (po->copy_thresh &&
2165 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2166 if (skb_shared(skb)) {
2167 copy_skb = skb_clone(skb, GFP_ATOMIC);
2169 copy_skb = skb_get(skb);
2170 skb_head = skb->data;
2173 skb_set_owner_r(copy_skb, sk);
2175 snaplen = po->rx_ring.frame_size - macoff;
2176 if ((int)snaplen < 0)
2179 } else if (unlikely(macoff + snaplen >
2180 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2183 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2184 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2185 snaplen, nval, macoff);
2187 if (unlikely((int)snaplen < 0)) {
2189 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2192 spin_lock(&sk->sk_receive_queue.lock);
2193 h.raw = packet_current_rx_frame(po, skb,
2194 TP_STATUS_KERNEL, (macoff+snaplen));
2197 if (po->tp_version <= TPACKET_V2) {
2198 packet_increment_rx_head(po, &po->rx_ring);
2200 * LOSING will be reported till you read the stats,
2201 * because it's COR - Clear On Read.
2202 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2205 if (po->stats.stats1.tp_drops)
2206 status |= TP_STATUS_LOSING;
2208 po->stats.stats1.tp_packets++;
2210 status |= TP_STATUS_COPY;
2211 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2213 spin_unlock(&sk->sk_receive_queue.lock);
2215 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2217 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2218 getnstimeofday(&ts);
2220 status |= ts_status;
2222 switch (po->tp_version) {
2224 h.h1->tp_len = skb->len;
2225 h.h1->tp_snaplen = snaplen;
2226 h.h1->tp_mac = macoff;
2227 h.h1->tp_net = netoff;
2228 h.h1->tp_sec = ts.tv_sec;
2229 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2230 hdrlen = sizeof(*h.h1);
2233 h.h2->tp_len = skb->len;
2234 h.h2->tp_snaplen = snaplen;
2235 h.h2->tp_mac = macoff;
2236 h.h2->tp_net = netoff;
2237 h.h2->tp_sec = ts.tv_sec;
2238 h.h2->tp_nsec = ts.tv_nsec;
2239 if (skb_vlan_tag_present(skb)) {
2240 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2241 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2242 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2244 h.h2->tp_vlan_tci = 0;
2245 h.h2->tp_vlan_tpid = 0;
2247 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2248 hdrlen = sizeof(*h.h2);
2251 /* tp_nxt_offset,vlan are already populated above.
2252 * So DONT clear those fields here
2254 h.h3->tp_status |= status;
2255 h.h3->tp_len = skb->len;
2256 h.h3->tp_snaplen = snaplen;
2257 h.h3->tp_mac = macoff;
2258 h.h3->tp_net = netoff;
2259 h.h3->tp_sec = ts.tv_sec;
2260 h.h3->tp_nsec = ts.tv_nsec;
2261 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2262 hdrlen = sizeof(*h.h3);
2268 sll = h.raw + TPACKET_ALIGN(hdrlen);
2269 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2270 sll->sll_family = AF_PACKET;
2271 sll->sll_hatype = dev->type;
2272 sll->sll_protocol = skb->protocol;
2273 sll->sll_pkttype = skb->pkt_type;
2274 if (unlikely(po->origdev))
2275 sll->sll_ifindex = orig_dev->ifindex;
2277 sll->sll_ifindex = dev->ifindex;
2281 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2282 if (po->tp_version <= TPACKET_V2) {
2285 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2288 for (start = h.raw; start < end; start += PAGE_SIZE)
2289 flush_dcache_page(pgv_to_page(start));
2294 if (po->tp_version <= TPACKET_V2) {
2295 __packet_set_status(po, h.raw, status);
2296 sk->sk_data_ready(sk);
2298 prb_clear_blk_fill_status(&po->rx_ring);
2302 if (skb_head != skb->data && skb_shared(skb)) {
2303 skb->data = skb_head;
2311 po->stats.stats1.tp_drops++;
2312 spin_unlock(&sk->sk_receive_queue.lock);
2314 sk->sk_data_ready(sk);
2315 kfree_skb(copy_skb);
2316 goto drop_n_restore;
2319 static void tpacket_destruct_skb(struct sk_buff *skb)
2321 struct packet_sock *po = pkt_sk(skb->sk);
2323 if (likely(po->tx_ring.pg_vec)) {
2327 ph = skb_shinfo(skb)->destructor_arg;
2328 packet_dec_pending(&po->tx_ring);
2330 ts = __packet_set_timestamp(po, ph, skb);
2331 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2337 static void tpacket_set_protocol(const struct net_device *dev,
2338 struct sk_buff *skb)
2340 if (dev->type == ARPHRD_ETHER) {
2341 skb_reset_mac_header(skb);
2342 skb->protocol = eth_hdr(skb)->h_proto;
2346 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2347 void *frame, struct net_device *dev, int size_max,
2348 __be16 proto, unsigned char *addr, int hlen)
2350 union tpacket_uhdr ph;
2351 int to_write, offset, len, tp_len, nr_frags, len_max;
2352 struct socket *sock = po->sk.sk_socket;
2359 skb->protocol = proto;
2361 skb->priority = po->sk.sk_priority;
2362 skb->mark = po->sk.sk_mark;
2363 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2364 skb_shinfo(skb)->destructor_arg = ph.raw;
2366 switch (po->tp_version) {
2368 tp_len = ph.h2->tp_len;
2371 tp_len = ph.h1->tp_len;
2374 if (unlikely(tp_len > size_max)) {
2375 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2379 skb_reserve(skb, hlen);
2380 skb_reset_network_header(skb);
2382 if (unlikely(po->tp_tx_has_off)) {
2383 int off_min, off_max, off;
2384 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2385 off_max = po->tx_ring.frame_size - tp_len;
2386 if (sock->type == SOCK_DGRAM) {
2387 switch (po->tp_version) {
2389 off = ph.h2->tp_net;
2392 off = ph.h1->tp_net;
2396 switch (po->tp_version) {
2398 off = ph.h2->tp_mac;
2401 off = ph.h1->tp_mac;
2405 if (unlikely((off < off_min) || (off_max < off)))
2407 data = ph.raw + off;
2409 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2413 if (sock->type == SOCK_DGRAM) {
2414 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2416 if (unlikely(err < 0))
2418 } else if (dev->hard_header_len) {
2419 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2421 skb_push(skb, dev->hard_header_len);
2422 err = skb_store_bits(skb, 0, data, hdrlen);
2425 if (!dev_validate_header(dev, skb->data, hdrlen))
2428 tpacket_set_protocol(dev, skb);
2434 offset = offset_in_page(data);
2435 len_max = PAGE_SIZE - offset;
2436 len = ((to_write > len_max) ? len_max : to_write);
2438 skb->data_len = to_write;
2439 skb->len += to_write;
2440 skb->truesize += to_write;
2441 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2443 while (likely(to_write)) {
2444 nr_frags = skb_shinfo(skb)->nr_frags;
2446 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2447 pr_err("Packet exceed the number of skb frags(%lu)\n",
2452 page = pgv_to_page(data);
2454 flush_dcache_page(page);
2456 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2459 len_max = PAGE_SIZE;
2460 len = ((to_write > len_max) ? len_max : to_write);
2463 skb_probe_transport_header(skb, 0);
2468 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2470 struct sk_buff *skb;
2471 struct net_device *dev;
2473 int err, reserve = 0;
2475 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2476 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2477 int tp_len, size_max;
2478 unsigned char *addr;
2480 int status = TP_STATUS_AVAILABLE;
2483 mutex_lock(&po->pg_vec_lock);
2485 if (likely(saddr == NULL)) {
2486 dev = packet_cached_dev_get(po);
2491 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2493 if (msg->msg_namelen < (saddr->sll_halen
2494 + offsetof(struct sockaddr_ll,
2497 proto = saddr->sll_protocol;
2498 addr = saddr->sll_addr;
2499 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2503 if (unlikely(dev == NULL))
2506 if (unlikely(!(dev->flags & IFF_UP)))
2509 if (po->sk.sk_socket->type == SOCK_RAW)
2510 reserve = dev->hard_header_len;
2511 size_max = po->tx_ring.frame_size
2512 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2514 if (size_max > dev->mtu + reserve + VLAN_HLEN)
2515 size_max = dev->mtu + reserve + VLAN_HLEN;
2518 ph = packet_current_frame(po, &po->tx_ring,
2519 TP_STATUS_SEND_REQUEST);
2520 if (unlikely(ph == NULL)) {
2521 if (need_wait && need_resched())
2526 status = TP_STATUS_SEND_REQUEST;
2527 hlen = LL_RESERVED_SPACE(dev);
2528 tlen = dev->needed_tailroom;
2529 skb = sock_alloc_send_skb(&po->sk,
2530 hlen + tlen + sizeof(struct sockaddr_ll),
2533 if (unlikely(skb == NULL)) {
2534 /* we assume the socket was initially writeable ... */
2535 if (likely(len_sum > 0))
2539 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2541 if (likely(tp_len >= 0) &&
2542 tp_len > dev->mtu + reserve &&
2543 !packet_extra_vlan_len_allowed(dev, skb))
2546 if (unlikely(tp_len < 0)) {
2548 __packet_set_status(po, ph,
2549 TP_STATUS_AVAILABLE);
2550 packet_increment_head(&po->tx_ring);
2554 status = TP_STATUS_WRONG_FORMAT;
2560 packet_pick_tx_queue(dev, skb);
2562 skb->destructor = tpacket_destruct_skb;
2563 __packet_set_status(po, ph, TP_STATUS_SENDING);
2564 packet_inc_pending(&po->tx_ring);
2566 status = TP_STATUS_SEND_REQUEST;
2567 err = po->xmit(skb);
2568 if (unlikely(err > 0)) {
2569 err = net_xmit_errno(err);
2570 if (err && __packet_get_status(po, ph) ==
2571 TP_STATUS_AVAILABLE) {
2572 /* skb was destructed already */
2577 * skb was dropped but not destructed yet;
2578 * let's treat it like congestion or err < 0
2582 packet_increment_head(&po->tx_ring);
2584 } while (likely((ph != NULL) ||
2585 /* Note: packet_read_pending() might be slow if we have
2586 * to call it as it's per_cpu variable, but in fast-path
2587 * we already short-circuit the loop with the first
2588 * condition, and luckily don't have to go that path
2591 (need_wait && packet_read_pending(&po->tx_ring))));
2597 __packet_set_status(po, ph, status);
2602 mutex_unlock(&po->pg_vec_lock);
2606 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2607 size_t reserve, size_t len,
2608 size_t linear, int noblock,
2611 struct sk_buff *skb;
2613 /* Under a page? Don't bother with paged skb. */
2614 if (prepad + len < PAGE_SIZE || !linear)
2617 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2622 skb_reserve(skb, reserve);
2623 skb_put(skb, linear);
2624 skb->data_len = len - linear;
2625 skb->len += len - linear;
2630 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2632 struct sock *sk = sock->sk;
2633 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2634 struct sk_buff *skb;
2635 struct net_device *dev;
2637 unsigned char *addr;
2638 int err, reserve = 0;
2639 struct sockcm_cookie sockc;
2640 struct virtio_net_hdr vnet_hdr = { 0 };
2643 struct packet_sock *po = pkt_sk(sk);
2644 unsigned short gso_type = 0;
2645 int hlen, tlen, linear;
2650 * Get and verify the address.
2653 if (likely(saddr == NULL)) {
2654 dev = packet_cached_dev_get(po);
2659 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2661 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2663 proto = saddr->sll_protocol;
2664 addr = saddr->sll_addr;
2665 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2669 if (unlikely(dev == NULL))
2672 if (unlikely(!(dev->flags & IFF_UP)))
2675 sockc.mark = sk->sk_mark;
2676 if (msg->msg_controllen) {
2677 err = sock_cmsg_send(sk, msg, &sockc);
2682 if (sock->type == SOCK_RAW)
2683 reserve = dev->hard_header_len;
2684 if (po->has_vnet_hdr) {
2685 vnet_hdr_len = sizeof(vnet_hdr);
2688 if (len < vnet_hdr_len)
2691 len -= vnet_hdr_len;
2694 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2695 if (n != vnet_hdr_len)
2698 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2699 (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2700 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2701 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2702 vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2703 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2704 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2707 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2710 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2711 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2712 case VIRTIO_NET_HDR_GSO_TCPV4:
2713 gso_type = SKB_GSO_TCPV4;
2715 case VIRTIO_NET_HDR_GSO_TCPV6:
2716 gso_type = SKB_GSO_TCPV6;
2718 case VIRTIO_NET_HDR_GSO_UDP:
2719 gso_type = SKB_GSO_UDP;
2725 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2726 gso_type |= SKB_GSO_TCP_ECN;
2728 if (vnet_hdr.gso_size == 0)
2734 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2735 if (!netif_supports_nofcs(dev)) {
2736 err = -EPROTONOSUPPORT;
2739 extra_len = 4; /* We're doing our own CRC */
2743 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2747 hlen = LL_RESERVED_SPACE(dev);
2748 tlen = dev->needed_tailroom;
2749 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2750 linear = max(linear, min_t(int, len, dev->hard_header_len));
2751 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2752 msg->msg_flags & MSG_DONTWAIT, &err);
2756 skb_set_network_header(skb, reserve);
2759 if (sock->type == SOCK_DGRAM) {
2760 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2761 if (unlikely(offset < 0))
2765 /* Returns -EFAULT on error */
2766 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2770 if (sock->type == SOCK_RAW &&
2771 !dev_validate_header(dev, skb->data, len)) {
2776 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2778 if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2779 !packet_extra_vlan_len_allowed(dev, skb)) {
2784 skb->protocol = proto;
2786 skb->priority = sk->sk_priority;
2787 skb->mark = sockc.mark;
2789 packet_pick_tx_queue(dev, skb);
2791 if (po->has_vnet_hdr) {
2792 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2793 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2794 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2795 if (!skb_partial_csum_set(skb, s, o)) {
2801 skb_shinfo(skb)->gso_size =
2802 __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2803 skb_shinfo(skb)->gso_type = gso_type;
2805 /* Header must be checked, and gso_segs computed. */
2806 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2807 skb_shinfo(skb)->gso_segs = 0;
2809 len += vnet_hdr_len;
2812 skb_probe_transport_header(skb, reserve);
2814 if (unlikely(extra_len == 4))
2817 err = po->xmit(skb);
2818 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2834 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2836 struct sock *sk = sock->sk;
2837 struct packet_sock *po = pkt_sk(sk);
2839 if (po->tx_ring.pg_vec)
2840 return tpacket_snd(po, msg);
2842 return packet_snd(sock, msg, len);
2846 * Close a PACKET socket. This is fairly simple. We immediately go
2847 * to 'closed' state and remove our protocol entry in the device list.
2850 static int packet_release(struct socket *sock)
2852 struct sock *sk = sock->sk;
2853 struct packet_sock *po;
2855 union tpacket_req_u req_u;
2863 mutex_lock(&net->packet.sklist_lock);
2864 sk_del_node_init_rcu(sk);
2865 mutex_unlock(&net->packet.sklist_lock);
2868 sock_prot_inuse_add(net, sk->sk_prot, -1);
2871 spin_lock(&po->bind_lock);
2872 unregister_prot_hook(sk, false);
2873 packet_cached_dev_reset(po);
2875 if (po->prot_hook.dev) {
2876 dev_put(po->prot_hook.dev);
2877 po->prot_hook.dev = NULL;
2879 spin_unlock(&po->bind_lock);
2881 packet_flush_mclist(sk);
2883 if (po->rx_ring.pg_vec) {
2884 memset(&req_u, 0, sizeof(req_u));
2885 packet_set_ring(sk, &req_u, 1, 0);
2888 if (po->tx_ring.pg_vec) {
2889 memset(&req_u, 0, sizeof(req_u));
2890 packet_set_ring(sk, &req_u, 1, 1);
2897 * Now the socket is dead. No more input will appear.
2904 skb_queue_purge(&sk->sk_receive_queue);
2905 packet_free_pending(po);
2906 sk_refcnt_debug_release(sk);
2913 * Attach a packet hook.
2916 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2919 struct packet_sock *po = pkt_sk(sk);
2920 struct net_device *dev_curr;
2923 struct net_device *dev = NULL;
2925 bool unlisted = false;
2931 spin_lock(&po->bind_lock);
2935 dev = dev_get_by_name_rcu(sock_net(sk), name);
2940 } else if (ifindex) {
2941 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2951 proto_curr = po->prot_hook.type;
2952 dev_curr = po->prot_hook.dev;
2954 need_rehook = proto_curr != proto || dev_curr != dev;
2959 __unregister_prot_hook(sk, true);
2961 dev_curr = po->prot_hook.dev;
2963 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2968 po->prot_hook.type = proto;
2970 if (unlikely(unlisted)) {
2972 po->prot_hook.dev = NULL;
2974 packet_cached_dev_reset(po);
2976 po->prot_hook.dev = dev;
2977 po->ifindex = dev ? dev->ifindex : 0;
2978 packet_cached_dev_assign(po, dev);
2984 if (proto == 0 || !need_rehook)
2987 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2988 register_prot_hook(sk);
2990 sk->sk_err = ENETDOWN;
2991 if (!sock_flag(sk, SOCK_DEAD))
2992 sk->sk_error_report(sk);
2997 spin_unlock(&po->bind_lock);
3003 * Bind a packet socket to a device
3006 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3009 struct sock *sk = sock->sk;
3016 if (addr_len != sizeof(struct sockaddr))
3018 strlcpy(name, uaddr->sa_data, sizeof(name));
3020 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3023 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3025 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3026 struct sock *sk = sock->sk;
3032 if (addr_len < sizeof(struct sockaddr_ll))
3034 if (sll->sll_family != AF_PACKET)
3037 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3038 sll->sll_protocol ? : pkt_sk(sk)->num);
3041 static struct proto packet_proto = {
3043 .owner = THIS_MODULE,
3044 .obj_size = sizeof(struct packet_sock),
3048 * Create a packet of type SOCK_PACKET.
3051 static int packet_create(struct net *net, struct socket *sock, int protocol,
3055 struct packet_sock *po;
3056 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3059 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3061 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3062 sock->type != SOCK_PACKET)
3063 return -ESOCKTNOSUPPORT;
3065 sock->state = SS_UNCONNECTED;
3068 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3072 sock->ops = &packet_ops;
3073 if (sock->type == SOCK_PACKET)
3074 sock->ops = &packet_ops_spkt;
3076 sock_init_data(sock, sk);
3079 sk->sk_family = PF_PACKET;
3081 po->xmit = dev_queue_xmit;
3083 err = packet_alloc_pending(po);
3087 packet_cached_dev_reset(po);
3089 sk->sk_destruct = packet_sock_destruct;
3090 sk_refcnt_debug_inc(sk);
3093 * Attach a protocol block
3096 spin_lock_init(&po->bind_lock);
3097 mutex_init(&po->pg_vec_lock);
3098 po->rollover = NULL;
3099 po->prot_hook.func = packet_rcv;
3101 if (sock->type == SOCK_PACKET)
3102 po->prot_hook.func = packet_rcv_spkt;
3104 po->prot_hook.af_packet_priv = sk;
3107 po->prot_hook.type = proto;
3108 register_prot_hook(sk);
3111 mutex_lock(&net->packet.sklist_lock);
3112 sk_add_node_rcu(sk, &net->packet.sklist);
3113 mutex_unlock(&net->packet.sklist_lock);
3116 sock_prot_inuse_add(net, &packet_proto, 1);
3127 * Pull a packet from our receive queue and hand it to the user.
3128 * If necessary we block.
3131 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3134 struct sock *sk = sock->sk;
3135 struct sk_buff *skb;
3137 int vnet_hdr_len = 0;
3138 unsigned int origlen = 0;
3141 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3145 /* What error should we return now? EUNATTACH? */
3146 if (pkt_sk(sk)->ifindex < 0)
3150 if (flags & MSG_ERRQUEUE) {
3151 err = sock_recv_errqueue(sk, msg, len,
3152 SOL_PACKET, PACKET_TX_TIMESTAMP);
3157 * Call the generic datagram receiver. This handles all sorts
3158 * of horrible races and re-entrancy so we can forget about it
3159 * in the protocol layers.
3161 * Now it will return ENETDOWN, if device have just gone down,
3162 * but then it will block.
3165 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3168 * An error occurred so return it. Because skb_recv_datagram()
3169 * handles the blocking we don't see and worry about blocking
3176 if (pkt_sk(sk)->pressure)
3177 packet_rcv_has_room(pkt_sk(sk), NULL);
3179 if (pkt_sk(sk)->has_vnet_hdr) {
3180 struct virtio_net_hdr vnet_hdr = { 0 };
3183 vnet_hdr_len = sizeof(vnet_hdr);
3184 if (len < vnet_hdr_len)
3187 len -= vnet_hdr_len;
3189 if (skb_is_gso(skb)) {
3190 struct skb_shared_info *sinfo = skb_shinfo(skb);
3192 /* This is a hint as to how much should be linear. */
3194 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3196 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3197 if (sinfo->gso_type & SKB_GSO_TCPV4)
3198 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3199 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3200 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3201 else if (sinfo->gso_type & SKB_GSO_UDP)
3202 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3203 else if (sinfo->gso_type & SKB_GSO_FCOE)
3207 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3208 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3210 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3212 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3213 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3214 vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3215 skb_checksum_start_offset(skb));
3216 vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3218 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3219 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3220 } /* else everything is zero */
3222 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3227 /* You lose any data beyond the buffer you gave. If it worries
3228 * a user program they can ask the device for its MTU
3234 msg->msg_flags |= MSG_TRUNC;
3237 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3241 if (sock->type != SOCK_PACKET) {
3242 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3244 /* Original length was stored in sockaddr_ll fields */
3245 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3246 sll->sll_family = AF_PACKET;
3247 sll->sll_protocol = skb->protocol;
3250 sock_recv_ts_and_drops(msg, sk, skb);
3252 if (msg->msg_name) {
3253 /* If the address length field is there to be filled
3254 * in, we fill it in now.
3256 if (sock->type == SOCK_PACKET) {
3257 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3258 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3260 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3262 msg->msg_namelen = sll->sll_halen +
3263 offsetof(struct sockaddr_ll, sll_addr);
3265 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3269 if (pkt_sk(sk)->auxdata) {
3270 struct tpacket_auxdata aux;
3272 aux.tp_status = TP_STATUS_USER;
3273 if (skb->ip_summed == CHECKSUM_PARTIAL)
3274 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3275 else if (skb->pkt_type != PACKET_OUTGOING &&
3276 (skb->ip_summed == CHECKSUM_COMPLETE ||
3277 skb_csum_unnecessary(skb)))
3278 aux.tp_status |= TP_STATUS_CSUM_VALID;
3280 aux.tp_len = origlen;
3281 aux.tp_snaplen = skb->len;
3283 aux.tp_net = skb_network_offset(skb);
3284 if (skb_vlan_tag_present(skb)) {
3285 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3286 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3287 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3289 aux.tp_vlan_tci = 0;
3290 aux.tp_vlan_tpid = 0;
3292 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3296 * Free or return the buffer as appropriate. Again this
3297 * hides all the races and re-entrancy issues from us.
3299 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3302 skb_free_datagram(sk, skb);
3307 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3308 int *uaddr_len, int peer)
3310 struct net_device *dev;
3311 struct sock *sk = sock->sk;
3316 uaddr->sa_family = AF_PACKET;
3317 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3319 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3321 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3323 *uaddr_len = sizeof(*uaddr);
3328 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3329 int *uaddr_len, int peer)
3331 struct net_device *dev;
3332 struct sock *sk = sock->sk;
3333 struct packet_sock *po = pkt_sk(sk);
3334 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3339 sll->sll_family = AF_PACKET;
3340 sll->sll_ifindex = po->ifindex;
3341 sll->sll_protocol = po->num;
3342 sll->sll_pkttype = 0;
3344 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3346 sll->sll_hatype = dev->type;
3347 sll->sll_halen = dev->addr_len;
3348 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3350 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3354 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3359 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3363 case PACKET_MR_MULTICAST:
3364 if (i->alen != dev->addr_len)
3367 return dev_mc_add(dev, i->addr);
3369 return dev_mc_del(dev, i->addr);
3371 case PACKET_MR_PROMISC:
3372 return dev_set_promiscuity(dev, what);
3373 case PACKET_MR_ALLMULTI:
3374 return dev_set_allmulti(dev, what);
3375 case PACKET_MR_UNICAST:
3376 if (i->alen != dev->addr_len)
3379 return dev_uc_add(dev, i->addr);
3381 return dev_uc_del(dev, i->addr);
3389 static void packet_dev_mclist_delete(struct net_device *dev,
3390 struct packet_mclist **mlp)
3392 struct packet_mclist *ml;
3394 while ((ml = *mlp) != NULL) {
3395 if (ml->ifindex == dev->ifindex) {
3396 packet_dev_mc(dev, ml, -1);
3404 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3406 struct packet_sock *po = pkt_sk(sk);
3407 struct packet_mclist *ml, *i;
3408 struct net_device *dev;
3414 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3419 if (mreq->mr_alen > dev->addr_len)
3423 i = kmalloc(sizeof(*i), GFP_KERNEL);
3428 for (ml = po->mclist; ml; ml = ml->next) {
3429 if (ml->ifindex == mreq->mr_ifindex &&
3430 ml->type == mreq->mr_type &&
3431 ml->alen == mreq->mr_alen &&
3432 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3434 /* Free the new element ... */
3440 i->type = mreq->mr_type;
3441 i->ifindex = mreq->mr_ifindex;
3442 i->alen = mreq->mr_alen;
3443 memcpy(i->addr, mreq->mr_address, i->alen);
3444 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3446 i->next = po->mclist;
3448 err = packet_dev_mc(dev, i, 1);
3450 po->mclist = i->next;
3459 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3461 struct packet_mclist *ml, **mlp;
3465 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3466 if (ml->ifindex == mreq->mr_ifindex &&
3467 ml->type == mreq->mr_type &&
3468 ml->alen == mreq->mr_alen &&
3469 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3470 if (--ml->count == 0) {
3471 struct net_device *dev;
3473 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3475 packet_dev_mc(dev, ml, -1);
3485 static void packet_flush_mclist(struct sock *sk)
3487 struct packet_sock *po = pkt_sk(sk);
3488 struct packet_mclist *ml;
3494 while ((ml = po->mclist) != NULL) {
3495 struct net_device *dev;
3497 po->mclist = ml->next;
3498 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3500 packet_dev_mc(dev, ml, -1);
3507 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3509 struct sock *sk = sock->sk;
3510 struct packet_sock *po = pkt_sk(sk);
3513 if (level != SOL_PACKET)
3514 return -ENOPROTOOPT;
3517 case PACKET_ADD_MEMBERSHIP:
3518 case PACKET_DROP_MEMBERSHIP:
3520 struct packet_mreq_max mreq;
3522 memset(&mreq, 0, sizeof(mreq));
3523 if (len < sizeof(struct packet_mreq))
3525 if (len > sizeof(mreq))
3527 if (copy_from_user(&mreq, optval, len))
3529 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3531 if (optname == PACKET_ADD_MEMBERSHIP)
3532 ret = packet_mc_add(sk, &mreq);
3534 ret = packet_mc_drop(sk, &mreq);
3538 case PACKET_RX_RING:
3539 case PACKET_TX_RING:
3541 union tpacket_req_u req_u;
3544 switch (po->tp_version) {
3547 len = sizeof(req_u.req);
3551 len = sizeof(req_u.req3);
3556 if (pkt_sk(sk)->has_vnet_hdr)
3558 if (copy_from_user(&req_u.req, optval, len))
3560 return packet_set_ring(sk, &req_u, 0,
3561 optname == PACKET_TX_RING);
3563 case PACKET_COPY_THRESH:
3567 if (optlen != sizeof(val))
3569 if (copy_from_user(&val, optval, sizeof(val)))
3572 pkt_sk(sk)->copy_thresh = val;
3575 case PACKET_VERSION:
3579 if (optlen != sizeof(val))
3581 if (copy_from_user(&val, optval, sizeof(val)))
3592 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3595 po->tp_version = val;
3601 case PACKET_RESERVE:
3605 if (optlen != sizeof(val))
3607 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3609 if (copy_from_user(&val, optval, sizeof(val)))
3611 po->tp_reserve = val;
3618 if (optlen != sizeof(val))
3620 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3622 if (copy_from_user(&val, optval, sizeof(val)))
3624 po->tp_loss = !!val;
3627 case PACKET_AUXDATA:
3631 if (optlen < sizeof(val))
3633 if (copy_from_user(&val, optval, sizeof(val)))
3636 po->auxdata = !!val;
3639 case PACKET_ORIGDEV:
3643 if (optlen < sizeof(val))
3645 if (copy_from_user(&val, optval, sizeof(val)))
3648 po->origdev = !!val;
3651 case PACKET_VNET_HDR:
3655 if (sock->type != SOCK_RAW)
3657 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3659 if (optlen < sizeof(val))
3661 if (copy_from_user(&val, optval, sizeof(val)))
3664 po->has_vnet_hdr = !!val;
3667 case PACKET_TIMESTAMP:
3671 if (optlen != sizeof(val))
3673 if (copy_from_user(&val, optval, sizeof(val)))
3676 po->tp_tstamp = val;
3683 if (optlen != sizeof(val))
3685 if (copy_from_user(&val, optval, sizeof(val)))
3688 return fanout_add(sk, val & 0xffff, val >> 16);
3690 case PACKET_FANOUT_DATA:
3695 return fanout_set_data(po, optval, optlen);
3697 case PACKET_TX_HAS_OFF:
3701 if (optlen != sizeof(val))
3703 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3705 if (copy_from_user(&val, optval, sizeof(val)))
3707 po->tp_tx_has_off = !!val;
3710 case PACKET_QDISC_BYPASS:
3714 if (optlen != sizeof(val))
3716 if (copy_from_user(&val, optval, sizeof(val)))
3719 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3723 return -ENOPROTOOPT;
3727 static int packet_getsockopt(struct socket *sock, int level, int optname,
3728 char __user *optval, int __user *optlen)
3731 int val, lv = sizeof(val);
3732 struct sock *sk = sock->sk;
3733 struct packet_sock *po = pkt_sk(sk);
3735 union tpacket_stats_u st;
3736 struct tpacket_rollover_stats rstats;
3738 if (level != SOL_PACKET)
3739 return -ENOPROTOOPT;
3741 if (get_user(len, optlen))
3748 case PACKET_STATISTICS:
3749 spin_lock_bh(&sk->sk_receive_queue.lock);
3750 memcpy(&st, &po->stats, sizeof(st));
3751 memset(&po->stats, 0, sizeof(po->stats));
3752 spin_unlock_bh(&sk->sk_receive_queue.lock);
3754 if (po->tp_version == TPACKET_V3) {
3755 lv = sizeof(struct tpacket_stats_v3);
3756 st.stats3.tp_packets += st.stats3.tp_drops;
3759 lv = sizeof(struct tpacket_stats);
3760 st.stats1.tp_packets += st.stats1.tp_drops;
3765 case PACKET_AUXDATA:
3768 case PACKET_ORIGDEV:
3771 case PACKET_VNET_HDR:
3772 val = po->has_vnet_hdr;
3774 case PACKET_VERSION:
3775 val = po->tp_version;
3778 if (len > sizeof(int))
3780 if (copy_from_user(&val, optval, len))
3784 val = sizeof(struct tpacket_hdr);
3787 val = sizeof(struct tpacket2_hdr);
3790 val = sizeof(struct tpacket3_hdr);
3796 case PACKET_RESERVE:
3797 val = po->tp_reserve;
3802 case PACKET_TIMESTAMP:
3803 val = po->tp_tstamp;
3807 ((u32)po->fanout->id |
3808 ((u32)po->fanout->type << 16) |
3809 ((u32)po->fanout->flags << 24)) :
3812 case PACKET_ROLLOVER_STATS:
3815 rstats.tp_all = atomic_long_read(&po->rollover->num);
3816 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3817 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3819 lv = sizeof(rstats);
3821 case PACKET_TX_HAS_OFF:
3822 val = po->tp_tx_has_off;
3824 case PACKET_QDISC_BYPASS:
3825 val = packet_use_direct_xmit(po);
3828 return -ENOPROTOOPT;
3833 if (put_user(len, optlen))
3835 if (copy_to_user(optval, data, len))
3841 static int packet_notifier(struct notifier_block *this,
3842 unsigned long msg, void *ptr)
3845 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3846 struct net *net = dev_net(dev);
3849 sk_for_each_rcu(sk, &net->packet.sklist) {
3850 struct packet_sock *po = pkt_sk(sk);
3853 case NETDEV_UNREGISTER:
3855 packet_dev_mclist_delete(dev, &po->mclist);
3859 if (dev->ifindex == po->ifindex) {
3860 spin_lock(&po->bind_lock);
3862 __unregister_prot_hook(sk, false);
3863 sk->sk_err = ENETDOWN;
3864 if (!sock_flag(sk, SOCK_DEAD))
3865 sk->sk_error_report(sk);
3867 if (msg == NETDEV_UNREGISTER) {
3868 packet_cached_dev_reset(po);
3871 if (po->prot_hook.dev)
3872 dev_put(po->prot_hook.dev);
3873 po->prot_hook.dev = NULL;
3875 spin_unlock(&po->bind_lock);
3879 if (dev->ifindex == po->ifindex) {
3880 spin_lock(&po->bind_lock);
3882 register_prot_hook(sk);
3883 spin_unlock(&po->bind_lock);
3893 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3896 struct sock *sk = sock->sk;
3901 int amount = sk_wmem_alloc_get(sk);
3903 return put_user(amount, (int __user *)arg);
3907 struct sk_buff *skb;
3910 spin_lock_bh(&sk->sk_receive_queue.lock);
3911 skb = skb_peek(&sk->sk_receive_queue);
3914 spin_unlock_bh(&sk->sk_receive_queue.lock);
3915 return put_user(amount, (int __user *)arg);
3918 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3920 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3930 case SIOCGIFBRDADDR:
3931 case SIOCSIFBRDADDR:
3932 case SIOCGIFNETMASK:
3933 case SIOCSIFNETMASK:
3934 case SIOCGIFDSTADDR:
3935 case SIOCSIFDSTADDR:
3937 return inet_dgram_ops.ioctl(sock, cmd, arg);
3941 return -ENOIOCTLCMD;
3946 static unsigned int packet_poll(struct file *file, struct socket *sock,
3949 struct sock *sk = sock->sk;
3950 struct packet_sock *po = pkt_sk(sk);
3951 unsigned int mask = datagram_poll(file, sock, wait);
3953 spin_lock_bh(&sk->sk_receive_queue.lock);
3954 if (po->rx_ring.pg_vec) {
3955 if (!packet_previous_rx_frame(po, &po->rx_ring,
3957 mask |= POLLIN | POLLRDNORM;
3959 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3961 spin_unlock_bh(&sk->sk_receive_queue.lock);
3962 spin_lock_bh(&sk->sk_write_queue.lock);
3963 if (po->tx_ring.pg_vec) {
3964 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3965 mask |= POLLOUT | POLLWRNORM;
3967 spin_unlock_bh(&sk->sk_write_queue.lock);
3972 /* Dirty? Well, I still did not learn better way to account
3976 static void packet_mm_open(struct vm_area_struct *vma)
3978 struct file *file = vma->vm_file;
3979 struct socket *sock = file->private_data;
3980 struct sock *sk = sock->sk;
3983 atomic_inc(&pkt_sk(sk)->mapped);
3986 static void packet_mm_close(struct vm_area_struct *vma)
3988 struct file *file = vma->vm_file;
3989 struct socket *sock = file->private_data;
3990 struct sock *sk = sock->sk;
3993 atomic_dec(&pkt_sk(sk)->mapped);
3996 static const struct vm_operations_struct packet_mmap_ops = {
3997 .open = packet_mm_open,
3998 .close = packet_mm_close,
4001 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4006 for (i = 0; i < len; i++) {
4007 if (likely(pg_vec[i].buffer)) {
4008 if (is_vmalloc_addr(pg_vec[i].buffer))
4009 vfree(pg_vec[i].buffer);
4011 free_pages((unsigned long)pg_vec[i].buffer,
4013 pg_vec[i].buffer = NULL;
4019 static char *alloc_one_pg_vec_page(unsigned long order)
4022 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4023 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4025 buffer = (char *) __get_free_pages(gfp_flags, order);
4029 /* __get_free_pages failed, fall back to vmalloc */
4030 buffer = vzalloc((1 << order) * PAGE_SIZE);
4034 /* vmalloc failed, lets dig into swap here */
4035 gfp_flags &= ~__GFP_NORETRY;
4036 buffer = (char *) __get_free_pages(gfp_flags, order);
4040 /* complete and utter failure */
4044 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4046 unsigned int block_nr = req->tp_block_nr;
4050 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4051 if (unlikely(!pg_vec))
4054 for (i = 0; i < block_nr; i++) {
4055 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4056 if (unlikely(!pg_vec[i].buffer))
4057 goto out_free_pgvec;
4064 free_pg_vec(pg_vec, order, block_nr);
4069 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4070 int closing, int tx_ring)
4072 struct pgv *pg_vec = NULL;
4073 struct packet_sock *po = pkt_sk(sk);
4074 int was_running, order = 0;
4075 struct packet_ring_buffer *rb;
4076 struct sk_buff_head *rb_queue;
4079 /* Added to avoid minimal code churn */
4080 struct tpacket_req *req = &req_u->req;
4083 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4084 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4085 WARN(1, "Tx-ring is not supported.\n");
4089 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4090 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4094 if (atomic_read(&po->mapped))
4096 if (packet_read_pending(rb))
4100 if (req->tp_block_nr) {
4101 /* Sanity tests and some calculations */
4103 if (unlikely(rb->pg_vec))
4106 switch (po->tp_version) {
4108 po->tp_hdrlen = TPACKET_HDRLEN;
4111 po->tp_hdrlen = TPACKET2_HDRLEN;
4114 po->tp_hdrlen = TPACKET3_HDRLEN;
4119 if (unlikely((int)req->tp_block_size <= 0))
4121 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4123 if (po->tp_version >= TPACKET_V3 &&
4124 (int)(req->tp_block_size -
4125 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4127 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4130 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4133 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4134 if (unlikely(rb->frames_per_block == 0))
4136 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4141 order = get_order(req->tp_block_size);
4142 pg_vec = alloc_pg_vec(req, order);
4143 if (unlikely(!pg_vec))
4145 switch (po->tp_version) {
4147 /* Transmit path is not supported. We checked
4148 * it above but just being paranoid
4151 init_prb_bdqc(po, rb, pg_vec, req_u);
4160 if (unlikely(req->tp_frame_nr))
4165 /* Detach socket from network */
4166 spin_lock(&po->bind_lock);
4167 was_running = po->running;
4171 __unregister_prot_hook(sk, false);
4173 spin_unlock(&po->bind_lock);
4178 mutex_lock(&po->pg_vec_lock);
4179 if (closing || atomic_read(&po->mapped) == 0) {
4181 spin_lock_bh(&rb_queue->lock);
4182 swap(rb->pg_vec, pg_vec);
4183 rb->frame_max = (req->tp_frame_nr - 1);
4185 rb->frame_size = req->tp_frame_size;
4186 spin_unlock_bh(&rb_queue->lock);
4188 swap(rb->pg_vec_order, order);
4189 swap(rb->pg_vec_len, req->tp_block_nr);
4191 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4192 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4193 tpacket_rcv : packet_rcv;
4194 skb_queue_purge(rb_queue);
4195 if (atomic_read(&po->mapped))
4196 pr_err("packet_mmap: vma is busy: %d\n",
4197 atomic_read(&po->mapped));
4199 mutex_unlock(&po->pg_vec_lock);
4201 spin_lock(&po->bind_lock);
4204 register_prot_hook(sk);
4206 spin_unlock(&po->bind_lock);
4207 if (closing && (po->tp_version > TPACKET_V2)) {
4208 /* Because we don't support block-based V3 on tx-ring */
4210 prb_shutdown_retire_blk_timer(po, rb_queue);
4214 free_pg_vec(pg_vec, order, req->tp_block_nr);
4220 static int packet_mmap(struct file *file, struct socket *sock,
4221 struct vm_area_struct *vma)
4223 struct sock *sk = sock->sk;
4224 struct packet_sock *po = pkt_sk(sk);
4225 unsigned long size, expected_size;
4226 struct packet_ring_buffer *rb;
4227 unsigned long start;
4234 mutex_lock(&po->pg_vec_lock);
4237 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4239 expected_size += rb->pg_vec_len
4245 if (expected_size == 0)
4248 size = vma->vm_end - vma->vm_start;
4249 if (size != expected_size)
4252 start = vma->vm_start;
4253 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4254 if (rb->pg_vec == NULL)
4257 for (i = 0; i < rb->pg_vec_len; i++) {
4259 void *kaddr = rb->pg_vec[i].buffer;
4262 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4263 page = pgv_to_page(kaddr);
4264 err = vm_insert_page(vma, start, page);
4273 atomic_inc(&po->mapped);
4274 vma->vm_ops = &packet_mmap_ops;
4278 mutex_unlock(&po->pg_vec_lock);
4282 static const struct proto_ops packet_ops_spkt = {
4283 .family = PF_PACKET,
4284 .owner = THIS_MODULE,
4285 .release = packet_release,
4286 .bind = packet_bind_spkt,
4287 .connect = sock_no_connect,
4288 .socketpair = sock_no_socketpair,
4289 .accept = sock_no_accept,
4290 .getname = packet_getname_spkt,
4291 .poll = datagram_poll,
4292 .ioctl = packet_ioctl,
4293 .listen = sock_no_listen,
4294 .shutdown = sock_no_shutdown,
4295 .setsockopt = sock_no_setsockopt,
4296 .getsockopt = sock_no_getsockopt,
4297 .sendmsg = packet_sendmsg_spkt,
4298 .recvmsg = packet_recvmsg,
4299 .mmap = sock_no_mmap,
4300 .sendpage = sock_no_sendpage,
4303 static const struct proto_ops packet_ops = {
4304 .family = PF_PACKET,
4305 .owner = THIS_MODULE,
4306 .release = packet_release,
4307 .bind = packet_bind,
4308 .connect = sock_no_connect,
4309 .socketpair = sock_no_socketpair,
4310 .accept = sock_no_accept,
4311 .getname = packet_getname,
4312 .poll = packet_poll,
4313 .ioctl = packet_ioctl,
4314 .listen = sock_no_listen,
4315 .shutdown = sock_no_shutdown,
4316 .setsockopt = packet_setsockopt,
4317 .getsockopt = packet_getsockopt,
4318 .sendmsg = packet_sendmsg,
4319 .recvmsg = packet_recvmsg,
4320 .mmap = packet_mmap,
4321 .sendpage = sock_no_sendpage,
4324 static const struct net_proto_family packet_family_ops = {
4325 .family = PF_PACKET,
4326 .create = packet_create,
4327 .owner = THIS_MODULE,
4330 static struct notifier_block packet_netdev_notifier = {
4331 .notifier_call = packet_notifier,
4334 #ifdef CONFIG_PROC_FS
4336 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4339 struct net *net = seq_file_net(seq);
4342 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4345 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4347 struct net *net = seq_file_net(seq);
4348 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4351 static void packet_seq_stop(struct seq_file *seq, void *v)
4357 static int packet_seq_show(struct seq_file *seq, void *v)
4359 if (v == SEQ_START_TOKEN)
4360 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4362 struct sock *s = sk_entry(v);
4363 const struct packet_sock *po = pkt_sk(s);
4366 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4368 atomic_read(&s->sk_refcnt),
4373 atomic_read(&s->sk_rmem_alloc),
4374 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4381 static const struct seq_operations packet_seq_ops = {
4382 .start = packet_seq_start,
4383 .next = packet_seq_next,
4384 .stop = packet_seq_stop,
4385 .show = packet_seq_show,
4388 static int packet_seq_open(struct inode *inode, struct file *file)
4390 return seq_open_net(inode, file, &packet_seq_ops,
4391 sizeof(struct seq_net_private));
4394 static const struct file_operations packet_seq_fops = {
4395 .owner = THIS_MODULE,
4396 .open = packet_seq_open,
4398 .llseek = seq_lseek,
4399 .release = seq_release_net,
4404 static int __net_init packet_net_init(struct net *net)
4406 mutex_init(&net->packet.sklist_lock);
4407 INIT_HLIST_HEAD(&net->packet.sklist);
4409 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4415 static void __net_exit packet_net_exit(struct net *net)
4417 remove_proc_entry("packet", net->proc_net);
4420 static struct pernet_operations packet_net_ops = {
4421 .init = packet_net_init,
4422 .exit = packet_net_exit,
4426 static void __exit packet_exit(void)
4428 unregister_netdevice_notifier(&packet_netdev_notifier);
4429 unregister_pernet_subsys(&packet_net_ops);
4430 sock_unregister(PF_PACKET);
4431 proto_unregister(&packet_proto);
4434 static int __init packet_init(void)
4436 int rc = proto_register(&packet_proto, 0);
4441 sock_register(&packet_family_ops);
4442 register_pernet_subsys(&packet_net_ops);
4443 register_netdevice_notifier(&packet_netdev_notifier);
4448 module_init(packet_init);
4449 module_exit(packet_exit);
4450 MODULE_LICENSE("GPL");
4451 MODULE_ALIAS_NETPROTO(PF_PACKET);