2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
218 struct packet_skb_cb {
220 struct sockaddr_pkt pkt;
222 /* Trick: alias skb original length with
223 * ll.sll_family and ll.protocol in order
226 unsigned int origlen;
227 struct sockaddr_ll ll;
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 static int packet_direct_xmit(struct sk_buff *skb)
248 struct net_device *dev = skb->dev;
249 netdev_features_t features;
250 struct netdev_queue *txq;
251 int ret = NETDEV_TX_BUSY;
253 if (unlikely(!netif_running(dev) ||
254 !netif_carrier_ok(dev)))
257 features = netif_skb_features(skb);
258 if (skb_needs_linearize(skb, features) &&
259 __skb_linearize(skb))
262 txq = skb_get_tx_queue(dev, skb);
266 HARD_TX_LOCK(dev, txq, smp_processor_id());
267 if (!netif_xmit_frozen_or_drv_stopped(txq))
268 ret = netdev_start_xmit(skb, dev, txq, false);
269 HARD_TX_UNLOCK(dev, txq);
273 if (!dev_xmit_complete(ret))
278 atomic_long_inc(&dev->tx_dropped);
280 return NET_XMIT_DROP;
283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
285 struct net_device *dev;
288 dev = rcu_dereference(po->cached_dev);
296 static void packet_cached_dev_assign(struct packet_sock *po,
297 struct net_device *dev)
299 rcu_assign_pointer(po->cached_dev, dev);
302 static void packet_cached_dev_reset(struct packet_sock *po)
304 RCU_INIT_POINTER(po->cached_dev, NULL);
307 static bool packet_use_direct_xmit(const struct packet_sock *po)
309 return po->xmit == packet_direct_xmit;
312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
314 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
319 const struct net_device_ops *ops = dev->netdev_ops;
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL,
324 __packet_pick_tx_queue);
325 queue_index = netdev_cap_txqueue(dev, queue_index);
327 queue_index = __packet_pick_tx_queue(dev, skb);
330 skb_set_queue_mapping(skb, queue_index);
333 /* register_prot_hook must be invoked with the po->bind_lock held,
334 * or from a context in which asynchronous accesses to the packet
335 * socket is not possible (packet_create()).
337 static void register_prot_hook(struct sock *sk)
339 struct packet_sock *po = pkt_sk(sk);
343 __fanout_link(sk, po);
345 dev_add_pack(&po->prot_hook);
352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
353 * held. If the sync parameter is true, we will temporarily drop
354 * the po->bind_lock and do a synchronize_net to make sure no
355 * asynchronous packet processing paths still refer to the elements
356 * of po->prot_hook. If the sync parameter is false, it is the
357 * callers responsibility to take care of this.
359 static void __unregister_prot_hook(struct sock *sk, bool sync)
361 struct packet_sock *po = pkt_sk(sk);
366 __fanout_unlink(sk, po);
368 __dev_remove_pack(&po->prot_hook);
373 spin_unlock(&po->bind_lock);
375 spin_lock(&po->bind_lock);
379 static void unregister_prot_hook(struct sock *sk, bool sync)
381 struct packet_sock *po = pkt_sk(sk);
384 __unregister_prot_hook(sk, sync);
387 static inline struct page * __pure pgv_to_page(void *addr)
389 if (is_vmalloc_addr(addr))
390 return vmalloc_to_page(addr);
391 return virt_to_page(addr);
394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
396 union tpacket_uhdr h;
399 switch (po->tp_version) {
401 h.h1->tp_status = status;
402 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 h.h2->tp_status = status;
406 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 WARN(1, "TPACKET version not supported.\n");
417 static int __packet_get_status(struct packet_sock *po, void *frame)
419 union tpacket_uhdr h;
424 switch (po->tp_version) {
426 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
427 return h.h1->tp_status;
429 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
430 return h.h2->tp_status;
433 WARN(1, "TPACKET version not supported.\n");
439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
442 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
445 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
447 return TP_STATUS_TS_RAW_HARDWARE;
449 if (ktime_to_timespec_cond(skb->tstamp, ts))
450 return TP_STATUS_TS_SOFTWARE;
455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
458 union tpacket_uhdr h;
462 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 switch (po->tp_version) {
468 h.h1->tp_sec = ts.tv_sec;
469 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
472 h.h2->tp_sec = ts.tv_sec;
473 h.h2->tp_nsec = ts.tv_nsec;
477 WARN(1, "TPACKET version not supported.\n");
481 /* one flush is safe, as both fields always lie on the same cacheline */
482 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
488 static void *packet_lookup_frame(struct packet_sock *po,
489 struct packet_ring_buffer *rb,
490 unsigned int position,
493 unsigned int pg_vec_pos, frame_offset;
494 union tpacket_uhdr h;
496 pg_vec_pos = position / rb->frames_per_block;
497 frame_offset = position % rb->frames_per_block;
499 h.raw = rb->pg_vec[pg_vec_pos].buffer +
500 (frame_offset * rb->frame_size);
502 if (status != __packet_get_status(po, h.raw))
508 static void *packet_current_frame(struct packet_sock *po,
509 struct packet_ring_buffer *rb,
512 return packet_lookup_frame(po, rb, rb->head, status);
515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
517 del_timer_sync(&pkc->retire_blk_timer);
520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
522 struct sk_buff_head *rb_queue)
524 struct tpacket_kbdq_core *pkc;
526 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
527 GET_PBDQC_FROM_RB(&po->rx_ring);
529 spin_lock_bh(&rb_queue->lock);
530 pkc->delete_blk_timer = 1;
531 spin_unlock_bh(&rb_queue->lock);
533 prb_del_retire_blk_timer(pkc);
536 static void prb_init_blk_timer(struct packet_sock *po,
537 struct tpacket_kbdq_core *pkc,
538 void (*func) (unsigned long))
540 init_timer(&pkc->retire_blk_timer);
541 pkc->retire_blk_timer.data = (long)po;
542 pkc->retire_blk_timer.function = func;
543 pkc->retire_blk_timer.expires = jiffies;
546 static void prb_setup_retire_blk_timer(struct packet_sock *po)
548 struct tpacket_kbdq_core *pkc;
550 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
551 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
554 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
555 int blk_size_in_bytes)
557 struct net_device *dev;
558 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
559 struct ethtool_cmd ecmd;
564 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
565 if (unlikely(!dev)) {
567 return DEFAULT_PRB_RETIRE_TOV;
569 err = __ethtool_get_settings(dev, &ecmd);
570 speed = ethtool_cmd_speed(&ecmd);
574 * If the link speed is so slow you don't really
575 * need to worry about perf anyways
577 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
578 return DEFAULT_PRB_RETIRE_TOV;
585 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598 union tpacket_req_u *req_u)
600 p1->feature_req_word = req_u->req3.tp_feature_req_word;
603 static void init_prb_bdqc(struct packet_sock *po,
604 struct packet_ring_buffer *rb,
606 union tpacket_req_u *req_u)
608 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609 struct tpacket_block_desc *pbd;
611 memset(p1, 0x0, sizeof(*p1));
613 p1->knxt_seq_num = 1;
615 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616 p1->pkblk_start = pg_vec[0].buffer;
617 p1->kblk_size = req_u->req3.tp_block_size;
618 p1->knum_blocks = req_u->req3.tp_block_nr;
619 p1->hdrlen = po->tp_hdrlen;
620 p1->version = po->tp_version;
621 p1->last_kactive_blk_num = 0;
622 po->stats.stats3.tp_freeze_q_cnt = 0;
623 if (req_u->req3.tp_retire_blk_tov)
624 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627 req_u->req3.tp_block_size);
628 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632 prb_init_ft_ops(p1, req_u);
633 prb_setup_retire_blk_timer(po);
634 prb_open_block(p1, pbd);
637 /* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
642 mod_timer(&pkc->retire_blk_timer,
643 jiffies + pkc->tov_in_jiffies);
644 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
672 struct packet_sock *po = (struct packet_sock *)data;
673 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
675 struct tpacket_block_desc *pbd;
677 spin_lock(&po->sk.sk_receive_queue.lock);
679 frozen = prb_queue_frozen(pkc);
680 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682 if (unlikely(pkc->delete_blk_timer))
685 /* We only need to plug the race when the block is partially filled.
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
694 if (BLOCK_NUM_PKTS(pbd)) {
695 while (atomic_read(&pkc->blk_fill_in_prog)) {
696 /* Waiting for skb_copy_bits to finish... */
701 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 if (!BLOCK_NUM_PKTS(pbd)) {
704 /* An empty block. Just refresh the timer. */
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
713 /* Case 1. Queue was frozen because user-space was
716 if (prb_curr_blk_in_use(pkc, pbd)) {
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
730 prb_open_block(pkc, pbd);
737 _prb_refresh_rx_retire_blk_timer(pkc);
740 spin_unlock(&po->sk.sk_receive_queue.lock);
743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744 struct tpacket_block_desc *pbd1, __u32 status)
746 /* Flush everything minus the block header */
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753 /* Skip the block header(we know header WILL fit in 4K) */
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
763 /* Now update the block status. */
765 BLOCK_STATUS(pbd1) = status;
767 /* Flush the block header */
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 flush_dcache_page(pgv_to_page(start));
781 * 2) Increment active_blk_num
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
788 struct packet_sock *po, unsigned int stat)
790 __u32 status = TP_STATUS_USER | stat;
792 struct tpacket3_hdr *last_pkt;
793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794 struct sock *sk = &po->sk;
796 if (po->stats.stats3.tp_drops)
797 status |= TP_STATUS_LOSING;
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
807 /* Ok, we tmo'd - so get the current time.
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
814 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
820 /* Flush the block */
821 prb_flush_block(pkc1, pbd1, status);
823 sk->sk_data_ready(sk);
825 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
830 pkc->reset_pending_on_curr_blk = 0;
834 * Side effect of opening a block:
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841 struct tpacket_block_desc *pbd1)
844 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
852 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853 BLOCK_NUM_PKTS(pbd1) = 0;
854 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
858 h1->ts_first_pkt.ts_sec = ts.tv_sec;
859 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 pkc1->pkblk_start = (char *)pbd1;
862 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 pbd1->version = pkc1->version;
868 pkc1->prev = pkc1->nxt_offset;
869 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 prb_thaw_queue(pkc1);
872 _prb_refresh_rx_retire_blk_timer(pkc1);
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po)
903 pkc->reset_pending_on_curr_blk = 1;
904 po->stats.stats3.tp_freeze_q_cnt++;
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 struct packet_sock *po)
918 struct tpacket_block_desc *pbd;
922 /* 1. Get current block num */
923 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 prb_freeze_queue(pkc, po);
933 * open this block and return the offset where the first packet
934 * needs to get stored.
936 prb_open_block(pkc, pbd);
937 return (void *)pkc->nxt_offset;
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 struct packet_sock *po, unsigned int status)
943 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
956 if (!(status & TP_STATUS_BLK_TMO)) {
957 while (atomic_read(&pkc->blk_fill_in_prog)) {
958 /* Waiting for skb_copy_bits to finish... */
962 prb_close_block(pkc, pbd, po, status);
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968 struct tpacket_block_desc *pbd)
970 return TP_STATUS_USER & BLOCK_STATUS(pbd);
973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
975 return pkc->reset_pending_on_curr_blk;
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
981 atomic_dec(&pkc->blk_fill_in_prog);
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 struct tpacket3_hdr *ppd)
987 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 struct tpacket3_hdr *ppd)
993 ppd->hv1.tp_rxhash = 0;
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 struct tpacket3_hdr *ppd)
999 if (skb_vlan_tag_present(pkc->skb)) {
1000 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004 ppd->hv1.tp_vlan_tci = 0;
1005 ppd->hv1.tp_vlan_tpid = 0;
1006 ppd->tp_status = TP_STATUS_AVAILABLE;
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1013 ppd->hv1.tp_padding = 0;
1014 prb_fill_vlan_info(pkc, ppd);
1016 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017 prb_fill_rxhash(pkc, ppd);
1019 prb_clear_rxhash(pkc, ppd);
1022 static void prb_fill_curr_block(char *curr,
1023 struct tpacket_kbdq_core *pkc,
1024 struct tpacket_block_desc *pbd,
1027 struct tpacket3_hdr *ppd;
1029 ppd = (struct tpacket3_hdr *)curr;
1030 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_NUM_PKTS(pbd) += 1;
1035 atomic_inc(&pkc->blk_fill_in_prog);
1036 prb_run_all_ft_ops(pkc, ppd);
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041 struct sk_buff *skb,
1046 struct tpacket_kbdq_core *pkc;
1047 struct tpacket_block_desc *pbd;
1050 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc)) {
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1059 if (prb_curr_blk_in_use(pkc, pbd)) {
1060 /* Can't record this packet */
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1069 prb_open_block(pkc, pbd);
1074 curr = pkc->nxt_offset;
1076 end = (char *)pbd + pkc->kblk_size;
1078 /* first try the current block */
1079 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080 prb_fill_curr_block(curr, pkc, pbd, len);
1081 return (void *)curr;
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc, po, 0);
1087 /* Now, try to dispatch the next block */
1088 curr = (char *)prb_dispatch_next_block(pkc, po);
1090 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103 struct sk_buff *skb,
1104 int status, unsigned int len)
1107 switch (po->tp_version) {
1110 curr = packet_lookup_frame(po, &po->rx_ring,
1111 po->rx_ring.head, status);
1114 return __packet_lookup_frame_in_block(po, skb, status, len);
1116 WARN(1, "TPACKET version not supported\n");
1122 static void *prb_lookup_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1127 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1128 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130 if (status != BLOCK_STATUS(pbd))
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1138 if (rb->prb_bdqc.kactive_blk_num)
1139 prev = rb->prb_bdqc.kactive_blk_num-1;
1141 prev = rb->prb_bdqc.knum_blocks-1;
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1150 unsigned int previous = prb_previous_blk_num(rb);
1151 return prb_lookup_block(po, rb, previous, status);
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 if (po->tp_version <= TPACKET_V2)
1159 return packet_previous_frame(po, rb, status);
1161 return __prb_previous_block(po, rb, status);
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165 struct packet_ring_buffer *rb)
1167 switch (po->tp_version) {
1170 return packet_increment_head(rb);
1173 WARN(1, "TPACKET version not supported.\n");
1179 static void *packet_previous_frame(struct packet_sock *po,
1180 struct packet_ring_buffer *rb,
1183 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184 return packet_lookup_frame(po, rb, previous, status);
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 this_cpu_inc(*rb->pending_refcnt);
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 this_cpu_dec(*rb->pending_refcnt);
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 unsigned int refcnt = 0;
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb->pending_refcnt == NULL)
1211 for_each_possible_cpu(cpu)
1212 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217 static int packet_alloc_pending(struct packet_sock *po)
1219 po->rx_ring.pending_refcnt = NULL;
1221 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1228 static void packet_free_pending(struct packet_sock *po)
1230 free_percpu(po->tx_ring.pending_refcnt);
1233 #define ROOM_POW_OFF 2
1234 #define ROOM_NONE 0x0
1235 #define ROOM_LOW 0x1
1236 #define ROOM_NORMAL 0x2
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1242 len = po->rx_ring.frame_max + 1;
1243 idx = po->rx_ring.head;
1245 idx += len >> pow_off;
1248 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1255 len = po->rx_ring.prb_bdqc.knum_blocks;
1256 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 idx += len >> pow_off;
1261 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 struct sock *sk = &po->sk;
1267 int ret = ROOM_NONE;
1269 if (po->prot_hook.func != tpacket_rcv) {
1270 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271 - (skb ? skb->truesize : 0);
1272 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280 if (po->tp_version == TPACKET_V3) {
1281 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 else if (__tpacket_v3_has_room(po, 0))
1286 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_has_room(po, 0))
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1300 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301 ret = __packet_rcv_has_room(po, skb);
1302 has_room = ret == ROOM_NORMAL;
1303 if (po->pressure == has_room)
1304 po->pressure = !has_room;
1305 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1310 static void packet_sock_destruct(struct sock *sk)
1312 skb_queue_purge(&sk->sk_error_queue);
1314 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317 if (!sock_flag(sk, SOCK_DEAD)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk);
1322 sk_refcnt_debug_dec(sk);
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1330 rxhash = skb_get_hash(skb);
1331 for (i = 0; i < ROLLOVER_HLEN; i++)
1332 if (po->rollover->history[i] == rxhash)
1335 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336 return count > (ROLLOVER_HLEN >> 1);
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340 struct sk_buff *skb,
1343 return reciprocal_scale(skb_get_hash(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 bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1417 return f->flags & (flag >> 8);
1420 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1421 struct packet_type *pt, struct net_device *orig_dev)
1423 struct packet_fanout *f = pt->af_packet_priv;
1424 unsigned int num = READ_ONCE(f->num_members);
1425 struct packet_sock *po;
1428 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1434 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1435 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1440 case PACKET_FANOUT_HASH:
1442 idx = fanout_demux_hash(f, skb, num);
1444 case PACKET_FANOUT_LB:
1445 idx = fanout_demux_lb(f, skb, num);
1447 case PACKET_FANOUT_CPU:
1448 idx = fanout_demux_cpu(f, skb, num);
1450 case PACKET_FANOUT_RND:
1451 idx = fanout_demux_rnd(f, skb, num);
1453 case PACKET_FANOUT_QM:
1454 idx = fanout_demux_qm(f, skb, num);
1456 case PACKET_FANOUT_ROLLOVER:
1457 idx = fanout_demux_rollover(f, skb, 0, false, num);
1461 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1462 idx = fanout_demux_rollover(f, skb, idx, true, num);
1464 po = pkt_sk(f->arr[idx]);
1465 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1468 DEFINE_MUTEX(fanout_mutex);
1469 EXPORT_SYMBOL_GPL(fanout_mutex);
1470 static LIST_HEAD(fanout_list);
1472 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1474 struct packet_fanout *f = po->fanout;
1476 spin_lock(&f->lock);
1477 f->arr[f->num_members] = sk;
1480 spin_unlock(&f->lock);
1483 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1485 struct packet_fanout *f = po->fanout;
1488 spin_lock(&f->lock);
1489 for (i = 0; i < f->num_members; i++) {
1490 if (f->arr[i] == sk)
1493 BUG_ON(i >= f->num_members);
1494 f->arr[i] = f->arr[f->num_members - 1];
1496 spin_unlock(&f->lock);
1499 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1501 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1507 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1509 struct packet_sock *po = pkt_sk(sk);
1510 struct packet_fanout *f, *match;
1511 u8 type = type_flags & 0xff;
1512 u8 flags = type_flags >> 8;
1516 case PACKET_FANOUT_ROLLOVER:
1517 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1519 case PACKET_FANOUT_HASH:
1520 case PACKET_FANOUT_LB:
1521 case PACKET_FANOUT_CPU:
1522 case PACKET_FANOUT_RND:
1523 case PACKET_FANOUT_QM:
1535 if (type == PACKET_FANOUT_ROLLOVER ||
1536 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1537 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1540 atomic_long_set(&po->rollover->num, 0);
1541 atomic_long_set(&po->rollover->num_huge, 0);
1542 atomic_long_set(&po->rollover->num_failed, 0);
1545 mutex_lock(&fanout_mutex);
1547 list_for_each_entry(f, &fanout_list, list) {
1549 read_pnet(&f->net) == sock_net(sk)) {
1555 if (match && match->flags != flags)
1559 match = kzalloc(sizeof(*match), GFP_KERNEL);
1562 write_pnet(&match->net, sock_net(sk));
1565 match->flags = flags;
1566 atomic_set(&match->rr_cur, 0);
1567 INIT_LIST_HEAD(&match->list);
1568 spin_lock_init(&match->lock);
1569 atomic_set(&match->sk_ref, 0);
1570 match->prot_hook.type = po->prot_hook.type;
1571 match->prot_hook.dev = po->prot_hook.dev;
1572 match->prot_hook.func = packet_rcv_fanout;
1573 match->prot_hook.af_packet_priv = match;
1574 match->prot_hook.id_match = match_fanout_group;
1575 dev_add_pack(&match->prot_hook);
1576 list_add(&match->list, &fanout_list);
1579 if (match->type == type &&
1580 match->prot_hook.type == po->prot_hook.type &&
1581 match->prot_hook.dev == po->prot_hook.dev) {
1583 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1584 __dev_remove_pack(&po->prot_hook);
1586 atomic_inc(&match->sk_ref);
1587 __fanout_link(sk, po);
1592 mutex_unlock(&fanout_mutex);
1594 kfree(po->rollover);
1595 po->rollover = NULL;
1600 static void fanout_release(struct sock *sk)
1602 struct packet_sock *po = pkt_sk(sk);
1603 struct packet_fanout *f;
1609 mutex_lock(&fanout_mutex);
1612 if (atomic_dec_and_test(&f->sk_ref)) {
1614 dev_remove_pack(&f->prot_hook);
1617 mutex_unlock(&fanout_mutex);
1620 kfree_rcu(po->rollover, rcu);
1623 static const struct proto_ops packet_ops;
1625 static const struct proto_ops packet_ops_spkt;
1627 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1628 struct packet_type *pt, struct net_device *orig_dev)
1631 struct sockaddr_pkt *spkt;
1634 * When we registered the protocol we saved the socket in the data
1635 * field for just this event.
1638 sk = pt->af_packet_priv;
1641 * Yank back the headers [hope the device set this
1642 * right or kerboom...]
1644 * Incoming packets have ll header pulled,
1647 * For outgoing ones skb->data == skb_mac_header(skb)
1648 * so that this procedure is noop.
1651 if (skb->pkt_type == PACKET_LOOPBACK)
1654 if (!net_eq(dev_net(dev), sock_net(sk)))
1657 skb = skb_share_check(skb, GFP_ATOMIC);
1661 /* drop any routing info */
1664 /* drop conntrack reference */
1667 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1669 skb_push(skb, skb->data - skb_mac_header(skb));
1672 * The SOCK_PACKET socket receives _all_ frames.
1675 spkt->spkt_family = dev->type;
1676 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1677 spkt->spkt_protocol = skb->protocol;
1680 * Charge the memory to the socket. This is done specifically
1681 * to prevent sockets using all the memory up.
1684 if (sock_queue_rcv_skb(sk, skb) == 0)
1695 * Output a raw packet to a device layer. This bypasses all the other
1696 * protocol layers and you must therefore supply it with a complete frame
1699 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1702 struct sock *sk = sock->sk;
1703 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1704 struct sk_buff *skb = NULL;
1705 struct net_device *dev;
1711 * Get and verify the address.
1715 if (msg->msg_namelen < sizeof(struct sockaddr))
1717 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1718 proto = saddr->spkt_protocol;
1720 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1723 * Find the device first to size check it
1726 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1729 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1735 if (!(dev->flags & IFF_UP))
1739 * You may not queue a frame bigger than the mtu. This is the lowest level
1740 * raw protocol and you must do your own fragmentation at this level.
1743 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1744 if (!netif_supports_nofcs(dev)) {
1745 err = -EPROTONOSUPPORT;
1748 extra_len = 4; /* We're doing our own CRC */
1752 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1756 size_t reserved = LL_RESERVED_SPACE(dev);
1757 int tlen = dev->needed_tailroom;
1758 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1761 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1764 /* FIXME: Save some space for broken drivers that write a hard
1765 * header at transmission time by themselves. PPP is the notable
1766 * one here. This should really be fixed at the driver level.
1768 skb_reserve(skb, reserved);
1769 skb_reset_network_header(skb);
1771 /* Try to align data part correctly */
1776 skb_reset_network_header(skb);
1778 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1784 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1785 /* Earlier code assumed this would be a VLAN pkt,
1786 * double-check this now that we have the actual
1789 struct ethhdr *ehdr;
1790 skb_reset_mac_header(skb);
1791 ehdr = eth_hdr(skb);
1792 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1798 skb->protocol = proto;
1800 skb->priority = sk->sk_priority;
1801 skb->mark = sk->sk_mark;
1803 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1805 if (unlikely(extra_len == 4))
1808 skb_probe_transport_header(skb, 0);
1810 dev_queue_xmit(skb);
1821 static unsigned int run_filter(const struct sk_buff *skb,
1822 const struct sock *sk,
1825 struct sk_filter *filter;
1828 filter = rcu_dereference(sk->sk_filter);
1830 res = SK_RUN_FILTER(filter, skb);
1837 * This function makes lazy skb cloning in hope that most of packets
1838 * are discarded by BPF.
1840 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1841 * and skb->cb are mangled. It works because (and until) packets
1842 * falling here are owned by current CPU. Output packets are cloned
1843 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1844 * sequencially, so that if we return skb to original state on exit,
1845 * we will not harm anyone.
1848 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1849 struct packet_type *pt, struct net_device *orig_dev)
1852 struct sockaddr_ll *sll;
1853 struct packet_sock *po;
1854 u8 *skb_head = skb->data;
1855 int skb_len = skb->len;
1856 unsigned int snaplen, res;
1858 if (skb->pkt_type == PACKET_LOOPBACK)
1861 sk = pt->af_packet_priv;
1864 if (!net_eq(dev_net(dev), sock_net(sk)))
1869 if (dev->header_ops) {
1870 /* The device has an explicit notion of ll header,
1871 * exported to higher levels.
1873 * Otherwise, the device hides details of its frame
1874 * structure, so that corresponding packet head is
1875 * never delivered to user.
1877 if (sk->sk_type != SOCK_DGRAM)
1878 skb_push(skb, skb->data - skb_mac_header(skb));
1879 else if (skb->pkt_type == PACKET_OUTGOING) {
1880 /* Special case: outgoing packets have ll header at head */
1881 skb_pull(skb, skb_network_offset(skb));
1887 res = run_filter(skb, sk, snaplen);
1889 goto drop_n_restore;
1893 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1896 if (skb_shared(skb)) {
1897 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1901 if (skb_head != skb->data) {
1902 skb->data = skb_head;
1909 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1911 sll = &PACKET_SKB_CB(skb)->sa.ll;
1912 sll->sll_hatype = dev->type;
1913 sll->sll_pkttype = skb->pkt_type;
1914 if (unlikely(po->origdev))
1915 sll->sll_ifindex = orig_dev->ifindex;
1917 sll->sll_ifindex = dev->ifindex;
1919 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1921 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1922 * Use their space for storing the original skb length.
1924 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1926 if (pskb_trim(skb, snaplen))
1929 skb_set_owner_r(skb, sk);
1933 /* drop conntrack reference */
1936 spin_lock(&sk->sk_receive_queue.lock);
1937 po->stats.stats1.tp_packets++;
1938 sock_skb_set_dropcount(sk, skb);
1939 __skb_queue_tail(&sk->sk_receive_queue, skb);
1940 spin_unlock(&sk->sk_receive_queue.lock);
1941 sk->sk_data_ready(sk);
1945 spin_lock(&sk->sk_receive_queue.lock);
1946 po->stats.stats1.tp_drops++;
1947 atomic_inc(&sk->sk_drops);
1948 spin_unlock(&sk->sk_receive_queue.lock);
1951 if (skb_head != skb->data && skb_shared(skb)) {
1952 skb->data = skb_head;
1960 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1961 struct packet_type *pt, struct net_device *orig_dev)
1964 struct packet_sock *po;
1965 struct sockaddr_ll *sll;
1966 union tpacket_uhdr h;
1967 u8 *skb_head = skb->data;
1968 int skb_len = skb->len;
1969 unsigned int snaplen, res;
1970 unsigned long status = TP_STATUS_USER;
1971 unsigned short macoff, netoff, hdrlen;
1972 struct sk_buff *copy_skb = NULL;
1976 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1977 * We may add members to them until current aligned size without forcing
1978 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1980 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1981 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1983 if (skb->pkt_type == PACKET_LOOPBACK)
1986 sk = pt->af_packet_priv;
1989 if (!net_eq(dev_net(dev), sock_net(sk)))
1992 if (dev->header_ops) {
1993 if (sk->sk_type != SOCK_DGRAM)
1994 skb_push(skb, skb->data - skb_mac_header(skb));
1995 else if (skb->pkt_type == PACKET_OUTGOING) {
1996 /* Special case: outgoing packets have ll header at head */
1997 skb_pull(skb, skb_network_offset(skb));
2003 res = run_filter(skb, sk, snaplen);
2005 goto drop_n_restore;
2007 if (skb->ip_summed == CHECKSUM_PARTIAL)
2008 status |= TP_STATUS_CSUMNOTREADY;
2009 else if (skb->pkt_type != PACKET_OUTGOING &&
2010 (skb->ip_summed == CHECKSUM_COMPLETE ||
2011 skb_csum_unnecessary(skb)))
2012 status |= TP_STATUS_CSUM_VALID;
2017 if (sk->sk_type == SOCK_DGRAM) {
2018 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2021 unsigned int maclen = skb_network_offset(skb);
2022 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2023 (maclen < 16 ? 16 : maclen)) +
2025 macoff = netoff - maclen;
2027 if (po->tp_version <= TPACKET_V2) {
2028 if (macoff + snaplen > po->rx_ring.frame_size) {
2029 if (po->copy_thresh &&
2030 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2031 if (skb_shared(skb)) {
2032 copy_skb = skb_clone(skb, GFP_ATOMIC);
2034 copy_skb = skb_get(skb);
2035 skb_head = skb->data;
2038 skb_set_owner_r(copy_skb, sk);
2040 snaplen = po->rx_ring.frame_size - macoff;
2041 if ((int)snaplen < 0)
2044 } else if (unlikely(macoff + snaplen >
2045 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2048 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2049 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2050 snaplen, nval, macoff);
2052 if (unlikely((int)snaplen < 0)) {
2054 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2057 spin_lock(&sk->sk_receive_queue.lock);
2058 h.raw = packet_current_rx_frame(po, skb,
2059 TP_STATUS_KERNEL, (macoff+snaplen));
2062 if (po->tp_version <= TPACKET_V2) {
2063 packet_increment_rx_head(po, &po->rx_ring);
2065 * LOSING will be reported till you read the stats,
2066 * because it's COR - Clear On Read.
2067 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2070 if (po->stats.stats1.tp_drops)
2071 status |= TP_STATUS_LOSING;
2073 po->stats.stats1.tp_packets++;
2075 status |= TP_STATUS_COPY;
2076 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2078 spin_unlock(&sk->sk_receive_queue.lock);
2080 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2082 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2083 getnstimeofday(&ts);
2085 status |= ts_status;
2087 switch (po->tp_version) {
2089 h.h1->tp_len = skb->len;
2090 h.h1->tp_snaplen = snaplen;
2091 h.h1->tp_mac = macoff;
2092 h.h1->tp_net = netoff;
2093 h.h1->tp_sec = ts.tv_sec;
2094 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2095 hdrlen = sizeof(*h.h1);
2098 h.h2->tp_len = skb->len;
2099 h.h2->tp_snaplen = snaplen;
2100 h.h2->tp_mac = macoff;
2101 h.h2->tp_net = netoff;
2102 h.h2->tp_sec = ts.tv_sec;
2103 h.h2->tp_nsec = ts.tv_nsec;
2104 if (skb_vlan_tag_present(skb)) {
2105 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2106 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2107 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2109 h.h2->tp_vlan_tci = 0;
2110 h.h2->tp_vlan_tpid = 0;
2112 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2113 hdrlen = sizeof(*h.h2);
2116 /* tp_nxt_offset,vlan are already populated above.
2117 * So DONT clear those fields here
2119 h.h3->tp_status |= status;
2120 h.h3->tp_len = skb->len;
2121 h.h3->tp_snaplen = snaplen;
2122 h.h3->tp_mac = macoff;
2123 h.h3->tp_net = netoff;
2124 h.h3->tp_sec = ts.tv_sec;
2125 h.h3->tp_nsec = ts.tv_nsec;
2126 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2127 hdrlen = sizeof(*h.h3);
2133 sll = h.raw + TPACKET_ALIGN(hdrlen);
2134 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2135 sll->sll_family = AF_PACKET;
2136 sll->sll_hatype = dev->type;
2137 sll->sll_protocol = skb->protocol;
2138 sll->sll_pkttype = skb->pkt_type;
2139 if (unlikely(po->origdev))
2140 sll->sll_ifindex = orig_dev->ifindex;
2142 sll->sll_ifindex = dev->ifindex;
2146 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2147 if (po->tp_version <= TPACKET_V2) {
2150 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2153 for (start = h.raw; start < end; start += PAGE_SIZE)
2154 flush_dcache_page(pgv_to_page(start));
2159 if (po->tp_version <= TPACKET_V2) {
2160 __packet_set_status(po, h.raw, status);
2161 sk->sk_data_ready(sk);
2163 prb_clear_blk_fill_status(&po->rx_ring);
2167 if (skb_head != skb->data && skb_shared(skb)) {
2168 skb->data = skb_head;
2176 po->stats.stats1.tp_drops++;
2177 spin_unlock(&sk->sk_receive_queue.lock);
2179 sk->sk_data_ready(sk);
2180 kfree_skb(copy_skb);
2181 goto drop_n_restore;
2184 static void tpacket_destruct_skb(struct sk_buff *skb)
2186 struct packet_sock *po = pkt_sk(skb->sk);
2188 if (likely(po->tx_ring.pg_vec)) {
2192 ph = skb_shinfo(skb)->destructor_arg;
2193 packet_dec_pending(&po->tx_ring);
2195 ts = __packet_set_timestamp(po, ph, skb);
2196 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2202 static bool ll_header_truncated(const struct net_device *dev, int len)
2204 /* net device doesn't like empty head */
2205 if (unlikely(len <= dev->hard_header_len)) {
2206 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2207 current->comm, len, dev->hard_header_len);
2214 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2215 void *frame, struct net_device *dev, int size_max,
2216 __be16 proto, unsigned char *addr, int hlen)
2218 union tpacket_uhdr ph;
2219 int to_write, offset, len, tp_len, nr_frags, len_max;
2220 struct socket *sock = po->sk.sk_socket;
2227 skb->protocol = proto;
2229 skb->priority = po->sk.sk_priority;
2230 skb->mark = po->sk.sk_mark;
2231 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2232 skb_shinfo(skb)->destructor_arg = ph.raw;
2234 switch (po->tp_version) {
2236 tp_len = ph.h2->tp_len;
2239 tp_len = ph.h1->tp_len;
2242 if (unlikely(tp_len > size_max)) {
2243 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2247 skb_reserve(skb, hlen);
2248 skb_reset_network_header(skb);
2250 if (!packet_use_direct_xmit(po))
2251 skb_probe_transport_header(skb, 0);
2252 if (unlikely(po->tp_tx_has_off)) {
2253 int off_min, off_max, off;
2254 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2255 off_max = po->tx_ring.frame_size - tp_len;
2256 if (sock->type == SOCK_DGRAM) {
2257 switch (po->tp_version) {
2259 off = ph.h2->tp_net;
2262 off = ph.h1->tp_net;
2266 switch (po->tp_version) {
2268 off = ph.h2->tp_mac;
2271 off = ph.h1->tp_mac;
2275 if (unlikely((off < off_min) || (off_max < off)))
2277 data = ph.raw + off;
2279 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2283 if (sock->type == SOCK_DGRAM) {
2284 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2286 if (unlikely(err < 0))
2288 } else if (dev->hard_header_len) {
2289 if (ll_header_truncated(dev, tp_len))
2292 skb_push(skb, dev->hard_header_len);
2293 err = skb_store_bits(skb, 0, data,
2294 dev->hard_header_len);
2298 data += dev->hard_header_len;
2299 to_write -= dev->hard_header_len;
2302 offset = offset_in_page(data);
2303 len_max = PAGE_SIZE - offset;
2304 len = ((to_write > len_max) ? len_max : to_write);
2306 skb->data_len = to_write;
2307 skb->len += to_write;
2308 skb->truesize += to_write;
2309 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2311 while (likely(to_write)) {
2312 nr_frags = skb_shinfo(skb)->nr_frags;
2314 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2315 pr_err("Packet exceed the number of skb frags(%lu)\n",
2320 page = pgv_to_page(data);
2322 flush_dcache_page(page);
2324 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2327 len_max = PAGE_SIZE;
2328 len = ((to_write > len_max) ? len_max : to_write);
2334 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2336 struct sk_buff *skb;
2337 struct net_device *dev;
2339 int err, reserve = 0;
2341 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2342 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2343 int tp_len, size_max;
2344 unsigned char *addr;
2346 int status = TP_STATUS_AVAILABLE;
2349 mutex_lock(&po->pg_vec_lock);
2351 if (likely(saddr == NULL)) {
2352 dev = packet_cached_dev_get(po);
2357 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2359 if (msg->msg_namelen < (saddr->sll_halen
2360 + offsetof(struct sockaddr_ll,
2363 proto = saddr->sll_protocol;
2364 addr = saddr->sll_addr;
2365 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2369 if (unlikely(dev == NULL))
2372 if (unlikely(!(dev->flags & IFF_UP)))
2375 reserve = dev->hard_header_len + VLAN_HLEN;
2376 size_max = po->tx_ring.frame_size
2377 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2379 if (size_max > dev->mtu + reserve)
2380 size_max = dev->mtu + reserve;
2383 ph = packet_current_frame(po, &po->tx_ring,
2384 TP_STATUS_SEND_REQUEST);
2385 if (unlikely(ph == NULL)) {
2386 if (need_wait && need_resched())
2391 status = TP_STATUS_SEND_REQUEST;
2392 hlen = LL_RESERVED_SPACE(dev);
2393 tlen = dev->needed_tailroom;
2394 skb = sock_alloc_send_skb(&po->sk,
2395 hlen + tlen + sizeof(struct sockaddr_ll),
2398 if (unlikely(skb == NULL)) {
2399 /* we assume the socket was initially writeable ... */
2400 if (likely(len_sum > 0))
2404 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2406 if (tp_len > dev->mtu + dev->hard_header_len) {
2407 struct ethhdr *ehdr;
2408 /* Earlier code assumed this would be a VLAN pkt,
2409 * double-check this now that we have the actual
2413 skb_reset_mac_header(skb);
2414 ehdr = eth_hdr(skb);
2415 if (ehdr->h_proto != htons(ETH_P_8021Q))
2418 if (unlikely(tp_len < 0)) {
2420 __packet_set_status(po, ph,
2421 TP_STATUS_AVAILABLE);
2422 packet_increment_head(&po->tx_ring);
2426 status = TP_STATUS_WRONG_FORMAT;
2432 packet_pick_tx_queue(dev, skb);
2434 skb->destructor = tpacket_destruct_skb;
2435 __packet_set_status(po, ph, TP_STATUS_SENDING);
2436 packet_inc_pending(&po->tx_ring);
2438 status = TP_STATUS_SEND_REQUEST;
2439 err = po->xmit(skb);
2440 if (unlikely(err > 0)) {
2441 err = net_xmit_errno(err);
2442 if (err && __packet_get_status(po, ph) ==
2443 TP_STATUS_AVAILABLE) {
2444 /* skb was destructed already */
2449 * skb was dropped but not destructed yet;
2450 * let's treat it like congestion or err < 0
2454 packet_increment_head(&po->tx_ring);
2456 } while (likely((ph != NULL) ||
2457 /* Note: packet_read_pending() might be slow if we have
2458 * to call it as it's per_cpu variable, but in fast-path
2459 * we already short-circuit the loop with the first
2460 * condition, and luckily don't have to go that path
2463 (need_wait && packet_read_pending(&po->tx_ring))));
2469 __packet_set_status(po, ph, status);
2474 mutex_unlock(&po->pg_vec_lock);
2478 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2479 size_t reserve, size_t len,
2480 size_t linear, int noblock,
2483 struct sk_buff *skb;
2485 /* Under a page? Don't bother with paged skb. */
2486 if (prepad + len < PAGE_SIZE || !linear)
2489 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2494 skb_reserve(skb, reserve);
2495 skb_put(skb, linear);
2496 skb->data_len = len - linear;
2497 skb->len += len - linear;
2502 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2504 struct sock *sk = sock->sk;
2505 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2506 struct sk_buff *skb;
2507 struct net_device *dev;
2509 unsigned char *addr;
2510 int err, reserve = 0;
2511 struct virtio_net_hdr vnet_hdr = { 0 };
2514 struct packet_sock *po = pkt_sk(sk);
2515 unsigned short gso_type = 0;
2521 * Get and verify the address.
2524 if (likely(saddr == NULL)) {
2525 dev = packet_cached_dev_get(po);
2530 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2532 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2534 proto = saddr->sll_protocol;
2535 addr = saddr->sll_addr;
2536 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2540 if (unlikely(dev == NULL))
2543 if (unlikely(!(dev->flags & IFF_UP)))
2546 if (sock->type == SOCK_RAW)
2547 reserve = dev->hard_header_len;
2548 if (po->has_vnet_hdr) {
2549 vnet_hdr_len = sizeof(vnet_hdr);
2552 if (len < vnet_hdr_len)
2555 len -= vnet_hdr_len;
2558 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2559 if (n != vnet_hdr_len)
2562 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2563 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2564 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2565 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2566 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2567 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2568 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2571 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2574 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2575 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2576 case VIRTIO_NET_HDR_GSO_TCPV4:
2577 gso_type = SKB_GSO_TCPV4;
2579 case VIRTIO_NET_HDR_GSO_TCPV6:
2580 gso_type = SKB_GSO_TCPV6;
2582 case VIRTIO_NET_HDR_GSO_UDP:
2583 gso_type = SKB_GSO_UDP;
2589 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2590 gso_type |= SKB_GSO_TCP_ECN;
2592 if (vnet_hdr.gso_size == 0)
2598 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2599 if (!netif_supports_nofcs(dev)) {
2600 err = -EPROTONOSUPPORT;
2603 extra_len = 4; /* We're doing our own CRC */
2607 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2611 hlen = LL_RESERVED_SPACE(dev);
2612 tlen = dev->needed_tailroom;
2613 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2614 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2615 msg->msg_flags & MSG_DONTWAIT, &err);
2619 skb_set_network_header(skb, reserve);
2622 if (sock->type == SOCK_DGRAM) {
2623 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2624 if (unlikely(offset < 0))
2627 if (ll_header_truncated(dev, len))
2631 /* Returns -EFAULT on error */
2632 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2636 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2638 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2639 /* Earlier code assumed this would be a VLAN pkt,
2640 * double-check this now that we have the actual
2643 struct ethhdr *ehdr;
2644 skb_reset_mac_header(skb);
2645 ehdr = eth_hdr(skb);
2646 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2652 skb->protocol = proto;
2654 skb->priority = sk->sk_priority;
2655 skb->mark = sk->sk_mark;
2657 packet_pick_tx_queue(dev, skb);
2659 if (po->has_vnet_hdr) {
2660 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2661 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2662 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2663 if (!skb_partial_csum_set(skb, s, o)) {
2669 skb_shinfo(skb)->gso_size =
2670 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2671 skb_shinfo(skb)->gso_type = gso_type;
2673 /* Header must be checked, and gso_segs computed. */
2674 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2675 skb_shinfo(skb)->gso_segs = 0;
2677 len += vnet_hdr_len;
2680 if (!packet_use_direct_xmit(po))
2681 skb_probe_transport_header(skb, reserve);
2682 if (unlikely(extra_len == 4))
2685 err = po->xmit(skb);
2686 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2702 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2704 struct sock *sk = sock->sk;
2705 struct packet_sock *po = pkt_sk(sk);
2707 if (po->tx_ring.pg_vec)
2708 return tpacket_snd(po, msg);
2710 return packet_snd(sock, msg, len);
2714 * Close a PACKET socket. This is fairly simple. We immediately go
2715 * to 'closed' state and remove our protocol entry in the device list.
2718 static int packet_release(struct socket *sock)
2720 struct sock *sk = sock->sk;
2721 struct packet_sock *po;
2723 union tpacket_req_u req_u;
2731 mutex_lock(&net->packet.sklist_lock);
2732 sk_del_node_init_rcu(sk);
2733 mutex_unlock(&net->packet.sklist_lock);
2736 sock_prot_inuse_add(net, sk->sk_prot, -1);
2739 spin_lock(&po->bind_lock);
2740 unregister_prot_hook(sk, false);
2741 packet_cached_dev_reset(po);
2743 if (po->prot_hook.dev) {
2744 dev_put(po->prot_hook.dev);
2745 po->prot_hook.dev = NULL;
2747 spin_unlock(&po->bind_lock);
2749 packet_flush_mclist(sk);
2751 if (po->rx_ring.pg_vec) {
2752 memset(&req_u, 0, sizeof(req_u));
2753 packet_set_ring(sk, &req_u, 1, 0);
2756 if (po->tx_ring.pg_vec) {
2757 memset(&req_u, 0, sizeof(req_u));
2758 packet_set_ring(sk, &req_u, 1, 1);
2765 * Now the socket is dead. No more input will appear.
2772 skb_queue_purge(&sk->sk_receive_queue);
2773 packet_free_pending(po);
2774 sk_refcnt_debug_release(sk);
2781 * Attach a packet hook.
2784 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2786 struct packet_sock *po = pkt_sk(sk);
2787 const struct net_device *dev_curr;
2799 spin_lock(&po->bind_lock);
2801 proto_curr = po->prot_hook.type;
2802 dev_curr = po->prot_hook.dev;
2804 need_rehook = proto_curr != proto || dev_curr != dev;
2807 unregister_prot_hook(sk, true);
2810 po->prot_hook.type = proto;
2812 if (po->prot_hook.dev)
2813 dev_put(po->prot_hook.dev);
2815 po->prot_hook.dev = dev;
2817 po->ifindex = dev ? dev->ifindex : 0;
2818 packet_cached_dev_assign(po, dev);
2821 if (proto == 0 || !need_rehook)
2824 if (!dev || (dev->flags & IFF_UP)) {
2825 register_prot_hook(sk);
2827 sk->sk_err = ENETDOWN;
2828 if (!sock_flag(sk, SOCK_DEAD))
2829 sk->sk_error_report(sk);
2833 spin_unlock(&po->bind_lock);
2839 * Bind a packet socket to a device
2842 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2845 struct sock *sk = sock->sk;
2847 struct net_device *dev;
2854 if (addr_len != sizeof(struct sockaddr))
2856 strlcpy(name, uaddr->sa_data, sizeof(name));
2858 dev = dev_get_by_name(sock_net(sk), name);
2860 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2864 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2866 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2867 struct sock *sk = sock->sk;
2868 struct net_device *dev = NULL;
2876 if (addr_len < sizeof(struct sockaddr_ll))
2878 if (sll->sll_family != AF_PACKET)
2881 if (sll->sll_ifindex) {
2883 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2887 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2893 static struct proto packet_proto = {
2895 .owner = THIS_MODULE,
2896 .obj_size = sizeof(struct packet_sock),
2900 * Create a packet of type SOCK_PACKET.
2903 static int packet_create(struct net *net, struct socket *sock, int protocol,
2907 struct packet_sock *po;
2908 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2911 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2913 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2914 sock->type != SOCK_PACKET)
2915 return -ESOCKTNOSUPPORT;
2917 sock->state = SS_UNCONNECTED;
2920 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2924 sock->ops = &packet_ops;
2925 if (sock->type == SOCK_PACKET)
2926 sock->ops = &packet_ops_spkt;
2928 sock_init_data(sock, sk);
2931 sk->sk_family = PF_PACKET;
2933 po->xmit = dev_queue_xmit;
2935 err = packet_alloc_pending(po);
2939 packet_cached_dev_reset(po);
2941 sk->sk_destruct = packet_sock_destruct;
2942 sk_refcnt_debug_inc(sk);
2945 * Attach a protocol block
2948 spin_lock_init(&po->bind_lock);
2949 mutex_init(&po->pg_vec_lock);
2950 po->rollover = NULL;
2951 po->prot_hook.func = packet_rcv;
2953 if (sock->type == SOCK_PACKET)
2954 po->prot_hook.func = packet_rcv_spkt;
2956 po->prot_hook.af_packet_priv = sk;
2959 po->prot_hook.type = proto;
2960 register_prot_hook(sk);
2963 mutex_lock(&net->packet.sklist_lock);
2964 sk_add_node_rcu(sk, &net->packet.sklist);
2965 mutex_unlock(&net->packet.sklist_lock);
2968 sock_prot_inuse_add(net, &packet_proto, 1);
2979 * Pull a packet from our receive queue and hand it to the user.
2980 * If necessary we block.
2983 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2986 struct sock *sk = sock->sk;
2987 struct sk_buff *skb;
2989 int vnet_hdr_len = 0;
2990 unsigned int origlen = 0;
2993 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2997 /* What error should we return now? EUNATTACH? */
2998 if (pkt_sk(sk)->ifindex < 0)
3002 if (flags & MSG_ERRQUEUE) {
3003 err = sock_recv_errqueue(sk, msg, len,
3004 SOL_PACKET, PACKET_TX_TIMESTAMP);
3009 * Call the generic datagram receiver. This handles all sorts
3010 * of horrible races and re-entrancy so we can forget about it
3011 * in the protocol layers.
3013 * Now it will return ENETDOWN, if device have just gone down,
3014 * but then it will block.
3017 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3020 * An error occurred so return it. Because skb_recv_datagram()
3021 * handles the blocking we don't see and worry about blocking
3028 if (pkt_sk(sk)->pressure)
3029 packet_rcv_has_room(pkt_sk(sk), NULL);
3031 if (pkt_sk(sk)->has_vnet_hdr) {
3032 struct virtio_net_hdr vnet_hdr = { 0 };
3035 vnet_hdr_len = sizeof(vnet_hdr);
3036 if (len < vnet_hdr_len)
3039 len -= vnet_hdr_len;
3041 if (skb_is_gso(skb)) {
3042 struct skb_shared_info *sinfo = skb_shinfo(skb);
3044 /* This is a hint as to how much should be linear. */
3046 __cpu_to_virtio16(false, skb_headlen(skb));
3048 __cpu_to_virtio16(false, sinfo->gso_size);
3049 if (sinfo->gso_type & SKB_GSO_TCPV4)
3050 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3051 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3052 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3053 else if (sinfo->gso_type & SKB_GSO_UDP)
3054 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3055 else if (sinfo->gso_type & SKB_GSO_FCOE)
3059 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3060 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3062 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3064 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3065 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3066 vnet_hdr.csum_start = __cpu_to_virtio16(false,
3067 skb_checksum_start_offset(skb));
3068 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3070 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3071 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3072 } /* else everything is zero */
3074 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3079 /* You lose any data beyond the buffer you gave. If it worries
3080 * a user program they can ask the device for its MTU
3086 msg->msg_flags |= MSG_TRUNC;
3089 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3093 if (sock->type != SOCK_PACKET) {
3094 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3096 /* Original length was stored in sockaddr_ll fields */
3097 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3098 sll->sll_family = AF_PACKET;
3099 sll->sll_protocol = skb->protocol;
3102 sock_recv_ts_and_drops(msg, sk, skb);
3104 if (msg->msg_name) {
3105 /* If the address length field is there to be filled
3106 * in, we fill it in now.
3108 if (sock->type == SOCK_PACKET) {
3109 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3110 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3112 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3114 msg->msg_namelen = sll->sll_halen +
3115 offsetof(struct sockaddr_ll, sll_addr);
3117 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3121 if (pkt_sk(sk)->auxdata) {
3122 struct tpacket_auxdata aux;
3124 aux.tp_status = TP_STATUS_USER;
3125 if (skb->ip_summed == CHECKSUM_PARTIAL)
3126 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3127 else if (skb->pkt_type != PACKET_OUTGOING &&
3128 (skb->ip_summed == CHECKSUM_COMPLETE ||
3129 skb_csum_unnecessary(skb)))
3130 aux.tp_status |= TP_STATUS_CSUM_VALID;
3132 aux.tp_len = origlen;
3133 aux.tp_snaplen = skb->len;
3135 aux.tp_net = skb_network_offset(skb);
3136 if (skb_vlan_tag_present(skb)) {
3137 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3138 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3139 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3141 aux.tp_vlan_tci = 0;
3142 aux.tp_vlan_tpid = 0;
3144 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3148 * Free or return the buffer as appropriate. Again this
3149 * hides all the races and re-entrancy issues from us.
3151 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3154 skb_free_datagram(sk, skb);
3159 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3160 int *uaddr_len, int peer)
3162 struct net_device *dev;
3163 struct sock *sk = sock->sk;
3168 uaddr->sa_family = AF_PACKET;
3169 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3171 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3173 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3175 *uaddr_len = sizeof(*uaddr);
3180 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3181 int *uaddr_len, int peer)
3183 struct net_device *dev;
3184 struct sock *sk = sock->sk;
3185 struct packet_sock *po = pkt_sk(sk);
3186 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3191 sll->sll_family = AF_PACKET;
3192 sll->sll_ifindex = po->ifindex;
3193 sll->sll_protocol = po->num;
3194 sll->sll_pkttype = 0;
3196 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3198 sll->sll_hatype = dev->type;
3199 sll->sll_halen = dev->addr_len;
3200 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3202 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3206 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3211 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3215 case PACKET_MR_MULTICAST:
3216 if (i->alen != dev->addr_len)
3219 return dev_mc_add(dev, i->addr);
3221 return dev_mc_del(dev, i->addr);
3223 case PACKET_MR_PROMISC:
3224 return dev_set_promiscuity(dev, what);
3225 case PACKET_MR_ALLMULTI:
3226 return dev_set_allmulti(dev, what);
3227 case PACKET_MR_UNICAST:
3228 if (i->alen != dev->addr_len)
3231 return dev_uc_add(dev, i->addr);
3233 return dev_uc_del(dev, i->addr);
3241 static void packet_dev_mclist_delete(struct net_device *dev,
3242 struct packet_mclist **mlp)
3244 struct packet_mclist *ml;
3246 while ((ml = *mlp) != NULL) {
3247 if (ml->ifindex == dev->ifindex) {
3248 packet_dev_mc(dev, ml, -1);
3256 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3258 struct packet_sock *po = pkt_sk(sk);
3259 struct packet_mclist *ml, *i;
3260 struct net_device *dev;
3266 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3271 if (mreq->mr_alen > dev->addr_len)
3275 i = kmalloc(sizeof(*i), GFP_KERNEL);
3280 for (ml = po->mclist; ml; ml = ml->next) {
3281 if (ml->ifindex == mreq->mr_ifindex &&
3282 ml->type == mreq->mr_type &&
3283 ml->alen == mreq->mr_alen &&
3284 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3286 /* Free the new element ... */
3292 i->type = mreq->mr_type;
3293 i->ifindex = mreq->mr_ifindex;
3294 i->alen = mreq->mr_alen;
3295 memcpy(i->addr, mreq->mr_address, i->alen);
3297 i->next = po->mclist;
3299 err = packet_dev_mc(dev, i, 1);
3301 po->mclist = i->next;
3310 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3312 struct packet_mclist *ml, **mlp;
3316 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3317 if (ml->ifindex == mreq->mr_ifindex &&
3318 ml->type == mreq->mr_type &&
3319 ml->alen == mreq->mr_alen &&
3320 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3321 if (--ml->count == 0) {
3322 struct net_device *dev;
3324 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3326 packet_dev_mc(dev, ml, -1);
3336 static void packet_flush_mclist(struct sock *sk)
3338 struct packet_sock *po = pkt_sk(sk);
3339 struct packet_mclist *ml;
3345 while ((ml = po->mclist) != NULL) {
3346 struct net_device *dev;
3348 po->mclist = ml->next;
3349 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3351 packet_dev_mc(dev, ml, -1);
3358 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3360 struct sock *sk = sock->sk;
3361 struct packet_sock *po = pkt_sk(sk);
3364 if (level != SOL_PACKET)
3365 return -ENOPROTOOPT;
3368 case PACKET_ADD_MEMBERSHIP:
3369 case PACKET_DROP_MEMBERSHIP:
3371 struct packet_mreq_max mreq;
3373 memset(&mreq, 0, sizeof(mreq));
3374 if (len < sizeof(struct packet_mreq))
3376 if (len > sizeof(mreq))
3378 if (copy_from_user(&mreq, optval, len))
3380 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3382 if (optname == PACKET_ADD_MEMBERSHIP)
3383 ret = packet_mc_add(sk, &mreq);
3385 ret = packet_mc_drop(sk, &mreq);
3389 case PACKET_RX_RING:
3390 case PACKET_TX_RING:
3392 union tpacket_req_u req_u;
3395 switch (po->tp_version) {
3398 len = sizeof(req_u.req);
3402 len = sizeof(req_u.req3);
3407 if (pkt_sk(sk)->has_vnet_hdr)
3409 if (copy_from_user(&req_u.req, optval, len))
3411 return packet_set_ring(sk, &req_u, 0,
3412 optname == PACKET_TX_RING);
3414 case PACKET_COPY_THRESH:
3418 if (optlen != sizeof(val))
3420 if (copy_from_user(&val, optval, sizeof(val)))
3423 pkt_sk(sk)->copy_thresh = val;
3426 case PACKET_VERSION:
3430 if (optlen != sizeof(val))
3432 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3434 if (copy_from_user(&val, optval, sizeof(val)))
3440 po->tp_version = val;
3446 case PACKET_RESERVE:
3450 if (optlen != sizeof(val))
3452 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3454 if (copy_from_user(&val, optval, sizeof(val)))
3456 po->tp_reserve = val;
3463 if (optlen != sizeof(val))
3465 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3467 if (copy_from_user(&val, optval, sizeof(val)))
3469 po->tp_loss = !!val;
3472 case PACKET_AUXDATA:
3476 if (optlen < sizeof(val))
3478 if (copy_from_user(&val, optval, sizeof(val)))
3481 po->auxdata = !!val;
3484 case PACKET_ORIGDEV:
3488 if (optlen < sizeof(val))
3490 if (copy_from_user(&val, optval, sizeof(val)))
3493 po->origdev = !!val;
3496 case PACKET_VNET_HDR:
3500 if (sock->type != SOCK_RAW)
3502 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3504 if (optlen < sizeof(val))
3506 if (copy_from_user(&val, optval, sizeof(val)))
3509 po->has_vnet_hdr = !!val;
3512 case PACKET_TIMESTAMP:
3516 if (optlen != sizeof(val))
3518 if (copy_from_user(&val, optval, sizeof(val)))
3521 po->tp_tstamp = val;
3528 if (optlen != sizeof(val))
3530 if (copy_from_user(&val, optval, sizeof(val)))
3533 return fanout_add(sk, val & 0xffff, val >> 16);
3535 case PACKET_TX_HAS_OFF:
3539 if (optlen != sizeof(val))
3541 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3543 if (copy_from_user(&val, optval, sizeof(val)))
3545 po->tp_tx_has_off = !!val;
3548 case PACKET_QDISC_BYPASS:
3552 if (optlen != sizeof(val))
3554 if (copy_from_user(&val, optval, sizeof(val)))
3557 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3561 return -ENOPROTOOPT;
3565 static int packet_getsockopt(struct socket *sock, int level, int optname,
3566 char __user *optval, int __user *optlen)
3569 int val, lv = sizeof(val);
3570 struct sock *sk = sock->sk;
3571 struct packet_sock *po = pkt_sk(sk);
3573 union tpacket_stats_u st;
3574 struct tpacket_rollover_stats rstats;
3576 if (level != SOL_PACKET)
3577 return -ENOPROTOOPT;
3579 if (get_user(len, optlen))
3586 case PACKET_STATISTICS:
3587 spin_lock_bh(&sk->sk_receive_queue.lock);
3588 memcpy(&st, &po->stats, sizeof(st));
3589 memset(&po->stats, 0, sizeof(po->stats));
3590 spin_unlock_bh(&sk->sk_receive_queue.lock);
3592 if (po->tp_version == TPACKET_V3) {
3593 lv = sizeof(struct tpacket_stats_v3);
3594 st.stats3.tp_packets += st.stats3.tp_drops;
3597 lv = sizeof(struct tpacket_stats);
3598 st.stats1.tp_packets += st.stats1.tp_drops;
3603 case PACKET_AUXDATA:
3606 case PACKET_ORIGDEV:
3609 case PACKET_VNET_HDR:
3610 val = po->has_vnet_hdr;
3612 case PACKET_VERSION:
3613 val = po->tp_version;
3616 if (len > sizeof(int))
3618 if (copy_from_user(&val, optval, len))
3622 val = sizeof(struct tpacket_hdr);
3625 val = sizeof(struct tpacket2_hdr);
3628 val = sizeof(struct tpacket3_hdr);
3634 case PACKET_RESERVE:
3635 val = po->tp_reserve;
3640 case PACKET_TIMESTAMP:
3641 val = po->tp_tstamp;
3645 ((u32)po->fanout->id |
3646 ((u32)po->fanout->type << 16) |
3647 ((u32)po->fanout->flags << 24)) :
3650 case PACKET_ROLLOVER_STATS:
3653 rstats.tp_all = atomic_long_read(&po->rollover->num);
3654 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3655 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3657 lv = sizeof(rstats);
3659 case PACKET_TX_HAS_OFF:
3660 val = po->tp_tx_has_off;
3662 case PACKET_QDISC_BYPASS:
3663 val = packet_use_direct_xmit(po);
3666 return -ENOPROTOOPT;
3671 if (put_user(len, optlen))
3673 if (copy_to_user(optval, data, len))
3679 static int packet_notifier(struct notifier_block *this,
3680 unsigned long msg, void *ptr)
3683 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3684 struct net *net = dev_net(dev);
3687 sk_for_each_rcu(sk, &net->packet.sklist) {
3688 struct packet_sock *po = pkt_sk(sk);
3691 case NETDEV_UNREGISTER:
3693 packet_dev_mclist_delete(dev, &po->mclist);
3697 if (dev->ifindex == po->ifindex) {
3698 spin_lock(&po->bind_lock);
3700 __unregister_prot_hook(sk, false);
3701 sk->sk_err = ENETDOWN;
3702 if (!sock_flag(sk, SOCK_DEAD))
3703 sk->sk_error_report(sk);
3705 if (msg == NETDEV_UNREGISTER) {
3706 packet_cached_dev_reset(po);
3708 if (po->prot_hook.dev)
3709 dev_put(po->prot_hook.dev);
3710 po->prot_hook.dev = NULL;
3712 spin_unlock(&po->bind_lock);
3716 if (dev->ifindex == po->ifindex) {
3717 spin_lock(&po->bind_lock);
3719 register_prot_hook(sk);
3720 spin_unlock(&po->bind_lock);
3730 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3733 struct sock *sk = sock->sk;
3738 int amount = sk_wmem_alloc_get(sk);
3740 return put_user(amount, (int __user *)arg);
3744 struct sk_buff *skb;
3747 spin_lock_bh(&sk->sk_receive_queue.lock);
3748 skb = skb_peek(&sk->sk_receive_queue);
3751 spin_unlock_bh(&sk->sk_receive_queue.lock);
3752 return put_user(amount, (int __user *)arg);
3755 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3757 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3767 case SIOCGIFBRDADDR:
3768 case SIOCSIFBRDADDR:
3769 case SIOCGIFNETMASK:
3770 case SIOCSIFNETMASK:
3771 case SIOCGIFDSTADDR:
3772 case SIOCSIFDSTADDR:
3774 return inet_dgram_ops.ioctl(sock, cmd, arg);
3778 return -ENOIOCTLCMD;
3783 static unsigned int packet_poll(struct file *file, struct socket *sock,
3786 struct sock *sk = sock->sk;
3787 struct packet_sock *po = pkt_sk(sk);
3788 unsigned int mask = datagram_poll(file, sock, wait);
3790 spin_lock_bh(&sk->sk_receive_queue.lock);
3791 if (po->rx_ring.pg_vec) {
3792 if (!packet_previous_rx_frame(po, &po->rx_ring,
3794 mask |= POLLIN | POLLRDNORM;
3796 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3798 spin_unlock_bh(&sk->sk_receive_queue.lock);
3799 spin_lock_bh(&sk->sk_write_queue.lock);
3800 if (po->tx_ring.pg_vec) {
3801 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3802 mask |= POLLOUT | POLLWRNORM;
3804 spin_unlock_bh(&sk->sk_write_queue.lock);
3809 /* Dirty? Well, I still did not learn better way to account
3813 static void packet_mm_open(struct vm_area_struct *vma)
3815 struct file *file = vma->vm_file;
3816 struct socket *sock = file->private_data;
3817 struct sock *sk = sock->sk;
3820 atomic_inc(&pkt_sk(sk)->mapped);
3823 static void packet_mm_close(struct vm_area_struct *vma)
3825 struct file *file = vma->vm_file;
3826 struct socket *sock = file->private_data;
3827 struct sock *sk = sock->sk;
3830 atomic_dec(&pkt_sk(sk)->mapped);
3833 static const struct vm_operations_struct packet_mmap_ops = {
3834 .open = packet_mm_open,
3835 .close = packet_mm_close,
3838 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3843 for (i = 0; i < len; i++) {
3844 if (likely(pg_vec[i].buffer)) {
3845 if (is_vmalloc_addr(pg_vec[i].buffer))
3846 vfree(pg_vec[i].buffer);
3848 free_pages((unsigned long)pg_vec[i].buffer,
3850 pg_vec[i].buffer = NULL;
3856 static char *alloc_one_pg_vec_page(unsigned long order)
3859 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3860 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3862 buffer = (char *) __get_free_pages(gfp_flags, order);
3866 /* __get_free_pages failed, fall back to vmalloc */
3867 buffer = vzalloc((1 << order) * PAGE_SIZE);
3871 /* vmalloc failed, lets dig into swap here */
3872 gfp_flags &= ~__GFP_NORETRY;
3873 buffer = (char *) __get_free_pages(gfp_flags, order);
3877 /* complete and utter failure */
3881 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3883 unsigned int block_nr = req->tp_block_nr;
3887 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3888 if (unlikely(!pg_vec))
3891 for (i = 0; i < block_nr; i++) {
3892 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3893 if (unlikely(!pg_vec[i].buffer))
3894 goto out_free_pgvec;
3901 free_pg_vec(pg_vec, order, block_nr);
3906 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3907 int closing, int tx_ring)
3909 struct pgv *pg_vec = NULL;
3910 struct packet_sock *po = pkt_sk(sk);
3911 int was_running, order = 0;
3912 struct packet_ring_buffer *rb;
3913 struct sk_buff_head *rb_queue;
3916 /* Added to avoid minimal code churn */
3917 struct tpacket_req *req = &req_u->req;
3919 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3920 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3921 WARN(1, "Tx-ring is not supported.\n");
3925 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3926 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3930 if (atomic_read(&po->mapped))
3932 if (packet_read_pending(rb))
3936 if (req->tp_block_nr) {
3937 /* Sanity tests and some calculations */
3939 if (unlikely(rb->pg_vec))
3942 switch (po->tp_version) {
3944 po->tp_hdrlen = TPACKET_HDRLEN;
3947 po->tp_hdrlen = TPACKET2_HDRLEN;
3950 po->tp_hdrlen = TPACKET3_HDRLEN;
3955 if (unlikely((int)req->tp_block_size <= 0))
3957 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3959 if (po->tp_version >= TPACKET_V3 &&
3960 (int)(req->tp_block_size -
3961 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3963 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3966 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3969 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3970 if (unlikely(rb->frames_per_block <= 0))
3972 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3977 order = get_order(req->tp_block_size);
3978 pg_vec = alloc_pg_vec(req, order);
3979 if (unlikely(!pg_vec))
3981 switch (po->tp_version) {
3983 /* Transmit path is not supported. We checked
3984 * it above but just being paranoid
3987 init_prb_bdqc(po, rb, pg_vec, req_u);
3996 if (unlikely(req->tp_frame_nr))
4002 /* Detach socket from network */
4003 spin_lock(&po->bind_lock);
4004 was_running = po->running;
4008 __unregister_prot_hook(sk, false);
4010 spin_unlock(&po->bind_lock);
4015 mutex_lock(&po->pg_vec_lock);
4016 if (closing || atomic_read(&po->mapped) == 0) {
4018 spin_lock_bh(&rb_queue->lock);
4019 swap(rb->pg_vec, pg_vec);
4020 rb->frame_max = (req->tp_frame_nr - 1);
4022 rb->frame_size = req->tp_frame_size;
4023 spin_unlock_bh(&rb_queue->lock);
4025 swap(rb->pg_vec_order, order);
4026 swap(rb->pg_vec_len, req->tp_block_nr);
4028 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4029 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4030 tpacket_rcv : packet_rcv;
4031 skb_queue_purge(rb_queue);
4032 if (atomic_read(&po->mapped))
4033 pr_err("packet_mmap: vma is busy: %d\n",
4034 atomic_read(&po->mapped));
4036 mutex_unlock(&po->pg_vec_lock);
4038 spin_lock(&po->bind_lock);
4041 register_prot_hook(sk);
4043 spin_unlock(&po->bind_lock);
4044 if (closing && (po->tp_version > TPACKET_V2)) {
4045 /* Because we don't support block-based V3 on tx-ring */
4047 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4052 free_pg_vec(pg_vec, order, req->tp_block_nr);
4057 static int packet_mmap(struct file *file, struct socket *sock,
4058 struct vm_area_struct *vma)
4060 struct sock *sk = sock->sk;
4061 struct packet_sock *po = pkt_sk(sk);
4062 unsigned long size, expected_size;
4063 struct packet_ring_buffer *rb;
4064 unsigned long start;
4071 mutex_lock(&po->pg_vec_lock);
4074 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4076 expected_size += rb->pg_vec_len
4082 if (expected_size == 0)
4085 size = vma->vm_end - vma->vm_start;
4086 if (size != expected_size)
4089 start = vma->vm_start;
4090 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4091 if (rb->pg_vec == NULL)
4094 for (i = 0; i < rb->pg_vec_len; i++) {
4096 void *kaddr = rb->pg_vec[i].buffer;
4099 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4100 page = pgv_to_page(kaddr);
4101 err = vm_insert_page(vma, start, page);
4110 atomic_inc(&po->mapped);
4111 vma->vm_ops = &packet_mmap_ops;
4115 mutex_unlock(&po->pg_vec_lock);
4119 static const struct proto_ops packet_ops_spkt = {
4120 .family = PF_PACKET,
4121 .owner = THIS_MODULE,
4122 .release = packet_release,
4123 .bind = packet_bind_spkt,
4124 .connect = sock_no_connect,
4125 .socketpair = sock_no_socketpair,
4126 .accept = sock_no_accept,
4127 .getname = packet_getname_spkt,
4128 .poll = datagram_poll,
4129 .ioctl = packet_ioctl,
4130 .listen = sock_no_listen,
4131 .shutdown = sock_no_shutdown,
4132 .setsockopt = sock_no_setsockopt,
4133 .getsockopt = sock_no_getsockopt,
4134 .sendmsg = packet_sendmsg_spkt,
4135 .recvmsg = packet_recvmsg,
4136 .mmap = sock_no_mmap,
4137 .sendpage = sock_no_sendpage,
4140 static const struct proto_ops packet_ops = {
4141 .family = PF_PACKET,
4142 .owner = THIS_MODULE,
4143 .release = packet_release,
4144 .bind = packet_bind,
4145 .connect = sock_no_connect,
4146 .socketpair = sock_no_socketpair,
4147 .accept = sock_no_accept,
4148 .getname = packet_getname,
4149 .poll = packet_poll,
4150 .ioctl = packet_ioctl,
4151 .listen = sock_no_listen,
4152 .shutdown = sock_no_shutdown,
4153 .setsockopt = packet_setsockopt,
4154 .getsockopt = packet_getsockopt,
4155 .sendmsg = packet_sendmsg,
4156 .recvmsg = packet_recvmsg,
4157 .mmap = packet_mmap,
4158 .sendpage = sock_no_sendpage,
4161 static const struct net_proto_family packet_family_ops = {
4162 .family = PF_PACKET,
4163 .create = packet_create,
4164 .owner = THIS_MODULE,
4167 static struct notifier_block packet_netdev_notifier = {
4168 .notifier_call = packet_notifier,
4171 #ifdef CONFIG_PROC_FS
4173 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4176 struct net *net = seq_file_net(seq);
4179 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4182 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4184 struct net *net = seq_file_net(seq);
4185 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4188 static void packet_seq_stop(struct seq_file *seq, void *v)
4194 static int packet_seq_show(struct seq_file *seq, void *v)
4196 if (v == SEQ_START_TOKEN)
4197 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4199 struct sock *s = sk_entry(v);
4200 const struct packet_sock *po = pkt_sk(s);
4203 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4205 atomic_read(&s->sk_refcnt),
4210 atomic_read(&s->sk_rmem_alloc),
4211 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4218 static const struct seq_operations packet_seq_ops = {
4219 .start = packet_seq_start,
4220 .next = packet_seq_next,
4221 .stop = packet_seq_stop,
4222 .show = packet_seq_show,
4225 static int packet_seq_open(struct inode *inode, struct file *file)
4227 return seq_open_net(inode, file, &packet_seq_ops,
4228 sizeof(struct seq_net_private));
4231 static const struct file_operations packet_seq_fops = {
4232 .owner = THIS_MODULE,
4233 .open = packet_seq_open,
4235 .llseek = seq_lseek,
4236 .release = seq_release_net,
4241 static int __net_init packet_net_init(struct net *net)
4243 mutex_init(&net->packet.sklist_lock);
4244 INIT_HLIST_HEAD(&net->packet.sklist);
4246 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4252 static void __net_exit packet_net_exit(struct net *net)
4254 remove_proc_entry("packet", net->proc_net);
4257 static struct pernet_operations packet_net_ops = {
4258 .init = packet_net_init,
4259 .exit = packet_net_exit,
4263 static void __exit packet_exit(void)
4265 unregister_netdevice_notifier(&packet_netdev_notifier);
4266 unregister_pernet_subsys(&packet_net_ops);
4267 sock_unregister(PF_PACKET);
4268 proto_unregister(&packet_proto);
4271 static int __init packet_init(void)
4273 int rc = proto_register(&packet_proto, 0);
4278 sock_register(&packet_family_ops);
4279 register_pernet_subsys(&packet_net_ops);
4280 register_netdevice_notifier(&packet_netdev_notifier);
4285 module_init(packet_init);
4286 module_exit(packet_exit);
4287 MODULE_LICENSE("GPL");
4288 MODULE_ALIAS_NETPROTO(PF_PACKET);