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 int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1327 int x = atomic_read(&f->rr_cur) + 1;
1335 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1340 rxhash = skb_get_hash(skb);
1341 for (i = 0; i < ROLLOVER_HLEN; i++)
1342 if (po->rollover->history[i] == rxhash)
1345 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1346 return count > (ROLLOVER_HLEN >> 1);
1349 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1350 struct sk_buff *skb,
1353 return reciprocal_scale(skb_get_hash(skb), num);
1356 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1357 struct sk_buff *skb,
1362 cur = atomic_read(&f->rr_cur);
1363 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1364 fanout_rr_next(f, num))) != cur)
1369 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1370 struct sk_buff *skb,
1373 return smp_processor_id() % num;
1376 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1377 struct sk_buff *skb,
1380 return prandom_u32_max(num);
1383 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1384 struct sk_buff *skb,
1385 unsigned int idx, bool try_self,
1388 struct packet_sock *po, *po_next, *po_skip = NULL;
1389 unsigned int i, j, room = ROOM_NONE;
1391 po = pkt_sk(f->arr[idx]);
1394 room = packet_rcv_has_room(po, skb);
1395 if (room == ROOM_NORMAL ||
1396 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1401 i = j = min_t(int, po->rollover->sock, num - 1);
1403 po_next = pkt_sk(f->arr[i]);
1404 if (po_next != po_skip && !po_next->pressure &&
1405 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1407 po->rollover->sock = i;
1408 atomic_long_inc(&po->rollover->num);
1409 if (room == ROOM_LOW)
1410 atomic_long_inc(&po->rollover->num_huge);
1418 atomic_long_inc(&po->rollover->num_failed);
1422 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1423 struct sk_buff *skb,
1426 return skb_get_queue_mapping(skb) % num;
1429 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1431 return f->flags & (flag >> 8);
1434 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1435 struct packet_type *pt, struct net_device *orig_dev)
1437 struct packet_fanout *f = pt->af_packet_priv;
1438 unsigned int num = f->num_members;
1439 struct packet_sock *po;
1442 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1448 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1449 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1454 case PACKET_FANOUT_HASH:
1456 idx = fanout_demux_hash(f, skb, num);
1458 case PACKET_FANOUT_LB:
1459 idx = fanout_demux_lb(f, skb, num);
1461 case PACKET_FANOUT_CPU:
1462 idx = fanout_demux_cpu(f, skb, num);
1464 case PACKET_FANOUT_RND:
1465 idx = fanout_demux_rnd(f, skb, num);
1467 case PACKET_FANOUT_QM:
1468 idx = fanout_demux_qm(f, skb, num);
1470 case PACKET_FANOUT_ROLLOVER:
1471 idx = fanout_demux_rollover(f, skb, 0, false, num);
1475 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1476 idx = fanout_demux_rollover(f, skb, idx, true, num);
1478 po = pkt_sk(f->arr[idx]);
1479 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1482 DEFINE_MUTEX(fanout_mutex);
1483 EXPORT_SYMBOL_GPL(fanout_mutex);
1484 static LIST_HEAD(fanout_list);
1486 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1488 struct packet_fanout *f = po->fanout;
1490 spin_lock(&f->lock);
1491 f->arr[f->num_members] = sk;
1494 spin_unlock(&f->lock);
1497 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1499 struct packet_fanout *f = po->fanout;
1502 spin_lock(&f->lock);
1503 for (i = 0; i < f->num_members; i++) {
1504 if (f->arr[i] == sk)
1507 BUG_ON(i >= f->num_members);
1508 f->arr[i] = f->arr[f->num_members - 1];
1510 spin_unlock(&f->lock);
1513 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1515 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1521 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1523 struct packet_sock *po = pkt_sk(sk);
1524 struct packet_fanout *f, *match;
1525 u8 type = type_flags & 0xff;
1526 u8 flags = type_flags >> 8;
1530 case PACKET_FANOUT_ROLLOVER:
1531 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1533 case PACKET_FANOUT_HASH:
1534 case PACKET_FANOUT_LB:
1535 case PACKET_FANOUT_CPU:
1536 case PACKET_FANOUT_RND:
1537 case PACKET_FANOUT_QM:
1549 if (type == PACKET_FANOUT_ROLLOVER ||
1550 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1551 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1554 atomic_long_set(&po->rollover->num, 0);
1555 atomic_long_set(&po->rollover->num_huge, 0);
1556 atomic_long_set(&po->rollover->num_failed, 0);
1559 mutex_lock(&fanout_mutex);
1561 list_for_each_entry(f, &fanout_list, list) {
1563 read_pnet(&f->net) == sock_net(sk)) {
1569 if (match && match->flags != flags)
1573 match = kzalloc(sizeof(*match), GFP_KERNEL);
1576 write_pnet(&match->net, sock_net(sk));
1579 match->flags = flags;
1580 atomic_set(&match->rr_cur, 0);
1581 INIT_LIST_HEAD(&match->list);
1582 spin_lock_init(&match->lock);
1583 atomic_set(&match->sk_ref, 0);
1584 match->prot_hook.type = po->prot_hook.type;
1585 match->prot_hook.dev = po->prot_hook.dev;
1586 match->prot_hook.func = packet_rcv_fanout;
1587 match->prot_hook.af_packet_priv = match;
1588 match->prot_hook.id_match = match_fanout_group;
1589 dev_add_pack(&match->prot_hook);
1590 list_add(&match->list, &fanout_list);
1593 if (match->type == type &&
1594 match->prot_hook.type == po->prot_hook.type &&
1595 match->prot_hook.dev == po->prot_hook.dev) {
1597 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1598 __dev_remove_pack(&po->prot_hook);
1600 atomic_inc(&match->sk_ref);
1601 __fanout_link(sk, po);
1606 mutex_unlock(&fanout_mutex);
1608 kfree(po->rollover);
1609 po->rollover = NULL;
1614 static void fanout_release(struct sock *sk)
1616 struct packet_sock *po = pkt_sk(sk);
1617 struct packet_fanout *f;
1623 mutex_lock(&fanout_mutex);
1626 if (atomic_dec_and_test(&f->sk_ref)) {
1628 dev_remove_pack(&f->prot_hook);
1631 mutex_unlock(&fanout_mutex);
1634 kfree_rcu(po->rollover, rcu);
1637 static const struct proto_ops packet_ops;
1639 static const struct proto_ops packet_ops_spkt;
1641 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1642 struct packet_type *pt, struct net_device *orig_dev)
1645 struct sockaddr_pkt *spkt;
1648 * When we registered the protocol we saved the socket in the data
1649 * field for just this event.
1652 sk = pt->af_packet_priv;
1655 * Yank back the headers [hope the device set this
1656 * right or kerboom...]
1658 * Incoming packets have ll header pulled,
1661 * For outgoing ones skb->data == skb_mac_header(skb)
1662 * so that this procedure is noop.
1665 if (skb->pkt_type == PACKET_LOOPBACK)
1668 if (!net_eq(dev_net(dev), sock_net(sk)))
1671 skb = skb_share_check(skb, GFP_ATOMIC);
1675 /* drop any routing info */
1678 /* drop conntrack reference */
1681 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1683 skb_push(skb, skb->data - skb_mac_header(skb));
1686 * The SOCK_PACKET socket receives _all_ frames.
1689 spkt->spkt_family = dev->type;
1690 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1691 spkt->spkt_protocol = skb->protocol;
1694 * Charge the memory to the socket. This is done specifically
1695 * to prevent sockets using all the memory up.
1698 if (sock_queue_rcv_skb(sk, skb) == 0)
1709 * Output a raw packet to a device layer. This bypasses all the other
1710 * protocol layers and you must therefore supply it with a complete frame
1713 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1716 struct sock *sk = sock->sk;
1717 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1718 struct sk_buff *skb = NULL;
1719 struct net_device *dev;
1725 * Get and verify the address.
1729 if (msg->msg_namelen < sizeof(struct sockaddr))
1731 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1732 proto = saddr->spkt_protocol;
1734 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1737 * Find the device first to size check it
1740 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1743 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1749 if (!(dev->flags & IFF_UP))
1753 * You may not queue a frame bigger than the mtu. This is the lowest level
1754 * raw protocol and you must do your own fragmentation at this level.
1757 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1758 if (!netif_supports_nofcs(dev)) {
1759 err = -EPROTONOSUPPORT;
1762 extra_len = 4; /* We're doing our own CRC */
1766 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1770 size_t reserved = LL_RESERVED_SPACE(dev);
1771 int tlen = dev->needed_tailroom;
1772 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1775 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1778 /* FIXME: Save some space for broken drivers that write a hard
1779 * header at transmission time by themselves. PPP is the notable
1780 * one here. This should really be fixed at the driver level.
1782 skb_reserve(skb, reserved);
1783 skb_reset_network_header(skb);
1785 /* Try to align data part correctly */
1790 skb_reset_network_header(skb);
1792 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1798 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1799 /* Earlier code assumed this would be a VLAN pkt,
1800 * double-check this now that we have the actual
1803 struct ethhdr *ehdr;
1804 skb_reset_mac_header(skb);
1805 ehdr = eth_hdr(skb);
1806 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1812 skb->protocol = proto;
1814 skb->priority = sk->sk_priority;
1815 skb->mark = sk->sk_mark;
1817 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1819 if (unlikely(extra_len == 4))
1822 skb_probe_transport_header(skb, 0);
1824 dev_queue_xmit(skb);
1835 static unsigned int run_filter(const struct sk_buff *skb,
1836 const struct sock *sk,
1839 struct sk_filter *filter;
1842 filter = rcu_dereference(sk->sk_filter);
1844 res = SK_RUN_FILTER(filter, skb);
1851 * This function makes lazy skb cloning in hope that most of packets
1852 * are discarded by BPF.
1854 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1855 * and skb->cb are mangled. It works because (and until) packets
1856 * falling here are owned by current CPU. Output packets are cloned
1857 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1858 * sequencially, so that if we return skb to original state on exit,
1859 * we will not harm anyone.
1862 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1863 struct packet_type *pt, struct net_device *orig_dev)
1866 struct sockaddr_ll *sll;
1867 struct packet_sock *po;
1868 u8 *skb_head = skb->data;
1869 int skb_len = skb->len;
1870 unsigned int snaplen, res;
1872 if (skb->pkt_type == PACKET_LOOPBACK)
1875 sk = pt->af_packet_priv;
1878 if (!net_eq(dev_net(dev), sock_net(sk)))
1883 if (dev->header_ops) {
1884 /* The device has an explicit notion of ll header,
1885 * exported to higher levels.
1887 * Otherwise, the device hides details of its frame
1888 * structure, so that corresponding packet head is
1889 * never delivered to user.
1891 if (sk->sk_type != SOCK_DGRAM)
1892 skb_push(skb, skb->data - skb_mac_header(skb));
1893 else if (skb->pkt_type == PACKET_OUTGOING) {
1894 /* Special case: outgoing packets have ll header at head */
1895 skb_pull(skb, skb_network_offset(skb));
1901 res = run_filter(skb, sk, snaplen);
1903 goto drop_n_restore;
1907 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1910 if (skb_shared(skb)) {
1911 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1915 if (skb_head != skb->data) {
1916 skb->data = skb_head;
1923 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1925 sll = &PACKET_SKB_CB(skb)->sa.ll;
1926 sll->sll_hatype = dev->type;
1927 sll->sll_pkttype = skb->pkt_type;
1928 if (unlikely(po->origdev))
1929 sll->sll_ifindex = orig_dev->ifindex;
1931 sll->sll_ifindex = dev->ifindex;
1933 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1935 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1936 * Use their space for storing the original skb length.
1938 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1940 if (pskb_trim(skb, snaplen))
1943 skb_set_owner_r(skb, sk);
1947 /* drop conntrack reference */
1950 spin_lock(&sk->sk_receive_queue.lock);
1951 po->stats.stats1.tp_packets++;
1952 sock_skb_set_dropcount(sk, skb);
1953 __skb_queue_tail(&sk->sk_receive_queue, skb);
1954 spin_unlock(&sk->sk_receive_queue.lock);
1955 sk->sk_data_ready(sk);
1959 spin_lock(&sk->sk_receive_queue.lock);
1960 po->stats.stats1.tp_drops++;
1961 atomic_inc(&sk->sk_drops);
1962 spin_unlock(&sk->sk_receive_queue.lock);
1965 if (skb_head != skb->data && skb_shared(skb)) {
1966 skb->data = skb_head;
1974 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1975 struct packet_type *pt, struct net_device *orig_dev)
1978 struct packet_sock *po;
1979 struct sockaddr_ll *sll;
1980 union tpacket_uhdr h;
1981 u8 *skb_head = skb->data;
1982 int skb_len = skb->len;
1983 unsigned int snaplen, res;
1984 unsigned long status = TP_STATUS_USER;
1985 unsigned short macoff, netoff, hdrlen;
1986 struct sk_buff *copy_skb = NULL;
1990 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1991 * We may add members to them until current aligned size without forcing
1992 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1994 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1995 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1997 if (skb->pkt_type == PACKET_LOOPBACK)
2000 sk = pt->af_packet_priv;
2003 if (!net_eq(dev_net(dev), sock_net(sk)))
2006 if (dev->header_ops) {
2007 if (sk->sk_type != SOCK_DGRAM)
2008 skb_push(skb, skb->data - skb_mac_header(skb));
2009 else if (skb->pkt_type == PACKET_OUTGOING) {
2010 /* Special case: outgoing packets have ll header at head */
2011 skb_pull(skb, skb_network_offset(skb));
2017 res = run_filter(skb, sk, snaplen);
2019 goto drop_n_restore;
2021 if (skb->ip_summed == CHECKSUM_PARTIAL)
2022 status |= TP_STATUS_CSUMNOTREADY;
2023 else if (skb->pkt_type != PACKET_OUTGOING &&
2024 (skb->ip_summed == CHECKSUM_COMPLETE ||
2025 skb_csum_unnecessary(skb)))
2026 status |= TP_STATUS_CSUM_VALID;
2031 if (sk->sk_type == SOCK_DGRAM) {
2032 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2035 unsigned int maclen = skb_network_offset(skb);
2036 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2037 (maclen < 16 ? 16 : maclen)) +
2039 macoff = netoff - maclen;
2041 if (po->tp_version <= TPACKET_V2) {
2042 if (macoff + snaplen > po->rx_ring.frame_size) {
2043 if (po->copy_thresh &&
2044 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2045 if (skb_shared(skb)) {
2046 copy_skb = skb_clone(skb, GFP_ATOMIC);
2048 copy_skb = skb_get(skb);
2049 skb_head = skb->data;
2052 skb_set_owner_r(copy_skb, sk);
2054 snaplen = po->rx_ring.frame_size - macoff;
2055 if ((int)snaplen < 0)
2058 } else if (unlikely(macoff + snaplen >
2059 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2062 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2063 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2064 snaplen, nval, macoff);
2066 if (unlikely((int)snaplen < 0)) {
2068 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2071 spin_lock(&sk->sk_receive_queue.lock);
2072 h.raw = packet_current_rx_frame(po, skb,
2073 TP_STATUS_KERNEL, (macoff+snaplen));
2076 if (po->tp_version <= TPACKET_V2) {
2077 packet_increment_rx_head(po, &po->rx_ring);
2079 * LOSING will be reported till you read the stats,
2080 * because it's COR - Clear On Read.
2081 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2084 if (po->stats.stats1.tp_drops)
2085 status |= TP_STATUS_LOSING;
2087 po->stats.stats1.tp_packets++;
2089 status |= TP_STATUS_COPY;
2090 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2092 spin_unlock(&sk->sk_receive_queue.lock);
2094 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2096 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2097 getnstimeofday(&ts);
2099 status |= ts_status;
2101 switch (po->tp_version) {
2103 h.h1->tp_len = skb->len;
2104 h.h1->tp_snaplen = snaplen;
2105 h.h1->tp_mac = macoff;
2106 h.h1->tp_net = netoff;
2107 h.h1->tp_sec = ts.tv_sec;
2108 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2109 hdrlen = sizeof(*h.h1);
2112 h.h2->tp_len = skb->len;
2113 h.h2->tp_snaplen = snaplen;
2114 h.h2->tp_mac = macoff;
2115 h.h2->tp_net = netoff;
2116 h.h2->tp_sec = ts.tv_sec;
2117 h.h2->tp_nsec = ts.tv_nsec;
2118 if (skb_vlan_tag_present(skb)) {
2119 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2120 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2121 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2123 h.h2->tp_vlan_tci = 0;
2124 h.h2->tp_vlan_tpid = 0;
2126 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2127 hdrlen = sizeof(*h.h2);
2130 /* tp_nxt_offset,vlan are already populated above.
2131 * So DONT clear those fields here
2133 h.h3->tp_status |= status;
2134 h.h3->tp_len = skb->len;
2135 h.h3->tp_snaplen = snaplen;
2136 h.h3->tp_mac = macoff;
2137 h.h3->tp_net = netoff;
2138 h.h3->tp_sec = ts.tv_sec;
2139 h.h3->tp_nsec = ts.tv_nsec;
2140 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2141 hdrlen = sizeof(*h.h3);
2147 sll = h.raw + TPACKET_ALIGN(hdrlen);
2148 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2149 sll->sll_family = AF_PACKET;
2150 sll->sll_hatype = dev->type;
2151 sll->sll_protocol = skb->protocol;
2152 sll->sll_pkttype = skb->pkt_type;
2153 if (unlikely(po->origdev))
2154 sll->sll_ifindex = orig_dev->ifindex;
2156 sll->sll_ifindex = dev->ifindex;
2160 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2161 if (po->tp_version <= TPACKET_V2) {
2164 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2167 for (start = h.raw; start < end; start += PAGE_SIZE)
2168 flush_dcache_page(pgv_to_page(start));
2173 if (po->tp_version <= TPACKET_V2) {
2174 __packet_set_status(po, h.raw, status);
2175 sk->sk_data_ready(sk);
2177 prb_clear_blk_fill_status(&po->rx_ring);
2181 if (skb_head != skb->data && skb_shared(skb)) {
2182 skb->data = skb_head;
2190 po->stats.stats1.tp_drops++;
2191 spin_unlock(&sk->sk_receive_queue.lock);
2193 sk->sk_data_ready(sk);
2194 kfree_skb(copy_skb);
2195 goto drop_n_restore;
2198 static void tpacket_destruct_skb(struct sk_buff *skb)
2200 struct packet_sock *po = pkt_sk(skb->sk);
2202 if (likely(po->tx_ring.pg_vec)) {
2206 ph = skb_shinfo(skb)->destructor_arg;
2207 packet_dec_pending(&po->tx_ring);
2209 ts = __packet_set_timestamp(po, ph, skb);
2210 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2216 static bool ll_header_truncated(const struct net_device *dev, int len)
2218 /* net device doesn't like empty head */
2219 if (unlikely(len <= dev->hard_header_len)) {
2220 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2221 current->comm, len, dev->hard_header_len);
2228 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2229 void *frame, struct net_device *dev, int size_max,
2230 __be16 proto, unsigned char *addr, int hlen)
2232 union tpacket_uhdr ph;
2233 int to_write, offset, len, tp_len, nr_frags, len_max;
2234 struct socket *sock = po->sk.sk_socket;
2241 skb->protocol = proto;
2243 skb->priority = po->sk.sk_priority;
2244 skb->mark = po->sk.sk_mark;
2245 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2246 skb_shinfo(skb)->destructor_arg = ph.raw;
2248 switch (po->tp_version) {
2250 tp_len = ph.h2->tp_len;
2253 tp_len = ph.h1->tp_len;
2256 if (unlikely(tp_len > size_max)) {
2257 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2261 skb_reserve(skb, hlen);
2262 skb_reset_network_header(skb);
2264 if (!packet_use_direct_xmit(po))
2265 skb_probe_transport_header(skb, 0);
2266 if (unlikely(po->tp_tx_has_off)) {
2267 int off_min, off_max, off;
2268 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2269 off_max = po->tx_ring.frame_size - tp_len;
2270 if (sock->type == SOCK_DGRAM) {
2271 switch (po->tp_version) {
2273 off = ph.h2->tp_net;
2276 off = ph.h1->tp_net;
2280 switch (po->tp_version) {
2282 off = ph.h2->tp_mac;
2285 off = ph.h1->tp_mac;
2289 if (unlikely((off < off_min) || (off_max < off)))
2291 data = ph.raw + off;
2293 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2297 if (sock->type == SOCK_DGRAM) {
2298 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2300 if (unlikely(err < 0))
2302 } else if (dev->hard_header_len) {
2303 if (ll_header_truncated(dev, tp_len))
2306 skb_push(skb, dev->hard_header_len);
2307 err = skb_store_bits(skb, 0, data,
2308 dev->hard_header_len);
2312 data += dev->hard_header_len;
2313 to_write -= dev->hard_header_len;
2316 offset = offset_in_page(data);
2317 len_max = PAGE_SIZE - offset;
2318 len = ((to_write > len_max) ? len_max : to_write);
2320 skb->data_len = to_write;
2321 skb->len += to_write;
2322 skb->truesize += to_write;
2323 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2325 while (likely(to_write)) {
2326 nr_frags = skb_shinfo(skb)->nr_frags;
2328 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2329 pr_err("Packet exceed the number of skb frags(%lu)\n",
2334 page = pgv_to_page(data);
2336 flush_dcache_page(page);
2338 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2341 len_max = PAGE_SIZE;
2342 len = ((to_write > len_max) ? len_max : to_write);
2348 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2350 struct sk_buff *skb;
2351 struct net_device *dev;
2353 int err, reserve = 0;
2355 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2356 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2357 int tp_len, size_max;
2358 unsigned char *addr;
2360 int status = TP_STATUS_AVAILABLE;
2363 mutex_lock(&po->pg_vec_lock);
2365 if (likely(saddr == NULL)) {
2366 dev = packet_cached_dev_get(po);
2371 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2373 if (msg->msg_namelen < (saddr->sll_halen
2374 + offsetof(struct sockaddr_ll,
2377 proto = saddr->sll_protocol;
2378 addr = saddr->sll_addr;
2379 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2383 if (unlikely(dev == NULL))
2386 if (unlikely(!(dev->flags & IFF_UP)))
2389 reserve = dev->hard_header_len + VLAN_HLEN;
2390 size_max = po->tx_ring.frame_size
2391 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2393 if (size_max > dev->mtu + reserve)
2394 size_max = dev->mtu + reserve;
2397 ph = packet_current_frame(po, &po->tx_ring,
2398 TP_STATUS_SEND_REQUEST);
2399 if (unlikely(ph == NULL)) {
2400 if (need_wait && need_resched())
2405 status = TP_STATUS_SEND_REQUEST;
2406 hlen = LL_RESERVED_SPACE(dev);
2407 tlen = dev->needed_tailroom;
2408 skb = sock_alloc_send_skb(&po->sk,
2409 hlen + tlen + sizeof(struct sockaddr_ll),
2412 if (unlikely(skb == NULL)) {
2413 /* we assume the socket was initially writeable ... */
2414 if (likely(len_sum > 0))
2418 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2420 if (tp_len > dev->mtu + dev->hard_header_len) {
2421 struct ethhdr *ehdr;
2422 /* Earlier code assumed this would be a VLAN pkt,
2423 * double-check this now that we have the actual
2427 skb_reset_mac_header(skb);
2428 ehdr = eth_hdr(skb);
2429 if (ehdr->h_proto != htons(ETH_P_8021Q))
2432 if (unlikely(tp_len < 0)) {
2434 __packet_set_status(po, ph,
2435 TP_STATUS_AVAILABLE);
2436 packet_increment_head(&po->tx_ring);
2440 status = TP_STATUS_WRONG_FORMAT;
2446 packet_pick_tx_queue(dev, skb);
2448 skb->destructor = tpacket_destruct_skb;
2449 __packet_set_status(po, ph, TP_STATUS_SENDING);
2450 packet_inc_pending(&po->tx_ring);
2452 status = TP_STATUS_SEND_REQUEST;
2453 err = po->xmit(skb);
2454 if (unlikely(err > 0)) {
2455 err = net_xmit_errno(err);
2456 if (err && __packet_get_status(po, ph) ==
2457 TP_STATUS_AVAILABLE) {
2458 /* skb was destructed already */
2463 * skb was dropped but not destructed yet;
2464 * let's treat it like congestion or err < 0
2468 packet_increment_head(&po->tx_ring);
2470 } while (likely((ph != NULL) ||
2471 /* Note: packet_read_pending() might be slow if we have
2472 * to call it as it's per_cpu variable, but in fast-path
2473 * we already short-circuit the loop with the first
2474 * condition, and luckily don't have to go that path
2477 (need_wait && packet_read_pending(&po->tx_ring))));
2483 __packet_set_status(po, ph, status);
2488 mutex_unlock(&po->pg_vec_lock);
2492 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2493 size_t reserve, size_t len,
2494 size_t linear, int noblock,
2497 struct sk_buff *skb;
2499 /* Under a page? Don't bother with paged skb. */
2500 if (prepad + len < PAGE_SIZE || !linear)
2503 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2508 skb_reserve(skb, reserve);
2509 skb_put(skb, linear);
2510 skb->data_len = len - linear;
2511 skb->len += len - linear;
2516 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2518 struct sock *sk = sock->sk;
2519 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2520 struct sk_buff *skb;
2521 struct net_device *dev;
2523 unsigned char *addr;
2524 int err, reserve = 0;
2525 struct virtio_net_hdr vnet_hdr = { 0 };
2528 struct packet_sock *po = pkt_sk(sk);
2529 unsigned short gso_type = 0;
2535 * Get and verify the address.
2538 if (likely(saddr == NULL)) {
2539 dev = packet_cached_dev_get(po);
2544 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2546 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2548 proto = saddr->sll_protocol;
2549 addr = saddr->sll_addr;
2550 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2554 if (unlikely(dev == NULL))
2557 if (unlikely(!(dev->flags & IFF_UP)))
2560 if (sock->type == SOCK_RAW)
2561 reserve = dev->hard_header_len;
2562 if (po->has_vnet_hdr) {
2563 vnet_hdr_len = sizeof(vnet_hdr);
2566 if (len < vnet_hdr_len)
2569 len -= vnet_hdr_len;
2572 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2573 if (n != vnet_hdr_len)
2576 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2577 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2578 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2579 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2580 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2581 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2582 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2585 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2588 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2589 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2590 case VIRTIO_NET_HDR_GSO_TCPV4:
2591 gso_type = SKB_GSO_TCPV4;
2593 case VIRTIO_NET_HDR_GSO_TCPV6:
2594 gso_type = SKB_GSO_TCPV6;
2596 case VIRTIO_NET_HDR_GSO_UDP:
2597 gso_type = SKB_GSO_UDP;
2603 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2604 gso_type |= SKB_GSO_TCP_ECN;
2606 if (vnet_hdr.gso_size == 0)
2612 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2613 if (!netif_supports_nofcs(dev)) {
2614 err = -EPROTONOSUPPORT;
2617 extra_len = 4; /* We're doing our own CRC */
2621 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2625 hlen = LL_RESERVED_SPACE(dev);
2626 tlen = dev->needed_tailroom;
2627 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2628 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2629 msg->msg_flags & MSG_DONTWAIT, &err);
2633 skb_set_network_header(skb, reserve);
2636 if (sock->type == SOCK_DGRAM) {
2637 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2638 if (unlikely(offset < 0))
2641 if (ll_header_truncated(dev, len))
2645 /* Returns -EFAULT on error */
2646 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2650 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2652 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2653 /* Earlier code assumed this would be a VLAN pkt,
2654 * double-check this now that we have the actual
2657 struct ethhdr *ehdr;
2658 skb_reset_mac_header(skb);
2659 ehdr = eth_hdr(skb);
2660 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2666 skb->protocol = proto;
2668 skb->priority = sk->sk_priority;
2669 skb->mark = sk->sk_mark;
2671 packet_pick_tx_queue(dev, skb);
2673 if (po->has_vnet_hdr) {
2674 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2675 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2676 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2677 if (!skb_partial_csum_set(skb, s, o)) {
2683 skb_shinfo(skb)->gso_size =
2684 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2685 skb_shinfo(skb)->gso_type = gso_type;
2687 /* Header must be checked, and gso_segs computed. */
2688 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2689 skb_shinfo(skb)->gso_segs = 0;
2691 len += vnet_hdr_len;
2694 if (!packet_use_direct_xmit(po))
2695 skb_probe_transport_header(skb, reserve);
2696 if (unlikely(extra_len == 4))
2699 err = po->xmit(skb);
2700 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2716 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2718 struct sock *sk = sock->sk;
2719 struct packet_sock *po = pkt_sk(sk);
2721 if (po->tx_ring.pg_vec)
2722 return tpacket_snd(po, msg);
2724 return packet_snd(sock, msg, len);
2728 * Close a PACKET socket. This is fairly simple. We immediately go
2729 * to 'closed' state and remove our protocol entry in the device list.
2732 static int packet_release(struct socket *sock)
2734 struct sock *sk = sock->sk;
2735 struct packet_sock *po;
2737 union tpacket_req_u req_u;
2745 mutex_lock(&net->packet.sklist_lock);
2746 sk_del_node_init_rcu(sk);
2747 mutex_unlock(&net->packet.sklist_lock);
2750 sock_prot_inuse_add(net, sk->sk_prot, -1);
2753 spin_lock(&po->bind_lock);
2754 unregister_prot_hook(sk, false);
2755 packet_cached_dev_reset(po);
2757 if (po->prot_hook.dev) {
2758 dev_put(po->prot_hook.dev);
2759 po->prot_hook.dev = NULL;
2761 spin_unlock(&po->bind_lock);
2763 packet_flush_mclist(sk);
2765 if (po->rx_ring.pg_vec) {
2766 memset(&req_u, 0, sizeof(req_u));
2767 packet_set_ring(sk, &req_u, 1, 0);
2770 if (po->tx_ring.pg_vec) {
2771 memset(&req_u, 0, sizeof(req_u));
2772 packet_set_ring(sk, &req_u, 1, 1);
2779 * Now the socket is dead. No more input will appear.
2786 skb_queue_purge(&sk->sk_receive_queue);
2787 packet_free_pending(po);
2788 sk_refcnt_debug_release(sk);
2795 * Attach a packet hook.
2798 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2800 struct packet_sock *po = pkt_sk(sk);
2801 const struct net_device *dev_curr;
2813 spin_lock(&po->bind_lock);
2815 proto_curr = po->prot_hook.type;
2816 dev_curr = po->prot_hook.dev;
2818 need_rehook = proto_curr != proto || dev_curr != dev;
2821 unregister_prot_hook(sk, true);
2824 po->prot_hook.type = proto;
2826 if (po->prot_hook.dev)
2827 dev_put(po->prot_hook.dev);
2829 po->prot_hook.dev = dev;
2831 po->ifindex = dev ? dev->ifindex : 0;
2832 packet_cached_dev_assign(po, dev);
2835 if (proto == 0 || !need_rehook)
2838 if (!dev || (dev->flags & IFF_UP)) {
2839 register_prot_hook(sk);
2841 sk->sk_err = ENETDOWN;
2842 if (!sock_flag(sk, SOCK_DEAD))
2843 sk->sk_error_report(sk);
2847 spin_unlock(&po->bind_lock);
2853 * Bind a packet socket to a device
2856 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2859 struct sock *sk = sock->sk;
2861 struct net_device *dev;
2868 if (addr_len != sizeof(struct sockaddr))
2870 strlcpy(name, uaddr->sa_data, sizeof(name));
2872 dev = dev_get_by_name(sock_net(sk), name);
2874 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2878 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2880 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2881 struct sock *sk = sock->sk;
2882 struct net_device *dev = NULL;
2890 if (addr_len < sizeof(struct sockaddr_ll))
2892 if (sll->sll_family != AF_PACKET)
2895 if (sll->sll_ifindex) {
2897 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2901 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2907 static struct proto packet_proto = {
2909 .owner = THIS_MODULE,
2910 .obj_size = sizeof(struct packet_sock),
2914 * Create a packet of type SOCK_PACKET.
2917 static int packet_create(struct net *net, struct socket *sock, int protocol,
2921 struct packet_sock *po;
2922 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2925 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2927 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2928 sock->type != SOCK_PACKET)
2929 return -ESOCKTNOSUPPORT;
2931 sock->state = SS_UNCONNECTED;
2934 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2938 sock->ops = &packet_ops;
2939 if (sock->type == SOCK_PACKET)
2940 sock->ops = &packet_ops_spkt;
2942 sock_init_data(sock, sk);
2945 sk->sk_family = PF_PACKET;
2947 po->xmit = dev_queue_xmit;
2949 err = packet_alloc_pending(po);
2953 packet_cached_dev_reset(po);
2955 sk->sk_destruct = packet_sock_destruct;
2956 sk_refcnt_debug_inc(sk);
2959 * Attach a protocol block
2962 spin_lock_init(&po->bind_lock);
2963 mutex_init(&po->pg_vec_lock);
2964 po->rollover = NULL;
2965 po->prot_hook.func = packet_rcv;
2967 if (sock->type == SOCK_PACKET)
2968 po->prot_hook.func = packet_rcv_spkt;
2970 po->prot_hook.af_packet_priv = sk;
2973 po->prot_hook.type = proto;
2974 register_prot_hook(sk);
2977 mutex_lock(&net->packet.sklist_lock);
2978 sk_add_node_rcu(sk, &net->packet.sklist);
2979 mutex_unlock(&net->packet.sklist_lock);
2982 sock_prot_inuse_add(net, &packet_proto, 1);
2993 * Pull a packet from our receive queue and hand it to the user.
2994 * If necessary we block.
2997 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3000 struct sock *sk = sock->sk;
3001 struct sk_buff *skb;
3003 int vnet_hdr_len = 0;
3004 unsigned int origlen = 0;
3007 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3011 /* What error should we return now? EUNATTACH? */
3012 if (pkt_sk(sk)->ifindex < 0)
3016 if (flags & MSG_ERRQUEUE) {
3017 err = sock_recv_errqueue(sk, msg, len,
3018 SOL_PACKET, PACKET_TX_TIMESTAMP);
3023 * Call the generic datagram receiver. This handles all sorts
3024 * of horrible races and re-entrancy so we can forget about it
3025 * in the protocol layers.
3027 * Now it will return ENETDOWN, if device have just gone down,
3028 * but then it will block.
3031 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3034 * An error occurred so return it. Because skb_recv_datagram()
3035 * handles the blocking we don't see and worry about blocking
3042 if (pkt_sk(sk)->pressure)
3043 packet_rcv_has_room(pkt_sk(sk), NULL);
3045 if (pkt_sk(sk)->has_vnet_hdr) {
3046 struct virtio_net_hdr vnet_hdr = { 0 };
3049 vnet_hdr_len = sizeof(vnet_hdr);
3050 if (len < vnet_hdr_len)
3053 len -= vnet_hdr_len;
3055 if (skb_is_gso(skb)) {
3056 struct skb_shared_info *sinfo = skb_shinfo(skb);
3058 /* This is a hint as to how much should be linear. */
3060 __cpu_to_virtio16(false, skb_headlen(skb));
3062 __cpu_to_virtio16(false, sinfo->gso_size);
3063 if (sinfo->gso_type & SKB_GSO_TCPV4)
3064 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3065 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3066 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3067 else if (sinfo->gso_type & SKB_GSO_UDP)
3068 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3069 else if (sinfo->gso_type & SKB_GSO_FCOE)
3073 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3074 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3076 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3078 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3079 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3080 vnet_hdr.csum_start = __cpu_to_virtio16(false,
3081 skb_checksum_start_offset(skb));
3082 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3084 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3085 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3086 } /* else everything is zero */
3088 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3093 /* You lose any data beyond the buffer you gave. If it worries
3094 * a user program they can ask the device for its MTU
3100 msg->msg_flags |= MSG_TRUNC;
3103 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3107 if (sock->type != SOCK_PACKET) {
3108 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3110 /* Original length was stored in sockaddr_ll fields */
3111 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3112 sll->sll_family = AF_PACKET;
3113 sll->sll_protocol = skb->protocol;
3116 sock_recv_ts_and_drops(msg, sk, skb);
3118 if (msg->msg_name) {
3119 /* If the address length field is there to be filled
3120 * in, we fill it in now.
3122 if (sock->type == SOCK_PACKET) {
3123 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3124 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3126 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3128 msg->msg_namelen = sll->sll_halen +
3129 offsetof(struct sockaddr_ll, sll_addr);
3131 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3135 if (pkt_sk(sk)->auxdata) {
3136 struct tpacket_auxdata aux;
3138 aux.tp_status = TP_STATUS_USER;
3139 if (skb->ip_summed == CHECKSUM_PARTIAL)
3140 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3141 else if (skb->pkt_type != PACKET_OUTGOING &&
3142 (skb->ip_summed == CHECKSUM_COMPLETE ||
3143 skb_csum_unnecessary(skb)))
3144 aux.tp_status |= TP_STATUS_CSUM_VALID;
3146 aux.tp_len = origlen;
3147 aux.tp_snaplen = skb->len;
3149 aux.tp_net = skb_network_offset(skb);
3150 if (skb_vlan_tag_present(skb)) {
3151 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3152 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3153 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3155 aux.tp_vlan_tci = 0;
3156 aux.tp_vlan_tpid = 0;
3158 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3162 * Free or return the buffer as appropriate. Again this
3163 * hides all the races and re-entrancy issues from us.
3165 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3168 skb_free_datagram(sk, skb);
3173 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3174 int *uaddr_len, int peer)
3176 struct net_device *dev;
3177 struct sock *sk = sock->sk;
3182 uaddr->sa_family = AF_PACKET;
3183 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3185 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3187 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3189 *uaddr_len = sizeof(*uaddr);
3194 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3195 int *uaddr_len, int peer)
3197 struct net_device *dev;
3198 struct sock *sk = sock->sk;
3199 struct packet_sock *po = pkt_sk(sk);
3200 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3205 sll->sll_family = AF_PACKET;
3206 sll->sll_ifindex = po->ifindex;
3207 sll->sll_protocol = po->num;
3208 sll->sll_pkttype = 0;
3210 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3212 sll->sll_hatype = dev->type;
3213 sll->sll_halen = dev->addr_len;
3214 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3216 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3220 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3225 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3229 case PACKET_MR_MULTICAST:
3230 if (i->alen != dev->addr_len)
3233 return dev_mc_add(dev, i->addr);
3235 return dev_mc_del(dev, i->addr);
3237 case PACKET_MR_PROMISC:
3238 return dev_set_promiscuity(dev, what);
3239 case PACKET_MR_ALLMULTI:
3240 return dev_set_allmulti(dev, what);
3241 case PACKET_MR_UNICAST:
3242 if (i->alen != dev->addr_len)
3245 return dev_uc_add(dev, i->addr);
3247 return dev_uc_del(dev, i->addr);
3255 static void packet_dev_mclist_delete(struct net_device *dev,
3256 struct packet_mclist **mlp)
3258 struct packet_mclist *ml;
3260 while ((ml = *mlp) != NULL) {
3261 if (ml->ifindex == dev->ifindex) {
3262 packet_dev_mc(dev, ml, -1);
3270 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3272 struct packet_sock *po = pkt_sk(sk);
3273 struct packet_mclist *ml, *i;
3274 struct net_device *dev;
3280 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3285 if (mreq->mr_alen > dev->addr_len)
3289 i = kmalloc(sizeof(*i), GFP_KERNEL);
3294 for (ml = po->mclist; ml; ml = ml->next) {
3295 if (ml->ifindex == mreq->mr_ifindex &&
3296 ml->type == mreq->mr_type &&
3297 ml->alen == mreq->mr_alen &&
3298 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3300 /* Free the new element ... */
3306 i->type = mreq->mr_type;
3307 i->ifindex = mreq->mr_ifindex;
3308 i->alen = mreq->mr_alen;
3309 memcpy(i->addr, mreq->mr_address, i->alen);
3311 i->next = po->mclist;
3313 err = packet_dev_mc(dev, i, 1);
3315 po->mclist = i->next;
3324 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3326 struct packet_mclist *ml, **mlp;
3330 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3331 if (ml->ifindex == mreq->mr_ifindex &&
3332 ml->type == mreq->mr_type &&
3333 ml->alen == mreq->mr_alen &&
3334 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3335 if (--ml->count == 0) {
3336 struct net_device *dev;
3338 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3340 packet_dev_mc(dev, ml, -1);
3350 static void packet_flush_mclist(struct sock *sk)
3352 struct packet_sock *po = pkt_sk(sk);
3353 struct packet_mclist *ml;
3359 while ((ml = po->mclist) != NULL) {
3360 struct net_device *dev;
3362 po->mclist = ml->next;
3363 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3365 packet_dev_mc(dev, ml, -1);
3372 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3374 struct sock *sk = sock->sk;
3375 struct packet_sock *po = pkt_sk(sk);
3378 if (level != SOL_PACKET)
3379 return -ENOPROTOOPT;
3382 case PACKET_ADD_MEMBERSHIP:
3383 case PACKET_DROP_MEMBERSHIP:
3385 struct packet_mreq_max mreq;
3387 memset(&mreq, 0, sizeof(mreq));
3388 if (len < sizeof(struct packet_mreq))
3390 if (len > sizeof(mreq))
3392 if (copy_from_user(&mreq, optval, len))
3394 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3396 if (optname == PACKET_ADD_MEMBERSHIP)
3397 ret = packet_mc_add(sk, &mreq);
3399 ret = packet_mc_drop(sk, &mreq);
3403 case PACKET_RX_RING:
3404 case PACKET_TX_RING:
3406 union tpacket_req_u req_u;
3409 switch (po->tp_version) {
3412 len = sizeof(req_u.req);
3416 len = sizeof(req_u.req3);
3421 if (pkt_sk(sk)->has_vnet_hdr)
3423 if (copy_from_user(&req_u.req, optval, len))
3425 return packet_set_ring(sk, &req_u, 0,
3426 optname == PACKET_TX_RING);
3428 case PACKET_COPY_THRESH:
3432 if (optlen != sizeof(val))
3434 if (copy_from_user(&val, optval, sizeof(val)))
3437 pkt_sk(sk)->copy_thresh = val;
3440 case PACKET_VERSION:
3444 if (optlen != sizeof(val))
3446 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3448 if (copy_from_user(&val, optval, sizeof(val)))
3454 po->tp_version = val;
3460 case PACKET_RESERVE:
3464 if (optlen != sizeof(val))
3466 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3468 if (copy_from_user(&val, optval, sizeof(val)))
3470 po->tp_reserve = val;
3477 if (optlen != sizeof(val))
3479 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3481 if (copy_from_user(&val, optval, sizeof(val)))
3483 po->tp_loss = !!val;
3486 case PACKET_AUXDATA:
3490 if (optlen < sizeof(val))
3492 if (copy_from_user(&val, optval, sizeof(val)))
3495 po->auxdata = !!val;
3498 case PACKET_ORIGDEV:
3502 if (optlen < sizeof(val))
3504 if (copy_from_user(&val, optval, sizeof(val)))
3507 po->origdev = !!val;
3510 case PACKET_VNET_HDR:
3514 if (sock->type != SOCK_RAW)
3516 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3518 if (optlen < sizeof(val))
3520 if (copy_from_user(&val, optval, sizeof(val)))
3523 po->has_vnet_hdr = !!val;
3526 case PACKET_TIMESTAMP:
3530 if (optlen != sizeof(val))
3532 if (copy_from_user(&val, optval, sizeof(val)))
3535 po->tp_tstamp = val;
3542 if (optlen != sizeof(val))
3544 if (copy_from_user(&val, optval, sizeof(val)))
3547 return fanout_add(sk, val & 0xffff, val >> 16);
3549 case PACKET_TX_HAS_OFF:
3553 if (optlen != sizeof(val))
3555 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3557 if (copy_from_user(&val, optval, sizeof(val)))
3559 po->tp_tx_has_off = !!val;
3562 case PACKET_QDISC_BYPASS:
3566 if (optlen != sizeof(val))
3568 if (copy_from_user(&val, optval, sizeof(val)))
3571 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3575 return -ENOPROTOOPT;
3579 static int packet_getsockopt(struct socket *sock, int level, int optname,
3580 char __user *optval, int __user *optlen)
3583 int val, lv = sizeof(val);
3584 struct sock *sk = sock->sk;
3585 struct packet_sock *po = pkt_sk(sk);
3587 union tpacket_stats_u st;
3588 struct tpacket_rollover_stats rstats;
3590 if (level != SOL_PACKET)
3591 return -ENOPROTOOPT;
3593 if (get_user(len, optlen))
3600 case PACKET_STATISTICS:
3601 spin_lock_bh(&sk->sk_receive_queue.lock);
3602 memcpy(&st, &po->stats, sizeof(st));
3603 memset(&po->stats, 0, sizeof(po->stats));
3604 spin_unlock_bh(&sk->sk_receive_queue.lock);
3606 if (po->tp_version == TPACKET_V3) {
3607 lv = sizeof(struct tpacket_stats_v3);
3608 st.stats3.tp_packets += st.stats3.tp_drops;
3611 lv = sizeof(struct tpacket_stats);
3612 st.stats1.tp_packets += st.stats1.tp_drops;
3617 case PACKET_AUXDATA:
3620 case PACKET_ORIGDEV:
3623 case PACKET_VNET_HDR:
3624 val = po->has_vnet_hdr;
3626 case PACKET_VERSION:
3627 val = po->tp_version;
3630 if (len > sizeof(int))
3632 if (copy_from_user(&val, optval, len))
3636 val = sizeof(struct tpacket_hdr);
3639 val = sizeof(struct tpacket2_hdr);
3642 val = sizeof(struct tpacket3_hdr);
3648 case PACKET_RESERVE:
3649 val = po->tp_reserve;
3654 case PACKET_TIMESTAMP:
3655 val = po->tp_tstamp;
3659 ((u32)po->fanout->id |
3660 ((u32)po->fanout->type << 16) |
3661 ((u32)po->fanout->flags << 24)) :
3664 case PACKET_ROLLOVER_STATS:
3667 rstats.tp_all = atomic_long_read(&po->rollover->num);
3668 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3669 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3671 lv = sizeof(rstats);
3673 case PACKET_TX_HAS_OFF:
3674 val = po->tp_tx_has_off;
3676 case PACKET_QDISC_BYPASS:
3677 val = packet_use_direct_xmit(po);
3680 return -ENOPROTOOPT;
3685 if (put_user(len, optlen))
3687 if (copy_to_user(optval, data, len))
3693 static int packet_notifier(struct notifier_block *this,
3694 unsigned long msg, void *ptr)
3697 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3698 struct net *net = dev_net(dev);
3701 sk_for_each_rcu(sk, &net->packet.sklist) {
3702 struct packet_sock *po = pkt_sk(sk);
3705 case NETDEV_UNREGISTER:
3707 packet_dev_mclist_delete(dev, &po->mclist);
3711 if (dev->ifindex == po->ifindex) {
3712 spin_lock(&po->bind_lock);
3714 __unregister_prot_hook(sk, false);
3715 sk->sk_err = ENETDOWN;
3716 if (!sock_flag(sk, SOCK_DEAD))
3717 sk->sk_error_report(sk);
3719 if (msg == NETDEV_UNREGISTER) {
3720 packet_cached_dev_reset(po);
3722 if (po->prot_hook.dev)
3723 dev_put(po->prot_hook.dev);
3724 po->prot_hook.dev = NULL;
3726 spin_unlock(&po->bind_lock);
3730 if (dev->ifindex == po->ifindex) {
3731 spin_lock(&po->bind_lock);
3733 register_prot_hook(sk);
3734 spin_unlock(&po->bind_lock);
3744 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3747 struct sock *sk = sock->sk;
3752 int amount = sk_wmem_alloc_get(sk);
3754 return put_user(amount, (int __user *)arg);
3758 struct sk_buff *skb;
3761 spin_lock_bh(&sk->sk_receive_queue.lock);
3762 skb = skb_peek(&sk->sk_receive_queue);
3765 spin_unlock_bh(&sk->sk_receive_queue.lock);
3766 return put_user(amount, (int __user *)arg);
3769 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3771 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3781 case SIOCGIFBRDADDR:
3782 case SIOCSIFBRDADDR:
3783 case SIOCGIFNETMASK:
3784 case SIOCSIFNETMASK:
3785 case SIOCGIFDSTADDR:
3786 case SIOCSIFDSTADDR:
3788 return inet_dgram_ops.ioctl(sock, cmd, arg);
3792 return -ENOIOCTLCMD;
3797 static unsigned int packet_poll(struct file *file, struct socket *sock,
3800 struct sock *sk = sock->sk;
3801 struct packet_sock *po = pkt_sk(sk);
3802 unsigned int mask = datagram_poll(file, sock, wait);
3804 spin_lock_bh(&sk->sk_receive_queue.lock);
3805 if (po->rx_ring.pg_vec) {
3806 if (!packet_previous_rx_frame(po, &po->rx_ring,
3808 mask |= POLLIN | POLLRDNORM;
3810 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3812 spin_unlock_bh(&sk->sk_receive_queue.lock);
3813 spin_lock_bh(&sk->sk_write_queue.lock);
3814 if (po->tx_ring.pg_vec) {
3815 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3816 mask |= POLLOUT | POLLWRNORM;
3818 spin_unlock_bh(&sk->sk_write_queue.lock);
3823 /* Dirty? Well, I still did not learn better way to account
3827 static void packet_mm_open(struct vm_area_struct *vma)
3829 struct file *file = vma->vm_file;
3830 struct socket *sock = file->private_data;
3831 struct sock *sk = sock->sk;
3834 atomic_inc(&pkt_sk(sk)->mapped);
3837 static void packet_mm_close(struct vm_area_struct *vma)
3839 struct file *file = vma->vm_file;
3840 struct socket *sock = file->private_data;
3841 struct sock *sk = sock->sk;
3844 atomic_dec(&pkt_sk(sk)->mapped);
3847 static const struct vm_operations_struct packet_mmap_ops = {
3848 .open = packet_mm_open,
3849 .close = packet_mm_close,
3852 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3857 for (i = 0; i < len; i++) {
3858 if (likely(pg_vec[i].buffer)) {
3859 if (is_vmalloc_addr(pg_vec[i].buffer))
3860 vfree(pg_vec[i].buffer);
3862 free_pages((unsigned long)pg_vec[i].buffer,
3864 pg_vec[i].buffer = NULL;
3870 static char *alloc_one_pg_vec_page(unsigned long order)
3873 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3874 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3876 buffer = (char *) __get_free_pages(gfp_flags, order);
3880 /* __get_free_pages failed, fall back to vmalloc */
3881 buffer = vzalloc((1 << order) * PAGE_SIZE);
3885 /* vmalloc failed, lets dig into swap here */
3886 gfp_flags &= ~__GFP_NORETRY;
3887 buffer = (char *) __get_free_pages(gfp_flags, order);
3891 /* complete and utter failure */
3895 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3897 unsigned int block_nr = req->tp_block_nr;
3901 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3902 if (unlikely(!pg_vec))
3905 for (i = 0; i < block_nr; i++) {
3906 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3907 if (unlikely(!pg_vec[i].buffer))
3908 goto out_free_pgvec;
3915 free_pg_vec(pg_vec, order, block_nr);
3920 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3921 int closing, int tx_ring)
3923 struct pgv *pg_vec = NULL;
3924 struct packet_sock *po = pkt_sk(sk);
3925 int was_running, order = 0;
3926 struct packet_ring_buffer *rb;
3927 struct sk_buff_head *rb_queue;
3930 /* Added to avoid minimal code churn */
3931 struct tpacket_req *req = &req_u->req;
3933 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3934 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3935 WARN(1, "Tx-ring is not supported.\n");
3939 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3940 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3944 if (atomic_read(&po->mapped))
3946 if (packet_read_pending(rb))
3950 if (req->tp_block_nr) {
3951 /* Sanity tests and some calculations */
3953 if (unlikely(rb->pg_vec))
3956 switch (po->tp_version) {
3958 po->tp_hdrlen = TPACKET_HDRLEN;
3961 po->tp_hdrlen = TPACKET2_HDRLEN;
3964 po->tp_hdrlen = TPACKET3_HDRLEN;
3969 if (unlikely((int)req->tp_block_size <= 0))
3971 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3973 if (po->tp_version >= TPACKET_V3 &&
3974 (int)(req->tp_block_size -
3975 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3977 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3980 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3983 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3984 if (unlikely(rb->frames_per_block <= 0))
3986 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3991 order = get_order(req->tp_block_size);
3992 pg_vec = alloc_pg_vec(req, order);
3993 if (unlikely(!pg_vec))
3995 switch (po->tp_version) {
3997 /* Transmit path is not supported. We checked
3998 * it above but just being paranoid
4001 init_prb_bdqc(po, rb, pg_vec, req_u);
4010 if (unlikely(req->tp_frame_nr))
4016 /* Detach socket from network */
4017 spin_lock(&po->bind_lock);
4018 was_running = po->running;
4022 __unregister_prot_hook(sk, false);
4024 spin_unlock(&po->bind_lock);
4029 mutex_lock(&po->pg_vec_lock);
4030 if (closing || atomic_read(&po->mapped) == 0) {
4032 spin_lock_bh(&rb_queue->lock);
4033 swap(rb->pg_vec, pg_vec);
4034 rb->frame_max = (req->tp_frame_nr - 1);
4036 rb->frame_size = req->tp_frame_size;
4037 spin_unlock_bh(&rb_queue->lock);
4039 swap(rb->pg_vec_order, order);
4040 swap(rb->pg_vec_len, req->tp_block_nr);
4042 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4043 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4044 tpacket_rcv : packet_rcv;
4045 skb_queue_purge(rb_queue);
4046 if (atomic_read(&po->mapped))
4047 pr_err("packet_mmap: vma is busy: %d\n",
4048 atomic_read(&po->mapped));
4050 mutex_unlock(&po->pg_vec_lock);
4052 spin_lock(&po->bind_lock);
4055 register_prot_hook(sk);
4057 spin_unlock(&po->bind_lock);
4058 if (closing && (po->tp_version > TPACKET_V2)) {
4059 /* Because we don't support block-based V3 on tx-ring */
4061 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4066 free_pg_vec(pg_vec, order, req->tp_block_nr);
4071 static int packet_mmap(struct file *file, struct socket *sock,
4072 struct vm_area_struct *vma)
4074 struct sock *sk = sock->sk;
4075 struct packet_sock *po = pkt_sk(sk);
4076 unsigned long size, expected_size;
4077 struct packet_ring_buffer *rb;
4078 unsigned long start;
4085 mutex_lock(&po->pg_vec_lock);
4088 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4090 expected_size += rb->pg_vec_len
4096 if (expected_size == 0)
4099 size = vma->vm_end - vma->vm_start;
4100 if (size != expected_size)
4103 start = vma->vm_start;
4104 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4105 if (rb->pg_vec == NULL)
4108 for (i = 0; i < rb->pg_vec_len; i++) {
4110 void *kaddr = rb->pg_vec[i].buffer;
4113 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4114 page = pgv_to_page(kaddr);
4115 err = vm_insert_page(vma, start, page);
4124 atomic_inc(&po->mapped);
4125 vma->vm_ops = &packet_mmap_ops;
4129 mutex_unlock(&po->pg_vec_lock);
4133 static const struct proto_ops packet_ops_spkt = {
4134 .family = PF_PACKET,
4135 .owner = THIS_MODULE,
4136 .release = packet_release,
4137 .bind = packet_bind_spkt,
4138 .connect = sock_no_connect,
4139 .socketpair = sock_no_socketpair,
4140 .accept = sock_no_accept,
4141 .getname = packet_getname_spkt,
4142 .poll = datagram_poll,
4143 .ioctl = packet_ioctl,
4144 .listen = sock_no_listen,
4145 .shutdown = sock_no_shutdown,
4146 .setsockopt = sock_no_setsockopt,
4147 .getsockopt = sock_no_getsockopt,
4148 .sendmsg = packet_sendmsg_spkt,
4149 .recvmsg = packet_recvmsg,
4150 .mmap = sock_no_mmap,
4151 .sendpage = sock_no_sendpage,
4154 static const struct proto_ops packet_ops = {
4155 .family = PF_PACKET,
4156 .owner = THIS_MODULE,
4157 .release = packet_release,
4158 .bind = packet_bind,
4159 .connect = sock_no_connect,
4160 .socketpair = sock_no_socketpair,
4161 .accept = sock_no_accept,
4162 .getname = packet_getname,
4163 .poll = packet_poll,
4164 .ioctl = packet_ioctl,
4165 .listen = sock_no_listen,
4166 .shutdown = sock_no_shutdown,
4167 .setsockopt = packet_setsockopt,
4168 .getsockopt = packet_getsockopt,
4169 .sendmsg = packet_sendmsg,
4170 .recvmsg = packet_recvmsg,
4171 .mmap = packet_mmap,
4172 .sendpage = sock_no_sendpage,
4175 static const struct net_proto_family packet_family_ops = {
4176 .family = PF_PACKET,
4177 .create = packet_create,
4178 .owner = THIS_MODULE,
4181 static struct notifier_block packet_netdev_notifier = {
4182 .notifier_call = packet_notifier,
4185 #ifdef CONFIG_PROC_FS
4187 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4190 struct net *net = seq_file_net(seq);
4193 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4196 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4198 struct net *net = seq_file_net(seq);
4199 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4202 static void packet_seq_stop(struct seq_file *seq, void *v)
4208 static int packet_seq_show(struct seq_file *seq, void *v)
4210 if (v == SEQ_START_TOKEN)
4211 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4213 struct sock *s = sk_entry(v);
4214 const struct packet_sock *po = pkt_sk(s);
4217 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4219 atomic_read(&s->sk_refcnt),
4224 atomic_read(&s->sk_rmem_alloc),
4225 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4232 static const struct seq_operations packet_seq_ops = {
4233 .start = packet_seq_start,
4234 .next = packet_seq_next,
4235 .stop = packet_seq_stop,
4236 .show = packet_seq_show,
4239 static int packet_seq_open(struct inode *inode, struct file *file)
4241 return seq_open_net(inode, file, &packet_seq_ops,
4242 sizeof(struct seq_net_private));
4245 static const struct file_operations packet_seq_fops = {
4246 .owner = THIS_MODULE,
4247 .open = packet_seq_open,
4249 .llseek = seq_lseek,
4250 .release = seq_release_net,
4255 static int __net_init packet_net_init(struct net *net)
4257 mutex_init(&net->packet.sklist_lock);
4258 INIT_HLIST_HEAD(&net->packet.sklist);
4260 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4266 static void __net_exit packet_net_exit(struct net *net)
4268 remove_proc_entry("packet", net->proc_net);
4271 static struct pernet_operations packet_net_ops = {
4272 .init = packet_net_init,
4273 .exit = packet_net_exit,
4277 static void __exit packet_exit(void)
4279 unregister_netdevice_notifier(&packet_netdev_notifier);
4280 unregister_pernet_subsys(&packet_net_ops);
4281 sock_unregister(PF_PACKET);
4282 proto_unregister(&packet_proto);
4285 static int __init packet_init(void)
4287 int rc = proto_register(&packet_proto, 0);
4292 sock_register(&packet_family_ops);
4293 register_pernet_subsys(&packet_net_ops);
4294 register_netdevice_notifier(&packet_netdev_notifier);
4299 module_init(packet_init);
4300 module_exit(packet_exit);
4301 MODULE_LICENSE("GPL");
4302 MODULE_ALIAS_NETPROTO(PF_PACKET);