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 <linux/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>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 struct packet_type *pt, struct net_device *orig_dev);
195 static void *packet_previous_frame(struct packet_sock *po,
196 struct packet_ring_buffer *rb,
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211 struct tpacket_kbdq_core *,
212 void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217 struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
220 struct packet_skb_cb {
222 struct sockaddr_pkt pkt;
224 /* Trick: alias skb original length with
225 * ll.sll_family and ll.protocol in order
228 unsigned int origlen;
229 struct sockaddr_ll ll;
234 #define vio_le() virtio_legacy_is_little_endian()
236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid) \
240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245 ((x)->kactive_blk_num+1) : 0)
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
250 static int packet_direct_xmit(struct sk_buff *skb)
252 struct net_device *dev = skb->dev;
253 struct sk_buff *orig_skb = skb;
254 struct netdev_queue *txq;
255 int ret = NETDEV_TX_BUSY;
257 if (unlikely(!netif_running(dev) ||
258 !netif_carrier_ok(dev)))
261 skb = validate_xmit_skb_list(skb, dev);
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
369 __fanout_unlink(sk, po);
371 __dev_remove_pack(&po->prot_hook);
376 spin_unlock(&po->bind_lock);
378 spin_lock(&po->bind_lock);
382 static void unregister_prot_hook(struct sock *sk, bool sync)
384 struct packet_sock *po = pkt_sk(sk);
387 __unregister_prot_hook(sk, sync);
390 static inline struct page * __pure pgv_to_page(void *addr)
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
399 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
413 WARN(1, "TPACKET version not supported.\n");
420 static int __packet_get_status(struct packet_sock *po, void *frame)
422 union tpacket_uhdr h;
427 switch (po->tp_version) {
429 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430 return h.h1->tp_status;
432 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433 return h.h2->tp_status;
436 WARN(1, "TPACKET version not supported.\n");
442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
445 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
448 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
452 if (ktime_to_timespec_cond(skb->tstamp, ts))
453 return TP_STATUS_TS_SOFTWARE;
458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
461 union tpacket_uhdr h;
465 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
469 switch (po->tp_version) {
471 h.h1->tp_sec = ts.tv_sec;
472 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
475 h.h2->tp_sec = ts.tv_sec;
476 h.h2->tp_nsec = ts.tv_nsec;
480 WARN(1, "TPACKET version not supported.\n");
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
491 static void *packet_lookup_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 unsigned int position,
496 unsigned int pg_vec_pos, frame_offset;
497 union tpacket_uhdr h;
499 pg_vec_pos = position / rb->frames_per_block;
500 frame_offset = position % rb->frames_per_block;
502 h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 (frame_offset * rb->frame_size);
505 if (status != __packet_get_status(po, h.raw))
511 static void *packet_current_frame(struct packet_sock *po,
512 struct packet_ring_buffer *rb,
515 return packet_lookup_frame(po, rb, rb->head, status);
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
520 del_timer_sync(&pkc->retire_blk_timer);
523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524 struct sk_buff_head *rb_queue)
526 struct tpacket_kbdq_core *pkc;
528 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
530 spin_lock_bh(&rb_queue->lock);
531 pkc->delete_blk_timer = 1;
532 spin_unlock_bh(&rb_queue->lock);
534 prb_del_retire_blk_timer(pkc);
537 static void prb_init_blk_timer(struct packet_sock *po,
538 struct tpacket_kbdq_core *pkc,
539 void (*func) (unsigned long))
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
549 struct tpacket_kbdq_core *pkc;
551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 int blk_size_in_bytes)
558 struct net_device *dev;
559 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560 struct ethtool_link_ksettings 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_link_ksettings(dev, &ecmd);
573 * If the link speed is so slow you don't really
574 * need to worry about perf anyways
576 if (ecmd.base.speed < SPEED_1000 ||
577 ecmd.base.speed == SPEED_UNKNOWN) {
578 return DEFAULT_PRB_RETIRE_TOV;
581 div = ecmd.base.speed / 1000;
585 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598 union tpacket_req_u *req_u)
600 p1->feature_req_word = req_u->req3.tp_feature_req_word;
603 static void init_prb_bdqc(struct packet_sock *po,
604 struct packet_ring_buffer *rb,
606 union tpacket_req_u *req_u)
608 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609 struct tpacket_block_desc *pbd;
611 memset(p1, 0x0, sizeof(*p1));
613 p1->knxt_seq_num = 1;
615 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616 p1->pkblk_start = pg_vec[0].buffer;
617 p1->kblk_size = req_u->req3.tp_block_size;
618 p1->knum_blocks = req_u->req3.tp_block_nr;
619 p1->hdrlen = po->tp_hdrlen;
620 p1->version = po->tp_version;
621 p1->last_kactive_blk_num = 0;
622 po->stats.stats3.tp_freeze_q_cnt = 0;
623 if (req_u->req3.tp_retire_blk_tov)
624 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627 req_u->req3.tp_block_size);
628 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632 prb_init_ft_ops(p1, req_u);
633 prb_setup_retire_blk_timer(po);
634 prb_open_block(p1, pbd);
637 /* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
642 mod_timer(&pkc->retire_blk_timer,
643 jiffies + pkc->tov_in_jiffies);
644 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
672 struct packet_sock *po = (struct packet_sock *)data;
673 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
675 struct tpacket_block_desc *pbd;
677 spin_lock(&po->sk.sk_receive_queue.lock);
679 frozen = prb_queue_frozen(pkc);
680 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682 if (unlikely(pkc->delete_blk_timer))
685 /* We only need to plug the race when the block is partially filled.
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
694 if (BLOCK_NUM_PKTS(pbd)) {
695 while (atomic_read(&pkc->blk_fill_in_prog)) {
696 /* Waiting for skb_copy_bits to finish... */
701 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 if (!BLOCK_NUM_PKTS(pbd)) {
704 /* An empty block. Just refresh the timer. */
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
713 /* Case 1. Queue was frozen because user-space was
716 if (prb_curr_blk_in_use(pkc, pbd)) {
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
730 prb_open_block(pkc, pbd);
737 _prb_refresh_rx_retire_blk_timer(pkc);
740 spin_unlock(&po->sk.sk_receive_queue.lock);
743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744 struct tpacket_block_desc *pbd1, __u32 status)
746 /* Flush everything minus the block header */
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753 /* Skip the block header(we know header WILL fit in 4K) */
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
763 /* Now update the block status. */
765 BLOCK_STATUS(pbd1) = status;
767 /* Flush the block header */
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 flush_dcache_page(pgv_to_page(start));
781 * 2) Increment active_blk_num
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
788 struct packet_sock *po, unsigned int stat)
790 __u32 status = TP_STATUS_USER | stat;
792 struct tpacket3_hdr *last_pkt;
793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794 struct sock *sk = &po->sk;
796 if (po->stats.stats3.tp_drops)
797 status |= TP_STATUS_LOSING;
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
807 /* Ok, we tmo'd - so get the current time.
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
814 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
820 /* Flush the block */
821 prb_flush_block(pkc1, pbd1, status);
823 sk->sk_data_ready(sk);
825 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
830 pkc->reset_pending_on_curr_blk = 0;
834 * Side effect of opening a block:
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841 struct tpacket_block_desc *pbd1)
844 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
852 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853 BLOCK_NUM_PKTS(pbd1) = 0;
854 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
858 h1->ts_first_pkt.ts_sec = ts.tv_sec;
859 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 pkc1->pkblk_start = (char *)pbd1;
862 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 pbd1->version = pkc1->version;
868 pkc1->prev = pkc1->nxt_offset;
869 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 prb_thaw_queue(pkc1);
872 _prb_refresh_rx_retire_blk_timer(pkc1);
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po)
903 pkc->reset_pending_on_curr_blk = 1;
904 po->stats.stats3.tp_freeze_q_cnt++;
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 struct packet_sock *po)
918 struct tpacket_block_desc *pbd;
922 /* 1. Get current block num */
923 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 prb_freeze_queue(pkc, po);
933 * open this block and return the offset where the first packet
934 * needs to get stored.
936 prb_open_block(pkc, pbd);
937 return (void *)pkc->nxt_offset;
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 struct packet_sock *po, unsigned int status)
943 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
956 if (!(status & TP_STATUS_BLK_TMO)) {
957 while (atomic_read(&pkc->blk_fill_in_prog)) {
958 /* Waiting for skb_copy_bits to finish... */
962 prb_close_block(pkc, pbd, po, status);
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968 struct tpacket_block_desc *pbd)
970 return TP_STATUS_USER & BLOCK_STATUS(pbd);
973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
975 return pkc->reset_pending_on_curr_blk;
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
981 atomic_dec(&pkc->blk_fill_in_prog);
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 struct tpacket3_hdr *ppd)
987 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 struct tpacket3_hdr *ppd)
993 ppd->hv1.tp_rxhash = 0;
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 struct tpacket3_hdr *ppd)
999 if (skb_vlan_tag_present(pkc->skb)) {
1000 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004 ppd->hv1.tp_vlan_tci = 0;
1005 ppd->hv1.tp_vlan_tpid = 0;
1006 ppd->tp_status = TP_STATUS_AVAILABLE;
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1013 ppd->hv1.tp_padding = 0;
1014 prb_fill_vlan_info(pkc, ppd);
1016 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017 prb_fill_rxhash(pkc, ppd);
1019 prb_clear_rxhash(pkc, ppd);
1022 static void prb_fill_curr_block(char *curr,
1023 struct tpacket_kbdq_core *pkc,
1024 struct tpacket_block_desc *pbd,
1027 struct tpacket3_hdr *ppd;
1029 ppd = (struct tpacket3_hdr *)curr;
1030 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_NUM_PKTS(pbd) += 1;
1035 atomic_inc(&pkc->blk_fill_in_prog);
1036 prb_run_all_ft_ops(pkc, ppd);
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041 struct sk_buff *skb,
1046 struct tpacket_kbdq_core *pkc;
1047 struct tpacket_block_desc *pbd;
1050 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc)) {
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1059 if (prb_curr_blk_in_use(pkc, pbd)) {
1060 /* Can't record this packet */
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1069 prb_open_block(pkc, pbd);
1074 curr = pkc->nxt_offset;
1076 end = (char *)pbd + pkc->kblk_size;
1078 /* first try the current block */
1079 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080 prb_fill_curr_block(curr, pkc, pbd, len);
1081 return (void *)curr;
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc, po, 0);
1087 /* Now, try to dispatch the next block */
1088 curr = (char *)prb_dispatch_next_block(pkc, po);
1090 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103 struct sk_buff *skb,
1104 int status, unsigned int len)
1107 switch (po->tp_version) {
1110 curr = packet_lookup_frame(po, &po->rx_ring,
1111 po->rx_ring.head, status);
1114 return __packet_lookup_frame_in_block(po, skb, status, len);
1116 WARN(1, "TPACKET version not supported\n");
1122 static void *prb_lookup_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1127 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1128 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130 if (status != BLOCK_STATUS(pbd))
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1138 if (rb->prb_bdqc.kactive_blk_num)
1139 prev = rb->prb_bdqc.kactive_blk_num-1;
1141 prev = rb->prb_bdqc.knum_blocks-1;
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1150 unsigned int previous = prb_previous_blk_num(rb);
1151 return prb_lookup_block(po, rb, previous, status);
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 if (po->tp_version <= TPACKET_V2)
1159 return packet_previous_frame(po, rb, status);
1161 return __prb_previous_block(po, rb, status);
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165 struct packet_ring_buffer *rb)
1167 switch (po->tp_version) {
1170 return packet_increment_head(rb);
1173 WARN(1, "TPACKET version not supported.\n");
1179 static void *packet_previous_frame(struct packet_sock *po,
1180 struct packet_ring_buffer *rb,
1183 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184 return packet_lookup_frame(po, rb, previous, status);
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 this_cpu_inc(*rb->pending_refcnt);
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 this_cpu_dec(*rb->pending_refcnt);
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 unsigned int refcnt = 0;
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb->pending_refcnt == NULL)
1211 for_each_possible_cpu(cpu)
1212 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217 static int packet_alloc_pending(struct packet_sock *po)
1219 po->rx_ring.pending_refcnt = NULL;
1221 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1228 static void packet_free_pending(struct packet_sock *po)
1230 free_percpu(po->tx_ring.pending_refcnt);
1233 #define ROOM_POW_OFF 2
1234 #define ROOM_NONE 0x0
1235 #define ROOM_LOW 0x1
1236 #define ROOM_NORMAL 0x2
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1242 len = po->rx_ring.frame_max + 1;
1243 idx = po->rx_ring.head;
1245 idx += len >> pow_off;
1248 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1255 len = po->rx_ring.prb_bdqc.knum_blocks;
1256 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 idx += len >> pow_off;
1261 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 struct sock *sk = &po->sk;
1267 int ret = ROOM_NONE;
1269 if (po->prot_hook.func != tpacket_rcv) {
1270 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271 - (skb ? skb->truesize : 0);
1272 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280 if (po->tp_version == TPACKET_V3) {
1281 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 else if (__tpacket_v3_has_room(po, 0))
1286 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_has_room(po, 0))
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1300 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301 ret = __packet_rcv_has_room(po, skb);
1302 has_room = ret == ROOM_NORMAL;
1303 if (po->pressure == has_room)
1304 po->pressure = !has_room;
1305 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1310 static void packet_sock_destruct(struct sock *sk)
1312 skb_queue_purge(&sk->sk_error_queue);
1314 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317 if (!sock_flag(sk, SOCK_DEAD)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk);
1322 sk_refcnt_debug_dec(sk);
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1330 rxhash = skb_get_hash(skb);
1331 for (i = 0; i < ROLLOVER_HLEN; i++)
1332 if (po->rollover->history[i] == rxhash)
1335 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336 return count > (ROLLOVER_HLEN >> 1);
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340 struct sk_buff *skb,
1343 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 unsigned int val = atomic_inc_return(&f->rr_cur);
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356 struct sk_buff *skb,
1359 return smp_processor_id() % num;
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return prandom_u32_max(num);
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370 struct sk_buff *skb,
1371 unsigned int idx, bool try_self,
1374 struct packet_sock *po, *po_next, *po_skip = NULL;
1375 unsigned int i, j, room = ROOM_NONE;
1377 po = pkt_sk(f->arr[idx]);
1380 room = packet_rcv_has_room(po, skb);
1381 if (room == ROOM_NORMAL ||
1382 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1387 i = j = min_t(int, po->rollover->sock, num - 1);
1389 po_next = pkt_sk(f->arr[i]);
1390 if (po_next != po_skip && !po_next->pressure &&
1391 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1393 po->rollover->sock = i;
1394 atomic_long_inc(&po->rollover->num);
1395 if (room == ROOM_LOW)
1396 atomic_long_inc(&po->rollover->num_huge);
1404 atomic_long_inc(&po->rollover->num_failed);
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409 struct sk_buff *skb,
1412 return skb_get_queue_mapping(skb) % num;
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416 struct sk_buff *skb,
1419 struct bpf_prog *prog;
1420 unsigned int ret = 0;
1423 prog = rcu_dereference(f->bpf_prog);
1425 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433 return f->flags & (flag >> 8);
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437 struct packet_type *pt, struct net_device *orig_dev)
1439 struct packet_fanout *f = pt->af_packet_priv;
1440 unsigned int num = READ_ONCE(f->num_members);
1441 struct net *net = read_pnet(&f->net);
1442 struct packet_sock *po;
1445 if (!net_eq(dev_net(dev), net) || !num) {
1450 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1456 case PACKET_FANOUT_HASH:
1458 idx = fanout_demux_hash(f, skb, num);
1460 case PACKET_FANOUT_LB:
1461 idx = fanout_demux_lb(f, skb, num);
1463 case PACKET_FANOUT_CPU:
1464 idx = fanout_demux_cpu(f, skb, num);
1466 case PACKET_FANOUT_RND:
1467 idx = fanout_demux_rnd(f, skb, num);
1469 case PACKET_FANOUT_QM:
1470 idx = fanout_demux_qm(f, skb, num);
1472 case PACKET_FANOUT_ROLLOVER:
1473 idx = fanout_demux_rollover(f, skb, 0, false, num);
1475 case PACKET_FANOUT_CBPF:
1476 case PACKET_FANOUT_EBPF:
1477 idx = fanout_demux_bpf(f, skb, num);
1481 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482 idx = fanout_demux_rollover(f, skb, idx, true, num);
1484 po = pkt_sk(f->arr[idx]);
1485 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494 struct packet_fanout *f = po->fanout;
1496 spin_lock(&f->lock);
1497 f->arr[f->num_members] = sk;
1500 spin_unlock(&f->lock);
1503 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505 struct packet_fanout *f = po->fanout;
1508 spin_lock(&f->lock);
1509 for (i = 0; i < f->num_members; i++) {
1510 if (f->arr[i] == sk)
1513 BUG_ON(i >= f->num_members);
1514 f->arr[i] = f->arr[f->num_members - 1];
1516 spin_unlock(&f->lock);
1519 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1521 if (sk->sk_family != PF_PACKET)
1524 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1527 static void fanout_init_data(struct packet_fanout *f)
1530 case PACKET_FANOUT_LB:
1531 atomic_set(&f->rr_cur, 0);
1533 case PACKET_FANOUT_CBPF:
1534 case PACKET_FANOUT_EBPF:
1535 RCU_INIT_POINTER(f->bpf_prog, NULL);
1540 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542 struct bpf_prog *old;
1544 spin_lock(&f->lock);
1545 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546 rcu_assign_pointer(f->bpf_prog, new);
1547 spin_unlock(&f->lock);
1551 bpf_prog_destroy(old);
1555 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1558 struct bpf_prog *new;
1559 struct sock_fprog fprog;
1562 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564 if (len != sizeof(fprog))
1566 if (copy_from_user(&fprog, data, len))
1569 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1573 __fanout_set_data_bpf(po->fanout, new);
1577 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1580 struct bpf_prog *new;
1583 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585 if (len != sizeof(fd))
1587 if (copy_from_user(&fd, data, len))
1590 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1592 return PTR_ERR(new);
1594 __fanout_set_data_bpf(po->fanout, new);
1598 static int fanout_set_data(struct packet_sock *po, char __user *data,
1601 switch (po->fanout->type) {
1602 case PACKET_FANOUT_CBPF:
1603 return fanout_set_data_cbpf(po, data, len);
1604 case PACKET_FANOUT_EBPF:
1605 return fanout_set_data_ebpf(po, data, len);
1611 static void fanout_release_data(struct packet_fanout *f)
1614 case PACKET_FANOUT_CBPF:
1615 case PACKET_FANOUT_EBPF:
1616 __fanout_set_data_bpf(f, NULL);
1620 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1622 struct packet_rollover *rollover = NULL;
1623 struct packet_sock *po = pkt_sk(sk);
1624 struct packet_fanout *f, *match;
1625 u8 type = type_flags & 0xff;
1626 u8 flags = type_flags >> 8;
1630 case PACKET_FANOUT_ROLLOVER:
1631 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1633 case PACKET_FANOUT_HASH:
1634 case PACKET_FANOUT_LB:
1635 case PACKET_FANOUT_CPU:
1636 case PACKET_FANOUT_RND:
1637 case PACKET_FANOUT_QM:
1638 case PACKET_FANOUT_CBPF:
1639 case PACKET_FANOUT_EBPF:
1645 mutex_lock(&fanout_mutex);
1655 if (type == PACKET_FANOUT_ROLLOVER ||
1656 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1658 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1661 atomic_long_set(&rollover->num, 0);
1662 atomic_long_set(&rollover->num_huge, 0);
1663 atomic_long_set(&rollover->num_failed, 0);
1664 po->rollover = rollover;
1668 list_for_each_entry(f, &fanout_list, list) {
1670 read_pnet(&f->net) == sock_net(sk)) {
1676 if (match && match->flags != flags)
1680 match = kzalloc(sizeof(*match), GFP_KERNEL);
1683 write_pnet(&match->net, sock_net(sk));
1686 match->flags = flags;
1687 INIT_LIST_HEAD(&match->list);
1688 spin_lock_init(&match->lock);
1689 atomic_set(&match->sk_ref, 0);
1690 fanout_init_data(match);
1691 match->prot_hook.type = po->prot_hook.type;
1692 match->prot_hook.dev = po->prot_hook.dev;
1693 match->prot_hook.func = packet_rcv_fanout;
1694 match->prot_hook.af_packet_priv = match;
1695 match->prot_hook.id_match = match_fanout_group;
1696 dev_add_pack(&match->prot_hook);
1697 list_add(&match->list, &fanout_list);
1700 if (match->type == type &&
1701 match->prot_hook.type == po->prot_hook.type &&
1702 match->prot_hook.dev == po->prot_hook.dev) {
1704 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1705 __dev_remove_pack(&po->prot_hook);
1707 atomic_inc(&match->sk_ref);
1708 __fanout_link(sk, po);
1713 if (err && rollover) {
1715 po->rollover = NULL;
1717 mutex_unlock(&fanout_mutex);
1721 static void fanout_release(struct sock *sk)
1723 struct packet_sock *po = pkt_sk(sk);
1724 struct packet_fanout *f;
1726 mutex_lock(&fanout_mutex);
1731 if (atomic_dec_and_test(&f->sk_ref)) {
1733 dev_remove_pack(&f->prot_hook);
1734 fanout_release_data(f);
1739 kfree_rcu(po->rollover, rcu);
1741 mutex_unlock(&fanout_mutex);
1744 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1745 struct sk_buff *skb)
1747 /* Earlier code assumed this would be a VLAN pkt, double-check
1748 * this now that we have the actual packet in hand. We can only
1749 * do this check on Ethernet devices.
1751 if (unlikely(dev->type != ARPHRD_ETHER))
1754 skb_reset_mac_header(skb);
1755 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1758 static const struct proto_ops packet_ops;
1760 static const struct proto_ops packet_ops_spkt;
1762 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1763 struct packet_type *pt, struct net_device *orig_dev)
1766 struct sockaddr_pkt *spkt;
1769 * When we registered the protocol we saved the socket in the data
1770 * field for just this event.
1773 sk = pt->af_packet_priv;
1776 * Yank back the headers [hope the device set this
1777 * right or kerboom...]
1779 * Incoming packets have ll header pulled,
1782 * For outgoing ones skb->data == skb_mac_header(skb)
1783 * so that this procedure is noop.
1786 if (skb->pkt_type == PACKET_LOOPBACK)
1789 if (!net_eq(dev_net(dev), sock_net(sk)))
1792 skb = skb_share_check(skb, GFP_ATOMIC);
1796 /* drop any routing info */
1799 /* drop conntrack reference */
1802 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1804 skb_push(skb, skb->data - skb_mac_header(skb));
1807 * The SOCK_PACKET socket receives _all_ frames.
1810 spkt->spkt_family = dev->type;
1811 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1812 spkt->spkt_protocol = skb->protocol;
1815 * Charge the memory to the socket. This is done specifically
1816 * to prevent sockets using all the memory up.
1819 if (sock_queue_rcv_skb(sk, skb) == 0)
1830 * Output a raw packet to a device layer. This bypasses all the other
1831 * protocol layers and you must therefore supply it with a complete frame
1834 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1837 struct sock *sk = sock->sk;
1838 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1839 struct sk_buff *skb = NULL;
1840 struct net_device *dev;
1841 struct sockcm_cookie sockc;
1847 * Get and verify the address.
1851 if (msg->msg_namelen < sizeof(struct sockaddr))
1853 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1854 proto = saddr->spkt_protocol;
1856 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1859 * Find the device first to size check it
1862 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1865 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1871 if (!(dev->flags & IFF_UP))
1875 * You may not queue a frame bigger than the mtu. This is the lowest level
1876 * raw protocol and you must do your own fragmentation at this level.
1879 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1880 if (!netif_supports_nofcs(dev)) {
1881 err = -EPROTONOSUPPORT;
1884 extra_len = 4; /* We're doing our own CRC */
1888 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1892 size_t reserved = LL_RESERVED_SPACE(dev);
1893 int tlen = dev->needed_tailroom;
1894 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1897 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1900 /* FIXME: Save some space for broken drivers that write a hard
1901 * header at transmission time by themselves. PPP is the notable
1902 * one here. This should really be fixed at the driver level.
1904 skb_reserve(skb, reserved);
1905 skb_reset_network_header(skb);
1907 /* Try to align data part correctly */
1912 skb_reset_network_header(skb);
1914 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1920 if (!dev_validate_header(dev, skb->data, len)) {
1924 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1925 !packet_extra_vlan_len_allowed(dev, skb)) {
1930 sockc.tsflags = sk->sk_tsflags;
1931 if (msg->msg_controllen) {
1932 err = sock_cmsg_send(sk, msg, &sockc);
1937 skb->protocol = proto;
1939 skb->priority = sk->sk_priority;
1940 skb->mark = sk->sk_mark;
1942 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1944 if (unlikely(extra_len == 4))
1947 skb_probe_transport_header(skb, 0);
1949 dev_queue_xmit(skb);
1960 static unsigned int run_filter(struct sk_buff *skb,
1961 const struct sock *sk,
1964 struct sk_filter *filter;
1967 filter = rcu_dereference(sk->sk_filter);
1969 res = bpf_prog_run_clear_cb(filter->prog, skb);
1975 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1978 struct virtio_net_hdr vnet_hdr;
1980 if (*len < sizeof(vnet_hdr))
1982 *len -= sizeof(vnet_hdr);
1984 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
1987 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
1991 * This function makes lazy skb cloning in hope that most of packets
1992 * are discarded by BPF.
1994 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1995 * and skb->cb are mangled. It works because (and until) packets
1996 * falling here are owned by current CPU. Output packets are cloned
1997 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1998 * sequencially, so that if we return skb to original state on exit,
1999 * we will not harm anyone.
2002 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2003 struct packet_type *pt, struct net_device *orig_dev)
2006 struct sockaddr_ll *sll;
2007 struct packet_sock *po;
2008 u8 *skb_head = skb->data;
2009 int skb_len = skb->len;
2010 unsigned int snaplen, res;
2011 bool is_drop_n_account = false;
2013 if (skb->pkt_type == PACKET_LOOPBACK)
2016 sk = pt->af_packet_priv;
2019 if (!net_eq(dev_net(dev), sock_net(sk)))
2024 if (dev->header_ops) {
2025 /* The device has an explicit notion of ll header,
2026 * exported to higher levels.
2028 * Otherwise, the device hides details of its frame
2029 * structure, so that corresponding packet head is
2030 * never delivered to user.
2032 if (sk->sk_type != SOCK_DGRAM)
2033 skb_push(skb, skb->data - skb_mac_header(skb));
2034 else if (skb->pkt_type == PACKET_OUTGOING) {
2035 /* Special case: outgoing packets have ll header at head */
2036 skb_pull(skb, skb_network_offset(skb));
2042 res = run_filter(skb, sk, snaplen);
2044 goto drop_n_restore;
2048 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2051 if (skb_shared(skb)) {
2052 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2056 if (skb_head != skb->data) {
2057 skb->data = skb_head;
2064 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2066 sll = &PACKET_SKB_CB(skb)->sa.ll;
2067 sll->sll_hatype = dev->type;
2068 sll->sll_pkttype = skb->pkt_type;
2069 if (unlikely(po->origdev))
2070 sll->sll_ifindex = orig_dev->ifindex;
2072 sll->sll_ifindex = dev->ifindex;
2074 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2076 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2077 * Use their space for storing the original skb length.
2079 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2081 if (pskb_trim(skb, snaplen))
2084 skb_set_owner_r(skb, sk);
2088 /* drop conntrack reference */
2091 spin_lock(&sk->sk_receive_queue.lock);
2092 po->stats.stats1.tp_packets++;
2093 sock_skb_set_dropcount(sk, skb);
2094 __skb_queue_tail(&sk->sk_receive_queue, skb);
2095 spin_unlock(&sk->sk_receive_queue.lock);
2096 sk->sk_data_ready(sk);
2100 is_drop_n_account = true;
2101 spin_lock(&sk->sk_receive_queue.lock);
2102 po->stats.stats1.tp_drops++;
2103 atomic_inc(&sk->sk_drops);
2104 spin_unlock(&sk->sk_receive_queue.lock);
2107 if (skb_head != skb->data && skb_shared(skb)) {
2108 skb->data = skb_head;
2112 if (!is_drop_n_account)
2119 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2120 struct packet_type *pt, struct net_device *orig_dev)
2123 struct packet_sock *po;
2124 struct sockaddr_ll *sll;
2125 union tpacket_uhdr h;
2126 u8 *skb_head = skb->data;
2127 int skb_len = skb->len;
2128 unsigned int snaplen, res;
2129 unsigned long status = TP_STATUS_USER;
2130 unsigned short macoff, netoff, hdrlen;
2131 struct sk_buff *copy_skb = NULL;
2134 bool is_drop_n_account = false;
2136 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2137 * We may add members to them until current aligned size without forcing
2138 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2140 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2141 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2143 if (skb->pkt_type == PACKET_LOOPBACK)
2146 sk = pt->af_packet_priv;
2149 if (!net_eq(dev_net(dev), sock_net(sk)))
2152 if (dev->header_ops) {
2153 if (sk->sk_type != SOCK_DGRAM)
2154 skb_push(skb, skb->data - skb_mac_header(skb));
2155 else if (skb->pkt_type == PACKET_OUTGOING) {
2156 /* Special case: outgoing packets have ll header at head */
2157 skb_pull(skb, skb_network_offset(skb));
2163 res = run_filter(skb, sk, snaplen);
2165 goto drop_n_restore;
2167 if (skb->ip_summed == CHECKSUM_PARTIAL)
2168 status |= TP_STATUS_CSUMNOTREADY;
2169 else if (skb->pkt_type != PACKET_OUTGOING &&
2170 (skb->ip_summed == CHECKSUM_COMPLETE ||
2171 skb_csum_unnecessary(skb)))
2172 status |= TP_STATUS_CSUM_VALID;
2177 if (sk->sk_type == SOCK_DGRAM) {
2178 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2181 unsigned int maclen = skb_network_offset(skb);
2182 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2183 (maclen < 16 ? 16 : maclen)) +
2185 if (po->has_vnet_hdr)
2186 netoff += sizeof(struct virtio_net_hdr);
2187 macoff = netoff - maclen;
2189 if (po->tp_version <= TPACKET_V2) {
2190 if (macoff + snaplen > po->rx_ring.frame_size) {
2191 if (po->copy_thresh &&
2192 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2193 if (skb_shared(skb)) {
2194 copy_skb = skb_clone(skb, GFP_ATOMIC);
2196 copy_skb = skb_get(skb);
2197 skb_head = skb->data;
2200 skb_set_owner_r(copy_skb, sk);
2202 snaplen = po->rx_ring.frame_size - macoff;
2203 if ((int)snaplen < 0)
2206 } else if (unlikely(macoff + snaplen >
2207 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2210 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2211 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2212 snaplen, nval, macoff);
2214 if (unlikely((int)snaplen < 0)) {
2216 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2219 spin_lock(&sk->sk_receive_queue.lock);
2220 h.raw = packet_current_rx_frame(po, skb,
2221 TP_STATUS_KERNEL, (macoff+snaplen));
2223 goto drop_n_account;
2224 if (po->tp_version <= TPACKET_V2) {
2225 packet_increment_rx_head(po, &po->rx_ring);
2227 * LOSING will be reported till you read the stats,
2228 * because it's COR - Clear On Read.
2229 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2232 if (po->stats.stats1.tp_drops)
2233 status |= TP_STATUS_LOSING;
2235 po->stats.stats1.tp_packets++;
2237 status |= TP_STATUS_COPY;
2238 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2240 spin_unlock(&sk->sk_receive_queue.lock);
2242 if (po->has_vnet_hdr) {
2243 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2244 sizeof(struct virtio_net_hdr),
2246 spin_lock(&sk->sk_receive_queue.lock);
2247 goto drop_n_account;
2251 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2253 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2254 getnstimeofday(&ts);
2256 status |= ts_status;
2258 switch (po->tp_version) {
2260 h.h1->tp_len = skb->len;
2261 h.h1->tp_snaplen = snaplen;
2262 h.h1->tp_mac = macoff;
2263 h.h1->tp_net = netoff;
2264 h.h1->tp_sec = ts.tv_sec;
2265 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2266 hdrlen = sizeof(*h.h1);
2269 h.h2->tp_len = skb->len;
2270 h.h2->tp_snaplen = snaplen;
2271 h.h2->tp_mac = macoff;
2272 h.h2->tp_net = netoff;
2273 h.h2->tp_sec = ts.tv_sec;
2274 h.h2->tp_nsec = ts.tv_nsec;
2275 if (skb_vlan_tag_present(skb)) {
2276 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2277 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2278 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2280 h.h2->tp_vlan_tci = 0;
2281 h.h2->tp_vlan_tpid = 0;
2283 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2284 hdrlen = sizeof(*h.h2);
2287 /* tp_nxt_offset,vlan are already populated above.
2288 * So DONT clear those fields here
2290 h.h3->tp_status |= status;
2291 h.h3->tp_len = skb->len;
2292 h.h3->tp_snaplen = snaplen;
2293 h.h3->tp_mac = macoff;
2294 h.h3->tp_net = netoff;
2295 h.h3->tp_sec = ts.tv_sec;
2296 h.h3->tp_nsec = ts.tv_nsec;
2297 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2298 hdrlen = sizeof(*h.h3);
2304 sll = h.raw + TPACKET_ALIGN(hdrlen);
2305 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2306 sll->sll_family = AF_PACKET;
2307 sll->sll_hatype = dev->type;
2308 sll->sll_protocol = skb->protocol;
2309 sll->sll_pkttype = skb->pkt_type;
2310 if (unlikely(po->origdev))
2311 sll->sll_ifindex = orig_dev->ifindex;
2313 sll->sll_ifindex = dev->ifindex;
2317 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2318 if (po->tp_version <= TPACKET_V2) {
2321 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2324 for (start = h.raw; start < end; start += PAGE_SIZE)
2325 flush_dcache_page(pgv_to_page(start));
2330 if (po->tp_version <= TPACKET_V2) {
2331 __packet_set_status(po, h.raw, status);
2332 sk->sk_data_ready(sk);
2334 prb_clear_blk_fill_status(&po->rx_ring);
2338 if (skb_head != skb->data && skb_shared(skb)) {
2339 skb->data = skb_head;
2343 if (!is_drop_n_account)
2350 is_drop_n_account = true;
2351 po->stats.stats1.tp_drops++;
2352 spin_unlock(&sk->sk_receive_queue.lock);
2354 sk->sk_data_ready(sk);
2355 kfree_skb(copy_skb);
2356 goto drop_n_restore;
2359 static void tpacket_destruct_skb(struct sk_buff *skb)
2361 struct packet_sock *po = pkt_sk(skb->sk);
2363 if (likely(po->tx_ring.pg_vec)) {
2367 ph = skb_shinfo(skb)->destructor_arg;
2368 packet_dec_pending(&po->tx_ring);
2370 ts = __packet_set_timestamp(po, ph, skb);
2371 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2377 static void tpacket_set_protocol(const struct net_device *dev,
2378 struct sk_buff *skb)
2380 if (dev->type == ARPHRD_ETHER) {
2381 skb_reset_mac_header(skb);
2382 skb->protocol = eth_hdr(skb)->h_proto;
2386 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2388 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2389 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2390 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2391 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2392 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2393 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2394 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2396 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2402 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2403 struct virtio_net_hdr *vnet_hdr)
2405 if (*len < sizeof(*vnet_hdr))
2407 *len -= sizeof(*vnet_hdr);
2409 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2412 return __packet_snd_vnet_parse(vnet_hdr, *len);
2415 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2416 void *frame, struct net_device *dev, void *data, int tp_len,
2417 __be16 proto, unsigned char *addr, int hlen, int copylen,
2418 const struct sockcm_cookie *sockc)
2420 union tpacket_uhdr ph;
2421 int to_write, offset, len, nr_frags, len_max;
2422 struct socket *sock = po->sk.sk_socket;
2428 skb->protocol = proto;
2430 skb->priority = po->sk.sk_priority;
2431 skb->mark = po->sk.sk_mark;
2432 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2433 skb_shinfo(skb)->destructor_arg = ph.raw;
2435 skb_reserve(skb, hlen);
2436 skb_reset_network_header(skb);
2440 if (sock->type == SOCK_DGRAM) {
2441 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2443 if (unlikely(err < 0))
2445 } else if (copylen) {
2446 int hdrlen = min_t(int, copylen, tp_len);
2448 skb_push(skb, dev->hard_header_len);
2449 skb_put(skb, copylen - dev->hard_header_len);
2450 err = skb_store_bits(skb, 0, data, hdrlen);
2453 if (!dev_validate_header(dev, skb->data, hdrlen))
2456 tpacket_set_protocol(dev, skb);
2462 offset = offset_in_page(data);
2463 len_max = PAGE_SIZE - offset;
2464 len = ((to_write > len_max) ? len_max : to_write);
2466 skb->data_len = to_write;
2467 skb->len += to_write;
2468 skb->truesize += to_write;
2469 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2471 while (likely(to_write)) {
2472 nr_frags = skb_shinfo(skb)->nr_frags;
2474 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2475 pr_err("Packet exceed the number of skb frags(%lu)\n",
2480 page = pgv_to_page(data);
2482 flush_dcache_page(page);
2484 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2487 len_max = PAGE_SIZE;
2488 len = ((to_write > len_max) ? len_max : to_write);
2491 skb_probe_transport_header(skb, 0);
2496 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2497 int size_max, void **data)
2499 union tpacket_uhdr ph;
2504 switch (po->tp_version) {
2506 tp_len = ph.h2->tp_len;
2509 tp_len = ph.h1->tp_len;
2512 if (unlikely(tp_len > size_max)) {
2513 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2517 if (unlikely(po->tp_tx_has_off)) {
2518 int off_min, off_max;
2520 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2521 off_max = po->tx_ring.frame_size - tp_len;
2522 if (po->sk.sk_type == SOCK_DGRAM) {
2523 switch (po->tp_version) {
2525 off = ph.h2->tp_net;
2528 off = ph.h1->tp_net;
2532 switch (po->tp_version) {
2534 off = ph.h2->tp_mac;
2537 off = ph.h1->tp_mac;
2541 if (unlikely((off < off_min) || (off_max < off)))
2544 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2547 *data = frame + off;
2551 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2553 struct sk_buff *skb;
2554 struct net_device *dev;
2555 struct virtio_net_hdr *vnet_hdr = NULL;
2556 struct sockcm_cookie sockc;
2558 int err, reserve = 0;
2560 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2561 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2562 int tp_len, size_max;
2563 unsigned char *addr;
2566 int status = TP_STATUS_AVAILABLE;
2567 int hlen, tlen, copylen = 0;
2569 mutex_lock(&po->pg_vec_lock);
2571 if (likely(saddr == NULL)) {
2572 dev = packet_cached_dev_get(po);
2577 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2579 if (msg->msg_namelen < (saddr->sll_halen
2580 + offsetof(struct sockaddr_ll,
2583 proto = saddr->sll_protocol;
2584 addr = saddr->sll_addr;
2585 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2588 sockc.tsflags = po->sk.sk_tsflags;
2589 if (msg->msg_controllen) {
2590 err = sock_cmsg_send(&po->sk, msg, &sockc);
2596 if (unlikely(dev == NULL))
2599 if (unlikely(!(dev->flags & IFF_UP)))
2602 if (po->sk.sk_socket->type == SOCK_RAW)
2603 reserve = dev->hard_header_len;
2604 size_max = po->tx_ring.frame_size
2605 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2607 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2608 size_max = dev->mtu + reserve + VLAN_HLEN;
2611 ph = packet_current_frame(po, &po->tx_ring,
2612 TP_STATUS_SEND_REQUEST);
2613 if (unlikely(ph == NULL)) {
2614 if (need_wait && need_resched())
2620 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2624 status = TP_STATUS_SEND_REQUEST;
2625 hlen = LL_RESERVED_SPACE(dev);
2626 tlen = dev->needed_tailroom;
2627 if (po->has_vnet_hdr) {
2629 data += sizeof(*vnet_hdr);
2630 tp_len -= sizeof(*vnet_hdr);
2632 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2636 copylen = __virtio16_to_cpu(vio_le(),
2639 copylen = max_t(int, copylen, dev->hard_header_len);
2640 skb = sock_alloc_send_skb(&po->sk,
2641 hlen + tlen + sizeof(struct sockaddr_ll) +
2642 (copylen - dev->hard_header_len),
2645 if (unlikely(skb == NULL)) {
2646 /* we assume the socket was initially writeable ... */
2647 if (likely(len_sum > 0))
2651 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2652 addr, hlen, copylen, &sockc);
2653 if (likely(tp_len >= 0) &&
2654 tp_len > dev->mtu + reserve &&
2655 !po->has_vnet_hdr &&
2656 !packet_extra_vlan_len_allowed(dev, skb))
2659 if (unlikely(tp_len < 0)) {
2662 __packet_set_status(po, ph,
2663 TP_STATUS_AVAILABLE);
2664 packet_increment_head(&po->tx_ring);
2668 status = TP_STATUS_WRONG_FORMAT;
2674 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2680 packet_pick_tx_queue(dev, skb);
2682 skb->destructor = tpacket_destruct_skb;
2683 __packet_set_status(po, ph, TP_STATUS_SENDING);
2684 packet_inc_pending(&po->tx_ring);
2686 status = TP_STATUS_SEND_REQUEST;
2687 err = po->xmit(skb);
2688 if (unlikely(err > 0)) {
2689 err = net_xmit_errno(err);
2690 if (err && __packet_get_status(po, ph) ==
2691 TP_STATUS_AVAILABLE) {
2692 /* skb was destructed already */
2697 * skb was dropped but not destructed yet;
2698 * let's treat it like congestion or err < 0
2702 packet_increment_head(&po->tx_ring);
2704 } while (likely((ph != NULL) ||
2705 /* Note: packet_read_pending() might be slow if we have
2706 * to call it as it's per_cpu variable, but in fast-path
2707 * we already short-circuit the loop with the first
2708 * condition, and luckily don't have to go that path
2711 (need_wait && packet_read_pending(&po->tx_ring))));
2717 __packet_set_status(po, ph, status);
2722 mutex_unlock(&po->pg_vec_lock);
2726 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2727 size_t reserve, size_t len,
2728 size_t linear, int noblock,
2731 struct sk_buff *skb;
2733 /* Under a page? Don't bother with paged skb. */
2734 if (prepad + len < PAGE_SIZE || !linear)
2737 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2742 skb_reserve(skb, reserve);
2743 skb_put(skb, linear);
2744 skb->data_len = len - linear;
2745 skb->len += len - linear;
2750 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2752 struct sock *sk = sock->sk;
2753 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2754 struct sk_buff *skb;
2755 struct net_device *dev;
2757 unsigned char *addr;
2758 int err, reserve = 0;
2759 struct sockcm_cookie sockc;
2760 struct virtio_net_hdr vnet_hdr = { 0 };
2762 struct packet_sock *po = pkt_sk(sk);
2763 int hlen, tlen, linear;
2767 * Get and verify the address.
2770 if (likely(saddr == NULL)) {
2771 dev = packet_cached_dev_get(po);
2776 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2778 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2780 proto = saddr->sll_protocol;
2781 addr = saddr->sll_addr;
2782 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2786 if (unlikely(dev == NULL))
2789 if (unlikely(!(dev->flags & IFF_UP)))
2792 sockc.tsflags = sk->sk_tsflags;
2793 sockc.mark = sk->sk_mark;
2794 if (msg->msg_controllen) {
2795 err = sock_cmsg_send(sk, msg, &sockc);
2800 if (sock->type == SOCK_RAW)
2801 reserve = dev->hard_header_len;
2802 if (po->has_vnet_hdr) {
2803 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2808 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2809 if (!netif_supports_nofcs(dev)) {
2810 err = -EPROTONOSUPPORT;
2813 extra_len = 4; /* We're doing our own CRC */
2817 if (!vnet_hdr.gso_type &&
2818 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2822 hlen = LL_RESERVED_SPACE(dev);
2823 tlen = dev->needed_tailroom;
2824 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2825 linear = max(linear, min_t(int, len, dev->hard_header_len));
2826 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2827 msg->msg_flags & MSG_DONTWAIT, &err);
2831 skb_set_network_header(skb, reserve);
2834 if (sock->type == SOCK_DGRAM) {
2835 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2836 if (unlikely(offset < 0))
2840 /* Returns -EFAULT on error */
2841 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2845 if (sock->type == SOCK_RAW &&
2846 !dev_validate_header(dev, skb->data, len)) {
2851 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2853 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2854 !packet_extra_vlan_len_allowed(dev, skb)) {
2859 skb->protocol = proto;
2861 skb->priority = sk->sk_priority;
2862 skb->mark = sockc.mark;
2864 packet_pick_tx_queue(dev, skb);
2866 if (po->has_vnet_hdr) {
2867 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2870 len += sizeof(vnet_hdr);
2873 skb_probe_transport_header(skb, reserve);
2875 if (unlikely(extra_len == 4))
2878 err = po->xmit(skb);
2879 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2895 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2897 struct sock *sk = sock->sk;
2898 struct packet_sock *po = pkt_sk(sk);
2900 if (po->tx_ring.pg_vec)
2901 return tpacket_snd(po, msg);
2903 return packet_snd(sock, msg, len);
2907 * Close a PACKET socket. This is fairly simple. We immediately go
2908 * to 'closed' state and remove our protocol entry in the device list.
2911 static int packet_release(struct socket *sock)
2913 struct sock *sk = sock->sk;
2914 struct packet_sock *po;
2916 union tpacket_req_u req_u;
2924 mutex_lock(&net->packet.sklist_lock);
2925 sk_del_node_init_rcu(sk);
2926 mutex_unlock(&net->packet.sklist_lock);
2929 sock_prot_inuse_add(net, sk->sk_prot, -1);
2932 spin_lock(&po->bind_lock);
2933 unregister_prot_hook(sk, false);
2934 packet_cached_dev_reset(po);
2936 if (po->prot_hook.dev) {
2937 dev_put(po->prot_hook.dev);
2938 po->prot_hook.dev = NULL;
2940 spin_unlock(&po->bind_lock);
2942 packet_flush_mclist(sk);
2944 if (po->rx_ring.pg_vec) {
2945 memset(&req_u, 0, sizeof(req_u));
2946 packet_set_ring(sk, &req_u, 1, 0);
2949 if (po->tx_ring.pg_vec) {
2950 memset(&req_u, 0, sizeof(req_u));
2951 packet_set_ring(sk, &req_u, 1, 1);
2958 * Now the socket is dead. No more input will appear.
2965 skb_queue_purge(&sk->sk_receive_queue);
2966 packet_free_pending(po);
2967 sk_refcnt_debug_release(sk);
2974 * Attach a packet hook.
2977 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2980 struct packet_sock *po = pkt_sk(sk);
2981 struct net_device *dev_curr;
2984 struct net_device *dev = NULL;
2986 bool unlisted = false;
2992 spin_lock(&po->bind_lock);
2996 dev = dev_get_by_name_rcu(sock_net(sk), name);
3001 } else if (ifindex) {
3002 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3012 proto_curr = po->prot_hook.type;
3013 dev_curr = po->prot_hook.dev;
3015 need_rehook = proto_curr != proto || dev_curr != dev;
3020 __unregister_prot_hook(sk, true);
3022 dev_curr = po->prot_hook.dev;
3024 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3029 po->prot_hook.type = proto;
3031 if (unlikely(unlisted)) {
3033 po->prot_hook.dev = NULL;
3035 packet_cached_dev_reset(po);
3037 po->prot_hook.dev = dev;
3038 po->ifindex = dev ? dev->ifindex : 0;
3039 packet_cached_dev_assign(po, dev);
3045 if (proto == 0 || !need_rehook)
3048 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3049 register_prot_hook(sk);
3051 sk->sk_err = ENETDOWN;
3052 if (!sock_flag(sk, SOCK_DEAD))
3053 sk->sk_error_report(sk);
3058 spin_unlock(&po->bind_lock);
3064 * Bind a packet socket to a device
3067 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3070 struct sock *sk = sock->sk;
3077 if (addr_len != sizeof(struct sockaddr))
3079 strlcpy(name, uaddr->sa_data, sizeof(name));
3081 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3084 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3086 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3087 struct sock *sk = sock->sk;
3093 if (addr_len < sizeof(struct sockaddr_ll))
3095 if (sll->sll_family != AF_PACKET)
3098 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3099 sll->sll_protocol ? : pkt_sk(sk)->num);
3102 static struct proto packet_proto = {
3104 .owner = THIS_MODULE,
3105 .obj_size = sizeof(struct packet_sock),
3109 * Create a packet of type SOCK_PACKET.
3112 static int packet_create(struct net *net, struct socket *sock, int protocol,
3116 struct packet_sock *po;
3117 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3120 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3122 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3123 sock->type != SOCK_PACKET)
3124 return -ESOCKTNOSUPPORT;
3126 sock->state = SS_UNCONNECTED;
3129 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3133 sock->ops = &packet_ops;
3134 if (sock->type == SOCK_PACKET)
3135 sock->ops = &packet_ops_spkt;
3137 sock_init_data(sock, sk);
3140 sk->sk_family = PF_PACKET;
3142 po->xmit = dev_queue_xmit;
3144 err = packet_alloc_pending(po);
3148 packet_cached_dev_reset(po);
3150 sk->sk_destruct = packet_sock_destruct;
3151 sk_refcnt_debug_inc(sk);
3154 * Attach a protocol block
3157 spin_lock_init(&po->bind_lock);
3158 mutex_init(&po->pg_vec_lock);
3159 po->rollover = NULL;
3160 po->prot_hook.func = packet_rcv;
3162 if (sock->type == SOCK_PACKET)
3163 po->prot_hook.func = packet_rcv_spkt;
3165 po->prot_hook.af_packet_priv = sk;
3168 po->prot_hook.type = proto;
3169 register_prot_hook(sk);
3172 mutex_lock(&net->packet.sklist_lock);
3173 sk_add_node_rcu(sk, &net->packet.sklist);
3174 mutex_unlock(&net->packet.sklist_lock);
3177 sock_prot_inuse_add(net, &packet_proto, 1);
3188 * Pull a packet from our receive queue and hand it to the user.
3189 * If necessary we block.
3192 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3195 struct sock *sk = sock->sk;
3196 struct sk_buff *skb;
3198 int vnet_hdr_len = 0;
3199 unsigned int origlen = 0;
3202 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3206 /* What error should we return now? EUNATTACH? */
3207 if (pkt_sk(sk)->ifindex < 0)
3211 if (flags & MSG_ERRQUEUE) {
3212 err = sock_recv_errqueue(sk, msg, len,
3213 SOL_PACKET, PACKET_TX_TIMESTAMP);
3218 * Call the generic datagram receiver. This handles all sorts
3219 * of horrible races and re-entrancy so we can forget about it
3220 * in the protocol layers.
3222 * Now it will return ENETDOWN, if device have just gone down,
3223 * but then it will block.
3226 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3229 * An error occurred so return it. Because skb_recv_datagram()
3230 * handles the blocking we don't see and worry about blocking
3237 if (pkt_sk(sk)->pressure)
3238 packet_rcv_has_room(pkt_sk(sk), NULL);
3240 if (pkt_sk(sk)->has_vnet_hdr) {
3241 err = packet_rcv_vnet(msg, skb, &len);
3244 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3247 /* You lose any data beyond the buffer you gave. If it worries
3248 * a user program they can ask the device for its MTU
3254 msg->msg_flags |= MSG_TRUNC;
3257 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3261 if (sock->type != SOCK_PACKET) {
3262 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3264 /* Original length was stored in sockaddr_ll fields */
3265 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3266 sll->sll_family = AF_PACKET;
3267 sll->sll_protocol = skb->protocol;
3270 sock_recv_ts_and_drops(msg, sk, skb);
3272 if (msg->msg_name) {
3273 /* If the address length field is there to be filled
3274 * in, we fill it in now.
3276 if (sock->type == SOCK_PACKET) {
3277 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3278 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3280 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3282 msg->msg_namelen = sll->sll_halen +
3283 offsetof(struct sockaddr_ll, sll_addr);
3285 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3289 if (pkt_sk(sk)->auxdata) {
3290 struct tpacket_auxdata aux;
3292 aux.tp_status = TP_STATUS_USER;
3293 if (skb->ip_summed == CHECKSUM_PARTIAL)
3294 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3295 else if (skb->pkt_type != PACKET_OUTGOING &&
3296 (skb->ip_summed == CHECKSUM_COMPLETE ||
3297 skb_csum_unnecessary(skb)))
3298 aux.tp_status |= TP_STATUS_CSUM_VALID;
3300 aux.tp_len = origlen;
3301 aux.tp_snaplen = skb->len;
3303 aux.tp_net = skb_network_offset(skb);
3304 if (skb_vlan_tag_present(skb)) {
3305 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3306 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3307 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3309 aux.tp_vlan_tci = 0;
3310 aux.tp_vlan_tpid = 0;
3312 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3316 * Free or return the buffer as appropriate. Again this
3317 * hides all the races and re-entrancy issues from us.
3319 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3322 skb_free_datagram(sk, skb);
3327 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3328 int *uaddr_len, int peer)
3330 struct net_device *dev;
3331 struct sock *sk = sock->sk;
3336 uaddr->sa_family = AF_PACKET;
3337 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3339 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3341 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3343 *uaddr_len = sizeof(*uaddr);
3348 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3349 int *uaddr_len, int peer)
3351 struct net_device *dev;
3352 struct sock *sk = sock->sk;
3353 struct packet_sock *po = pkt_sk(sk);
3354 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3359 sll->sll_family = AF_PACKET;
3360 sll->sll_ifindex = po->ifindex;
3361 sll->sll_protocol = po->num;
3362 sll->sll_pkttype = 0;
3364 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3366 sll->sll_hatype = dev->type;
3367 sll->sll_halen = dev->addr_len;
3368 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3370 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3374 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3379 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3383 case PACKET_MR_MULTICAST:
3384 if (i->alen != dev->addr_len)
3387 return dev_mc_add(dev, i->addr);
3389 return dev_mc_del(dev, i->addr);
3391 case PACKET_MR_PROMISC:
3392 return dev_set_promiscuity(dev, what);
3393 case PACKET_MR_ALLMULTI:
3394 return dev_set_allmulti(dev, what);
3395 case PACKET_MR_UNICAST:
3396 if (i->alen != dev->addr_len)
3399 return dev_uc_add(dev, i->addr);
3401 return dev_uc_del(dev, i->addr);
3409 static void packet_dev_mclist_delete(struct net_device *dev,
3410 struct packet_mclist **mlp)
3412 struct packet_mclist *ml;
3414 while ((ml = *mlp) != NULL) {
3415 if (ml->ifindex == dev->ifindex) {
3416 packet_dev_mc(dev, ml, -1);
3424 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3426 struct packet_sock *po = pkt_sk(sk);
3427 struct packet_mclist *ml, *i;
3428 struct net_device *dev;
3434 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3439 if (mreq->mr_alen > dev->addr_len)
3443 i = kmalloc(sizeof(*i), GFP_KERNEL);
3448 for (ml = po->mclist; ml; ml = ml->next) {
3449 if (ml->ifindex == mreq->mr_ifindex &&
3450 ml->type == mreq->mr_type &&
3451 ml->alen == mreq->mr_alen &&
3452 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3454 /* Free the new element ... */
3460 i->type = mreq->mr_type;
3461 i->ifindex = mreq->mr_ifindex;
3462 i->alen = mreq->mr_alen;
3463 memcpy(i->addr, mreq->mr_address, i->alen);
3464 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3466 i->next = po->mclist;
3468 err = packet_dev_mc(dev, i, 1);
3470 po->mclist = i->next;
3479 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3481 struct packet_mclist *ml, **mlp;
3485 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3486 if (ml->ifindex == mreq->mr_ifindex &&
3487 ml->type == mreq->mr_type &&
3488 ml->alen == mreq->mr_alen &&
3489 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3490 if (--ml->count == 0) {
3491 struct net_device *dev;
3493 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3495 packet_dev_mc(dev, ml, -1);
3505 static void packet_flush_mclist(struct sock *sk)
3507 struct packet_sock *po = pkt_sk(sk);
3508 struct packet_mclist *ml;
3514 while ((ml = po->mclist) != NULL) {
3515 struct net_device *dev;
3517 po->mclist = ml->next;
3518 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3520 packet_dev_mc(dev, ml, -1);
3527 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3529 struct sock *sk = sock->sk;
3530 struct packet_sock *po = pkt_sk(sk);
3533 if (level != SOL_PACKET)
3534 return -ENOPROTOOPT;
3537 case PACKET_ADD_MEMBERSHIP:
3538 case PACKET_DROP_MEMBERSHIP:
3540 struct packet_mreq_max mreq;
3542 memset(&mreq, 0, sizeof(mreq));
3543 if (len < sizeof(struct packet_mreq))
3545 if (len > sizeof(mreq))
3547 if (copy_from_user(&mreq, optval, len))
3549 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3551 if (optname == PACKET_ADD_MEMBERSHIP)
3552 ret = packet_mc_add(sk, &mreq);
3554 ret = packet_mc_drop(sk, &mreq);
3558 case PACKET_RX_RING:
3559 case PACKET_TX_RING:
3561 union tpacket_req_u req_u;
3564 switch (po->tp_version) {
3567 len = sizeof(req_u.req);
3571 len = sizeof(req_u.req3);
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 #ifdef CONFIG_COMPAT
3860 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3861 char __user *optval, unsigned int optlen)
3863 struct packet_sock *po = pkt_sk(sock->sk);
3865 if (level != SOL_PACKET)
3866 return -ENOPROTOOPT;
3868 if (optname == PACKET_FANOUT_DATA &&
3869 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3870 optval = (char __user *)get_compat_bpf_fprog(optval);
3873 optlen = sizeof(struct sock_fprog);
3876 return packet_setsockopt(sock, level, optname, optval, optlen);
3880 static int packet_notifier(struct notifier_block *this,
3881 unsigned long msg, void *ptr)
3884 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3885 struct net *net = dev_net(dev);
3888 sk_for_each_rcu(sk, &net->packet.sklist) {
3889 struct packet_sock *po = pkt_sk(sk);
3892 case NETDEV_UNREGISTER:
3894 packet_dev_mclist_delete(dev, &po->mclist);
3898 if (dev->ifindex == po->ifindex) {
3899 spin_lock(&po->bind_lock);
3901 __unregister_prot_hook(sk, false);
3902 sk->sk_err = ENETDOWN;
3903 if (!sock_flag(sk, SOCK_DEAD))
3904 sk->sk_error_report(sk);
3906 if (msg == NETDEV_UNREGISTER) {
3907 packet_cached_dev_reset(po);
3910 if (po->prot_hook.dev)
3911 dev_put(po->prot_hook.dev);
3912 po->prot_hook.dev = NULL;
3914 spin_unlock(&po->bind_lock);
3918 if (dev->ifindex == po->ifindex) {
3919 spin_lock(&po->bind_lock);
3921 register_prot_hook(sk);
3922 spin_unlock(&po->bind_lock);
3932 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3935 struct sock *sk = sock->sk;
3940 int amount = sk_wmem_alloc_get(sk);
3942 return put_user(amount, (int __user *)arg);
3946 struct sk_buff *skb;
3949 spin_lock_bh(&sk->sk_receive_queue.lock);
3950 skb = skb_peek(&sk->sk_receive_queue);
3953 spin_unlock_bh(&sk->sk_receive_queue.lock);
3954 return put_user(amount, (int __user *)arg);
3957 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3959 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3969 case SIOCGIFBRDADDR:
3970 case SIOCSIFBRDADDR:
3971 case SIOCGIFNETMASK:
3972 case SIOCSIFNETMASK:
3973 case SIOCGIFDSTADDR:
3974 case SIOCSIFDSTADDR:
3976 return inet_dgram_ops.ioctl(sock, cmd, arg);
3980 return -ENOIOCTLCMD;
3985 static unsigned int packet_poll(struct file *file, struct socket *sock,
3988 struct sock *sk = sock->sk;
3989 struct packet_sock *po = pkt_sk(sk);
3990 unsigned int mask = datagram_poll(file, sock, wait);
3992 spin_lock_bh(&sk->sk_receive_queue.lock);
3993 if (po->rx_ring.pg_vec) {
3994 if (!packet_previous_rx_frame(po, &po->rx_ring,
3996 mask |= POLLIN | POLLRDNORM;
3998 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4000 spin_unlock_bh(&sk->sk_receive_queue.lock);
4001 spin_lock_bh(&sk->sk_write_queue.lock);
4002 if (po->tx_ring.pg_vec) {
4003 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4004 mask |= POLLOUT | POLLWRNORM;
4006 spin_unlock_bh(&sk->sk_write_queue.lock);
4011 /* Dirty? Well, I still did not learn better way to account
4015 static void packet_mm_open(struct vm_area_struct *vma)
4017 struct file *file = vma->vm_file;
4018 struct socket *sock = file->private_data;
4019 struct sock *sk = sock->sk;
4022 atomic_inc(&pkt_sk(sk)->mapped);
4025 static void packet_mm_close(struct vm_area_struct *vma)
4027 struct file *file = vma->vm_file;
4028 struct socket *sock = file->private_data;
4029 struct sock *sk = sock->sk;
4032 atomic_dec(&pkt_sk(sk)->mapped);
4035 static const struct vm_operations_struct packet_mmap_ops = {
4036 .open = packet_mm_open,
4037 .close = packet_mm_close,
4040 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4045 for (i = 0; i < len; i++) {
4046 if (likely(pg_vec[i].buffer)) {
4047 if (is_vmalloc_addr(pg_vec[i].buffer))
4048 vfree(pg_vec[i].buffer);
4050 free_pages((unsigned long)pg_vec[i].buffer,
4052 pg_vec[i].buffer = NULL;
4058 static char *alloc_one_pg_vec_page(unsigned long order)
4061 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4062 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4064 buffer = (char *) __get_free_pages(gfp_flags, order);
4068 /* __get_free_pages failed, fall back to vmalloc */
4069 buffer = vzalloc((1 << order) * PAGE_SIZE);
4073 /* vmalloc failed, lets dig into swap here */
4074 gfp_flags &= ~__GFP_NORETRY;
4075 buffer = (char *) __get_free_pages(gfp_flags, order);
4079 /* complete and utter failure */
4083 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4085 unsigned int block_nr = req->tp_block_nr;
4089 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4090 if (unlikely(!pg_vec))
4093 for (i = 0; i < block_nr; i++) {
4094 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4095 if (unlikely(!pg_vec[i].buffer))
4096 goto out_free_pgvec;
4103 free_pg_vec(pg_vec, order, block_nr);
4108 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4109 int closing, int tx_ring)
4111 struct pgv *pg_vec = NULL;
4112 struct packet_sock *po = pkt_sk(sk);
4113 int was_running, order = 0;
4114 struct packet_ring_buffer *rb;
4115 struct sk_buff_head *rb_queue;
4118 /* Added to avoid minimal code churn */
4119 struct tpacket_req *req = &req_u->req;
4122 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4123 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4124 net_warn_ratelimited("Tx-ring is not supported.\n");
4128 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4129 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4133 if (atomic_read(&po->mapped))
4135 if (packet_read_pending(rb))
4139 if (req->tp_block_nr) {
4140 /* Sanity tests and some calculations */
4142 if (unlikely(rb->pg_vec))
4145 switch (po->tp_version) {
4147 po->tp_hdrlen = TPACKET_HDRLEN;
4150 po->tp_hdrlen = TPACKET2_HDRLEN;
4153 po->tp_hdrlen = TPACKET3_HDRLEN;
4158 if (unlikely((int)req->tp_block_size <= 0))
4160 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4162 if (po->tp_version >= TPACKET_V3 &&
4163 (int)(req->tp_block_size -
4164 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4166 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4169 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4172 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4173 if (unlikely(rb->frames_per_block == 0))
4175 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4180 order = get_order(req->tp_block_size);
4181 pg_vec = alloc_pg_vec(req, order);
4182 if (unlikely(!pg_vec))
4184 switch (po->tp_version) {
4186 /* Transmit path is not supported. We checked
4187 * it above but just being paranoid
4190 init_prb_bdqc(po, rb, pg_vec, req_u);
4199 if (unlikely(req->tp_frame_nr))
4204 /* Detach socket from network */
4205 spin_lock(&po->bind_lock);
4206 was_running = po->running;
4210 __unregister_prot_hook(sk, false);
4212 spin_unlock(&po->bind_lock);
4217 mutex_lock(&po->pg_vec_lock);
4218 if (closing || atomic_read(&po->mapped) == 0) {
4220 spin_lock_bh(&rb_queue->lock);
4221 swap(rb->pg_vec, pg_vec);
4222 rb->frame_max = (req->tp_frame_nr - 1);
4224 rb->frame_size = req->tp_frame_size;
4225 spin_unlock_bh(&rb_queue->lock);
4227 swap(rb->pg_vec_order, order);
4228 swap(rb->pg_vec_len, req->tp_block_nr);
4230 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4231 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4232 tpacket_rcv : packet_rcv;
4233 skb_queue_purge(rb_queue);
4234 if (atomic_read(&po->mapped))
4235 pr_err("packet_mmap: vma is busy: %d\n",
4236 atomic_read(&po->mapped));
4238 mutex_unlock(&po->pg_vec_lock);
4240 spin_lock(&po->bind_lock);
4243 register_prot_hook(sk);
4245 spin_unlock(&po->bind_lock);
4246 if (closing && (po->tp_version > TPACKET_V2)) {
4247 /* Because we don't support block-based V3 on tx-ring */
4249 prb_shutdown_retire_blk_timer(po, rb_queue);
4253 free_pg_vec(pg_vec, order, req->tp_block_nr);
4259 static int packet_mmap(struct file *file, struct socket *sock,
4260 struct vm_area_struct *vma)
4262 struct sock *sk = sock->sk;
4263 struct packet_sock *po = pkt_sk(sk);
4264 unsigned long size, expected_size;
4265 struct packet_ring_buffer *rb;
4266 unsigned long start;
4273 mutex_lock(&po->pg_vec_lock);
4276 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4278 expected_size += rb->pg_vec_len
4284 if (expected_size == 0)
4287 size = vma->vm_end - vma->vm_start;
4288 if (size != expected_size)
4291 start = vma->vm_start;
4292 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4293 if (rb->pg_vec == NULL)
4296 for (i = 0; i < rb->pg_vec_len; i++) {
4298 void *kaddr = rb->pg_vec[i].buffer;
4301 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4302 page = pgv_to_page(kaddr);
4303 err = vm_insert_page(vma, start, page);
4312 atomic_inc(&po->mapped);
4313 vma->vm_ops = &packet_mmap_ops;
4317 mutex_unlock(&po->pg_vec_lock);
4321 static const struct proto_ops packet_ops_spkt = {
4322 .family = PF_PACKET,
4323 .owner = THIS_MODULE,
4324 .release = packet_release,
4325 .bind = packet_bind_spkt,
4326 .connect = sock_no_connect,
4327 .socketpair = sock_no_socketpair,
4328 .accept = sock_no_accept,
4329 .getname = packet_getname_spkt,
4330 .poll = datagram_poll,
4331 .ioctl = packet_ioctl,
4332 .listen = sock_no_listen,
4333 .shutdown = sock_no_shutdown,
4334 .setsockopt = sock_no_setsockopt,
4335 .getsockopt = sock_no_getsockopt,
4336 .sendmsg = packet_sendmsg_spkt,
4337 .recvmsg = packet_recvmsg,
4338 .mmap = sock_no_mmap,
4339 .sendpage = sock_no_sendpage,
4342 static const struct proto_ops packet_ops = {
4343 .family = PF_PACKET,
4344 .owner = THIS_MODULE,
4345 .release = packet_release,
4346 .bind = packet_bind,
4347 .connect = sock_no_connect,
4348 .socketpair = sock_no_socketpair,
4349 .accept = sock_no_accept,
4350 .getname = packet_getname,
4351 .poll = packet_poll,
4352 .ioctl = packet_ioctl,
4353 .listen = sock_no_listen,
4354 .shutdown = sock_no_shutdown,
4355 .setsockopt = packet_setsockopt,
4356 .getsockopt = packet_getsockopt,
4357 #ifdef CONFIG_COMPAT
4358 .compat_setsockopt = compat_packet_setsockopt,
4360 .sendmsg = packet_sendmsg,
4361 .recvmsg = packet_recvmsg,
4362 .mmap = packet_mmap,
4363 .sendpage = sock_no_sendpage,
4366 static const struct net_proto_family packet_family_ops = {
4367 .family = PF_PACKET,
4368 .create = packet_create,
4369 .owner = THIS_MODULE,
4372 static struct notifier_block packet_netdev_notifier = {
4373 .notifier_call = packet_notifier,
4376 #ifdef CONFIG_PROC_FS
4378 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4381 struct net *net = seq_file_net(seq);
4384 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4387 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4389 struct net *net = seq_file_net(seq);
4390 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4393 static void packet_seq_stop(struct seq_file *seq, void *v)
4399 static int packet_seq_show(struct seq_file *seq, void *v)
4401 if (v == SEQ_START_TOKEN)
4402 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4404 struct sock *s = sk_entry(v);
4405 const struct packet_sock *po = pkt_sk(s);
4408 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4410 atomic_read(&s->sk_refcnt),
4415 atomic_read(&s->sk_rmem_alloc),
4416 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4423 static const struct seq_operations packet_seq_ops = {
4424 .start = packet_seq_start,
4425 .next = packet_seq_next,
4426 .stop = packet_seq_stop,
4427 .show = packet_seq_show,
4430 static int packet_seq_open(struct inode *inode, struct file *file)
4432 return seq_open_net(inode, file, &packet_seq_ops,
4433 sizeof(struct seq_net_private));
4436 static const struct file_operations packet_seq_fops = {
4437 .owner = THIS_MODULE,
4438 .open = packet_seq_open,
4440 .llseek = seq_lseek,
4441 .release = seq_release_net,
4446 static int __net_init packet_net_init(struct net *net)
4448 mutex_init(&net->packet.sklist_lock);
4449 INIT_HLIST_HEAD(&net->packet.sklist);
4451 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4457 static void __net_exit packet_net_exit(struct net *net)
4459 remove_proc_entry("packet", net->proc_net);
4462 static struct pernet_operations packet_net_ops = {
4463 .init = packet_net_init,
4464 .exit = packet_net_exit,
4468 static void __exit packet_exit(void)
4470 unregister_netdevice_notifier(&packet_netdev_notifier);
4471 unregister_pernet_subsys(&packet_net_ops);
4472 sock_unregister(PF_PACKET);
4473 proto_unregister(&packet_proto);
4476 static int __init packet_init(void)
4478 int rc = proto_register(&packet_proto, 0);
4483 sock_register(&packet_family_ops);
4484 register_pernet_subsys(&packet_net_ops);
4485 register_netdevice_notifier(&packet_netdev_notifier);
4490 module_init(packet_init);
4491 module_exit(packet_exit);
4492 MODULE_LICENSE("GPL");
4493 MODULE_ALIAS_NETPROTO(PF_PACKET);