2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define 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_rcv(struct sk_buff *skb, struct net_device *dev,
190 struct packet_type *pt, struct net_device *orig_dev);
192 static void *packet_previous_frame(struct packet_sock *po,
193 struct packet_ring_buffer *rb,
195 static void packet_increment_head(struct packet_ring_buffer *buff);
196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
198 struct packet_sock *);
199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
200 struct packet_sock *, unsigned int status);
201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
202 static void prb_open_block(struct tpacket_kbdq_core *,
203 struct tpacket_block_desc *);
204 static void prb_retire_rx_blk_timer_expired(unsigned long);
205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
206 static void prb_init_blk_timer(struct packet_sock *,
207 struct tpacket_kbdq_core *,
208 void (*func) (unsigned long));
209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
210 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
211 struct tpacket3_hdr *);
212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void packet_flush_mclist(struct sock *sk);
215 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb);
217 struct packet_skb_cb {
219 struct sockaddr_pkt pkt;
221 /* Trick: alias skb original length with
222 * ll.sll_family and ll.protocol in order
225 unsigned int origlen;
226 struct sockaddr_ll ll;
231 #define vio_le() virtio_legacy_is_little_endian()
233 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
235 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
236 #define GET_PBLOCK_DESC(x, bid) \
237 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
240 #define GET_NEXT_PRB_BLK_NUM(x) \
241 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
242 ((x)->kactive_blk_num+1) : 0)
244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
245 static void __fanout_link(struct sock *sk, struct packet_sock *po);
247 static int packet_direct_xmit(struct sk_buff *skb)
249 struct net_device *dev = skb->dev;
250 struct sk_buff *orig_skb = skb;
251 struct netdev_queue *txq;
252 int ret = NETDEV_TX_BUSY;
254 if (unlikely(!netif_running(dev) ||
255 !netif_carrier_ok(dev)))
258 skb = validate_xmit_skb_list(skb, dev);
262 packet_pick_tx_queue(dev, skb);
263 txq = skb_get_tx_queue(dev, skb);
267 HARD_TX_LOCK(dev, txq, smp_processor_id());
268 if (!netif_xmit_frozen_or_drv_stopped(txq))
269 ret = netdev_start_xmit(skb, dev, txq, false);
270 HARD_TX_UNLOCK(dev, txq);
274 if (!dev_xmit_complete(ret))
279 atomic_long_inc(&dev->tx_dropped);
281 return NET_XMIT_DROP;
284 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
286 struct net_device *dev;
289 dev = rcu_dereference(po->cached_dev);
297 static void packet_cached_dev_assign(struct packet_sock *po,
298 struct net_device *dev)
300 rcu_assign_pointer(po->cached_dev, dev);
303 static void packet_cached_dev_reset(struct packet_sock *po)
305 RCU_INIT_POINTER(po->cached_dev, NULL);
308 static bool packet_use_direct_xmit(const struct packet_sock *po)
310 return po->xmit == packet_direct_xmit;
313 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
315 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
318 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
320 const struct net_device_ops *ops = dev->netdev_ops;
323 if (ops->ndo_select_queue) {
324 queue_index = ops->ndo_select_queue(dev, skb, NULL,
325 __packet_pick_tx_queue);
326 queue_index = netdev_cap_txqueue(dev, queue_index);
328 queue_index = __packet_pick_tx_queue(dev, skb);
331 skb_set_queue_mapping(skb, queue_index);
334 /* __register_prot_hook must be invoked through register_prot_hook
335 * or from a context in which asynchronous accesses to the packet
336 * socket is not possible (packet_create()).
338 static void __register_prot_hook(struct sock *sk)
340 struct packet_sock *po = pkt_sk(sk);
344 __fanout_link(sk, po);
346 dev_add_pack(&po->prot_hook);
353 static void register_prot_hook(struct sock *sk)
355 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
356 __register_prot_hook(sk);
359 /* If the sync parameter is true, we will temporarily drop
360 * the po->bind_lock and do a synchronize_net to make sure no
361 * asynchronous packet processing paths still refer to the elements
362 * of po->prot_hook. If the sync parameter is false, it is the
363 * callers responsibility to take care of this.
365 static void __unregister_prot_hook(struct sock *sk, bool sync)
367 struct packet_sock *po = pkt_sk(sk);
369 lockdep_assert_held_once(&po->bind_lock);
374 __fanout_unlink(sk, po);
376 __dev_remove_pack(&po->prot_hook);
381 spin_unlock(&po->bind_lock);
383 spin_lock(&po->bind_lock);
387 static void unregister_prot_hook(struct sock *sk, bool sync)
389 struct packet_sock *po = pkt_sk(sk);
392 __unregister_prot_hook(sk, sync);
395 static inline struct page * __pure pgv_to_page(void *addr)
397 if (is_vmalloc_addr(addr))
398 return vmalloc_to_page(addr);
399 return virt_to_page(addr);
402 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
404 union tpacket_uhdr h;
407 switch (po->tp_version) {
409 h.h1->tp_status = status;
410 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
413 h.h2->tp_status = status;
414 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
417 h.h3->tp_status = status;
418 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
421 WARN(1, "TPACKET version not supported.\n");
428 static int __packet_get_status(struct packet_sock *po, void *frame)
430 union tpacket_uhdr h;
435 switch (po->tp_version) {
437 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
438 return h.h1->tp_status;
440 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
441 return h.h2->tp_status;
443 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
444 return h.h3->tp_status;
446 WARN(1, "TPACKET version not supported.\n");
452 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
455 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
458 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
459 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
460 return TP_STATUS_TS_RAW_HARDWARE;
462 if (ktime_to_timespec_cond(skb->tstamp, ts))
463 return TP_STATUS_TS_SOFTWARE;
468 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
471 union tpacket_uhdr h;
475 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
479 switch (po->tp_version) {
481 h.h1->tp_sec = ts.tv_sec;
482 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
485 h.h2->tp_sec = ts.tv_sec;
486 h.h2->tp_nsec = ts.tv_nsec;
489 h.h3->tp_sec = ts.tv_sec;
490 h.h3->tp_nsec = ts.tv_nsec;
493 WARN(1, "TPACKET version not supported.\n");
497 /* one flush is safe, as both fields always lie on the same cacheline */
498 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
504 static void *packet_lookup_frame(struct packet_sock *po,
505 struct packet_ring_buffer *rb,
506 unsigned int position,
509 unsigned int pg_vec_pos, frame_offset;
510 union tpacket_uhdr h;
512 pg_vec_pos = position / rb->frames_per_block;
513 frame_offset = position % rb->frames_per_block;
515 h.raw = rb->pg_vec[pg_vec_pos].buffer +
516 (frame_offset * rb->frame_size);
518 if (status != __packet_get_status(po, h.raw))
524 static void *packet_current_frame(struct packet_sock *po,
525 struct packet_ring_buffer *rb,
528 return packet_lookup_frame(po, rb, rb->head, status);
531 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
533 del_timer_sync(&pkc->retire_blk_timer);
536 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
537 struct sk_buff_head *rb_queue)
539 struct tpacket_kbdq_core *pkc;
541 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
543 spin_lock_bh(&rb_queue->lock);
544 pkc->delete_blk_timer = 1;
545 spin_unlock_bh(&rb_queue->lock);
547 prb_del_retire_blk_timer(pkc);
550 static void prb_init_blk_timer(struct packet_sock *po,
551 struct tpacket_kbdq_core *pkc,
552 void (*func) (unsigned long))
554 init_timer(&pkc->retire_blk_timer);
555 pkc->retire_blk_timer.data = (long)po;
556 pkc->retire_blk_timer.function = func;
557 pkc->retire_blk_timer.expires = jiffies;
560 static void prb_setup_retire_blk_timer(struct packet_sock *po)
562 struct tpacket_kbdq_core *pkc;
564 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
565 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
568 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
569 int blk_size_in_bytes)
571 struct net_device *dev;
572 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
573 struct ethtool_link_ksettings ecmd;
577 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
578 if (unlikely(!dev)) {
580 return DEFAULT_PRB_RETIRE_TOV;
582 err = __ethtool_get_link_ksettings(dev, &ecmd);
586 * If the link speed is so slow you don't really
587 * need to worry about perf anyways
589 if (ecmd.base.speed < SPEED_1000 ||
590 ecmd.base.speed == SPEED_UNKNOWN) {
591 return DEFAULT_PRB_RETIRE_TOV;
594 div = ecmd.base.speed / 1000;
597 return DEFAULT_PRB_RETIRE_TOV;
599 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
611 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
612 union tpacket_req_u *req_u)
614 p1->feature_req_word = req_u->req3.tp_feature_req_word;
617 static void init_prb_bdqc(struct packet_sock *po,
618 struct packet_ring_buffer *rb,
620 union tpacket_req_u *req_u)
622 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
623 struct tpacket_block_desc *pbd;
625 memset(p1, 0x0, sizeof(*p1));
627 p1->knxt_seq_num = 1;
629 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
630 p1->pkblk_start = pg_vec[0].buffer;
631 p1->kblk_size = req_u->req3.tp_block_size;
632 p1->knum_blocks = req_u->req3.tp_block_nr;
633 p1->hdrlen = po->tp_hdrlen;
634 p1->version = po->tp_version;
635 p1->last_kactive_blk_num = 0;
636 po->stats.stats3.tp_freeze_q_cnt = 0;
637 if (req_u->req3.tp_retire_blk_tov)
638 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
640 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
641 req_u->req3.tp_block_size);
642 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
643 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
645 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
646 prb_init_ft_ops(p1, req_u);
647 prb_setup_retire_blk_timer(po);
648 prb_open_block(p1, pbd);
651 /* Do NOT update the last_blk_num first.
652 * Assumes sk_buff_head lock is held.
654 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
656 mod_timer(&pkc->retire_blk_timer,
657 jiffies + pkc->tov_in_jiffies);
658 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
663 * 1) We refresh the timer only when we open a block.
664 * By doing this we don't waste cycles refreshing the timer
665 * on packet-by-packet basis.
667 * With a 1MB block-size, on a 1Gbps line, it will take
668 * i) ~8 ms to fill a block + ii) memcpy etc.
669 * In this cut we are not accounting for the memcpy time.
671 * So, if the user sets the 'tmo' to 10ms then the timer
672 * will never fire while the block is still getting filled
673 * (which is what we want). However, the user could choose
674 * to close a block early and that's fine.
676 * But when the timer does fire, we check whether or not to refresh it.
677 * Since the tmo granularity is in msecs, it is not too expensive
678 * to refresh the timer, lets say every '8' msecs.
679 * Either the user can set the 'tmo' or we can derive it based on
680 * a) line-speed and b) block-size.
681 * prb_calc_retire_blk_tmo() calculates the tmo.
684 static void prb_retire_rx_blk_timer_expired(unsigned long data)
686 struct packet_sock *po = (struct packet_sock *)data;
687 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
689 struct tpacket_block_desc *pbd;
691 spin_lock(&po->sk.sk_receive_queue.lock);
693 frozen = prb_queue_frozen(pkc);
694 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
696 if (unlikely(pkc->delete_blk_timer))
699 /* We only need to plug the race when the block is partially filled.
701 * lock(); increment BLOCK_NUM_PKTS; unlock()
702 * copy_bits() is in progress ...
703 * timer fires on other cpu:
704 * we can't retire the current block because copy_bits
708 if (BLOCK_NUM_PKTS(pbd)) {
709 while (atomic_read(&pkc->blk_fill_in_prog)) {
710 /* Waiting for skb_copy_bits to finish... */
715 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
717 if (!BLOCK_NUM_PKTS(pbd)) {
718 /* An empty block. Just refresh the timer. */
721 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
722 if (!prb_dispatch_next_block(pkc, po))
727 /* Case 1. Queue was frozen because user-space was
730 if (prb_curr_blk_in_use(pbd)) {
732 * Ok, user-space is still behind.
733 * So just refresh the timer.
737 /* Case 2. queue was frozen,user-space caught up,
738 * now the link went idle && the timer fired.
739 * We don't have a block to close.So we open this
740 * block and restart the timer.
741 * opening a block thaws the queue,restarts timer
742 * Thawing/timer-refresh is a side effect.
744 prb_open_block(pkc, pbd);
751 _prb_refresh_rx_retire_blk_timer(pkc);
754 spin_unlock(&po->sk.sk_receive_queue.lock);
757 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
758 struct tpacket_block_desc *pbd1, __u32 status)
760 /* Flush everything minus the block header */
762 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
767 /* Skip the block header(we know header WILL fit in 4K) */
770 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
771 for (; start < end; start += PAGE_SIZE)
772 flush_dcache_page(pgv_to_page(start));
777 /* Now update the block status. */
779 BLOCK_STATUS(pbd1) = status;
781 /* Flush the block header */
783 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
785 flush_dcache_page(pgv_to_page(start));
795 * 2) Increment active_blk_num
797 * Note:We DONT refresh the timer on purpose.
798 * Because almost always the next block will be opened.
800 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
801 struct tpacket_block_desc *pbd1,
802 struct packet_sock *po, unsigned int stat)
804 __u32 status = TP_STATUS_USER | stat;
806 struct tpacket3_hdr *last_pkt;
807 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
808 struct sock *sk = &po->sk;
810 if (po->stats.stats3.tp_drops)
811 status |= TP_STATUS_LOSING;
813 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
814 last_pkt->tp_next_offset = 0;
816 /* Get the ts of the last pkt */
817 if (BLOCK_NUM_PKTS(pbd1)) {
818 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
819 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
821 /* Ok, we tmo'd - so get the current time.
823 * It shouldn't really happen as we don't close empty
824 * blocks. See prb_retire_rx_blk_timer_expired().
828 h1->ts_last_pkt.ts_sec = ts.tv_sec;
829 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
834 /* Flush the block */
835 prb_flush_block(pkc1, pbd1, status);
837 sk->sk_data_ready(sk);
839 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
842 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
844 pkc->reset_pending_on_curr_blk = 0;
848 * Side effect of opening a block:
850 * 1) prb_queue is thawed.
851 * 2) retire_blk_timer is refreshed.
854 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
855 struct tpacket_block_desc *pbd1)
858 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
862 /* We could have just memset this but we will lose the
863 * flexibility of making the priv area sticky
866 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
867 BLOCK_NUM_PKTS(pbd1) = 0;
868 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 h1->ts_first_pkt.ts_sec = ts.tv_sec;
873 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
875 pkc1->pkblk_start = (char *)pbd1;
876 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
878 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
879 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
881 pbd1->version = pkc1->version;
882 pkc1->prev = pkc1->nxt_offset;
883 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
885 prb_thaw_queue(pkc1);
886 _prb_refresh_rx_retire_blk_timer(pkc1);
892 * Queue freeze logic:
893 * 1) Assume tp_block_nr = 8 blocks.
894 * 2) At time 't0', user opens Rx ring.
895 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
896 * 4) user-space is either sleeping or processing block '0'.
897 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
898 * it will close block-7,loop around and try to fill block '0'.
900 * __packet_lookup_frame_in_block
901 * prb_retire_current_block()
902 * prb_dispatch_next_block()
903 * |->(BLOCK_STATUS == USER) evaluates to true
904 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
905 * 6) Now there are two cases:
906 * 6.1) Link goes idle right after the queue is frozen.
907 * But remember, the last open_block() refreshed the timer.
908 * When this timer expires,it will refresh itself so that we can
909 * re-open block-0 in near future.
910 * 6.2) Link is busy and keeps on receiving packets. This is a simple
911 * case and __packet_lookup_frame_in_block will check if block-0
912 * is free and can now be re-used.
914 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
915 struct packet_sock *po)
917 pkc->reset_pending_on_curr_blk = 1;
918 po->stats.stats3.tp_freeze_q_cnt++;
921 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
924 * If the next block is free then we will dispatch it
925 * and return a good offset.
926 * Else, we will freeze the queue.
927 * So, caller must check the return value.
929 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
930 struct packet_sock *po)
932 struct tpacket_block_desc *pbd;
936 /* 1. Get current block num */
937 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
939 /* 2. If this block is currently in_use then freeze the queue */
940 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
941 prb_freeze_queue(pkc, po);
947 * open this block and return the offset where the first packet
948 * needs to get stored.
950 prb_open_block(pkc, pbd);
951 return (void *)pkc->nxt_offset;
954 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
955 struct packet_sock *po, unsigned int status)
957 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
959 /* retire/close the current block */
960 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
962 * Plug the case where copy_bits() is in progress on
963 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
964 * have space to copy the pkt in the current block and
965 * called prb_retire_current_block()
967 * We don't need to worry about the TMO case because
968 * the timer-handler already handled this case.
970 if (!(status & TP_STATUS_BLK_TMO)) {
971 while (atomic_read(&pkc->blk_fill_in_prog)) {
972 /* Waiting for skb_copy_bits to finish... */
976 prb_close_block(pkc, pbd, po, status);
981 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
983 return TP_STATUS_USER & BLOCK_STATUS(pbd);
986 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
988 return pkc->reset_pending_on_curr_blk;
991 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
992 __releases(&pkc->blk_fill_in_prog_lock)
994 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
995 atomic_dec(&pkc->blk_fill_in_prog);
998 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
999 struct tpacket3_hdr *ppd)
1001 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1004 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1005 struct tpacket3_hdr *ppd)
1007 ppd->hv1.tp_rxhash = 0;
1010 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1013 if (skb_vlan_tag_present(pkc->skb)) {
1014 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1015 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1016 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1018 ppd->hv1.tp_vlan_tci = 0;
1019 ppd->hv1.tp_vlan_tpid = 0;
1020 ppd->tp_status = TP_STATUS_AVAILABLE;
1024 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1025 struct tpacket3_hdr *ppd)
1027 ppd->hv1.tp_padding = 0;
1028 prb_fill_vlan_info(pkc, ppd);
1030 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1031 prb_fill_rxhash(pkc, ppd);
1033 prb_clear_rxhash(pkc, ppd);
1036 static void prb_fill_curr_block(char *curr,
1037 struct tpacket_kbdq_core *pkc,
1038 struct tpacket_block_desc *pbd,
1040 __acquires(&pkc->blk_fill_in_prog_lock)
1042 struct tpacket3_hdr *ppd;
1044 ppd = (struct tpacket3_hdr *)curr;
1045 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1047 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1048 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1049 BLOCK_NUM_PKTS(pbd) += 1;
1050 atomic_inc(&pkc->blk_fill_in_prog);
1051 prb_run_all_ft_ops(pkc, ppd);
1054 /* Assumes caller has the sk->rx_queue.lock */
1055 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1056 struct sk_buff *skb,
1061 struct tpacket_kbdq_core *pkc;
1062 struct tpacket_block_desc *pbd;
1065 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1066 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1068 /* Queue is frozen when user space is lagging behind */
1069 if (prb_queue_frozen(pkc)) {
1071 * Check if that last block which caused the queue to freeze,
1072 * is still in_use by user-space.
1074 if (prb_curr_blk_in_use(pbd)) {
1075 /* Can't record this packet */
1079 * Ok, the block was released by user-space.
1080 * Now let's open that block.
1081 * opening a block also thaws the queue.
1082 * Thawing is a side effect.
1084 prb_open_block(pkc, pbd);
1089 curr = pkc->nxt_offset;
1091 end = (char *)pbd + pkc->kblk_size;
1093 /* first try the current block */
1094 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1095 prb_fill_curr_block(curr, pkc, pbd, len);
1096 return (void *)curr;
1099 /* Ok, close the current block */
1100 prb_retire_current_block(pkc, po, 0);
1102 /* Now, try to dispatch the next block */
1103 curr = (char *)prb_dispatch_next_block(pkc, po);
1105 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1106 prb_fill_curr_block(curr, pkc, pbd, len);
1107 return (void *)curr;
1111 * No free blocks are available.user_space hasn't caught up yet.
1112 * Queue was just frozen and now this packet will get dropped.
1117 static void *packet_current_rx_frame(struct packet_sock *po,
1118 struct sk_buff *skb,
1119 int status, unsigned int len)
1122 switch (po->tp_version) {
1125 curr = packet_lookup_frame(po, &po->rx_ring,
1126 po->rx_ring.head, status);
1129 return __packet_lookup_frame_in_block(po, skb, status, len);
1131 WARN(1, "TPACKET version not supported\n");
1137 static void *prb_lookup_block(struct packet_sock *po,
1138 struct packet_ring_buffer *rb,
1142 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1143 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1145 if (status != BLOCK_STATUS(pbd))
1150 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1153 if (rb->prb_bdqc.kactive_blk_num)
1154 prev = rb->prb_bdqc.kactive_blk_num-1;
1156 prev = rb->prb_bdqc.knum_blocks-1;
1160 /* Assumes caller has held the rx_queue.lock */
1161 static void *__prb_previous_block(struct packet_sock *po,
1162 struct packet_ring_buffer *rb,
1165 unsigned int previous = prb_previous_blk_num(rb);
1166 return prb_lookup_block(po, rb, previous, status);
1169 static void *packet_previous_rx_frame(struct packet_sock *po,
1170 struct packet_ring_buffer *rb,
1173 if (po->tp_version <= TPACKET_V2)
1174 return packet_previous_frame(po, rb, status);
1176 return __prb_previous_block(po, rb, status);
1179 static void packet_increment_rx_head(struct packet_sock *po,
1180 struct packet_ring_buffer *rb)
1182 switch (po->tp_version) {
1185 return packet_increment_head(rb);
1188 WARN(1, "TPACKET version not supported.\n");
1194 static void *packet_previous_frame(struct packet_sock *po,
1195 struct packet_ring_buffer *rb,
1198 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1199 return packet_lookup_frame(po, rb, previous, status);
1202 static void packet_increment_head(struct packet_ring_buffer *buff)
1204 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1207 static void packet_inc_pending(struct packet_ring_buffer *rb)
1209 this_cpu_inc(*rb->pending_refcnt);
1212 static void packet_dec_pending(struct packet_ring_buffer *rb)
1214 this_cpu_dec(*rb->pending_refcnt);
1217 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1219 unsigned int refcnt = 0;
1222 /* We don't use pending refcount in rx_ring. */
1223 if (rb->pending_refcnt == NULL)
1226 for_each_possible_cpu(cpu)
1227 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1232 static int packet_alloc_pending(struct packet_sock *po)
1234 po->rx_ring.pending_refcnt = NULL;
1236 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1237 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1243 static void packet_free_pending(struct packet_sock *po)
1245 free_percpu(po->tx_ring.pending_refcnt);
1248 #define ROOM_POW_OFF 2
1249 #define ROOM_NONE 0x0
1250 #define ROOM_LOW 0x1
1251 #define ROOM_NORMAL 0x2
1253 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1257 len = po->rx_ring.frame_max + 1;
1258 idx = po->rx_ring.head;
1260 idx += len >> pow_off;
1263 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1266 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1270 len = po->rx_ring.prb_bdqc.knum_blocks;
1271 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1273 idx += len >> pow_off;
1276 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1279 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1281 struct sock *sk = &po->sk;
1282 int ret = ROOM_NONE;
1284 if (po->prot_hook.func != tpacket_rcv) {
1285 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1286 - (skb ? skb->truesize : 0);
1287 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1295 if (po->tp_version == TPACKET_V3) {
1296 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1298 else if (__tpacket_v3_has_room(po, 0))
1301 if (__tpacket_has_room(po, ROOM_POW_OFF))
1303 else if (__tpacket_has_room(po, 0))
1310 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1315 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1316 ret = __packet_rcv_has_room(po, skb);
1317 has_room = ret == ROOM_NORMAL;
1318 if (po->pressure == has_room)
1319 po->pressure = !has_room;
1320 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1325 static void packet_sock_destruct(struct sock *sk)
1327 skb_queue_purge(&sk->sk_error_queue);
1329 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1330 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1332 if (!sock_flag(sk, SOCK_DEAD)) {
1333 pr_err("Attempt to release alive packet socket: %p\n", sk);
1337 sk_refcnt_debug_dec(sk);
1340 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1342 u32 *history = po->rollover->history;
1346 rxhash = skb_get_hash(skb);
1347 for (i = 0; i < ROLLOVER_HLEN; i++)
1348 if (READ_ONCE(history[i]) == rxhash)
1351 victim = prandom_u32() % ROLLOVER_HLEN;
1353 /* Avoid dirtying the cache line if possible */
1354 if (READ_ONCE(history[victim]) != rxhash)
1355 WRITE_ONCE(history[victim], rxhash);
1357 return count > (ROLLOVER_HLEN >> 1);
1360 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1361 struct sk_buff *skb,
1364 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1367 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1368 struct sk_buff *skb,
1371 unsigned int val = atomic_inc_return(&f->rr_cur);
1376 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1377 struct sk_buff *skb,
1380 return smp_processor_id() % num;
1383 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1384 struct sk_buff *skb,
1387 return prandom_u32_max(num);
1390 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1391 struct sk_buff *skb,
1392 unsigned int idx, bool try_self,
1395 struct packet_sock *po, *po_next, *po_skip = NULL;
1396 unsigned int i, j, room = ROOM_NONE;
1398 po = pkt_sk(f->arr[idx]);
1401 room = packet_rcv_has_room(po, skb);
1402 if (room == ROOM_NORMAL ||
1403 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1408 i = j = min_t(int, po->rollover->sock, num - 1);
1410 po_next = pkt_sk(f->arr[i]);
1411 if (po_next != po_skip && !po_next->pressure &&
1412 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1414 po->rollover->sock = i;
1415 atomic_long_inc(&po->rollover->num);
1416 if (room == ROOM_LOW)
1417 atomic_long_inc(&po->rollover->num_huge);
1425 atomic_long_inc(&po->rollover->num_failed);
1429 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1430 struct sk_buff *skb,
1433 return skb_get_queue_mapping(skb) % num;
1436 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1437 struct sk_buff *skb,
1440 struct bpf_prog *prog;
1441 unsigned int ret = 0;
1444 prog = rcu_dereference(f->bpf_prog);
1446 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1452 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1454 return f->flags & (flag >> 8);
1457 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1458 struct packet_type *pt, struct net_device *orig_dev)
1460 struct packet_fanout *f = pt->af_packet_priv;
1461 unsigned int num = READ_ONCE(f->num_members);
1462 struct net *net = read_pnet(&f->net);
1463 struct packet_sock *po;
1466 if (!net_eq(dev_net(dev), net) || !num) {
1471 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1472 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1477 case PACKET_FANOUT_HASH:
1479 idx = fanout_demux_hash(f, skb, num);
1481 case PACKET_FANOUT_LB:
1482 idx = fanout_demux_lb(f, skb, num);
1484 case PACKET_FANOUT_CPU:
1485 idx = fanout_demux_cpu(f, skb, num);
1487 case PACKET_FANOUT_RND:
1488 idx = fanout_demux_rnd(f, skb, num);
1490 case PACKET_FANOUT_QM:
1491 idx = fanout_demux_qm(f, skb, num);
1493 case PACKET_FANOUT_ROLLOVER:
1494 idx = fanout_demux_rollover(f, skb, 0, false, num);
1496 case PACKET_FANOUT_CBPF:
1497 case PACKET_FANOUT_EBPF:
1498 idx = fanout_demux_bpf(f, skb, num);
1502 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1503 idx = fanout_demux_rollover(f, skb, idx, true, num);
1505 po = pkt_sk(f->arr[idx]);
1506 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1509 DEFINE_MUTEX(fanout_mutex);
1510 EXPORT_SYMBOL_GPL(fanout_mutex);
1511 static LIST_HEAD(fanout_list);
1512 static u16 fanout_next_id;
1514 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1516 struct packet_fanout *f = po->fanout;
1518 spin_lock(&f->lock);
1519 f->arr[f->num_members] = sk;
1522 if (f->num_members == 1)
1523 dev_add_pack(&f->prot_hook);
1524 spin_unlock(&f->lock);
1527 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1529 struct packet_fanout *f = po->fanout;
1532 spin_lock(&f->lock);
1533 for (i = 0; i < f->num_members; i++) {
1534 if (f->arr[i] == sk)
1537 BUG_ON(i >= f->num_members);
1538 f->arr[i] = f->arr[f->num_members - 1];
1540 if (f->num_members == 0)
1541 __dev_remove_pack(&f->prot_hook);
1542 spin_unlock(&f->lock);
1545 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1547 if (sk->sk_family != PF_PACKET)
1550 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1553 static void fanout_init_data(struct packet_fanout *f)
1556 case PACKET_FANOUT_LB:
1557 atomic_set(&f->rr_cur, 0);
1559 case PACKET_FANOUT_CBPF:
1560 case PACKET_FANOUT_EBPF:
1561 RCU_INIT_POINTER(f->bpf_prog, NULL);
1566 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1568 struct bpf_prog *old;
1570 spin_lock(&f->lock);
1571 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1572 rcu_assign_pointer(f->bpf_prog, new);
1573 spin_unlock(&f->lock);
1577 bpf_prog_destroy(old);
1581 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1584 struct bpf_prog *new;
1585 struct sock_fprog fprog;
1588 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1590 if (len != sizeof(fprog))
1592 if (copy_from_user(&fprog, data, len))
1595 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1599 __fanout_set_data_bpf(po->fanout, new);
1603 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1606 struct bpf_prog *new;
1609 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1611 if (len != sizeof(fd))
1613 if (copy_from_user(&fd, data, len))
1616 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1618 return PTR_ERR(new);
1620 __fanout_set_data_bpf(po->fanout, new);
1624 static int fanout_set_data(struct packet_sock *po, char __user *data,
1627 switch (po->fanout->type) {
1628 case PACKET_FANOUT_CBPF:
1629 return fanout_set_data_cbpf(po, data, len);
1630 case PACKET_FANOUT_EBPF:
1631 return fanout_set_data_ebpf(po, data, len);
1637 static void fanout_release_data(struct packet_fanout *f)
1640 case PACKET_FANOUT_CBPF:
1641 case PACKET_FANOUT_EBPF:
1642 __fanout_set_data_bpf(f, NULL);
1646 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1648 struct packet_fanout *f;
1650 list_for_each_entry(f, &fanout_list, list) {
1651 if (f->id == candidate_id &&
1652 read_pnet(&f->net) == sock_net(sk)) {
1659 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1661 u16 id = fanout_next_id;
1664 if (__fanout_id_is_free(sk, id)) {
1666 fanout_next_id = id + 1;
1671 } while (id != fanout_next_id);
1676 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1678 struct packet_rollover *rollover = NULL;
1679 struct packet_sock *po = pkt_sk(sk);
1680 struct packet_fanout *f, *match;
1681 u8 type = type_flags & 0xff;
1682 u8 flags = type_flags >> 8;
1686 case PACKET_FANOUT_ROLLOVER:
1687 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1689 case PACKET_FANOUT_HASH:
1690 case PACKET_FANOUT_LB:
1691 case PACKET_FANOUT_CPU:
1692 case PACKET_FANOUT_RND:
1693 case PACKET_FANOUT_QM:
1694 case PACKET_FANOUT_CBPF:
1695 case PACKET_FANOUT_EBPF:
1701 mutex_lock(&fanout_mutex);
1707 if (type == PACKET_FANOUT_ROLLOVER ||
1708 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1710 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1713 atomic_long_set(&rollover->num, 0);
1714 atomic_long_set(&rollover->num_huge, 0);
1715 atomic_long_set(&rollover->num_failed, 0);
1718 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1723 if (!fanout_find_new_id(sk, &id)) {
1727 /* ephemeral flag for the first socket in the group: drop it */
1728 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1732 list_for_each_entry(f, &fanout_list, list) {
1734 read_pnet(&f->net) == sock_net(sk)) {
1740 if (match && match->flags != flags)
1744 match = kzalloc(sizeof(*match), GFP_KERNEL);
1747 write_pnet(&match->net, sock_net(sk));
1750 match->flags = flags;
1751 INIT_LIST_HEAD(&match->list);
1752 spin_lock_init(&match->lock);
1753 refcount_set(&match->sk_ref, 0);
1754 fanout_init_data(match);
1755 match->prot_hook.type = po->prot_hook.type;
1756 match->prot_hook.dev = po->prot_hook.dev;
1757 match->prot_hook.func = packet_rcv_fanout;
1758 match->prot_hook.af_packet_priv = match;
1759 match->prot_hook.af_packet_net = read_pnet(&match->net);
1760 match->prot_hook.id_match = match_fanout_group;
1761 list_add(&match->list, &fanout_list);
1765 spin_lock(&po->bind_lock);
1767 match->type == type &&
1768 match->prot_hook.type == po->prot_hook.type &&
1769 match->prot_hook.dev == po->prot_hook.dev) {
1771 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1772 __dev_remove_pack(&po->prot_hook);
1774 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1775 WRITE_ONCE(po->fanout, match);
1777 po->rollover = rollover;
1779 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1780 __fanout_link(sk, po);
1784 spin_unlock(&po->bind_lock);
1786 if (err && !refcount_read(&match->sk_ref)) {
1787 list_del(&match->list);
1793 mutex_unlock(&fanout_mutex);
1797 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1798 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1799 * It is the responsibility of the caller to call fanout_release_data() and
1800 * free the returned packet_fanout (after synchronize_net())
1802 static struct packet_fanout *fanout_release(struct sock *sk)
1804 struct packet_sock *po = pkt_sk(sk);
1805 struct packet_fanout *f;
1807 mutex_lock(&fanout_mutex);
1812 if (refcount_dec_and_test(&f->sk_ref))
1817 mutex_unlock(&fanout_mutex);
1822 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1823 struct sk_buff *skb)
1825 /* Earlier code assumed this would be a VLAN pkt, double-check
1826 * this now that we have the actual packet in hand. We can only
1827 * do this check on Ethernet devices.
1829 if (unlikely(dev->type != ARPHRD_ETHER))
1832 skb_reset_mac_header(skb);
1833 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1836 static const struct proto_ops packet_ops;
1838 static const struct proto_ops packet_ops_spkt;
1840 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1841 struct packet_type *pt, struct net_device *orig_dev)
1844 struct sockaddr_pkt *spkt;
1847 * When we registered the protocol we saved the socket in the data
1848 * field for just this event.
1851 sk = pt->af_packet_priv;
1854 * Yank back the headers [hope the device set this
1855 * right or kerboom...]
1857 * Incoming packets have ll header pulled,
1860 * For outgoing ones skb->data == skb_mac_header(skb)
1861 * so that this procedure is noop.
1864 if (skb->pkt_type == PACKET_LOOPBACK)
1867 if (!net_eq(dev_net(dev), sock_net(sk)))
1870 skb = skb_share_check(skb, GFP_ATOMIC);
1874 /* drop any routing info */
1877 /* drop conntrack reference */
1880 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1882 skb_push(skb, skb->data - skb_mac_header(skb));
1885 * The SOCK_PACKET socket receives _all_ frames.
1888 spkt->spkt_family = dev->type;
1889 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1890 spkt->spkt_protocol = skb->protocol;
1893 * Charge the memory to the socket. This is done specifically
1894 * to prevent sockets using all the memory up.
1897 if (sock_queue_rcv_skb(sk, skb) == 0)
1908 * Output a raw packet to a device layer. This bypasses all the other
1909 * protocol layers and you must therefore supply it with a complete frame
1912 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1915 struct sock *sk = sock->sk;
1916 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1917 struct sk_buff *skb = NULL;
1918 struct net_device *dev;
1919 struct sockcm_cookie sockc;
1925 * Get and verify the address.
1929 if (msg->msg_namelen < sizeof(struct sockaddr))
1931 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1932 proto = saddr->spkt_protocol;
1934 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1937 * Find the device first to size check it
1940 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1943 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1949 if (!(dev->flags & IFF_UP))
1953 * You may not queue a frame bigger than the mtu. This is the lowest level
1954 * raw protocol and you must do your own fragmentation at this level.
1957 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1958 if (!netif_supports_nofcs(dev)) {
1959 err = -EPROTONOSUPPORT;
1962 extra_len = 4; /* We're doing our own CRC */
1966 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1970 size_t reserved = LL_RESERVED_SPACE(dev);
1971 int tlen = dev->needed_tailroom;
1972 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1975 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1978 /* FIXME: Save some space for broken drivers that write a hard
1979 * header at transmission time by themselves. PPP is the notable
1980 * one here. This should really be fixed at the driver level.
1982 skb_reserve(skb, reserved);
1983 skb_reset_network_header(skb);
1985 /* Try to align data part correctly */
1990 skb_reset_network_header(skb);
1992 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1998 if (!dev_validate_header(dev, skb->data, len)) {
2002 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2003 !packet_extra_vlan_len_allowed(dev, skb)) {
2008 sockc.tsflags = sk->sk_tsflags;
2009 if (msg->msg_controllen) {
2010 err = sock_cmsg_send(sk, msg, &sockc);
2015 skb->protocol = proto;
2017 skb->priority = sk->sk_priority;
2018 skb->mark = sk->sk_mark;
2020 skb_setup_tx_timestamp(skb, sockc.tsflags);
2022 if (unlikely(extra_len == 4))
2025 skb_probe_transport_header(skb, 0);
2027 dev_queue_xmit(skb);
2038 static unsigned int run_filter(struct sk_buff *skb,
2039 const struct sock *sk,
2042 struct sk_filter *filter;
2045 filter = rcu_dereference(sk->sk_filter);
2047 res = bpf_prog_run_clear_cb(filter->prog, skb);
2053 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2056 struct virtio_net_hdr vnet_hdr;
2058 if (*len < sizeof(vnet_hdr))
2060 *len -= sizeof(vnet_hdr);
2062 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2065 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2069 * This function makes lazy skb cloning in hope that most of packets
2070 * are discarded by BPF.
2072 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2073 * and skb->cb are mangled. It works because (and until) packets
2074 * falling here are owned by current CPU. Output packets are cloned
2075 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2076 * sequencially, so that if we return skb to original state on exit,
2077 * we will not harm anyone.
2080 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2081 struct packet_type *pt, struct net_device *orig_dev)
2084 struct sockaddr_ll *sll;
2085 struct packet_sock *po;
2086 u8 *skb_head = skb->data;
2087 int skb_len = skb->len;
2088 unsigned int snaplen, res;
2089 bool is_drop_n_account = false;
2091 if (skb->pkt_type == PACKET_LOOPBACK)
2094 sk = pt->af_packet_priv;
2097 if (!net_eq(dev_net(dev), sock_net(sk)))
2102 if (dev->header_ops) {
2103 /* The device has an explicit notion of ll header,
2104 * exported to higher levels.
2106 * Otherwise, the device hides details of its frame
2107 * structure, so that corresponding packet head is
2108 * never delivered to user.
2110 if (sk->sk_type != SOCK_DGRAM)
2111 skb_push(skb, skb->data - skb_mac_header(skb));
2112 else if (skb->pkt_type == PACKET_OUTGOING) {
2113 /* Special case: outgoing packets have ll header at head */
2114 skb_pull(skb, skb_network_offset(skb));
2120 res = run_filter(skb, sk, snaplen);
2122 goto drop_n_restore;
2126 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2129 if (skb_shared(skb)) {
2130 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2134 if (skb_head != skb->data) {
2135 skb->data = skb_head;
2142 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2144 sll = &PACKET_SKB_CB(skb)->sa.ll;
2145 sll->sll_hatype = dev->type;
2146 sll->sll_pkttype = skb->pkt_type;
2147 if (unlikely(po->origdev))
2148 sll->sll_ifindex = orig_dev->ifindex;
2150 sll->sll_ifindex = dev->ifindex;
2152 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2154 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2155 * Use their space for storing the original skb length.
2157 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2159 if (pskb_trim(skb, snaplen))
2162 skb_set_owner_r(skb, sk);
2166 /* drop conntrack reference */
2169 spin_lock(&sk->sk_receive_queue.lock);
2170 po->stats.stats1.tp_packets++;
2171 sock_skb_set_dropcount(sk, skb);
2172 __skb_queue_tail(&sk->sk_receive_queue, skb);
2173 spin_unlock(&sk->sk_receive_queue.lock);
2174 sk->sk_data_ready(sk);
2178 is_drop_n_account = true;
2179 spin_lock(&sk->sk_receive_queue.lock);
2180 po->stats.stats1.tp_drops++;
2181 atomic_inc(&sk->sk_drops);
2182 spin_unlock(&sk->sk_receive_queue.lock);
2185 if (skb_head != skb->data && skb_shared(skb)) {
2186 skb->data = skb_head;
2190 if (!is_drop_n_account)
2197 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2198 struct packet_type *pt, struct net_device *orig_dev)
2201 struct packet_sock *po;
2202 struct sockaddr_ll *sll;
2203 union tpacket_uhdr h;
2204 u8 *skb_head = skb->data;
2205 int skb_len = skb->len;
2206 unsigned int snaplen, res;
2207 unsigned long status = TP_STATUS_USER;
2208 unsigned short macoff, hdrlen;
2209 unsigned int netoff;
2210 struct sk_buff *copy_skb = NULL;
2213 bool is_drop_n_account = false;
2214 unsigned int slot_id = 0;
2215 bool do_vnet = false;
2217 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2218 * We may add members to them until current aligned size without forcing
2219 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2221 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2222 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2224 if (skb->pkt_type == PACKET_LOOPBACK)
2227 sk = pt->af_packet_priv;
2230 if (!net_eq(dev_net(dev), sock_net(sk)))
2233 if (dev->header_ops) {
2234 if (sk->sk_type != SOCK_DGRAM)
2235 skb_push(skb, skb->data - skb_mac_header(skb));
2236 else if (skb->pkt_type == PACKET_OUTGOING) {
2237 /* Special case: outgoing packets have ll header at head */
2238 skb_pull(skb, skb_network_offset(skb));
2244 res = run_filter(skb, sk, snaplen);
2246 goto drop_n_restore;
2248 if (skb->ip_summed == CHECKSUM_PARTIAL)
2249 status |= TP_STATUS_CSUMNOTREADY;
2250 else if (skb->pkt_type != PACKET_OUTGOING &&
2251 skb_csum_unnecessary(skb))
2252 status |= TP_STATUS_CSUM_VALID;
2257 if (sk->sk_type == SOCK_DGRAM) {
2258 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2261 unsigned int maclen = skb_network_offset(skb);
2262 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2263 (maclen < 16 ? 16 : maclen)) +
2265 if (po->has_vnet_hdr) {
2266 netoff += sizeof(struct virtio_net_hdr);
2269 macoff = netoff - maclen;
2271 if (netoff > USHRT_MAX) {
2272 spin_lock(&sk->sk_receive_queue.lock);
2273 po->stats.stats1.tp_drops++;
2274 spin_unlock(&sk->sk_receive_queue.lock);
2275 goto drop_n_restore;
2277 if (po->tp_version <= TPACKET_V2) {
2278 if (macoff + snaplen > po->rx_ring.frame_size) {
2279 if (po->copy_thresh &&
2280 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2281 if (skb_shared(skb)) {
2282 copy_skb = skb_clone(skb, GFP_ATOMIC);
2284 copy_skb = skb_get(skb);
2285 skb_head = skb->data;
2288 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2289 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2290 skb_set_owner_r(copy_skb, sk);
2293 snaplen = po->rx_ring.frame_size - macoff;
2294 if ((int)snaplen < 0) {
2299 } else if (unlikely(macoff + snaplen >
2300 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2303 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2304 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2305 snaplen, nval, macoff);
2307 if (unlikely((int)snaplen < 0)) {
2309 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2313 spin_lock(&sk->sk_receive_queue.lock);
2314 h.raw = packet_current_rx_frame(po, skb,
2315 TP_STATUS_KERNEL, (macoff+snaplen));
2317 goto drop_n_account;
2319 if (po->tp_version <= TPACKET_V2) {
2320 slot_id = po->rx_ring.head;
2321 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2322 goto drop_n_account;
2323 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2327 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2328 sizeof(struct virtio_net_hdr),
2329 vio_le(), true, 0)) {
2330 if (po->tp_version == TPACKET_V3)
2331 prb_clear_blk_fill_status(&po->rx_ring);
2332 goto drop_n_account;
2335 if (po->tp_version <= TPACKET_V2) {
2336 packet_increment_rx_head(po, &po->rx_ring);
2338 * LOSING will be reported till you read the stats,
2339 * because it's COR - Clear On Read.
2340 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2343 if (po->stats.stats1.tp_drops)
2344 status |= TP_STATUS_LOSING;
2347 po->stats.stats1.tp_packets++;
2349 status |= TP_STATUS_COPY;
2350 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2352 spin_unlock(&sk->sk_receive_queue.lock);
2354 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2356 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2357 getnstimeofday(&ts);
2359 status |= ts_status;
2361 switch (po->tp_version) {
2363 h.h1->tp_len = skb->len;
2364 h.h1->tp_snaplen = snaplen;
2365 h.h1->tp_mac = macoff;
2366 h.h1->tp_net = netoff;
2367 h.h1->tp_sec = ts.tv_sec;
2368 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2369 hdrlen = sizeof(*h.h1);
2372 h.h2->tp_len = skb->len;
2373 h.h2->tp_snaplen = snaplen;
2374 h.h2->tp_mac = macoff;
2375 h.h2->tp_net = netoff;
2376 h.h2->tp_sec = ts.tv_sec;
2377 h.h2->tp_nsec = ts.tv_nsec;
2378 if (skb_vlan_tag_present(skb)) {
2379 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2380 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2381 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2383 h.h2->tp_vlan_tci = 0;
2384 h.h2->tp_vlan_tpid = 0;
2386 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2387 hdrlen = sizeof(*h.h2);
2390 /* tp_nxt_offset,vlan are already populated above.
2391 * So DONT clear those fields here
2393 h.h3->tp_status |= status;
2394 h.h3->tp_len = skb->len;
2395 h.h3->tp_snaplen = snaplen;
2396 h.h3->tp_mac = macoff;
2397 h.h3->tp_net = netoff;
2398 h.h3->tp_sec = ts.tv_sec;
2399 h.h3->tp_nsec = ts.tv_nsec;
2400 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2401 hdrlen = sizeof(*h.h3);
2407 sll = h.raw + TPACKET_ALIGN(hdrlen);
2408 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2409 sll->sll_family = AF_PACKET;
2410 sll->sll_hatype = dev->type;
2411 sll->sll_protocol = skb->protocol;
2412 sll->sll_pkttype = skb->pkt_type;
2413 if (unlikely(po->origdev))
2414 sll->sll_ifindex = orig_dev->ifindex;
2416 sll->sll_ifindex = dev->ifindex;
2420 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2421 if (po->tp_version <= TPACKET_V2) {
2424 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2427 for (start = h.raw; start < end; start += PAGE_SIZE)
2428 flush_dcache_page(pgv_to_page(start));
2433 if (po->tp_version <= TPACKET_V2) {
2434 spin_lock(&sk->sk_receive_queue.lock);
2435 __packet_set_status(po, h.raw, status);
2436 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2437 spin_unlock(&sk->sk_receive_queue.lock);
2438 sk->sk_data_ready(sk);
2439 } else if (po->tp_version == TPACKET_V3) {
2440 prb_clear_blk_fill_status(&po->rx_ring);
2444 if (skb_head != skb->data && skb_shared(skb)) {
2445 skb->data = skb_head;
2449 if (!is_drop_n_account)
2456 is_drop_n_account = true;
2457 po->stats.stats1.tp_drops++;
2458 spin_unlock(&sk->sk_receive_queue.lock);
2460 sk->sk_data_ready(sk);
2461 kfree_skb(copy_skb);
2462 goto drop_n_restore;
2465 static void tpacket_destruct_skb(struct sk_buff *skb)
2467 struct packet_sock *po = pkt_sk(skb->sk);
2469 if (likely(po->tx_ring.pg_vec)) {
2473 ph = skb_zcopy_get_nouarg(skb);
2474 packet_dec_pending(&po->tx_ring);
2476 ts = __packet_set_timestamp(po, ph, skb);
2477 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2479 if (!packet_read_pending(&po->tx_ring))
2480 complete(&po->skb_completion);
2486 static void tpacket_set_protocol(const struct net_device *dev,
2487 struct sk_buff *skb)
2489 if (dev->type == ARPHRD_ETHER) {
2490 skb_reset_mac_header(skb);
2491 skb->protocol = eth_hdr(skb)->h_proto;
2495 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2497 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2498 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2499 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2500 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2501 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2502 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2503 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2505 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2511 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2512 struct virtio_net_hdr *vnet_hdr)
2514 if (*len < sizeof(*vnet_hdr))
2516 *len -= sizeof(*vnet_hdr);
2518 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2521 return __packet_snd_vnet_parse(vnet_hdr, *len);
2524 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2525 void *frame, struct net_device *dev, void *data, int tp_len,
2526 __be16 proto, unsigned char *addr, int hlen, int copylen,
2527 const struct sockcm_cookie *sockc)
2529 union tpacket_uhdr ph;
2530 int to_write, offset, len, nr_frags, len_max;
2531 struct socket *sock = po->sk.sk_socket;
2537 skb->protocol = proto;
2539 skb->priority = po->sk.sk_priority;
2540 skb->mark = po->sk.sk_mark;
2541 skb_setup_tx_timestamp(skb, sockc->tsflags);
2542 skb_zcopy_set_nouarg(skb, ph.raw);
2544 skb_reserve(skb, hlen);
2545 skb_reset_network_header(skb);
2549 if (sock->type == SOCK_DGRAM) {
2550 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2552 if (unlikely(err < 0))
2554 } else if (copylen) {
2555 int hdrlen = min_t(int, copylen, tp_len);
2557 skb_push(skb, dev->hard_header_len);
2558 skb_put(skb, copylen - dev->hard_header_len);
2559 err = skb_store_bits(skb, 0, data, hdrlen);
2562 if (!dev_validate_header(dev, skb->data, hdrlen))
2565 tpacket_set_protocol(dev, skb);
2571 offset = offset_in_page(data);
2572 len_max = PAGE_SIZE - offset;
2573 len = ((to_write > len_max) ? len_max : to_write);
2575 skb->data_len = to_write;
2576 skb->len += to_write;
2577 skb->truesize += to_write;
2578 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2580 while (likely(to_write)) {
2581 nr_frags = skb_shinfo(skb)->nr_frags;
2583 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2584 pr_err("Packet exceed the number of skb frags(%lu)\n",
2589 page = pgv_to_page(data);
2591 flush_dcache_page(page);
2593 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2596 len_max = PAGE_SIZE;
2597 len = ((to_write > len_max) ? len_max : to_write);
2600 skb_probe_transport_header(skb, 0);
2605 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2606 int size_max, void **data)
2608 union tpacket_uhdr ph;
2613 switch (po->tp_version) {
2615 if (ph.h3->tp_next_offset != 0) {
2616 pr_warn_once("variable sized slot not supported");
2619 tp_len = ph.h3->tp_len;
2622 tp_len = ph.h2->tp_len;
2625 tp_len = ph.h1->tp_len;
2628 if (unlikely(tp_len > size_max)) {
2629 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2633 if (unlikely(po->tp_tx_has_off)) {
2634 int off_min, off_max;
2636 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2637 off_max = po->tx_ring.frame_size - tp_len;
2638 if (po->sk.sk_type == SOCK_DGRAM) {
2639 switch (po->tp_version) {
2641 off = ph.h3->tp_net;
2644 off = ph.h2->tp_net;
2647 off = ph.h1->tp_net;
2651 switch (po->tp_version) {
2653 off = ph.h3->tp_mac;
2656 off = ph.h2->tp_mac;
2659 off = ph.h1->tp_mac;
2663 if (unlikely((off < off_min) || (off_max < off)))
2666 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2669 *data = frame + off;
2673 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2675 struct sk_buff *skb = NULL;
2676 struct net_device *dev;
2677 struct virtio_net_hdr *vnet_hdr = NULL;
2678 struct sockcm_cookie sockc;
2680 int err, reserve = 0;
2682 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2683 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2684 unsigned char *addr = NULL;
2685 int tp_len, size_max;
2688 int status = TP_STATUS_AVAILABLE;
2689 int hlen, tlen, copylen = 0;
2692 mutex_lock(&po->pg_vec_lock);
2694 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2695 * we need to confirm it under protection of pg_vec_lock.
2697 if (unlikely(!po->tx_ring.pg_vec)) {
2701 if (likely(saddr == NULL)) {
2702 dev = packet_cached_dev_get(po);
2703 proto = READ_ONCE(po->num);
2706 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2708 if (msg->msg_namelen < (saddr->sll_halen
2709 + offsetof(struct sockaddr_ll,
2712 proto = saddr->sll_protocol;
2713 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2714 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2715 if (dev && msg->msg_namelen < dev->addr_len +
2716 offsetof(struct sockaddr_ll, sll_addr))
2718 addr = saddr->sll_addr;
2723 if (unlikely(dev == NULL))
2726 if (unlikely(!(dev->flags & IFF_UP)))
2729 sockc.tsflags = po->sk.sk_tsflags;
2730 if (msg->msg_controllen) {
2731 err = sock_cmsg_send(&po->sk, msg, &sockc);
2736 if (po->sk.sk_socket->type == SOCK_RAW)
2737 reserve = dev->hard_header_len;
2738 size_max = po->tx_ring.frame_size
2739 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2741 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2742 size_max = dev->mtu + reserve + VLAN_HLEN;
2744 reinit_completion(&po->skb_completion);
2747 ph = packet_current_frame(po, &po->tx_ring,
2748 TP_STATUS_SEND_REQUEST);
2749 if (unlikely(ph == NULL)) {
2750 if (need_wait && skb) {
2751 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2752 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2754 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2758 /* check for additional frames */
2763 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2767 status = TP_STATUS_SEND_REQUEST;
2768 hlen = LL_RESERVED_SPACE(dev);
2769 tlen = dev->needed_tailroom;
2770 if (po->has_vnet_hdr) {
2772 data += sizeof(*vnet_hdr);
2773 tp_len -= sizeof(*vnet_hdr);
2775 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2779 copylen = __virtio16_to_cpu(vio_le(),
2782 copylen = max_t(int, copylen, dev->hard_header_len);
2783 skb = sock_alloc_send_skb(&po->sk,
2784 hlen + tlen + sizeof(struct sockaddr_ll) +
2785 (copylen - dev->hard_header_len),
2788 if (unlikely(skb == NULL)) {
2789 /* we assume the socket was initially writeable ... */
2790 if (likely(len_sum > 0))
2794 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2795 addr, hlen, copylen, &sockc);
2796 if (likely(tp_len >= 0) &&
2797 tp_len > dev->mtu + reserve &&
2798 !po->has_vnet_hdr &&
2799 !packet_extra_vlan_len_allowed(dev, skb))
2802 if (unlikely(tp_len < 0)) {
2805 __packet_set_status(po, ph,
2806 TP_STATUS_AVAILABLE);
2807 packet_increment_head(&po->tx_ring);
2811 status = TP_STATUS_WRONG_FORMAT;
2817 if (po->has_vnet_hdr) {
2818 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2822 virtio_net_hdr_set_proto(skb, vnet_hdr);
2825 skb->destructor = tpacket_destruct_skb;
2826 __packet_set_status(po, ph, TP_STATUS_SENDING);
2827 packet_inc_pending(&po->tx_ring);
2829 status = TP_STATUS_SEND_REQUEST;
2830 err = po->xmit(skb);
2831 if (unlikely(err != 0)) {
2833 err = net_xmit_errno(err);
2834 if (err && __packet_get_status(po, ph) ==
2835 TP_STATUS_AVAILABLE) {
2836 /* skb was destructed already */
2841 * skb was dropped but not destructed yet;
2842 * let's treat it like congestion or err < 0
2846 packet_increment_head(&po->tx_ring);
2848 } while (likely((ph != NULL) ||
2849 /* Note: packet_read_pending() might be slow if we have
2850 * to call it as it's per_cpu variable, but in fast-path
2851 * we already short-circuit the loop with the first
2852 * condition, and luckily don't have to go that path
2855 (need_wait && packet_read_pending(&po->tx_ring))));
2861 __packet_set_status(po, ph, status);
2866 mutex_unlock(&po->pg_vec_lock);
2870 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2871 size_t reserve, size_t len,
2872 size_t linear, int noblock,
2875 struct sk_buff *skb;
2877 /* Under a page? Don't bother with paged skb. */
2878 if (prepad + len < PAGE_SIZE || !linear)
2881 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2886 skb_reserve(skb, reserve);
2887 skb_put(skb, linear);
2888 skb->data_len = len - linear;
2889 skb->len += len - linear;
2894 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2896 struct sock *sk = sock->sk;
2897 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2898 struct sk_buff *skb;
2899 struct net_device *dev;
2901 unsigned char *addr = NULL;
2902 int err, reserve = 0;
2903 struct sockcm_cookie sockc;
2904 struct virtio_net_hdr vnet_hdr = { 0 };
2906 struct packet_sock *po = pkt_sk(sk);
2907 bool has_vnet_hdr = false;
2908 int hlen, tlen, linear;
2912 * Get and verify the address.
2915 if (likely(saddr == NULL)) {
2916 dev = packet_cached_dev_get(po);
2917 proto = READ_ONCE(po->num);
2920 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2922 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2924 proto = saddr->sll_protocol;
2925 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2926 if (sock->type == SOCK_DGRAM) {
2927 if (dev && msg->msg_namelen < dev->addr_len +
2928 offsetof(struct sockaddr_ll, sll_addr))
2930 addr = saddr->sll_addr;
2935 if (unlikely(dev == NULL))
2938 if (unlikely(!(dev->flags & IFF_UP)))
2941 sockc.tsflags = sk->sk_tsflags;
2942 sockc.mark = sk->sk_mark;
2943 if (msg->msg_controllen) {
2944 err = sock_cmsg_send(sk, msg, &sockc);
2949 if (sock->type == SOCK_RAW)
2950 reserve = dev->hard_header_len;
2951 if (po->has_vnet_hdr) {
2952 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2955 has_vnet_hdr = true;
2958 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2959 if (!netif_supports_nofcs(dev)) {
2960 err = -EPROTONOSUPPORT;
2963 extra_len = 4; /* We're doing our own CRC */
2967 if (!vnet_hdr.gso_type &&
2968 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2972 hlen = LL_RESERVED_SPACE(dev);
2973 tlen = dev->needed_tailroom;
2974 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2975 linear = max(linear, min_t(int, len, dev->hard_header_len));
2976 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2977 msg->msg_flags & MSG_DONTWAIT, &err);
2981 skb_reset_network_header(skb);
2984 if (sock->type == SOCK_DGRAM) {
2985 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2986 if (unlikely(offset < 0))
2988 } else if (reserve) {
2989 skb_reserve(skb, -reserve);
2991 skb_reset_network_header(skb);
2994 /* Returns -EFAULT on error */
2995 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2999 if (sock->type == SOCK_RAW &&
3000 !dev_validate_header(dev, skb->data, len)) {
3005 skb_setup_tx_timestamp(skb, sockc.tsflags);
3007 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3008 !packet_extra_vlan_len_allowed(dev, skb)) {
3013 skb->protocol = proto;
3015 skb->priority = sk->sk_priority;
3016 skb->mark = sockc.mark;
3019 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3022 len += sizeof(vnet_hdr);
3023 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3026 skb_probe_transport_header(skb, reserve);
3028 if (unlikely(extra_len == 4))
3031 err = po->xmit(skb);
3032 if (unlikely(err != 0)) {
3034 err = net_xmit_errno(err);
3052 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3054 struct sock *sk = sock->sk;
3055 struct packet_sock *po = pkt_sk(sk);
3057 if (po->tx_ring.pg_vec)
3058 return tpacket_snd(po, msg);
3060 return packet_snd(sock, msg, len);
3064 * Close a PACKET socket. This is fairly simple. We immediately go
3065 * to 'closed' state and remove our protocol entry in the device list.
3068 static int packet_release(struct socket *sock)
3070 struct sock *sk = sock->sk;
3071 struct packet_sock *po;
3072 struct packet_fanout *f;
3074 union tpacket_req_u req_u;
3082 mutex_lock(&net->packet.sklist_lock);
3083 sk_del_node_init_rcu(sk);
3084 mutex_unlock(&net->packet.sklist_lock);
3087 sock_prot_inuse_add(net, sk->sk_prot, -1);
3090 spin_lock(&po->bind_lock);
3091 unregister_prot_hook(sk, false);
3092 packet_cached_dev_reset(po);
3094 if (po->prot_hook.dev) {
3095 dev_put(po->prot_hook.dev);
3096 po->prot_hook.dev = NULL;
3098 spin_unlock(&po->bind_lock);
3100 packet_flush_mclist(sk);
3103 if (po->rx_ring.pg_vec) {
3104 memset(&req_u, 0, sizeof(req_u));
3105 packet_set_ring(sk, &req_u, 1, 0);
3108 if (po->tx_ring.pg_vec) {
3109 memset(&req_u, 0, sizeof(req_u));
3110 packet_set_ring(sk, &req_u, 1, 1);
3114 f = fanout_release(sk);
3119 kfree(po->rollover);
3120 fanout_release_data(f);
3124 * Now the socket is dead. No more input will appear.
3131 skb_queue_purge(&sk->sk_receive_queue);
3132 packet_free_pending(po);
3133 sk_refcnt_debug_release(sk);
3140 * Attach a packet hook.
3143 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3146 struct packet_sock *po = pkt_sk(sk);
3147 struct net_device *dev_curr;
3150 struct net_device *dev = NULL;
3152 bool unlisted = false;
3155 spin_lock(&po->bind_lock);
3164 dev = dev_get_by_name_rcu(sock_net(sk), name);
3169 } else if (ifindex) {
3170 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3180 proto_curr = po->prot_hook.type;
3181 dev_curr = po->prot_hook.dev;
3183 need_rehook = proto_curr != proto || dev_curr != dev;
3188 /* prevents packet_notifier() from calling
3189 * register_prot_hook()
3191 WRITE_ONCE(po->num, 0);
3192 __unregister_prot_hook(sk, true);
3194 dev_curr = po->prot_hook.dev;
3196 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3200 BUG_ON(po->running);
3201 WRITE_ONCE(po->num, proto);
3202 po->prot_hook.type = proto;
3204 if (unlikely(unlisted)) {
3206 po->prot_hook.dev = NULL;
3207 WRITE_ONCE(po->ifindex, -1);
3208 packet_cached_dev_reset(po);
3210 po->prot_hook.dev = dev;
3211 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3212 packet_cached_dev_assign(po, dev);
3218 if (proto == 0 || !need_rehook)
3221 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3222 register_prot_hook(sk);
3224 sk->sk_err = ENETDOWN;
3225 if (!sock_flag(sk, SOCK_DEAD))
3226 sk->sk_error_report(sk);
3231 spin_unlock(&po->bind_lock);
3237 * Bind a packet socket to a device
3240 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3243 struct sock *sk = sock->sk;
3244 char name[sizeof(uaddr->sa_data) + 1];
3250 if (addr_len != sizeof(struct sockaddr))
3252 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3255 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3256 name[sizeof(uaddr->sa_data)] = 0;
3258 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3261 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3263 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3264 struct sock *sk = sock->sk;
3270 if (addr_len < sizeof(struct sockaddr_ll))
3272 if (sll->sll_family != AF_PACKET)
3275 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3276 sll->sll_protocol ? : pkt_sk(sk)->num);
3279 static struct proto packet_proto = {
3281 .owner = THIS_MODULE,
3282 .obj_size = sizeof(struct packet_sock),
3286 * Create a packet of type SOCK_PACKET.
3289 static int packet_create(struct net *net, struct socket *sock, int protocol,
3293 struct packet_sock *po;
3294 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3297 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3299 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3300 sock->type != SOCK_PACKET)
3301 return -ESOCKTNOSUPPORT;
3303 sock->state = SS_UNCONNECTED;
3306 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3310 sock->ops = &packet_ops;
3311 if (sock->type == SOCK_PACKET)
3312 sock->ops = &packet_ops_spkt;
3314 sock_init_data(sock, sk);
3317 init_completion(&po->skb_completion);
3318 sk->sk_family = PF_PACKET;
3320 po->xmit = dev_queue_xmit;
3322 err = packet_alloc_pending(po);
3326 packet_cached_dev_reset(po);
3328 sk->sk_destruct = packet_sock_destruct;
3329 sk_refcnt_debug_inc(sk);
3332 * Attach a protocol block
3335 spin_lock_init(&po->bind_lock);
3336 mutex_init(&po->pg_vec_lock);
3337 po->rollover = NULL;
3338 po->prot_hook.func = packet_rcv;
3340 if (sock->type == SOCK_PACKET)
3341 po->prot_hook.func = packet_rcv_spkt;
3343 po->prot_hook.af_packet_priv = sk;
3344 po->prot_hook.af_packet_net = sock_net(sk);
3347 po->prot_hook.type = proto;
3348 __register_prot_hook(sk);
3351 mutex_lock(&net->packet.sklist_lock);
3352 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3353 mutex_unlock(&net->packet.sklist_lock);
3356 sock_prot_inuse_add(net, &packet_proto, 1);
3367 * Pull a packet from our receive queue and hand it to the user.
3368 * If necessary we block.
3371 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3374 struct sock *sk = sock->sk;
3375 struct sk_buff *skb;
3377 int vnet_hdr_len = 0;
3378 unsigned int origlen = 0;
3381 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3385 /* What error should we return now? EUNATTACH? */
3386 if (pkt_sk(sk)->ifindex < 0)
3390 if (flags & MSG_ERRQUEUE) {
3391 err = sock_recv_errqueue(sk, msg, len,
3392 SOL_PACKET, PACKET_TX_TIMESTAMP);
3397 * Call the generic datagram receiver. This handles all sorts
3398 * of horrible races and re-entrancy so we can forget about it
3399 * in the protocol layers.
3401 * Now it will return ENETDOWN, if device have just gone down,
3402 * but then it will block.
3405 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3408 * An error occurred so return it. Because skb_recv_datagram()
3409 * handles the blocking we don't see and worry about blocking
3416 if (pkt_sk(sk)->pressure)
3417 packet_rcv_has_room(pkt_sk(sk), NULL);
3419 if (pkt_sk(sk)->has_vnet_hdr) {
3420 err = packet_rcv_vnet(msg, skb, &len);
3423 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3426 /* You lose any data beyond the buffer you gave. If it worries
3427 * a user program they can ask the device for its MTU
3433 msg->msg_flags |= MSG_TRUNC;
3436 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3440 if (sock->type != SOCK_PACKET) {
3441 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3443 /* Original length was stored in sockaddr_ll fields */
3444 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3445 sll->sll_family = AF_PACKET;
3446 sll->sll_protocol = skb->protocol;
3449 sock_recv_ts_and_drops(msg, sk, skb);
3451 if (msg->msg_name) {
3452 const size_t max_len = min(sizeof(skb->cb),
3453 sizeof(struct sockaddr_storage));
3456 /* If the address length field is there to be filled
3457 * in, we fill it in now.
3459 if (sock->type == SOCK_PACKET) {
3460 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3461 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3462 copy_len = msg->msg_namelen;
3464 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3466 msg->msg_namelen = sll->sll_halen +
3467 offsetof(struct sockaddr_ll, sll_addr);
3468 copy_len = msg->msg_namelen;
3469 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3470 memset(msg->msg_name +
3471 offsetof(struct sockaddr_ll, sll_addr),
3472 0, sizeof(sll->sll_addr));
3473 msg->msg_namelen = sizeof(struct sockaddr_ll);
3476 if (WARN_ON_ONCE(copy_len > max_len)) {
3478 msg->msg_namelen = copy_len;
3480 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3483 if (pkt_sk(sk)->auxdata) {
3484 struct tpacket_auxdata aux;
3486 aux.tp_status = TP_STATUS_USER;
3487 if (skb->ip_summed == CHECKSUM_PARTIAL)
3488 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3489 else if (skb->pkt_type != PACKET_OUTGOING &&
3490 skb_csum_unnecessary(skb))
3491 aux.tp_status |= TP_STATUS_CSUM_VALID;
3493 aux.tp_len = origlen;
3494 aux.tp_snaplen = skb->len;
3496 aux.tp_net = skb_network_offset(skb);
3497 if (skb_vlan_tag_present(skb)) {
3498 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3499 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3500 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3502 aux.tp_vlan_tci = 0;
3503 aux.tp_vlan_tpid = 0;
3505 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3509 * Free or return the buffer as appropriate. Again this
3510 * hides all the races and re-entrancy issues from us.
3512 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3515 skb_free_datagram(sk, skb);
3520 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3521 int *uaddr_len, int peer)
3523 struct net_device *dev;
3524 struct sock *sk = sock->sk;
3529 uaddr->sa_family = AF_PACKET;
3530 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3532 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3534 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3536 *uaddr_len = sizeof(*uaddr);
3541 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3542 int *uaddr_len, int peer)
3544 struct net_device *dev;
3545 struct sock *sk = sock->sk;
3546 struct packet_sock *po = pkt_sk(sk);
3547 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3553 ifindex = READ_ONCE(po->ifindex);
3554 sll->sll_family = AF_PACKET;
3555 sll->sll_ifindex = ifindex;
3556 sll->sll_protocol = READ_ONCE(po->num);
3557 sll->sll_pkttype = 0;
3559 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3561 sll->sll_hatype = dev->type;
3562 sll->sll_halen = dev->addr_len;
3563 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3565 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3569 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3574 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3578 case PACKET_MR_MULTICAST:
3579 if (i->alen != dev->addr_len)
3582 return dev_mc_add(dev, i->addr);
3584 return dev_mc_del(dev, i->addr);
3586 case PACKET_MR_PROMISC:
3587 return dev_set_promiscuity(dev, what);
3588 case PACKET_MR_ALLMULTI:
3589 return dev_set_allmulti(dev, what);
3590 case PACKET_MR_UNICAST:
3591 if (i->alen != dev->addr_len)
3594 return dev_uc_add(dev, i->addr);
3596 return dev_uc_del(dev, i->addr);
3604 static void packet_dev_mclist_delete(struct net_device *dev,
3605 struct packet_mclist **mlp)
3607 struct packet_mclist *ml;
3609 while ((ml = *mlp) != NULL) {
3610 if (ml->ifindex == dev->ifindex) {
3611 packet_dev_mc(dev, ml, -1);
3619 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3621 struct packet_sock *po = pkt_sk(sk);
3622 struct packet_mclist *ml, *i;
3623 struct net_device *dev;
3629 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3634 if (mreq->mr_alen > dev->addr_len)
3638 i = kmalloc(sizeof(*i), GFP_KERNEL);
3643 for (ml = po->mclist; ml; ml = ml->next) {
3644 if (ml->ifindex == mreq->mr_ifindex &&
3645 ml->type == mreq->mr_type &&
3646 ml->alen == mreq->mr_alen &&
3647 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3649 /* Free the new element ... */
3655 i->type = mreq->mr_type;
3656 i->ifindex = mreq->mr_ifindex;
3657 i->alen = mreq->mr_alen;
3658 memcpy(i->addr, mreq->mr_address, i->alen);
3659 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3661 i->next = po->mclist;
3663 err = packet_dev_mc(dev, i, 1);
3665 po->mclist = i->next;
3674 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3676 struct packet_mclist *ml, **mlp;
3680 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3681 if (ml->ifindex == mreq->mr_ifindex &&
3682 ml->type == mreq->mr_type &&
3683 ml->alen == mreq->mr_alen &&
3684 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3685 if (--ml->count == 0) {
3686 struct net_device *dev;
3688 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3690 packet_dev_mc(dev, ml, -1);
3700 static void packet_flush_mclist(struct sock *sk)
3702 struct packet_sock *po = pkt_sk(sk);
3703 struct packet_mclist *ml;
3709 while ((ml = po->mclist) != NULL) {
3710 struct net_device *dev;
3712 po->mclist = ml->next;
3713 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3715 packet_dev_mc(dev, ml, -1);
3722 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3724 struct sock *sk = sock->sk;
3725 struct packet_sock *po = pkt_sk(sk);
3728 if (level != SOL_PACKET)
3729 return -ENOPROTOOPT;
3732 case PACKET_ADD_MEMBERSHIP:
3733 case PACKET_DROP_MEMBERSHIP:
3735 struct packet_mreq_max mreq;
3737 memset(&mreq, 0, sizeof(mreq));
3738 if (len < sizeof(struct packet_mreq))
3740 if (len > sizeof(mreq))
3742 if (copy_from_user(&mreq, optval, len))
3744 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3746 if (optname == PACKET_ADD_MEMBERSHIP)
3747 ret = packet_mc_add(sk, &mreq);
3749 ret = packet_mc_drop(sk, &mreq);
3753 case PACKET_RX_RING:
3754 case PACKET_TX_RING:
3756 union tpacket_req_u req_u;
3760 switch (po->tp_version) {
3763 len = sizeof(req_u.req);
3767 len = sizeof(req_u.req3);
3773 if (copy_from_user(&req_u.req, optval, len))
3776 ret = packet_set_ring(sk, &req_u, 0,
3777 optname == PACKET_TX_RING);
3782 case PACKET_COPY_THRESH:
3786 if (optlen != sizeof(val))
3788 if (copy_from_user(&val, optval, sizeof(val)))
3791 pkt_sk(sk)->copy_thresh = val;
3794 case PACKET_VERSION:
3798 if (optlen != sizeof(val))
3800 if (copy_from_user(&val, optval, sizeof(val)))
3811 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3814 po->tp_version = val;
3820 case PACKET_RESERVE:
3824 if (optlen != sizeof(val))
3826 if (copy_from_user(&val, optval, sizeof(val)))
3831 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3834 po->tp_reserve = val;
3844 if (optlen != sizeof(val))
3846 if (copy_from_user(&val, optval, sizeof(val)))
3850 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3853 po->tp_loss = !!val;
3859 case PACKET_AUXDATA:
3863 if (optlen < sizeof(val))
3865 if (copy_from_user(&val, optval, sizeof(val)))
3869 po->auxdata = !!val;
3873 case PACKET_ORIGDEV:
3877 if (optlen < sizeof(val))
3879 if (copy_from_user(&val, optval, sizeof(val)))
3883 po->origdev = !!val;
3887 case PACKET_VNET_HDR:
3891 if (sock->type != SOCK_RAW)
3893 if (optlen < sizeof(val))
3895 if (copy_from_user(&val, optval, sizeof(val)))
3899 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3902 po->has_vnet_hdr = !!val;
3908 case PACKET_TIMESTAMP:
3912 if (optlen != sizeof(val))
3914 if (copy_from_user(&val, optval, sizeof(val)))
3917 po->tp_tstamp = val;
3924 if (optlen != sizeof(val))
3926 if (copy_from_user(&val, optval, sizeof(val)))
3929 return fanout_add(sk, val & 0xffff, val >> 16);
3931 case PACKET_FANOUT_DATA:
3933 /* Paired with the WRITE_ONCE() in fanout_add() */
3934 if (!READ_ONCE(po->fanout))
3937 return fanout_set_data(po, optval, optlen);
3939 case PACKET_TX_HAS_OFF:
3943 if (optlen != sizeof(val))
3945 if (copy_from_user(&val, optval, sizeof(val)))
3949 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3952 po->tp_tx_has_off = !!val;
3958 case PACKET_QDISC_BYPASS:
3962 if (optlen != sizeof(val))
3964 if (copy_from_user(&val, optval, sizeof(val)))
3967 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3971 return -ENOPROTOOPT;
3975 static int packet_getsockopt(struct socket *sock, int level, int optname,
3976 char __user *optval, int __user *optlen)
3979 int val, lv = sizeof(val);
3980 struct sock *sk = sock->sk;
3981 struct packet_sock *po = pkt_sk(sk);
3983 union tpacket_stats_u st;
3984 struct tpacket_rollover_stats rstats;
3986 if (level != SOL_PACKET)
3987 return -ENOPROTOOPT;
3989 if (get_user(len, optlen))
3996 case PACKET_STATISTICS:
3997 spin_lock_bh(&sk->sk_receive_queue.lock);
3998 memcpy(&st, &po->stats, sizeof(st));
3999 memset(&po->stats, 0, sizeof(po->stats));
4000 spin_unlock_bh(&sk->sk_receive_queue.lock);
4002 if (po->tp_version == TPACKET_V3) {
4003 lv = sizeof(struct tpacket_stats_v3);
4004 st.stats3.tp_packets += st.stats3.tp_drops;
4007 lv = sizeof(struct tpacket_stats);
4008 st.stats1.tp_packets += st.stats1.tp_drops;
4013 case PACKET_AUXDATA:
4016 case PACKET_ORIGDEV:
4019 case PACKET_VNET_HDR:
4020 val = po->has_vnet_hdr;
4022 case PACKET_VERSION:
4023 val = po->tp_version;
4026 if (len > sizeof(int))
4028 if (len < sizeof(int))
4030 if (copy_from_user(&val, optval, len))
4034 val = sizeof(struct tpacket_hdr);
4037 val = sizeof(struct tpacket2_hdr);
4040 val = sizeof(struct tpacket3_hdr);
4046 case PACKET_RESERVE:
4047 val = po->tp_reserve;
4052 case PACKET_TIMESTAMP:
4053 val = po->tp_tstamp;
4057 ((u32)po->fanout->id |
4058 ((u32)po->fanout->type << 16) |
4059 ((u32)po->fanout->flags << 24)) :
4062 case PACKET_ROLLOVER_STATS:
4065 rstats.tp_all = atomic_long_read(&po->rollover->num);
4066 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4067 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4069 lv = sizeof(rstats);
4071 case PACKET_TX_HAS_OFF:
4072 val = po->tp_tx_has_off;
4074 case PACKET_QDISC_BYPASS:
4075 val = packet_use_direct_xmit(po);
4078 return -ENOPROTOOPT;
4083 if (put_user(len, optlen))
4085 if (copy_to_user(optval, data, len))
4091 #ifdef CONFIG_COMPAT
4092 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4093 char __user *optval, unsigned int optlen)
4095 struct packet_sock *po = pkt_sk(sock->sk);
4097 if (level != SOL_PACKET)
4098 return -ENOPROTOOPT;
4100 if (optname == PACKET_FANOUT_DATA &&
4101 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4102 optval = (char __user *)get_compat_bpf_fprog(optval);
4105 optlen = sizeof(struct sock_fprog);
4108 return packet_setsockopt(sock, level, optname, optval, optlen);
4112 static int packet_notifier(struct notifier_block *this,
4113 unsigned long msg, void *ptr)
4116 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4117 struct net *net = dev_net(dev);
4120 sk_for_each_rcu(sk, &net->packet.sklist) {
4121 struct packet_sock *po = pkt_sk(sk);
4124 case NETDEV_UNREGISTER:
4126 packet_dev_mclist_delete(dev, &po->mclist);
4130 if (dev->ifindex == po->ifindex) {
4131 spin_lock(&po->bind_lock);
4133 __unregister_prot_hook(sk, false);
4134 sk->sk_err = ENETDOWN;
4135 if (!sock_flag(sk, SOCK_DEAD))
4136 sk->sk_error_report(sk);
4138 if (msg == NETDEV_UNREGISTER) {
4139 packet_cached_dev_reset(po);
4140 WRITE_ONCE(po->ifindex, -1);
4141 if (po->prot_hook.dev)
4142 dev_put(po->prot_hook.dev);
4143 po->prot_hook.dev = NULL;
4145 spin_unlock(&po->bind_lock);
4149 if (dev->ifindex == po->ifindex) {
4150 spin_lock(&po->bind_lock);
4152 register_prot_hook(sk);
4153 spin_unlock(&po->bind_lock);
4163 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4166 struct sock *sk = sock->sk;
4171 int amount = sk_wmem_alloc_get(sk);
4173 return put_user(amount, (int __user *)arg);
4177 struct sk_buff *skb;
4180 spin_lock_bh(&sk->sk_receive_queue.lock);
4181 skb = skb_peek(&sk->sk_receive_queue);
4184 spin_unlock_bh(&sk->sk_receive_queue.lock);
4185 return put_user(amount, (int __user *)arg);
4188 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4190 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4200 case SIOCGIFBRDADDR:
4201 case SIOCSIFBRDADDR:
4202 case SIOCGIFNETMASK:
4203 case SIOCSIFNETMASK:
4204 case SIOCGIFDSTADDR:
4205 case SIOCSIFDSTADDR:
4207 return inet_dgram_ops.ioctl(sock, cmd, arg);
4211 return -ENOIOCTLCMD;
4216 static unsigned int packet_poll(struct file *file, struct socket *sock,
4219 struct sock *sk = sock->sk;
4220 struct packet_sock *po = pkt_sk(sk);
4221 unsigned int mask = datagram_poll(file, sock, wait);
4223 spin_lock_bh(&sk->sk_receive_queue.lock);
4224 if (po->rx_ring.pg_vec) {
4225 if (!packet_previous_rx_frame(po, &po->rx_ring,
4227 mask |= POLLIN | POLLRDNORM;
4229 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4231 spin_unlock_bh(&sk->sk_receive_queue.lock);
4232 spin_lock_bh(&sk->sk_write_queue.lock);
4233 if (po->tx_ring.pg_vec) {
4234 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4235 mask |= POLLOUT | POLLWRNORM;
4237 spin_unlock_bh(&sk->sk_write_queue.lock);
4242 /* Dirty? Well, I still did not learn better way to account
4246 static void packet_mm_open(struct vm_area_struct *vma)
4248 struct file *file = vma->vm_file;
4249 struct socket *sock = file->private_data;
4250 struct sock *sk = sock->sk;
4253 atomic_inc(&pkt_sk(sk)->mapped);
4256 static void packet_mm_close(struct vm_area_struct *vma)
4258 struct file *file = vma->vm_file;
4259 struct socket *sock = file->private_data;
4260 struct sock *sk = sock->sk;
4263 atomic_dec(&pkt_sk(sk)->mapped);
4266 static const struct vm_operations_struct packet_mmap_ops = {
4267 .open = packet_mm_open,
4268 .close = packet_mm_close,
4271 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4276 for (i = 0; i < len; i++) {
4277 if (likely(pg_vec[i].buffer)) {
4278 if (is_vmalloc_addr(pg_vec[i].buffer))
4279 vfree(pg_vec[i].buffer);
4281 free_pages((unsigned long)pg_vec[i].buffer,
4283 pg_vec[i].buffer = NULL;
4289 static char *alloc_one_pg_vec_page(unsigned long order)
4292 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4293 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4295 buffer = (char *) __get_free_pages(gfp_flags, order);
4299 /* __get_free_pages failed, fall back to vmalloc */
4300 buffer = vzalloc((1 << order) * PAGE_SIZE);
4304 /* vmalloc failed, lets dig into swap here */
4305 gfp_flags &= ~__GFP_NORETRY;
4306 buffer = (char *) __get_free_pages(gfp_flags, order);
4310 /* complete and utter failure */
4314 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4316 unsigned int block_nr = req->tp_block_nr;
4320 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4321 if (unlikely(!pg_vec))
4324 for (i = 0; i < block_nr; i++) {
4325 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4326 if (unlikely(!pg_vec[i].buffer))
4327 goto out_free_pgvec;
4334 free_pg_vec(pg_vec, order, block_nr);
4339 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4340 int closing, int tx_ring)
4342 struct pgv *pg_vec = NULL;
4343 struct packet_sock *po = pkt_sk(sk);
4344 unsigned long *rx_owner_map = NULL;
4345 int was_running, order = 0;
4346 struct packet_ring_buffer *rb;
4347 struct sk_buff_head *rb_queue;
4350 /* Added to avoid minimal code churn */
4351 struct tpacket_req *req = &req_u->req;
4353 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4354 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4358 if (atomic_read(&po->mapped))
4360 if (packet_read_pending(rb))
4364 if (req->tp_block_nr) {
4365 unsigned int min_frame_size;
4367 /* Sanity tests and some calculations */
4369 if (unlikely(rb->pg_vec))
4372 switch (po->tp_version) {
4374 po->tp_hdrlen = TPACKET_HDRLEN;
4377 po->tp_hdrlen = TPACKET2_HDRLEN;
4380 po->tp_hdrlen = TPACKET3_HDRLEN;
4385 if (unlikely((int)req->tp_block_size <= 0))
4387 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4389 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4390 if (po->tp_version >= TPACKET_V3 &&
4391 req->tp_block_size <
4392 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4394 if (unlikely(req->tp_frame_size < min_frame_size))
4396 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4399 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4400 if (unlikely(rb->frames_per_block == 0))
4402 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4404 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4409 order = get_order(req->tp_block_size);
4410 pg_vec = alloc_pg_vec(req, order);
4411 if (unlikely(!pg_vec))
4413 switch (po->tp_version) {
4415 /* Block transmit is not supported yet */
4417 init_prb_bdqc(po, rb, pg_vec, req_u);
4419 struct tpacket_req3 *req3 = &req_u->req3;
4421 if (req3->tp_retire_blk_tov ||
4422 req3->tp_sizeof_priv ||
4423 req3->tp_feature_req_word) {
4425 goto out_free_pg_vec;
4431 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4432 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4434 goto out_free_pg_vec;
4442 if (unlikely(req->tp_frame_nr))
4447 /* Detach socket from network */
4448 spin_lock(&po->bind_lock);
4449 was_running = po->running;
4452 WRITE_ONCE(po->num, 0);
4453 __unregister_prot_hook(sk, false);
4455 spin_unlock(&po->bind_lock);
4460 mutex_lock(&po->pg_vec_lock);
4461 if (closing || atomic_read(&po->mapped) == 0) {
4463 spin_lock_bh(&rb_queue->lock);
4464 swap(rb->pg_vec, pg_vec);
4465 if (po->tp_version <= TPACKET_V2)
4466 swap(rb->rx_owner_map, rx_owner_map);
4467 rb->frame_max = (req->tp_frame_nr - 1);
4469 rb->frame_size = req->tp_frame_size;
4470 spin_unlock_bh(&rb_queue->lock);
4472 swap(rb->pg_vec_order, order);
4473 swap(rb->pg_vec_len, req->tp_block_nr);
4475 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4476 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4477 tpacket_rcv : packet_rcv;
4478 skb_queue_purge(rb_queue);
4479 if (atomic_read(&po->mapped))
4480 pr_err("packet_mmap: vma is busy: %d\n",
4481 atomic_read(&po->mapped));
4483 mutex_unlock(&po->pg_vec_lock);
4485 spin_lock(&po->bind_lock);
4487 WRITE_ONCE(po->num, num);
4488 register_prot_hook(sk);
4490 spin_unlock(&po->bind_lock);
4491 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4492 /* Because we don't support block-based V3 on tx-ring */
4494 prb_shutdown_retire_blk_timer(po, rb_queue);
4499 bitmap_free(rx_owner_map);
4500 free_pg_vec(pg_vec, order, req->tp_block_nr);
4506 static int packet_mmap(struct file *file, struct socket *sock,
4507 struct vm_area_struct *vma)
4509 struct sock *sk = sock->sk;
4510 struct packet_sock *po = pkt_sk(sk);
4511 unsigned long size, expected_size;
4512 struct packet_ring_buffer *rb;
4513 unsigned long start;
4520 mutex_lock(&po->pg_vec_lock);
4523 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4525 expected_size += rb->pg_vec_len
4531 if (expected_size == 0)
4534 size = vma->vm_end - vma->vm_start;
4535 if (size != expected_size)
4538 start = vma->vm_start;
4539 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4540 if (rb->pg_vec == NULL)
4543 for (i = 0; i < rb->pg_vec_len; i++) {
4545 void *kaddr = rb->pg_vec[i].buffer;
4548 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4549 page = pgv_to_page(kaddr);
4550 err = vm_insert_page(vma, start, page);
4559 atomic_inc(&po->mapped);
4560 vma->vm_ops = &packet_mmap_ops;
4564 mutex_unlock(&po->pg_vec_lock);
4568 static const struct proto_ops packet_ops_spkt = {
4569 .family = PF_PACKET,
4570 .owner = THIS_MODULE,
4571 .release = packet_release,
4572 .bind = packet_bind_spkt,
4573 .connect = sock_no_connect,
4574 .socketpair = sock_no_socketpair,
4575 .accept = sock_no_accept,
4576 .getname = packet_getname_spkt,
4577 .poll = datagram_poll,
4578 .ioctl = packet_ioctl,
4579 .listen = sock_no_listen,
4580 .shutdown = sock_no_shutdown,
4581 .setsockopt = sock_no_setsockopt,
4582 .getsockopt = sock_no_getsockopt,
4583 .sendmsg = packet_sendmsg_spkt,
4584 .recvmsg = packet_recvmsg,
4585 .mmap = sock_no_mmap,
4586 .sendpage = sock_no_sendpage,
4589 static const struct proto_ops packet_ops = {
4590 .family = PF_PACKET,
4591 .owner = THIS_MODULE,
4592 .release = packet_release,
4593 .bind = packet_bind,
4594 .connect = sock_no_connect,
4595 .socketpair = sock_no_socketpair,
4596 .accept = sock_no_accept,
4597 .getname = packet_getname,
4598 .poll = packet_poll,
4599 .ioctl = packet_ioctl,
4600 .listen = sock_no_listen,
4601 .shutdown = sock_no_shutdown,
4602 .setsockopt = packet_setsockopt,
4603 .getsockopt = packet_getsockopt,
4604 #ifdef CONFIG_COMPAT
4605 .compat_setsockopt = compat_packet_setsockopt,
4607 .sendmsg = packet_sendmsg,
4608 .recvmsg = packet_recvmsg,
4609 .mmap = packet_mmap,
4610 .sendpage = sock_no_sendpage,
4613 static const struct net_proto_family packet_family_ops = {
4614 .family = PF_PACKET,
4615 .create = packet_create,
4616 .owner = THIS_MODULE,
4619 static struct notifier_block packet_netdev_notifier = {
4620 .notifier_call = packet_notifier,
4623 #ifdef CONFIG_PROC_FS
4625 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4628 struct net *net = seq_file_net(seq);
4631 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4634 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4636 struct net *net = seq_file_net(seq);
4637 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4640 static void packet_seq_stop(struct seq_file *seq, void *v)
4646 static int packet_seq_show(struct seq_file *seq, void *v)
4648 if (v == SEQ_START_TOKEN)
4649 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4651 struct sock *s = sk_entry(v);
4652 const struct packet_sock *po = pkt_sk(s);
4655 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4657 refcount_read(&s->sk_refcnt),
4659 ntohs(READ_ONCE(po->num)),
4660 READ_ONCE(po->ifindex),
4662 atomic_read(&s->sk_rmem_alloc),
4663 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4670 static const struct seq_operations packet_seq_ops = {
4671 .start = packet_seq_start,
4672 .next = packet_seq_next,
4673 .stop = packet_seq_stop,
4674 .show = packet_seq_show,
4677 static int packet_seq_open(struct inode *inode, struct file *file)
4679 return seq_open_net(inode, file, &packet_seq_ops,
4680 sizeof(struct seq_net_private));
4683 static const struct file_operations packet_seq_fops = {
4684 .owner = THIS_MODULE,
4685 .open = packet_seq_open,
4687 .llseek = seq_lseek,
4688 .release = seq_release_net,
4693 static int __net_init packet_net_init(struct net *net)
4695 mutex_init(&net->packet.sklist_lock);
4696 INIT_HLIST_HEAD(&net->packet.sklist);
4698 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4704 static void __net_exit packet_net_exit(struct net *net)
4706 remove_proc_entry("packet", net->proc_net);
4709 static struct pernet_operations packet_net_ops = {
4710 .init = packet_net_init,
4711 .exit = packet_net_exit,
4715 static void __exit packet_exit(void)
4717 unregister_netdevice_notifier(&packet_netdev_notifier);
4718 unregister_pernet_subsys(&packet_net_ops);
4719 sock_unregister(PF_PACKET);
4720 proto_unregister(&packet_proto);
4723 static int __init packet_init(void)
4727 rc = proto_register(&packet_proto, 0);
4730 rc = sock_register(&packet_family_ops);
4733 rc = register_pernet_subsys(&packet_net_ops);
4736 rc = register_netdevice_notifier(&packet_netdev_notifier);
4743 unregister_pernet_subsys(&packet_net_ops);
4745 sock_unregister(PF_PACKET);
4747 proto_unregister(&packet_proto);
4752 module_init(packet_init);
4753 module_exit(packet_exit);
4754 MODULE_LICENSE("GPL");
4755 MODULE_ALIAS_NETPROTO(PF_PACKET);