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 if (f->num_members == 1)
1501 dev_add_pack(&f->prot_hook);
1502 spin_unlock(&f->lock);
1505 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1507 struct packet_fanout *f = po->fanout;
1510 spin_lock(&f->lock);
1511 for (i = 0; i < f->num_members; i++) {
1512 if (f->arr[i] == sk)
1515 BUG_ON(i >= f->num_members);
1516 f->arr[i] = f->arr[f->num_members - 1];
1518 if (f->num_members == 0)
1519 __dev_remove_pack(&f->prot_hook);
1520 spin_unlock(&f->lock);
1523 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1525 if (sk->sk_family != PF_PACKET)
1528 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1531 static void fanout_init_data(struct packet_fanout *f)
1534 case PACKET_FANOUT_LB:
1535 atomic_set(&f->rr_cur, 0);
1537 case PACKET_FANOUT_CBPF:
1538 case PACKET_FANOUT_EBPF:
1539 RCU_INIT_POINTER(f->bpf_prog, NULL);
1544 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1546 struct bpf_prog *old;
1548 spin_lock(&f->lock);
1549 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1550 rcu_assign_pointer(f->bpf_prog, new);
1551 spin_unlock(&f->lock);
1555 bpf_prog_destroy(old);
1559 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1562 struct bpf_prog *new;
1563 struct sock_fprog fprog;
1566 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1568 if (len != sizeof(fprog))
1570 if (copy_from_user(&fprog, data, len))
1573 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1577 __fanout_set_data_bpf(po->fanout, new);
1581 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1584 struct bpf_prog *new;
1587 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1589 if (len != sizeof(fd))
1591 if (copy_from_user(&fd, data, len))
1594 new = bpf_prog_get(fd);
1596 return PTR_ERR(new);
1597 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1602 __fanout_set_data_bpf(po->fanout, new);
1606 static int fanout_set_data(struct packet_sock *po, char __user *data,
1609 switch (po->fanout->type) {
1610 case PACKET_FANOUT_CBPF:
1611 return fanout_set_data_cbpf(po, data, len);
1612 case PACKET_FANOUT_EBPF:
1613 return fanout_set_data_ebpf(po, data, len);
1619 static void fanout_release_data(struct packet_fanout *f)
1622 case PACKET_FANOUT_CBPF:
1623 case PACKET_FANOUT_EBPF:
1624 __fanout_set_data_bpf(f, NULL);
1628 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1630 struct packet_rollover *rollover = NULL;
1631 struct packet_sock *po = pkt_sk(sk);
1632 struct packet_fanout *f, *match;
1633 u8 type = type_flags & 0xff;
1634 u8 flags = type_flags >> 8;
1638 case PACKET_FANOUT_ROLLOVER:
1639 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1641 case PACKET_FANOUT_HASH:
1642 case PACKET_FANOUT_LB:
1643 case PACKET_FANOUT_CPU:
1644 case PACKET_FANOUT_RND:
1645 case PACKET_FANOUT_QM:
1646 case PACKET_FANOUT_CBPF:
1647 case PACKET_FANOUT_EBPF:
1653 mutex_lock(&fanout_mutex);
1663 if (type == PACKET_FANOUT_ROLLOVER ||
1664 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1666 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1669 atomic_long_set(&rollover->num, 0);
1670 atomic_long_set(&rollover->num_huge, 0);
1671 atomic_long_set(&rollover->num_failed, 0);
1672 po->rollover = rollover;
1676 list_for_each_entry(f, &fanout_list, list) {
1678 read_pnet(&f->net) == sock_net(sk)) {
1684 if (match && match->flags != flags)
1688 match = kzalloc(sizeof(*match), GFP_KERNEL);
1691 write_pnet(&match->net, sock_net(sk));
1694 match->flags = flags;
1695 INIT_LIST_HEAD(&match->list);
1696 spin_lock_init(&match->lock);
1697 atomic_set(&match->sk_ref, 0);
1698 fanout_init_data(match);
1699 match->prot_hook.type = po->prot_hook.type;
1700 match->prot_hook.dev = po->prot_hook.dev;
1701 match->prot_hook.func = packet_rcv_fanout;
1702 match->prot_hook.af_packet_priv = match;
1703 match->prot_hook.id_match = match_fanout_group;
1704 list_add(&match->list, &fanout_list);
1707 if (match->type == type &&
1708 match->prot_hook.type == po->prot_hook.type &&
1709 match->prot_hook.dev == po->prot_hook.dev) {
1711 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1712 __dev_remove_pack(&po->prot_hook);
1714 atomic_inc(&match->sk_ref);
1715 __fanout_link(sk, po);
1720 if (err && rollover) {
1722 po->rollover = NULL;
1724 mutex_unlock(&fanout_mutex);
1728 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1729 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1730 * It is the responsibility of the caller to call fanout_release_data() and
1731 * free the returned packet_fanout (after synchronize_net())
1733 static struct packet_fanout *fanout_release(struct sock *sk)
1735 struct packet_sock *po = pkt_sk(sk);
1736 struct packet_fanout *f;
1738 mutex_lock(&fanout_mutex);
1743 if (atomic_dec_and_test(&f->sk_ref))
1749 kfree_rcu(po->rollover, rcu);
1751 mutex_unlock(&fanout_mutex);
1756 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1757 struct sk_buff *skb)
1759 /* Earlier code assumed this would be a VLAN pkt, double-check
1760 * this now that we have the actual packet in hand. We can only
1761 * do this check on Ethernet devices.
1763 if (unlikely(dev->type != ARPHRD_ETHER))
1766 skb_reset_mac_header(skb);
1767 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1770 static const struct proto_ops packet_ops;
1772 static const struct proto_ops packet_ops_spkt;
1774 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1775 struct packet_type *pt, struct net_device *orig_dev)
1778 struct sockaddr_pkt *spkt;
1781 * When we registered the protocol we saved the socket in the data
1782 * field for just this event.
1785 sk = pt->af_packet_priv;
1788 * Yank back the headers [hope the device set this
1789 * right or kerboom...]
1791 * Incoming packets have ll header pulled,
1794 * For outgoing ones skb->data == skb_mac_header(skb)
1795 * so that this procedure is noop.
1798 if (skb->pkt_type == PACKET_LOOPBACK)
1801 if (!net_eq(dev_net(dev), sock_net(sk)))
1804 skb = skb_share_check(skb, GFP_ATOMIC);
1808 /* drop any routing info */
1811 /* drop conntrack reference */
1814 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1816 skb_push(skb, skb->data - skb_mac_header(skb));
1819 * The SOCK_PACKET socket receives _all_ frames.
1822 spkt->spkt_family = dev->type;
1823 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1824 spkt->spkt_protocol = skb->protocol;
1827 * Charge the memory to the socket. This is done specifically
1828 * to prevent sockets using all the memory up.
1831 if (sock_queue_rcv_skb(sk, skb) == 0)
1842 * Output a raw packet to a device layer. This bypasses all the other
1843 * protocol layers and you must therefore supply it with a complete frame
1846 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1849 struct sock *sk = sock->sk;
1850 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1851 struct sk_buff *skb = NULL;
1852 struct net_device *dev;
1858 * Get and verify the address.
1862 if (msg->msg_namelen < sizeof(struct sockaddr))
1864 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1865 proto = saddr->spkt_protocol;
1867 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1870 * Find the device first to size check it
1873 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1876 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1882 if (!(dev->flags & IFF_UP))
1886 * You may not queue a frame bigger than the mtu. This is the lowest level
1887 * raw protocol and you must do your own fragmentation at this level.
1890 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1891 if (!netif_supports_nofcs(dev)) {
1892 err = -EPROTONOSUPPORT;
1895 extra_len = 4; /* We're doing our own CRC */
1899 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1903 size_t reserved = LL_RESERVED_SPACE(dev);
1904 int tlen = dev->needed_tailroom;
1905 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1908 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1911 /* FIXME: Save some space for broken drivers that write a hard
1912 * header at transmission time by themselves. PPP is the notable
1913 * one here. This should really be fixed at the driver level.
1915 skb_reserve(skb, reserved);
1916 skb_reset_network_header(skb);
1918 /* Try to align data part correctly */
1923 skb_reset_network_header(skb);
1925 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1931 if (!dev_validate_header(dev, skb->data, len)) {
1935 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1936 !packet_extra_vlan_len_allowed(dev, skb)) {
1941 skb->protocol = proto;
1943 skb->priority = sk->sk_priority;
1944 skb->mark = sk->sk_mark;
1946 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1948 if (unlikely(extra_len == 4))
1951 skb_probe_transport_header(skb, 0);
1953 dev_queue_xmit(skb);
1964 static unsigned int run_filter(struct sk_buff *skb,
1965 const struct sock *sk,
1968 struct sk_filter *filter;
1971 filter = rcu_dereference(sk->sk_filter);
1973 res = bpf_prog_run_clear_cb(filter->prog, skb);
1980 * This function makes lazy skb cloning in hope that most of packets
1981 * are discarded by BPF.
1983 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1984 * and skb->cb are mangled. It works because (and until) packets
1985 * falling here are owned by current CPU. Output packets are cloned
1986 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1987 * sequencially, so that if we return skb to original state on exit,
1988 * we will not harm anyone.
1991 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1992 struct packet_type *pt, struct net_device *orig_dev)
1995 struct sockaddr_ll *sll;
1996 struct packet_sock *po;
1997 u8 *skb_head = skb->data;
1998 int skb_len = skb->len;
1999 unsigned int snaplen, res;
2001 if (skb->pkt_type == PACKET_LOOPBACK)
2004 sk = pt->af_packet_priv;
2007 if (!net_eq(dev_net(dev), sock_net(sk)))
2012 if (dev->header_ops) {
2013 /* The device has an explicit notion of ll header,
2014 * exported to higher levels.
2016 * Otherwise, the device hides details of its frame
2017 * structure, so that corresponding packet head is
2018 * never delivered to user.
2020 if (sk->sk_type != SOCK_DGRAM)
2021 skb_push(skb, skb->data - skb_mac_header(skb));
2022 else if (skb->pkt_type == PACKET_OUTGOING) {
2023 /* Special case: outgoing packets have ll header at head */
2024 skb_pull(skb, skb_network_offset(skb));
2030 res = run_filter(skb, sk, snaplen);
2032 goto drop_n_restore;
2036 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2039 if (skb_shared(skb)) {
2040 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2044 if (skb_head != skb->data) {
2045 skb->data = skb_head;
2052 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2054 sll = &PACKET_SKB_CB(skb)->sa.ll;
2055 sll->sll_hatype = dev->type;
2056 sll->sll_pkttype = skb->pkt_type;
2057 if (unlikely(po->origdev))
2058 sll->sll_ifindex = orig_dev->ifindex;
2060 sll->sll_ifindex = dev->ifindex;
2062 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2064 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2065 * Use their space for storing the original skb length.
2067 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2069 if (pskb_trim(skb, snaplen))
2072 skb_set_owner_r(skb, sk);
2076 /* drop conntrack reference */
2079 spin_lock(&sk->sk_receive_queue.lock);
2080 po->stats.stats1.tp_packets++;
2081 sock_skb_set_dropcount(sk, skb);
2082 __skb_queue_tail(&sk->sk_receive_queue, skb);
2083 spin_unlock(&sk->sk_receive_queue.lock);
2084 sk->sk_data_ready(sk);
2088 spin_lock(&sk->sk_receive_queue.lock);
2089 po->stats.stats1.tp_drops++;
2090 atomic_inc(&sk->sk_drops);
2091 spin_unlock(&sk->sk_receive_queue.lock);
2094 if (skb_head != skb->data && skb_shared(skb)) {
2095 skb->data = skb_head;
2103 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2104 struct packet_type *pt, struct net_device *orig_dev)
2107 struct packet_sock *po;
2108 struct sockaddr_ll *sll;
2109 union tpacket_uhdr h;
2110 u8 *skb_head = skb->data;
2111 int skb_len = skb->len;
2112 unsigned int snaplen, res;
2113 unsigned long status = TP_STATUS_USER;
2114 unsigned short macoff, netoff, hdrlen;
2115 struct sk_buff *copy_skb = NULL;
2119 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2120 * We may add members to them until current aligned size without forcing
2121 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2123 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2124 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2126 if (skb->pkt_type == PACKET_LOOPBACK)
2129 sk = pt->af_packet_priv;
2132 if (!net_eq(dev_net(dev), sock_net(sk)))
2135 if (dev->header_ops) {
2136 if (sk->sk_type != SOCK_DGRAM)
2137 skb_push(skb, skb->data - skb_mac_header(skb));
2138 else if (skb->pkt_type == PACKET_OUTGOING) {
2139 /* Special case: outgoing packets have ll header at head */
2140 skb_pull(skb, skb_network_offset(skb));
2146 res = run_filter(skb, sk, snaplen);
2148 goto drop_n_restore;
2150 if (skb->ip_summed == CHECKSUM_PARTIAL)
2151 status |= TP_STATUS_CSUMNOTREADY;
2152 else if (skb->pkt_type != PACKET_OUTGOING &&
2153 (skb->ip_summed == CHECKSUM_COMPLETE ||
2154 skb_csum_unnecessary(skb)))
2155 status |= TP_STATUS_CSUM_VALID;
2160 if (sk->sk_type == SOCK_DGRAM) {
2161 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2164 unsigned int maclen = skb_network_offset(skb);
2165 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2166 (maclen < 16 ? 16 : maclen)) +
2168 macoff = netoff - maclen;
2170 if (po->tp_version <= TPACKET_V2) {
2171 if (macoff + snaplen > po->rx_ring.frame_size) {
2172 if (po->copy_thresh &&
2173 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2174 if (skb_shared(skb)) {
2175 copy_skb = skb_clone(skb, GFP_ATOMIC);
2177 copy_skb = skb_get(skb);
2178 skb_head = skb->data;
2181 skb_set_owner_r(copy_skb, sk);
2183 snaplen = po->rx_ring.frame_size - macoff;
2184 if ((int)snaplen < 0)
2187 } else if (unlikely(macoff + snaplen >
2188 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2191 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2192 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2193 snaplen, nval, macoff);
2195 if (unlikely((int)snaplen < 0)) {
2197 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2200 spin_lock(&sk->sk_receive_queue.lock);
2201 h.raw = packet_current_rx_frame(po, skb,
2202 TP_STATUS_KERNEL, (macoff+snaplen));
2205 if (po->tp_version <= TPACKET_V2) {
2206 packet_increment_rx_head(po, &po->rx_ring);
2208 * LOSING will be reported till you read the stats,
2209 * because it's COR - Clear On Read.
2210 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2213 if (po->stats.stats1.tp_drops)
2214 status |= TP_STATUS_LOSING;
2216 po->stats.stats1.tp_packets++;
2218 status |= TP_STATUS_COPY;
2219 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2221 spin_unlock(&sk->sk_receive_queue.lock);
2223 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2225 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2226 getnstimeofday(&ts);
2228 status |= ts_status;
2230 switch (po->tp_version) {
2232 h.h1->tp_len = skb->len;
2233 h.h1->tp_snaplen = snaplen;
2234 h.h1->tp_mac = macoff;
2235 h.h1->tp_net = netoff;
2236 h.h1->tp_sec = ts.tv_sec;
2237 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2238 hdrlen = sizeof(*h.h1);
2241 h.h2->tp_len = skb->len;
2242 h.h2->tp_snaplen = snaplen;
2243 h.h2->tp_mac = macoff;
2244 h.h2->tp_net = netoff;
2245 h.h2->tp_sec = ts.tv_sec;
2246 h.h2->tp_nsec = ts.tv_nsec;
2247 if (skb_vlan_tag_present(skb)) {
2248 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2249 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2250 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2252 h.h2->tp_vlan_tci = 0;
2253 h.h2->tp_vlan_tpid = 0;
2255 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2256 hdrlen = sizeof(*h.h2);
2259 /* tp_nxt_offset,vlan are already populated above.
2260 * So DONT clear those fields here
2262 h.h3->tp_status |= status;
2263 h.h3->tp_len = skb->len;
2264 h.h3->tp_snaplen = snaplen;
2265 h.h3->tp_mac = macoff;
2266 h.h3->tp_net = netoff;
2267 h.h3->tp_sec = ts.tv_sec;
2268 h.h3->tp_nsec = ts.tv_nsec;
2269 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2270 hdrlen = sizeof(*h.h3);
2276 sll = h.raw + TPACKET_ALIGN(hdrlen);
2277 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2278 sll->sll_family = AF_PACKET;
2279 sll->sll_hatype = dev->type;
2280 sll->sll_protocol = skb->protocol;
2281 sll->sll_pkttype = skb->pkt_type;
2282 if (unlikely(po->origdev))
2283 sll->sll_ifindex = orig_dev->ifindex;
2285 sll->sll_ifindex = dev->ifindex;
2289 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2290 if (po->tp_version <= TPACKET_V2) {
2293 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2296 for (start = h.raw; start < end; start += PAGE_SIZE)
2297 flush_dcache_page(pgv_to_page(start));
2302 if (po->tp_version <= TPACKET_V2) {
2303 __packet_set_status(po, h.raw, status);
2304 sk->sk_data_ready(sk);
2306 prb_clear_blk_fill_status(&po->rx_ring);
2310 if (skb_head != skb->data && skb_shared(skb)) {
2311 skb->data = skb_head;
2319 po->stats.stats1.tp_drops++;
2320 spin_unlock(&sk->sk_receive_queue.lock);
2322 sk->sk_data_ready(sk);
2323 kfree_skb(copy_skb);
2324 goto drop_n_restore;
2327 static void tpacket_destruct_skb(struct sk_buff *skb)
2329 struct packet_sock *po = pkt_sk(skb->sk);
2331 if (likely(po->tx_ring.pg_vec)) {
2335 ph = skb_shinfo(skb)->destructor_arg;
2336 packet_dec_pending(&po->tx_ring);
2338 ts = __packet_set_timestamp(po, ph, skb);
2339 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2345 static void tpacket_set_protocol(const struct net_device *dev,
2346 struct sk_buff *skb)
2348 if (dev->type == ARPHRD_ETHER) {
2349 skb_reset_mac_header(skb);
2350 skb->protocol = eth_hdr(skb)->h_proto;
2354 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2355 void *frame, struct net_device *dev, int size_max,
2356 __be16 proto, unsigned char *addr, int hlen)
2358 union tpacket_uhdr ph;
2359 int to_write, offset, len, tp_len, nr_frags, len_max;
2360 struct socket *sock = po->sk.sk_socket;
2367 skb->protocol = proto;
2369 skb->priority = po->sk.sk_priority;
2370 skb->mark = po->sk.sk_mark;
2371 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2372 skb_shinfo(skb)->destructor_arg = ph.raw;
2374 switch (po->tp_version) {
2376 tp_len = ph.h2->tp_len;
2379 tp_len = ph.h1->tp_len;
2382 if (unlikely(tp_len > size_max)) {
2383 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2387 skb_reserve(skb, hlen);
2388 skb_reset_network_header(skb);
2390 if (unlikely(po->tp_tx_has_off)) {
2391 int off_min, off_max, off;
2392 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2393 off_max = po->tx_ring.frame_size - tp_len;
2394 if (sock->type == SOCK_DGRAM) {
2395 switch (po->tp_version) {
2397 off = ph.h2->tp_net;
2400 off = ph.h1->tp_net;
2404 switch (po->tp_version) {
2406 off = ph.h2->tp_mac;
2409 off = ph.h1->tp_mac;
2413 if (unlikely((off < off_min) || (off_max < off)))
2415 data = ph.raw + off;
2417 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2421 if (sock->type == SOCK_DGRAM) {
2422 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2424 if (unlikely(err < 0))
2426 } else if (dev->hard_header_len) {
2427 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2429 skb_push(skb, dev->hard_header_len);
2430 err = skb_store_bits(skb, 0, data, hdrlen);
2433 if (!dev_validate_header(dev, skb->data, hdrlen))
2436 tpacket_set_protocol(dev, skb);
2442 offset = offset_in_page(data);
2443 len_max = PAGE_SIZE - offset;
2444 len = ((to_write > len_max) ? len_max : to_write);
2446 skb->data_len = to_write;
2447 skb->len += to_write;
2448 skb->truesize += to_write;
2449 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2451 while (likely(to_write)) {
2452 nr_frags = skb_shinfo(skb)->nr_frags;
2454 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2455 pr_err("Packet exceed the number of skb frags(%lu)\n",
2460 page = pgv_to_page(data);
2462 flush_dcache_page(page);
2464 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2467 len_max = PAGE_SIZE;
2468 len = ((to_write > len_max) ? len_max : to_write);
2471 skb_probe_transport_header(skb, 0);
2476 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2478 struct sk_buff *skb;
2479 struct net_device *dev;
2481 int err, reserve = 0;
2483 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2484 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2485 int tp_len, size_max;
2486 unsigned char *addr;
2488 int status = TP_STATUS_AVAILABLE;
2491 mutex_lock(&po->pg_vec_lock);
2493 if (likely(saddr == NULL)) {
2494 dev = packet_cached_dev_get(po);
2499 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2501 if (msg->msg_namelen < (saddr->sll_halen
2502 + offsetof(struct sockaddr_ll,
2505 proto = saddr->sll_protocol;
2506 addr = saddr->sll_addr;
2507 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2511 if (unlikely(dev == NULL))
2514 if (unlikely(!(dev->flags & IFF_UP)))
2517 if (po->sk.sk_socket->type == SOCK_RAW)
2518 reserve = dev->hard_header_len;
2519 size_max = po->tx_ring.frame_size
2520 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2522 if (size_max > dev->mtu + reserve + VLAN_HLEN)
2523 size_max = dev->mtu + reserve + VLAN_HLEN;
2526 ph = packet_current_frame(po, &po->tx_ring,
2527 TP_STATUS_SEND_REQUEST);
2528 if (unlikely(ph == NULL)) {
2529 if (need_wait && need_resched())
2534 status = TP_STATUS_SEND_REQUEST;
2535 hlen = LL_RESERVED_SPACE(dev);
2536 tlen = dev->needed_tailroom;
2537 skb = sock_alloc_send_skb(&po->sk,
2538 hlen + tlen + sizeof(struct sockaddr_ll),
2541 if (unlikely(skb == NULL)) {
2542 /* we assume the socket was initially writeable ... */
2543 if (likely(len_sum > 0))
2547 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2549 if (likely(tp_len >= 0) &&
2550 tp_len > dev->mtu + reserve &&
2551 !packet_extra_vlan_len_allowed(dev, skb))
2554 if (unlikely(tp_len < 0)) {
2556 __packet_set_status(po, ph,
2557 TP_STATUS_AVAILABLE);
2558 packet_increment_head(&po->tx_ring);
2562 status = TP_STATUS_WRONG_FORMAT;
2568 packet_pick_tx_queue(dev, skb);
2570 skb->destructor = tpacket_destruct_skb;
2571 __packet_set_status(po, ph, TP_STATUS_SENDING);
2572 packet_inc_pending(&po->tx_ring);
2574 status = TP_STATUS_SEND_REQUEST;
2575 err = po->xmit(skb);
2576 if (unlikely(err > 0)) {
2577 err = net_xmit_errno(err);
2578 if (err && __packet_get_status(po, ph) ==
2579 TP_STATUS_AVAILABLE) {
2580 /* skb was destructed already */
2585 * skb was dropped but not destructed yet;
2586 * let's treat it like congestion or err < 0
2590 packet_increment_head(&po->tx_ring);
2592 } while (likely((ph != NULL) ||
2593 /* Note: packet_read_pending() might be slow if we have
2594 * to call it as it's per_cpu variable, but in fast-path
2595 * we already short-circuit the loop with the first
2596 * condition, and luckily don't have to go that path
2599 (need_wait && packet_read_pending(&po->tx_ring))));
2605 __packet_set_status(po, ph, status);
2610 mutex_unlock(&po->pg_vec_lock);
2614 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2615 size_t reserve, size_t len,
2616 size_t linear, int noblock,
2619 struct sk_buff *skb;
2621 /* Under a page? Don't bother with paged skb. */
2622 if (prepad + len < PAGE_SIZE || !linear)
2625 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2630 skb_reserve(skb, reserve);
2631 skb_put(skb, linear);
2632 skb->data_len = len - linear;
2633 skb->len += len - linear;
2638 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2640 struct sock *sk = sock->sk;
2641 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2642 struct sk_buff *skb;
2643 struct net_device *dev;
2645 unsigned char *addr;
2646 int err, reserve = 0;
2647 struct sockcm_cookie sockc;
2648 struct virtio_net_hdr vnet_hdr = { 0 };
2651 struct packet_sock *po = pkt_sk(sk);
2652 unsigned short gso_type = 0;
2653 int hlen, tlen, linear;
2658 * Get and verify the address.
2661 if (likely(saddr == NULL)) {
2662 dev = packet_cached_dev_get(po);
2667 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2669 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2671 proto = saddr->sll_protocol;
2672 addr = saddr->sll_addr;
2673 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2677 if (unlikely(dev == NULL))
2680 if (unlikely(!(dev->flags & IFF_UP)))
2683 sockc.mark = sk->sk_mark;
2684 if (msg->msg_controllen) {
2685 err = sock_cmsg_send(sk, msg, &sockc);
2690 if (sock->type == SOCK_RAW)
2691 reserve = dev->hard_header_len;
2692 if (po->has_vnet_hdr) {
2693 vnet_hdr_len = sizeof(vnet_hdr);
2696 if (len < vnet_hdr_len)
2699 len -= vnet_hdr_len;
2702 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2703 if (n != vnet_hdr_len)
2706 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2707 (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2708 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2709 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2710 vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2711 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2712 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2715 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2718 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2719 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2720 case VIRTIO_NET_HDR_GSO_TCPV4:
2721 gso_type = SKB_GSO_TCPV4;
2723 case VIRTIO_NET_HDR_GSO_TCPV6:
2724 gso_type = SKB_GSO_TCPV6;
2726 case VIRTIO_NET_HDR_GSO_UDP:
2727 gso_type = SKB_GSO_UDP;
2733 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2734 gso_type |= SKB_GSO_TCP_ECN;
2736 if (vnet_hdr.gso_size == 0)
2742 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2743 if (!netif_supports_nofcs(dev)) {
2744 err = -EPROTONOSUPPORT;
2747 extra_len = 4; /* We're doing our own CRC */
2751 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2755 hlen = LL_RESERVED_SPACE(dev);
2756 tlen = dev->needed_tailroom;
2757 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2758 linear = max(linear, min_t(int, len, dev->hard_header_len));
2759 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2760 msg->msg_flags & MSG_DONTWAIT, &err);
2764 skb_set_network_header(skb, reserve);
2767 if (sock->type == SOCK_DGRAM) {
2768 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2769 if (unlikely(offset < 0))
2773 /* Returns -EFAULT on error */
2774 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2778 if (sock->type == SOCK_RAW &&
2779 !dev_validate_header(dev, skb->data, len)) {
2784 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2786 if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2787 !packet_extra_vlan_len_allowed(dev, skb)) {
2792 skb->protocol = proto;
2794 skb->priority = sk->sk_priority;
2795 skb->mark = sockc.mark;
2797 packet_pick_tx_queue(dev, skb);
2799 if (po->has_vnet_hdr) {
2800 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2801 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2802 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2803 if (!skb_partial_csum_set(skb, s, o)) {
2809 skb_shinfo(skb)->gso_size =
2810 __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2811 skb_shinfo(skb)->gso_type = gso_type;
2813 /* Header must be checked, and gso_segs computed. */
2814 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2815 skb_shinfo(skb)->gso_segs = 0;
2817 len += vnet_hdr_len;
2820 skb_probe_transport_header(skb, reserve);
2822 if (unlikely(extra_len == 4))
2825 err = po->xmit(skb);
2826 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2842 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2844 struct sock *sk = sock->sk;
2845 struct packet_sock *po = pkt_sk(sk);
2847 if (po->tx_ring.pg_vec)
2848 return tpacket_snd(po, msg);
2850 return packet_snd(sock, msg, len);
2854 * Close a PACKET socket. This is fairly simple. We immediately go
2855 * to 'closed' state and remove our protocol entry in the device list.
2858 static int packet_release(struct socket *sock)
2860 struct sock *sk = sock->sk;
2861 struct packet_sock *po;
2862 struct packet_fanout *f;
2864 union tpacket_req_u req_u;
2872 mutex_lock(&net->packet.sklist_lock);
2873 sk_del_node_init_rcu(sk);
2874 mutex_unlock(&net->packet.sklist_lock);
2877 sock_prot_inuse_add(net, sk->sk_prot, -1);
2880 spin_lock(&po->bind_lock);
2881 unregister_prot_hook(sk, false);
2882 packet_cached_dev_reset(po);
2884 if (po->prot_hook.dev) {
2885 dev_put(po->prot_hook.dev);
2886 po->prot_hook.dev = NULL;
2888 spin_unlock(&po->bind_lock);
2890 packet_flush_mclist(sk);
2892 if (po->rx_ring.pg_vec) {
2893 memset(&req_u, 0, sizeof(req_u));
2894 packet_set_ring(sk, &req_u, 1, 0);
2897 if (po->tx_ring.pg_vec) {
2898 memset(&req_u, 0, sizeof(req_u));
2899 packet_set_ring(sk, &req_u, 1, 1);
2902 f = fanout_release(sk);
2907 fanout_release_data(f);
2911 * Now the socket is dead. No more input will appear.
2918 skb_queue_purge(&sk->sk_receive_queue);
2919 packet_free_pending(po);
2920 sk_refcnt_debug_release(sk);
2927 * Attach a packet hook.
2930 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2933 struct packet_sock *po = pkt_sk(sk);
2934 struct net_device *dev_curr;
2937 struct net_device *dev = NULL;
2939 bool unlisted = false;
2945 spin_lock(&po->bind_lock);
2949 dev = dev_get_by_name_rcu(sock_net(sk), name);
2954 } else if (ifindex) {
2955 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2965 proto_curr = po->prot_hook.type;
2966 dev_curr = po->prot_hook.dev;
2968 need_rehook = proto_curr != proto || dev_curr != dev;
2973 __unregister_prot_hook(sk, true);
2975 dev_curr = po->prot_hook.dev;
2977 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2982 po->prot_hook.type = proto;
2984 if (unlikely(unlisted)) {
2986 po->prot_hook.dev = NULL;
2988 packet_cached_dev_reset(po);
2990 po->prot_hook.dev = dev;
2991 po->ifindex = dev ? dev->ifindex : 0;
2992 packet_cached_dev_assign(po, dev);
2998 if (proto == 0 || !need_rehook)
3001 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3002 register_prot_hook(sk);
3004 sk->sk_err = ENETDOWN;
3005 if (!sock_flag(sk, SOCK_DEAD))
3006 sk->sk_error_report(sk);
3011 spin_unlock(&po->bind_lock);
3017 * Bind a packet socket to a device
3020 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3023 struct sock *sk = sock->sk;
3024 char name[sizeof(uaddr->sa_data) + 1];
3030 if (addr_len != sizeof(struct sockaddr))
3032 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3035 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3036 name[sizeof(uaddr->sa_data)] = 0;
3038 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3041 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3043 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3044 struct sock *sk = sock->sk;
3050 if (addr_len < sizeof(struct sockaddr_ll))
3052 if (sll->sll_family != AF_PACKET)
3055 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3056 sll->sll_protocol ? : pkt_sk(sk)->num);
3059 static struct proto packet_proto = {
3061 .owner = THIS_MODULE,
3062 .obj_size = sizeof(struct packet_sock),
3066 * Create a packet of type SOCK_PACKET.
3069 static int packet_create(struct net *net, struct socket *sock, int protocol,
3073 struct packet_sock *po;
3074 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3077 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3079 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3080 sock->type != SOCK_PACKET)
3081 return -ESOCKTNOSUPPORT;
3083 sock->state = SS_UNCONNECTED;
3086 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3090 sock->ops = &packet_ops;
3091 if (sock->type == SOCK_PACKET)
3092 sock->ops = &packet_ops_spkt;
3094 sock_init_data(sock, sk);
3097 sk->sk_family = PF_PACKET;
3099 po->xmit = dev_queue_xmit;
3101 err = packet_alloc_pending(po);
3105 packet_cached_dev_reset(po);
3107 sk->sk_destruct = packet_sock_destruct;
3108 sk_refcnt_debug_inc(sk);
3111 * Attach a protocol block
3114 spin_lock_init(&po->bind_lock);
3115 mutex_init(&po->pg_vec_lock);
3116 po->rollover = NULL;
3117 po->prot_hook.func = packet_rcv;
3119 if (sock->type == SOCK_PACKET)
3120 po->prot_hook.func = packet_rcv_spkt;
3122 po->prot_hook.af_packet_priv = sk;
3125 po->prot_hook.type = proto;
3126 register_prot_hook(sk);
3129 mutex_lock(&net->packet.sklist_lock);
3130 sk_add_node_rcu(sk, &net->packet.sklist);
3131 mutex_unlock(&net->packet.sklist_lock);
3134 sock_prot_inuse_add(net, &packet_proto, 1);
3145 * Pull a packet from our receive queue and hand it to the user.
3146 * If necessary we block.
3149 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3152 struct sock *sk = sock->sk;
3153 struct sk_buff *skb;
3155 int vnet_hdr_len = 0;
3156 unsigned int origlen = 0;
3159 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3163 /* What error should we return now? EUNATTACH? */
3164 if (pkt_sk(sk)->ifindex < 0)
3168 if (flags & MSG_ERRQUEUE) {
3169 err = sock_recv_errqueue(sk, msg, len,
3170 SOL_PACKET, PACKET_TX_TIMESTAMP);
3175 * Call the generic datagram receiver. This handles all sorts
3176 * of horrible races and re-entrancy so we can forget about it
3177 * in the protocol layers.
3179 * Now it will return ENETDOWN, if device have just gone down,
3180 * but then it will block.
3183 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3186 * An error occurred so return it. Because skb_recv_datagram()
3187 * handles the blocking we don't see and worry about blocking
3194 if (pkt_sk(sk)->pressure)
3195 packet_rcv_has_room(pkt_sk(sk), NULL);
3197 if (pkt_sk(sk)->has_vnet_hdr) {
3198 struct virtio_net_hdr vnet_hdr = { 0 };
3201 vnet_hdr_len = sizeof(vnet_hdr);
3202 if (len < vnet_hdr_len)
3205 len -= vnet_hdr_len;
3207 if (skb_is_gso(skb)) {
3208 struct skb_shared_info *sinfo = skb_shinfo(skb);
3210 /* This is a hint as to how much should be linear. */
3212 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3214 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3215 if (sinfo->gso_type & SKB_GSO_TCPV4)
3216 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3217 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3218 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3219 else if (sinfo->gso_type & SKB_GSO_UDP)
3220 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3221 else if (sinfo->gso_type & SKB_GSO_FCOE)
3225 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3226 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3228 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3230 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3231 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3232 vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3233 skb_checksum_start_offset(skb));
3234 vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3236 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3237 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3238 } /* else everything is zero */
3240 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3245 /* You lose any data beyond the buffer you gave. If it worries
3246 * a user program they can ask the device for its MTU
3252 msg->msg_flags |= MSG_TRUNC;
3255 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3259 if (sock->type != SOCK_PACKET) {
3260 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3262 /* Original length was stored in sockaddr_ll fields */
3263 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3264 sll->sll_family = AF_PACKET;
3265 sll->sll_protocol = skb->protocol;
3268 sock_recv_ts_and_drops(msg, sk, skb);
3270 if (msg->msg_name) {
3271 /* If the address length field is there to be filled
3272 * in, we fill it in now.
3274 if (sock->type == SOCK_PACKET) {
3275 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3276 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3278 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3280 msg->msg_namelen = sll->sll_halen +
3281 offsetof(struct sockaddr_ll, sll_addr);
3283 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3287 if (pkt_sk(sk)->auxdata) {
3288 struct tpacket_auxdata aux;
3290 aux.tp_status = TP_STATUS_USER;
3291 if (skb->ip_summed == CHECKSUM_PARTIAL)
3292 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3293 else if (skb->pkt_type != PACKET_OUTGOING &&
3294 (skb->ip_summed == CHECKSUM_COMPLETE ||
3295 skb_csum_unnecessary(skb)))
3296 aux.tp_status |= TP_STATUS_CSUM_VALID;
3298 aux.tp_len = origlen;
3299 aux.tp_snaplen = skb->len;
3301 aux.tp_net = skb_network_offset(skb);
3302 if (skb_vlan_tag_present(skb)) {
3303 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3304 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3305 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3307 aux.tp_vlan_tci = 0;
3308 aux.tp_vlan_tpid = 0;
3310 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3314 * Free or return the buffer as appropriate. Again this
3315 * hides all the races and re-entrancy issues from us.
3317 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3320 skb_free_datagram(sk, skb);
3325 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3326 int *uaddr_len, int peer)
3328 struct net_device *dev;
3329 struct sock *sk = sock->sk;
3334 uaddr->sa_family = AF_PACKET;
3335 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3337 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3339 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3341 *uaddr_len = sizeof(*uaddr);
3346 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3347 int *uaddr_len, int peer)
3349 struct net_device *dev;
3350 struct sock *sk = sock->sk;
3351 struct packet_sock *po = pkt_sk(sk);
3352 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3357 sll->sll_family = AF_PACKET;
3358 sll->sll_ifindex = po->ifindex;
3359 sll->sll_protocol = po->num;
3360 sll->sll_pkttype = 0;
3362 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3364 sll->sll_hatype = dev->type;
3365 sll->sll_halen = dev->addr_len;
3366 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3368 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3372 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3377 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3381 case PACKET_MR_MULTICAST:
3382 if (i->alen != dev->addr_len)
3385 return dev_mc_add(dev, i->addr);
3387 return dev_mc_del(dev, i->addr);
3389 case PACKET_MR_PROMISC:
3390 return dev_set_promiscuity(dev, what);
3391 case PACKET_MR_ALLMULTI:
3392 return dev_set_allmulti(dev, what);
3393 case PACKET_MR_UNICAST:
3394 if (i->alen != dev->addr_len)
3397 return dev_uc_add(dev, i->addr);
3399 return dev_uc_del(dev, i->addr);
3407 static void packet_dev_mclist_delete(struct net_device *dev,
3408 struct packet_mclist **mlp)
3410 struct packet_mclist *ml;
3412 while ((ml = *mlp) != NULL) {
3413 if (ml->ifindex == dev->ifindex) {
3414 packet_dev_mc(dev, ml, -1);
3422 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3424 struct packet_sock *po = pkt_sk(sk);
3425 struct packet_mclist *ml, *i;
3426 struct net_device *dev;
3432 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3437 if (mreq->mr_alen > dev->addr_len)
3441 i = kmalloc(sizeof(*i), GFP_KERNEL);
3446 for (ml = po->mclist; ml; ml = ml->next) {
3447 if (ml->ifindex == mreq->mr_ifindex &&
3448 ml->type == mreq->mr_type &&
3449 ml->alen == mreq->mr_alen &&
3450 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3452 /* Free the new element ... */
3458 i->type = mreq->mr_type;
3459 i->ifindex = mreq->mr_ifindex;
3460 i->alen = mreq->mr_alen;
3461 memcpy(i->addr, mreq->mr_address, i->alen);
3462 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3464 i->next = po->mclist;
3466 err = packet_dev_mc(dev, i, 1);
3468 po->mclist = i->next;
3477 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3479 struct packet_mclist *ml, **mlp;
3483 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3484 if (ml->ifindex == mreq->mr_ifindex &&
3485 ml->type == mreq->mr_type &&
3486 ml->alen == mreq->mr_alen &&
3487 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3488 if (--ml->count == 0) {
3489 struct net_device *dev;
3491 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3493 packet_dev_mc(dev, ml, -1);
3503 static void packet_flush_mclist(struct sock *sk)
3505 struct packet_sock *po = pkt_sk(sk);
3506 struct packet_mclist *ml;
3512 while ((ml = po->mclist) != NULL) {
3513 struct net_device *dev;
3515 po->mclist = ml->next;
3516 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3518 packet_dev_mc(dev, ml, -1);
3525 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3527 struct sock *sk = sock->sk;
3528 struct packet_sock *po = pkt_sk(sk);
3531 if (level != SOL_PACKET)
3532 return -ENOPROTOOPT;
3535 case PACKET_ADD_MEMBERSHIP:
3536 case PACKET_DROP_MEMBERSHIP:
3538 struct packet_mreq_max mreq;
3540 memset(&mreq, 0, sizeof(mreq));
3541 if (len < sizeof(struct packet_mreq))
3543 if (len > sizeof(mreq))
3545 if (copy_from_user(&mreq, optval, len))
3547 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3549 if (optname == PACKET_ADD_MEMBERSHIP)
3550 ret = packet_mc_add(sk, &mreq);
3552 ret = packet_mc_drop(sk, &mreq);
3556 case PACKET_RX_RING:
3557 case PACKET_TX_RING:
3559 union tpacket_req_u req_u;
3562 switch (po->tp_version) {
3565 len = sizeof(req_u.req);
3569 len = sizeof(req_u.req3);
3574 if (pkt_sk(sk)->has_vnet_hdr)
3576 if (copy_from_user(&req_u.req, optval, len))
3578 return packet_set_ring(sk, &req_u, 0,
3579 optname == PACKET_TX_RING);
3581 case PACKET_COPY_THRESH:
3585 if (optlen != sizeof(val))
3587 if (copy_from_user(&val, optval, sizeof(val)))
3590 pkt_sk(sk)->copy_thresh = val;
3593 case PACKET_VERSION:
3597 if (optlen != sizeof(val))
3599 if (copy_from_user(&val, optval, sizeof(val)))
3610 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3613 po->tp_version = val;
3619 case PACKET_RESERVE:
3623 if (optlen != sizeof(val))
3625 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3627 if (copy_from_user(&val, optval, sizeof(val)))
3629 po->tp_reserve = val;
3636 if (optlen != sizeof(val))
3638 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3640 if (copy_from_user(&val, optval, sizeof(val)))
3642 po->tp_loss = !!val;
3645 case PACKET_AUXDATA:
3649 if (optlen < sizeof(val))
3651 if (copy_from_user(&val, optval, sizeof(val)))
3654 po->auxdata = !!val;
3657 case PACKET_ORIGDEV:
3661 if (optlen < sizeof(val))
3663 if (copy_from_user(&val, optval, sizeof(val)))
3666 po->origdev = !!val;
3669 case PACKET_VNET_HDR:
3673 if (sock->type != SOCK_RAW)
3675 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3677 if (optlen < sizeof(val))
3679 if (copy_from_user(&val, optval, sizeof(val)))
3682 po->has_vnet_hdr = !!val;
3685 case PACKET_TIMESTAMP:
3689 if (optlen != sizeof(val))
3691 if (copy_from_user(&val, optval, sizeof(val)))
3694 po->tp_tstamp = val;
3701 if (optlen != sizeof(val))
3703 if (copy_from_user(&val, optval, sizeof(val)))
3706 return fanout_add(sk, val & 0xffff, val >> 16);
3708 case PACKET_FANOUT_DATA:
3713 return fanout_set_data(po, optval, optlen);
3715 case PACKET_TX_HAS_OFF:
3719 if (optlen != sizeof(val))
3721 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3723 if (copy_from_user(&val, optval, sizeof(val)))
3725 po->tp_tx_has_off = !!val;
3728 case PACKET_QDISC_BYPASS:
3732 if (optlen != sizeof(val))
3734 if (copy_from_user(&val, optval, sizeof(val)))
3737 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3741 return -ENOPROTOOPT;
3745 static int packet_getsockopt(struct socket *sock, int level, int optname,
3746 char __user *optval, int __user *optlen)
3749 int val, lv = sizeof(val);
3750 struct sock *sk = sock->sk;
3751 struct packet_sock *po = pkt_sk(sk);
3753 union tpacket_stats_u st;
3754 struct tpacket_rollover_stats rstats;
3756 if (level != SOL_PACKET)
3757 return -ENOPROTOOPT;
3759 if (get_user(len, optlen))
3766 case PACKET_STATISTICS:
3767 spin_lock_bh(&sk->sk_receive_queue.lock);
3768 memcpy(&st, &po->stats, sizeof(st));
3769 memset(&po->stats, 0, sizeof(po->stats));
3770 spin_unlock_bh(&sk->sk_receive_queue.lock);
3772 if (po->tp_version == TPACKET_V3) {
3773 lv = sizeof(struct tpacket_stats_v3);
3774 st.stats3.tp_packets += st.stats3.tp_drops;
3777 lv = sizeof(struct tpacket_stats);
3778 st.stats1.tp_packets += st.stats1.tp_drops;
3783 case PACKET_AUXDATA:
3786 case PACKET_ORIGDEV:
3789 case PACKET_VNET_HDR:
3790 val = po->has_vnet_hdr;
3792 case PACKET_VERSION:
3793 val = po->tp_version;
3796 if (len > sizeof(int))
3798 if (copy_from_user(&val, optval, len))
3802 val = sizeof(struct tpacket_hdr);
3805 val = sizeof(struct tpacket2_hdr);
3808 val = sizeof(struct tpacket3_hdr);
3814 case PACKET_RESERVE:
3815 val = po->tp_reserve;
3820 case PACKET_TIMESTAMP:
3821 val = po->tp_tstamp;
3825 ((u32)po->fanout->id |
3826 ((u32)po->fanout->type << 16) |
3827 ((u32)po->fanout->flags << 24)) :
3830 case PACKET_ROLLOVER_STATS:
3833 rstats.tp_all = atomic_long_read(&po->rollover->num);
3834 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3835 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3837 lv = sizeof(rstats);
3839 case PACKET_TX_HAS_OFF:
3840 val = po->tp_tx_has_off;
3842 case PACKET_QDISC_BYPASS:
3843 val = packet_use_direct_xmit(po);
3846 return -ENOPROTOOPT;
3851 if (put_user(len, optlen))
3853 if (copy_to_user(optval, data, len))
3859 static int packet_notifier(struct notifier_block *this,
3860 unsigned long msg, void *ptr)
3863 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3864 struct net *net = dev_net(dev);
3867 sk_for_each_rcu(sk, &net->packet.sklist) {
3868 struct packet_sock *po = pkt_sk(sk);
3871 case NETDEV_UNREGISTER:
3873 packet_dev_mclist_delete(dev, &po->mclist);
3877 if (dev->ifindex == po->ifindex) {
3878 spin_lock(&po->bind_lock);
3880 __unregister_prot_hook(sk, false);
3881 sk->sk_err = ENETDOWN;
3882 if (!sock_flag(sk, SOCK_DEAD))
3883 sk->sk_error_report(sk);
3885 if (msg == NETDEV_UNREGISTER) {
3886 packet_cached_dev_reset(po);
3888 if (po->prot_hook.dev)
3889 dev_put(po->prot_hook.dev);
3890 po->prot_hook.dev = NULL;
3892 spin_unlock(&po->bind_lock);
3896 if (dev->ifindex == po->ifindex) {
3897 spin_lock(&po->bind_lock);
3899 register_prot_hook(sk);
3900 spin_unlock(&po->bind_lock);
3910 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3913 struct sock *sk = sock->sk;
3918 int amount = sk_wmem_alloc_get(sk);
3920 return put_user(amount, (int __user *)arg);
3924 struct sk_buff *skb;
3927 spin_lock_bh(&sk->sk_receive_queue.lock);
3928 skb = skb_peek(&sk->sk_receive_queue);
3931 spin_unlock_bh(&sk->sk_receive_queue.lock);
3932 return put_user(amount, (int __user *)arg);
3935 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3937 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3947 case SIOCGIFBRDADDR:
3948 case SIOCSIFBRDADDR:
3949 case SIOCGIFNETMASK:
3950 case SIOCSIFNETMASK:
3951 case SIOCGIFDSTADDR:
3952 case SIOCSIFDSTADDR:
3954 return inet_dgram_ops.ioctl(sock, cmd, arg);
3958 return -ENOIOCTLCMD;
3963 static unsigned int packet_poll(struct file *file, struct socket *sock,
3966 struct sock *sk = sock->sk;
3967 struct packet_sock *po = pkt_sk(sk);
3968 unsigned int mask = datagram_poll(file, sock, wait);
3970 spin_lock_bh(&sk->sk_receive_queue.lock);
3971 if (po->rx_ring.pg_vec) {
3972 if (!packet_previous_rx_frame(po, &po->rx_ring,
3974 mask |= POLLIN | POLLRDNORM;
3976 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3978 spin_unlock_bh(&sk->sk_receive_queue.lock);
3979 spin_lock_bh(&sk->sk_write_queue.lock);
3980 if (po->tx_ring.pg_vec) {
3981 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3982 mask |= POLLOUT | POLLWRNORM;
3984 spin_unlock_bh(&sk->sk_write_queue.lock);
3989 /* Dirty? Well, I still did not learn better way to account
3993 static void packet_mm_open(struct vm_area_struct *vma)
3995 struct file *file = vma->vm_file;
3996 struct socket *sock = file->private_data;
3997 struct sock *sk = sock->sk;
4000 atomic_inc(&pkt_sk(sk)->mapped);
4003 static void packet_mm_close(struct vm_area_struct *vma)
4005 struct file *file = vma->vm_file;
4006 struct socket *sock = file->private_data;
4007 struct sock *sk = sock->sk;
4010 atomic_dec(&pkt_sk(sk)->mapped);
4013 static const struct vm_operations_struct packet_mmap_ops = {
4014 .open = packet_mm_open,
4015 .close = packet_mm_close,
4018 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4023 for (i = 0; i < len; i++) {
4024 if (likely(pg_vec[i].buffer)) {
4025 if (is_vmalloc_addr(pg_vec[i].buffer))
4026 vfree(pg_vec[i].buffer);
4028 free_pages((unsigned long)pg_vec[i].buffer,
4030 pg_vec[i].buffer = NULL;
4036 static char *alloc_one_pg_vec_page(unsigned long order)
4039 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4040 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4042 buffer = (char *) __get_free_pages(gfp_flags, order);
4046 /* __get_free_pages failed, fall back to vmalloc */
4047 buffer = vzalloc((1 << order) * PAGE_SIZE);
4051 /* vmalloc failed, lets dig into swap here */
4052 gfp_flags &= ~__GFP_NORETRY;
4053 buffer = (char *) __get_free_pages(gfp_flags, order);
4057 /* complete and utter failure */
4061 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4063 unsigned int block_nr = req->tp_block_nr;
4067 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4068 if (unlikely(!pg_vec))
4071 for (i = 0; i < block_nr; i++) {
4072 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4073 if (unlikely(!pg_vec[i].buffer))
4074 goto out_free_pgvec;
4081 free_pg_vec(pg_vec, order, block_nr);
4086 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4087 int closing, int tx_ring)
4089 struct pgv *pg_vec = NULL;
4090 struct packet_sock *po = pkt_sk(sk);
4091 int was_running, order = 0;
4092 struct packet_ring_buffer *rb;
4093 struct sk_buff_head *rb_queue;
4096 /* Added to avoid minimal code churn */
4097 struct tpacket_req *req = &req_u->req;
4100 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4101 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4102 WARN(1, "Tx-ring is not supported.\n");
4106 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4107 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4111 if (atomic_read(&po->mapped))
4113 if (packet_read_pending(rb))
4117 if (req->tp_block_nr) {
4118 /* Sanity tests and some calculations */
4120 if (unlikely(rb->pg_vec))
4123 switch (po->tp_version) {
4125 po->tp_hdrlen = TPACKET_HDRLEN;
4128 po->tp_hdrlen = TPACKET2_HDRLEN;
4131 po->tp_hdrlen = TPACKET3_HDRLEN;
4136 if (unlikely((int)req->tp_block_size <= 0))
4138 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4140 if (po->tp_version >= TPACKET_V3 &&
4141 (int)(req->tp_block_size -
4142 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4144 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4147 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4150 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4151 if (unlikely(rb->frames_per_block == 0))
4153 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4158 order = get_order(req->tp_block_size);
4159 pg_vec = alloc_pg_vec(req, order);
4160 if (unlikely(!pg_vec))
4162 switch (po->tp_version) {
4164 /* Transmit path is not supported. We checked
4165 * it above but just being paranoid
4168 init_prb_bdqc(po, rb, pg_vec, req_u);
4177 if (unlikely(req->tp_frame_nr))
4182 /* Detach socket from network */
4183 spin_lock(&po->bind_lock);
4184 was_running = po->running;
4188 __unregister_prot_hook(sk, false);
4190 spin_unlock(&po->bind_lock);
4195 mutex_lock(&po->pg_vec_lock);
4196 if (closing || atomic_read(&po->mapped) == 0) {
4198 spin_lock_bh(&rb_queue->lock);
4199 swap(rb->pg_vec, pg_vec);
4200 rb->frame_max = (req->tp_frame_nr - 1);
4202 rb->frame_size = req->tp_frame_size;
4203 spin_unlock_bh(&rb_queue->lock);
4205 swap(rb->pg_vec_order, order);
4206 swap(rb->pg_vec_len, req->tp_block_nr);
4208 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4209 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4210 tpacket_rcv : packet_rcv;
4211 skb_queue_purge(rb_queue);
4212 if (atomic_read(&po->mapped))
4213 pr_err("packet_mmap: vma is busy: %d\n",
4214 atomic_read(&po->mapped));
4216 mutex_unlock(&po->pg_vec_lock);
4218 spin_lock(&po->bind_lock);
4221 register_prot_hook(sk);
4223 spin_unlock(&po->bind_lock);
4224 if (closing && (po->tp_version > TPACKET_V2)) {
4225 /* Because we don't support block-based V3 on tx-ring */
4227 prb_shutdown_retire_blk_timer(po, rb_queue);
4231 free_pg_vec(pg_vec, order, req->tp_block_nr);
4237 static int packet_mmap(struct file *file, struct socket *sock,
4238 struct vm_area_struct *vma)
4240 struct sock *sk = sock->sk;
4241 struct packet_sock *po = pkt_sk(sk);
4242 unsigned long size, expected_size;
4243 struct packet_ring_buffer *rb;
4244 unsigned long start;
4251 mutex_lock(&po->pg_vec_lock);
4254 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4256 expected_size += rb->pg_vec_len
4262 if (expected_size == 0)
4265 size = vma->vm_end - vma->vm_start;
4266 if (size != expected_size)
4269 start = vma->vm_start;
4270 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4271 if (rb->pg_vec == NULL)
4274 for (i = 0; i < rb->pg_vec_len; i++) {
4276 void *kaddr = rb->pg_vec[i].buffer;
4279 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4280 page = pgv_to_page(kaddr);
4281 err = vm_insert_page(vma, start, page);
4290 atomic_inc(&po->mapped);
4291 vma->vm_ops = &packet_mmap_ops;
4295 mutex_unlock(&po->pg_vec_lock);
4299 static const struct proto_ops packet_ops_spkt = {
4300 .family = PF_PACKET,
4301 .owner = THIS_MODULE,
4302 .release = packet_release,
4303 .bind = packet_bind_spkt,
4304 .connect = sock_no_connect,
4305 .socketpair = sock_no_socketpair,
4306 .accept = sock_no_accept,
4307 .getname = packet_getname_spkt,
4308 .poll = datagram_poll,
4309 .ioctl = packet_ioctl,
4310 .listen = sock_no_listen,
4311 .shutdown = sock_no_shutdown,
4312 .setsockopt = sock_no_setsockopt,
4313 .getsockopt = sock_no_getsockopt,
4314 .sendmsg = packet_sendmsg_spkt,
4315 .recvmsg = packet_recvmsg,
4316 .mmap = sock_no_mmap,
4317 .sendpage = sock_no_sendpage,
4320 static const struct proto_ops packet_ops = {
4321 .family = PF_PACKET,
4322 .owner = THIS_MODULE,
4323 .release = packet_release,
4324 .bind = packet_bind,
4325 .connect = sock_no_connect,
4326 .socketpair = sock_no_socketpair,
4327 .accept = sock_no_accept,
4328 .getname = packet_getname,
4329 .poll = packet_poll,
4330 .ioctl = packet_ioctl,
4331 .listen = sock_no_listen,
4332 .shutdown = sock_no_shutdown,
4333 .setsockopt = packet_setsockopt,
4334 .getsockopt = packet_getsockopt,
4335 .sendmsg = packet_sendmsg,
4336 .recvmsg = packet_recvmsg,
4337 .mmap = packet_mmap,
4338 .sendpage = sock_no_sendpage,
4341 static const struct net_proto_family packet_family_ops = {
4342 .family = PF_PACKET,
4343 .create = packet_create,
4344 .owner = THIS_MODULE,
4347 static struct notifier_block packet_netdev_notifier = {
4348 .notifier_call = packet_notifier,
4351 #ifdef CONFIG_PROC_FS
4353 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4356 struct net *net = seq_file_net(seq);
4359 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4362 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4364 struct net *net = seq_file_net(seq);
4365 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4368 static void packet_seq_stop(struct seq_file *seq, void *v)
4374 static int packet_seq_show(struct seq_file *seq, void *v)
4376 if (v == SEQ_START_TOKEN)
4377 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4379 struct sock *s = sk_entry(v);
4380 const struct packet_sock *po = pkt_sk(s);
4383 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4385 atomic_read(&s->sk_refcnt),
4390 atomic_read(&s->sk_rmem_alloc),
4391 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4398 static const struct seq_operations packet_seq_ops = {
4399 .start = packet_seq_start,
4400 .next = packet_seq_next,
4401 .stop = packet_seq_stop,
4402 .show = packet_seq_show,
4405 static int packet_seq_open(struct inode *inode, struct file *file)
4407 return seq_open_net(inode, file, &packet_seq_ops,
4408 sizeof(struct seq_net_private));
4411 static const struct file_operations packet_seq_fops = {
4412 .owner = THIS_MODULE,
4413 .open = packet_seq_open,
4415 .llseek = seq_lseek,
4416 .release = seq_release_net,
4421 static int __net_init packet_net_init(struct net *net)
4423 mutex_init(&net->packet.sklist_lock);
4424 INIT_HLIST_HEAD(&net->packet.sklist);
4426 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4432 static void __net_exit packet_net_exit(struct net *net)
4434 remove_proc_entry("packet", net->proc_net);
4437 static struct pernet_operations packet_net_ops = {
4438 .init = packet_net_init,
4439 .exit = packet_net_exit,
4443 static void __exit packet_exit(void)
4445 unregister_netdevice_notifier(&packet_netdev_notifier);
4446 unregister_pernet_subsys(&packet_net_ops);
4447 sock_unregister(PF_PACKET);
4448 proto_unregister(&packet_proto);
4451 static int __init packet_init(void)
4453 int rc = proto_register(&packet_proto, 0);
4458 sock_register(&packet_family_ops);
4459 register_pernet_subsys(&packet_net_ops);
4460 register_netdevice_notifier(&packet_netdev_notifier);
4465 module_init(packet_init);
4466 module_exit(packet_exit);
4467 MODULE_LICENSE("GPL");
4468 MODULE_ALIAS_NETPROTO(PF_PACKET);