2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 struct packet_type *pt, struct net_device *orig_dev);
195 static void *packet_previous_frame(struct packet_sock *po,
196 struct packet_ring_buffer *rb,
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211 struct tpacket_kbdq_core *,
212 void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217 struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
220 struct packet_skb_cb {
222 struct sockaddr_pkt pkt;
224 /* Trick: alias skb original length with
225 * ll.sll_family and ll.protocol in order
228 unsigned int origlen;
229 struct sockaddr_ll ll;
234 #define vio_le() virtio_legacy_is_little_endian()
236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid) \
240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245 ((x)->kactive_blk_num+1) : 0)
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
250 static int packet_direct_xmit(struct sk_buff *skb)
252 struct net_device *dev = skb->dev;
253 struct sk_buff *orig_skb = skb;
254 struct netdev_queue *txq;
255 int ret = NETDEV_TX_BUSY;
257 if (unlikely(!netif_running(dev) ||
258 !netif_carrier_ok(dev)))
261 skb = validate_xmit_skb_list(skb, dev);
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
369 __fanout_unlink(sk, po);
371 __dev_remove_pack(&po->prot_hook);
376 spin_unlock(&po->bind_lock);
378 spin_lock(&po->bind_lock);
382 static void unregister_prot_hook(struct sock *sk, bool sync)
384 struct packet_sock *po = pkt_sk(sk);
387 __unregister_prot_hook(sk, sync);
390 static inline struct page * __pure pgv_to_page(void *addr)
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
399 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
412 h.h3->tp_status = status;
413 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
416 WARN(1, "TPACKET version not supported.\n");
423 static int __packet_get_status(struct packet_sock *po, void *frame)
425 union tpacket_uhdr h;
430 switch (po->tp_version) {
432 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
433 return h.h1->tp_status;
435 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
436 return h.h2->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
439 return h.h3->tp_status;
441 WARN(1, "TPACKET version not supported.\n");
447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
450 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
453 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
454 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
455 return TP_STATUS_TS_RAW_HARDWARE;
457 if (ktime_to_timespec_cond(skb->tstamp, ts))
458 return TP_STATUS_TS_SOFTWARE;
463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
466 union tpacket_uhdr h;
470 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
474 switch (po->tp_version) {
476 h.h1->tp_sec = ts.tv_sec;
477 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
480 h.h2->tp_sec = ts.tv_sec;
481 h.h2->tp_nsec = ts.tv_nsec;
485 WARN(1, "TPACKET version not supported.\n");
489 /* one flush is safe, as both fields always lie on the same cacheline */
490 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
496 static void *packet_lookup_frame(struct packet_sock *po,
497 struct packet_ring_buffer *rb,
498 unsigned int position,
501 unsigned int pg_vec_pos, frame_offset;
502 union tpacket_uhdr h;
504 pg_vec_pos = position / rb->frames_per_block;
505 frame_offset = position % rb->frames_per_block;
507 h.raw = rb->pg_vec[pg_vec_pos].buffer +
508 (frame_offset * rb->frame_size);
510 if (status != __packet_get_status(po, h.raw))
516 static void *packet_current_frame(struct packet_sock *po,
517 struct packet_ring_buffer *rb,
520 return packet_lookup_frame(po, rb, rb->head, status);
523 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
525 del_timer_sync(&pkc->retire_blk_timer);
528 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
529 struct sk_buff_head *rb_queue)
531 struct tpacket_kbdq_core *pkc;
533 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
535 spin_lock_bh(&rb_queue->lock);
536 pkc->delete_blk_timer = 1;
537 spin_unlock_bh(&rb_queue->lock);
539 prb_del_retire_blk_timer(pkc);
542 static void prb_init_blk_timer(struct packet_sock *po,
543 struct tpacket_kbdq_core *pkc,
544 void (*func) (unsigned long))
546 init_timer(&pkc->retire_blk_timer);
547 pkc->retire_blk_timer.data = (long)po;
548 pkc->retire_blk_timer.function = func;
549 pkc->retire_blk_timer.expires = jiffies;
552 static void prb_setup_retire_blk_timer(struct packet_sock *po)
554 struct tpacket_kbdq_core *pkc;
556 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
557 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
560 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
561 int blk_size_in_bytes)
563 struct net_device *dev;
564 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
565 struct ethtool_link_ksettings ecmd;
569 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
570 if (unlikely(!dev)) {
572 return DEFAULT_PRB_RETIRE_TOV;
574 err = __ethtool_get_link_ksettings(dev, &ecmd);
578 * If the link speed is so slow you don't really
579 * need to worry about perf anyways
581 if (ecmd.base.speed < SPEED_1000 ||
582 ecmd.base.speed == SPEED_UNKNOWN) {
583 return DEFAULT_PRB_RETIRE_TOV;
586 div = ecmd.base.speed / 1000;
590 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
602 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
603 union tpacket_req_u *req_u)
605 p1->feature_req_word = req_u->req3.tp_feature_req_word;
608 static void init_prb_bdqc(struct packet_sock *po,
609 struct packet_ring_buffer *rb,
611 union tpacket_req_u *req_u)
613 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
614 struct tpacket_block_desc *pbd;
616 memset(p1, 0x0, sizeof(*p1));
618 p1->knxt_seq_num = 1;
620 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
621 p1->pkblk_start = pg_vec[0].buffer;
622 p1->kblk_size = req_u->req3.tp_block_size;
623 p1->knum_blocks = req_u->req3.tp_block_nr;
624 p1->hdrlen = po->tp_hdrlen;
625 p1->version = po->tp_version;
626 p1->last_kactive_blk_num = 0;
627 po->stats.stats3.tp_freeze_q_cnt = 0;
628 if (req_u->req3.tp_retire_blk_tov)
629 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
631 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
632 req_u->req3.tp_block_size);
633 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
634 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
636 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
637 prb_init_ft_ops(p1, req_u);
638 prb_setup_retire_blk_timer(po);
639 prb_open_block(p1, pbd);
642 /* Do NOT update the last_blk_num first.
643 * Assumes sk_buff_head lock is held.
645 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
647 mod_timer(&pkc->retire_blk_timer,
648 jiffies + pkc->tov_in_jiffies);
649 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
654 * 1) We refresh the timer only when we open a block.
655 * By doing this we don't waste cycles refreshing the timer
656 * on packet-by-packet basis.
658 * With a 1MB block-size, on a 1Gbps line, it will take
659 * i) ~8 ms to fill a block + ii) memcpy etc.
660 * In this cut we are not accounting for the memcpy time.
662 * So, if the user sets the 'tmo' to 10ms then the timer
663 * will never fire while the block is still getting filled
664 * (which is what we want). However, the user could choose
665 * to close a block early and that's fine.
667 * But when the timer does fire, we check whether or not to refresh it.
668 * Since the tmo granularity is in msecs, it is not too expensive
669 * to refresh the timer, lets say every '8' msecs.
670 * Either the user can set the 'tmo' or we can derive it based on
671 * a) line-speed and b) block-size.
672 * prb_calc_retire_blk_tmo() calculates the tmo.
675 static void prb_retire_rx_blk_timer_expired(unsigned long data)
677 struct packet_sock *po = (struct packet_sock *)data;
678 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
680 struct tpacket_block_desc *pbd;
682 spin_lock(&po->sk.sk_receive_queue.lock);
684 frozen = prb_queue_frozen(pkc);
685 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
687 if (unlikely(pkc->delete_blk_timer))
690 /* We only need to plug the race when the block is partially filled.
692 * lock(); increment BLOCK_NUM_PKTS; unlock()
693 * copy_bits() is in progress ...
694 * timer fires on other cpu:
695 * we can't retire the current block because copy_bits
699 if (BLOCK_NUM_PKTS(pbd)) {
700 while (atomic_read(&pkc->blk_fill_in_prog)) {
701 /* Waiting for skb_copy_bits to finish... */
706 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
708 if (!BLOCK_NUM_PKTS(pbd)) {
709 /* An empty block. Just refresh the timer. */
712 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
713 if (!prb_dispatch_next_block(pkc, po))
718 /* Case 1. Queue was frozen because user-space was
721 if (prb_curr_blk_in_use(pkc, pbd)) {
723 * Ok, user-space is still behind.
724 * So just refresh the timer.
728 /* Case 2. queue was frozen,user-space caught up,
729 * now the link went idle && the timer fired.
730 * We don't have a block to close.So we open this
731 * block and restart the timer.
732 * opening a block thaws the queue,restarts timer
733 * Thawing/timer-refresh is a side effect.
735 prb_open_block(pkc, pbd);
742 _prb_refresh_rx_retire_blk_timer(pkc);
745 spin_unlock(&po->sk.sk_receive_queue.lock);
748 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
749 struct tpacket_block_desc *pbd1, __u32 status)
751 /* Flush everything minus the block header */
753 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
758 /* Skip the block header(we know header WILL fit in 4K) */
761 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
762 for (; start < end; start += PAGE_SIZE)
763 flush_dcache_page(pgv_to_page(start));
768 /* Now update the block status. */
770 BLOCK_STATUS(pbd1) = status;
772 /* Flush the block header */
774 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
776 flush_dcache_page(pgv_to_page(start));
786 * 2) Increment active_blk_num
788 * Note:We DONT refresh the timer on purpose.
789 * Because almost always the next block will be opened.
791 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
792 struct tpacket_block_desc *pbd1,
793 struct packet_sock *po, unsigned int stat)
795 __u32 status = TP_STATUS_USER | stat;
797 struct tpacket3_hdr *last_pkt;
798 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
799 struct sock *sk = &po->sk;
801 if (po->stats.stats3.tp_drops)
802 status |= TP_STATUS_LOSING;
804 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
805 last_pkt->tp_next_offset = 0;
807 /* Get the ts of the last pkt */
808 if (BLOCK_NUM_PKTS(pbd1)) {
809 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
810 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
812 /* Ok, we tmo'd - so get the current time.
814 * It shouldn't really happen as we don't close empty
815 * blocks. See prb_retire_rx_blk_timer_expired().
819 h1->ts_last_pkt.ts_sec = ts.tv_sec;
820 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
825 /* Flush the block */
826 prb_flush_block(pkc1, pbd1, status);
828 sk->sk_data_ready(sk);
830 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
833 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
835 pkc->reset_pending_on_curr_blk = 0;
839 * Side effect of opening a block:
841 * 1) prb_queue is thawed.
842 * 2) retire_blk_timer is refreshed.
845 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
846 struct tpacket_block_desc *pbd1)
849 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
853 /* We could have just memset this but we will lose the
854 * flexibility of making the priv area sticky
857 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
858 BLOCK_NUM_PKTS(pbd1) = 0;
859 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863 h1->ts_first_pkt.ts_sec = ts.tv_sec;
864 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
866 pkc1->pkblk_start = (char *)pbd1;
867 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
870 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
872 pbd1->version = pkc1->version;
873 pkc1->prev = pkc1->nxt_offset;
874 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
876 prb_thaw_queue(pkc1);
877 _prb_refresh_rx_retire_blk_timer(pkc1);
883 * Queue freeze logic:
884 * 1) Assume tp_block_nr = 8 blocks.
885 * 2) At time 't0', user opens Rx ring.
886 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
887 * 4) user-space is either sleeping or processing block '0'.
888 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
889 * it will close block-7,loop around and try to fill block '0'.
891 * __packet_lookup_frame_in_block
892 * prb_retire_current_block()
893 * prb_dispatch_next_block()
894 * |->(BLOCK_STATUS == USER) evaluates to true
895 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
896 * 6) Now there are two cases:
897 * 6.1) Link goes idle right after the queue is frozen.
898 * But remember, the last open_block() refreshed the timer.
899 * When this timer expires,it will refresh itself so that we can
900 * re-open block-0 in near future.
901 * 6.2) Link is busy and keeps on receiving packets. This is a simple
902 * case and __packet_lookup_frame_in_block will check if block-0
903 * is free and can now be re-used.
905 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
906 struct packet_sock *po)
908 pkc->reset_pending_on_curr_blk = 1;
909 po->stats.stats3.tp_freeze_q_cnt++;
912 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
915 * If the next block is free then we will dispatch it
916 * and return a good offset.
917 * Else, we will freeze the queue.
918 * So, caller must check the return value.
920 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
921 struct packet_sock *po)
923 struct tpacket_block_desc *pbd;
927 /* 1. Get current block num */
928 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
930 /* 2. If this block is currently in_use then freeze the queue */
931 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
932 prb_freeze_queue(pkc, po);
938 * open this block and return the offset where the first packet
939 * needs to get stored.
941 prb_open_block(pkc, pbd);
942 return (void *)pkc->nxt_offset;
945 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
946 struct packet_sock *po, unsigned int status)
948 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
950 /* retire/close the current block */
951 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
953 * Plug the case where copy_bits() is in progress on
954 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
955 * have space to copy the pkt in the current block and
956 * called prb_retire_current_block()
958 * We don't need to worry about the TMO case because
959 * the timer-handler already handled this case.
961 if (!(status & TP_STATUS_BLK_TMO)) {
962 while (atomic_read(&pkc->blk_fill_in_prog)) {
963 /* Waiting for skb_copy_bits to finish... */
967 prb_close_block(pkc, pbd, po, status);
972 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
973 struct tpacket_block_desc *pbd)
975 return TP_STATUS_USER & BLOCK_STATUS(pbd);
978 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
980 return pkc->reset_pending_on_curr_blk;
983 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
985 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
986 atomic_dec(&pkc->blk_fill_in_prog);
989 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
990 struct tpacket3_hdr *ppd)
992 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
995 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
996 struct tpacket3_hdr *ppd)
998 ppd->hv1.tp_rxhash = 0;
1001 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1002 struct tpacket3_hdr *ppd)
1004 if (skb_vlan_tag_present(pkc->skb)) {
1005 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1006 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1007 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1009 ppd->hv1.tp_vlan_tci = 0;
1010 ppd->hv1.tp_vlan_tpid = 0;
1011 ppd->tp_status = TP_STATUS_AVAILABLE;
1015 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1016 struct tpacket3_hdr *ppd)
1018 ppd->hv1.tp_padding = 0;
1019 prb_fill_vlan_info(pkc, ppd);
1021 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1022 prb_fill_rxhash(pkc, ppd);
1024 prb_clear_rxhash(pkc, ppd);
1027 static void prb_fill_curr_block(char *curr,
1028 struct tpacket_kbdq_core *pkc,
1029 struct tpacket_block_desc *pbd,
1032 struct tpacket3_hdr *ppd;
1034 ppd = (struct tpacket3_hdr *)curr;
1035 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1037 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1039 BLOCK_NUM_PKTS(pbd) += 1;
1040 atomic_inc(&pkc->blk_fill_in_prog);
1041 prb_run_all_ft_ops(pkc, ppd);
1044 /* Assumes caller has the sk->rx_queue.lock */
1045 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1046 struct sk_buff *skb,
1051 struct tpacket_kbdq_core *pkc;
1052 struct tpacket_block_desc *pbd;
1055 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1056 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058 /* Queue is frozen when user space is lagging behind */
1059 if (prb_queue_frozen(pkc)) {
1061 * Check if that last block which caused the queue to freeze,
1062 * is still in_use by user-space.
1064 if (prb_curr_blk_in_use(pkc, pbd)) {
1065 /* Can't record this packet */
1069 * Ok, the block was released by user-space.
1070 * Now let's open that block.
1071 * opening a block also thaws the queue.
1072 * Thawing is a side effect.
1074 prb_open_block(pkc, pbd);
1079 curr = pkc->nxt_offset;
1081 end = (char *)pbd + pkc->kblk_size;
1083 /* first try the current block */
1084 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1085 prb_fill_curr_block(curr, pkc, pbd, len);
1086 return (void *)curr;
1089 /* Ok, close the current block */
1090 prb_retire_current_block(pkc, po, 0);
1092 /* Now, try to dispatch the next block */
1093 curr = (char *)prb_dispatch_next_block(pkc, po);
1095 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1096 prb_fill_curr_block(curr, pkc, pbd, len);
1097 return (void *)curr;
1101 * No free blocks are available.user_space hasn't caught up yet.
1102 * Queue was just frozen and now this packet will get dropped.
1107 static void *packet_current_rx_frame(struct packet_sock *po,
1108 struct sk_buff *skb,
1109 int status, unsigned int len)
1112 switch (po->tp_version) {
1115 curr = packet_lookup_frame(po, &po->rx_ring,
1116 po->rx_ring.head, status);
1119 return __packet_lookup_frame_in_block(po, skb, status, len);
1121 WARN(1, "TPACKET version not supported\n");
1127 static void *prb_lookup_block(struct packet_sock *po,
1128 struct packet_ring_buffer *rb,
1132 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1133 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1135 if (status != BLOCK_STATUS(pbd))
1140 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1143 if (rb->prb_bdqc.kactive_blk_num)
1144 prev = rb->prb_bdqc.kactive_blk_num-1;
1146 prev = rb->prb_bdqc.knum_blocks-1;
1150 /* Assumes caller has held the rx_queue.lock */
1151 static void *__prb_previous_block(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1155 unsigned int previous = prb_previous_blk_num(rb);
1156 return prb_lookup_block(po, rb, previous, status);
1159 static void *packet_previous_rx_frame(struct packet_sock *po,
1160 struct packet_ring_buffer *rb,
1163 if (po->tp_version <= TPACKET_V2)
1164 return packet_previous_frame(po, rb, status);
1166 return __prb_previous_block(po, rb, status);
1169 static void packet_increment_rx_head(struct packet_sock *po,
1170 struct packet_ring_buffer *rb)
1172 switch (po->tp_version) {
1175 return packet_increment_head(rb);
1178 WARN(1, "TPACKET version not supported.\n");
1184 static void *packet_previous_frame(struct packet_sock *po,
1185 struct packet_ring_buffer *rb,
1188 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1189 return packet_lookup_frame(po, rb, previous, status);
1192 static void packet_increment_head(struct packet_ring_buffer *buff)
1194 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1197 static void packet_inc_pending(struct packet_ring_buffer *rb)
1199 this_cpu_inc(*rb->pending_refcnt);
1202 static void packet_dec_pending(struct packet_ring_buffer *rb)
1204 this_cpu_dec(*rb->pending_refcnt);
1207 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1209 unsigned int refcnt = 0;
1212 /* We don't use pending refcount in rx_ring. */
1213 if (rb->pending_refcnt == NULL)
1216 for_each_possible_cpu(cpu)
1217 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1222 static int packet_alloc_pending(struct packet_sock *po)
1224 po->rx_ring.pending_refcnt = NULL;
1226 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1227 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1233 static void packet_free_pending(struct packet_sock *po)
1235 free_percpu(po->tx_ring.pending_refcnt);
1238 #define ROOM_POW_OFF 2
1239 #define ROOM_NONE 0x0
1240 #define ROOM_LOW 0x1
1241 #define ROOM_NORMAL 0x2
1243 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1247 len = po->rx_ring.frame_max + 1;
1248 idx = po->rx_ring.head;
1250 idx += len >> pow_off;
1253 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1256 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1260 len = po->rx_ring.prb_bdqc.knum_blocks;
1261 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1263 idx += len >> pow_off;
1266 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1269 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1271 struct sock *sk = &po->sk;
1272 int ret = ROOM_NONE;
1274 if (po->prot_hook.func != tpacket_rcv) {
1275 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1276 - (skb ? skb->truesize : 0);
1277 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1285 if (po->tp_version == TPACKET_V3) {
1286 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_v3_has_room(po, 0))
1291 if (__tpacket_has_room(po, ROOM_POW_OFF))
1293 else if (__tpacket_has_room(po, 0))
1300 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1305 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1306 ret = __packet_rcv_has_room(po, skb);
1307 has_room = ret == ROOM_NORMAL;
1308 if (po->pressure == has_room)
1309 po->pressure = !has_room;
1310 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1315 static void packet_sock_destruct(struct sock *sk)
1317 skb_queue_purge(&sk->sk_error_queue);
1319 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1320 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1322 if (!sock_flag(sk, SOCK_DEAD)) {
1323 pr_err("Attempt to release alive packet socket: %p\n", sk);
1327 sk_refcnt_debug_dec(sk);
1330 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1335 rxhash = skb_get_hash(skb);
1336 for (i = 0; i < ROLLOVER_HLEN; i++)
1337 if (po->rollover->history[i] == rxhash)
1340 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1341 return count > (ROLLOVER_HLEN >> 1);
1344 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1345 struct sk_buff *skb,
1348 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1351 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1352 struct sk_buff *skb,
1355 unsigned int val = atomic_inc_return(&f->rr_cur);
1360 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1361 struct sk_buff *skb,
1364 return smp_processor_id() % num;
1367 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1368 struct sk_buff *skb,
1371 return prandom_u32_max(num);
1374 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1375 struct sk_buff *skb,
1376 unsigned int idx, bool try_self,
1379 struct packet_sock *po, *po_next, *po_skip = NULL;
1380 unsigned int i, j, room = ROOM_NONE;
1382 po = pkt_sk(f->arr[idx]);
1385 room = packet_rcv_has_room(po, skb);
1386 if (room == ROOM_NORMAL ||
1387 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1392 i = j = min_t(int, po->rollover->sock, num - 1);
1394 po_next = pkt_sk(f->arr[i]);
1395 if (po_next != po_skip && !po_next->pressure &&
1396 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1398 po->rollover->sock = i;
1399 atomic_long_inc(&po->rollover->num);
1400 if (room == ROOM_LOW)
1401 atomic_long_inc(&po->rollover->num_huge);
1409 atomic_long_inc(&po->rollover->num_failed);
1413 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1414 struct sk_buff *skb,
1417 return skb_get_queue_mapping(skb) % num;
1420 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1421 struct sk_buff *skb,
1424 struct bpf_prog *prog;
1425 unsigned int ret = 0;
1428 prog = rcu_dereference(f->bpf_prog);
1430 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1436 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1438 return f->flags & (flag >> 8);
1441 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1442 struct packet_type *pt, struct net_device *orig_dev)
1444 struct packet_fanout *f = pt->af_packet_priv;
1445 unsigned int num = READ_ONCE(f->num_members);
1446 struct net *net = read_pnet(&f->net);
1447 struct packet_sock *po;
1450 if (!net_eq(dev_net(dev), net) || !num) {
1455 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1456 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1461 case PACKET_FANOUT_HASH:
1463 idx = fanout_demux_hash(f, skb, num);
1465 case PACKET_FANOUT_LB:
1466 idx = fanout_demux_lb(f, skb, num);
1468 case PACKET_FANOUT_CPU:
1469 idx = fanout_demux_cpu(f, skb, num);
1471 case PACKET_FANOUT_RND:
1472 idx = fanout_demux_rnd(f, skb, num);
1474 case PACKET_FANOUT_QM:
1475 idx = fanout_demux_qm(f, skb, num);
1477 case PACKET_FANOUT_ROLLOVER:
1478 idx = fanout_demux_rollover(f, skb, 0, false, num);
1480 case PACKET_FANOUT_CBPF:
1481 case PACKET_FANOUT_EBPF:
1482 idx = fanout_demux_bpf(f, skb, num);
1486 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1487 idx = fanout_demux_rollover(f, skb, idx, true, num);
1489 po = pkt_sk(f->arr[idx]);
1490 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1493 DEFINE_MUTEX(fanout_mutex);
1494 EXPORT_SYMBOL_GPL(fanout_mutex);
1495 static LIST_HEAD(fanout_list);
1497 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1499 struct packet_fanout *f = po->fanout;
1501 spin_lock(&f->lock);
1502 f->arr[f->num_members] = sk;
1505 spin_unlock(&f->lock);
1508 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1510 struct packet_fanout *f = po->fanout;
1513 spin_lock(&f->lock);
1514 for (i = 0; i < f->num_members; i++) {
1515 if (f->arr[i] == sk)
1518 BUG_ON(i >= f->num_members);
1519 f->arr[i] = f->arr[f->num_members - 1];
1521 spin_unlock(&f->lock);
1524 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1526 if (sk->sk_family != PF_PACKET)
1529 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1532 static void fanout_init_data(struct packet_fanout *f)
1535 case PACKET_FANOUT_LB:
1536 atomic_set(&f->rr_cur, 0);
1538 case PACKET_FANOUT_CBPF:
1539 case PACKET_FANOUT_EBPF:
1540 RCU_INIT_POINTER(f->bpf_prog, NULL);
1545 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1547 struct bpf_prog *old;
1549 spin_lock(&f->lock);
1550 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1551 rcu_assign_pointer(f->bpf_prog, new);
1552 spin_unlock(&f->lock);
1556 bpf_prog_destroy(old);
1560 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1563 struct bpf_prog *new;
1564 struct sock_fprog fprog;
1567 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1569 if (len != sizeof(fprog))
1571 if (copy_from_user(&fprog, data, len))
1574 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1578 __fanout_set_data_bpf(po->fanout, new);
1582 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1585 struct bpf_prog *new;
1588 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1590 if (len != sizeof(fd))
1592 if (copy_from_user(&fd, data, len))
1595 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1597 return PTR_ERR(new);
1599 __fanout_set_data_bpf(po->fanout, new);
1603 static int fanout_set_data(struct packet_sock *po, char __user *data,
1606 switch (po->fanout->type) {
1607 case PACKET_FANOUT_CBPF:
1608 return fanout_set_data_cbpf(po, data, len);
1609 case PACKET_FANOUT_EBPF:
1610 return fanout_set_data_ebpf(po, data, len);
1616 static void fanout_release_data(struct packet_fanout *f)
1619 case PACKET_FANOUT_CBPF:
1620 case PACKET_FANOUT_EBPF:
1621 __fanout_set_data_bpf(f, NULL);
1625 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1627 struct packet_sock *po = pkt_sk(sk);
1628 struct packet_fanout *f, *match;
1629 u8 type = type_flags & 0xff;
1630 u8 flags = type_flags >> 8;
1634 case PACKET_FANOUT_ROLLOVER:
1635 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1637 case PACKET_FANOUT_HASH:
1638 case PACKET_FANOUT_LB:
1639 case PACKET_FANOUT_CPU:
1640 case PACKET_FANOUT_RND:
1641 case PACKET_FANOUT_QM:
1642 case PACKET_FANOUT_CBPF:
1643 case PACKET_FANOUT_EBPF:
1655 if (type == PACKET_FANOUT_ROLLOVER ||
1656 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1657 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1660 atomic_long_set(&po->rollover->num, 0);
1661 atomic_long_set(&po->rollover->num_huge, 0);
1662 atomic_long_set(&po->rollover->num_failed, 0);
1665 mutex_lock(&fanout_mutex);
1667 list_for_each_entry(f, &fanout_list, list) {
1669 read_pnet(&f->net) == sock_net(sk)) {
1675 if (match && match->flags != flags)
1679 match = kzalloc(sizeof(*match), GFP_KERNEL);
1682 write_pnet(&match->net, sock_net(sk));
1685 match->flags = flags;
1686 INIT_LIST_HEAD(&match->list);
1687 spin_lock_init(&match->lock);
1688 atomic_set(&match->sk_ref, 0);
1689 fanout_init_data(match);
1690 match->prot_hook.type = po->prot_hook.type;
1691 match->prot_hook.dev = po->prot_hook.dev;
1692 match->prot_hook.func = packet_rcv_fanout;
1693 match->prot_hook.af_packet_priv = match;
1694 match->prot_hook.id_match = match_fanout_group;
1695 dev_add_pack(&match->prot_hook);
1696 list_add(&match->list, &fanout_list);
1699 if (match->type == type &&
1700 match->prot_hook.type == po->prot_hook.type &&
1701 match->prot_hook.dev == po->prot_hook.dev) {
1703 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1704 __dev_remove_pack(&po->prot_hook);
1706 atomic_inc(&match->sk_ref);
1707 __fanout_link(sk, po);
1712 mutex_unlock(&fanout_mutex);
1714 kfree(po->rollover);
1715 po->rollover = NULL;
1720 static void fanout_release(struct sock *sk)
1722 struct packet_sock *po = pkt_sk(sk);
1723 struct packet_fanout *f;
1729 mutex_lock(&fanout_mutex);
1732 if (atomic_dec_and_test(&f->sk_ref)) {
1734 dev_remove_pack(&f->prot_hook);
1735 fanout_release_data(f);
1738 mutex_unlock(&fanout_mutex);
1741 kfree_rcu(po->rollover, rcu);
1744 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1745 struct sk_buff *skb)
1747 /* Earlier code assumed this would be a VLAN pkt, double-check
1748 * this now that we have the actual packet in hand. We can only
1749 * do this check on Ethernet devices.
1751 if (unlikely(dev->type != ARPHRD_ETHER))
1754 skb_reset_mac_header(skb);
1755 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1758 static const struct proto_ops packet_ops;
1760 static const struct proto_ops packet_ops_spkt;
1762 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1763 struct packet_type *pt, struct net_device *orig_dev)
1766 struct sockaddr_pkt *spkt;
1769 * When we registered the protocol we saved the socket in the data
1770 * field for just this event.
1773 sk = pt->af_packet_priv;
1776 * Yank back the headers [hope the device set this
1777 * right or kerboom...]
1779 * Incoming packets have ll header pulled,
1782 * For outgoing ones skb->data == skb_mac_header(skb)
1783 * so that this procedure is noop.
1786 if (skb->pkt_type == PACKET_LOOPBACK)
1789 if (!net_eq(dev_net(dev), sock_net(sk)))
1792 skb = skb_share_check(skb, GFP_ATOMIC);
1796 /* drop any routing info */
1799 /* drop conntrack reference */
1802 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1804 skb_push(skb, skb->data - skb_mac_header(skb));
1807 * The SOCK_PACKET socket receives _all_ frames.
1810 spkt->spkt_family = dev->type;
1811 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1812 spkt->spkt_protocol = skb->protocol;
1815 * Charge the memory to the socket. This is done specifically
1816 * to prevent sockets using all the memory up.
1819 if (sock_queue_rcv_skb(sk, skb) == 0)
1830 * Output a raw packet to a device layer. This bypasses all the other
1831 * protocol layers and you must therefore supply it with a complete frame
1834 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1837 struct sock *sk = sock->sk;
1838 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1839 struct sk_buff *skb = NULL;
1840 struct net_device *dev;
1841 struct sockcm_cookie sockc;
1847 * Get and verify the address.
1851 if (msg->msg_namelen < sizeof(struct sockaddr))
1853 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1854 proto = saddr->spkt_protocol;
1856 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1859 * Find the device first to size check it
1862 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1865 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1871 if (!(dev->flags & IFF_UP))
1875 * You may not queue a frame bigger than the mtu. This is the lowest level
1876 * raw protocol and you must do your own fragmentation at this level.
1879 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1880 if (!netif_supports_nofcs(dev)) {
1881 err = -EPROTONOSUPPORT;
1884 extra_len = 4; /* We're doing our own CRC */
1888 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1892 size_t reserved = LL_RESERVED_SPACE(dev);
1893 int tlen = dev->needed_tailroom;
1894 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1897 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1900 /* FIXME: Save some space for broken drivers that write a hard
1901 * header at transmission time by themselves. PPP is the notable
1902 * one here. This should really be fixed at the driver level.
1904 skb_reserve(skb, reserved);
1905 skb_reset_network_header(skb);
1907 /* Try to align data part correctly */
1912 skb_reset_network_header(skb);
1914 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1920 if (!dev_validate_header(dev, skb->data, len)) {
1924 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1925 !packet_extra_vlan_len_allowed(dev, skb)) {
1930 sockc.tsflags = sk->sk_tsflags;
1931 if (msg->msg_controllen) {
1932 err = sock_cmsg_send(sk, msg, &sockc);
1937 skb->protocol = proto;
1939 skb->priority = sk->sk_priority;
1940 skb->mark = sk->sk_mark;
1942 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1944 if (unlikely(extra_len == 4))
1947 skb_probe_transport_header(skb, 0);
1949 dev_queue_xmit(skb);
1960 static unsigned int run_filter(struct sk_buff *skb,
1961 const struct sock *sk,
1964 struct sk_filter *filter;
1967 filter = rcu_dereference(sk->sk_filter);
1969 res = bpf_prog_run_clear_cb(filter->prog, skb);
1975 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1978 struct virtio_net_hdr vnet_hdr;
1980 if (*len < sizeof(vnet_hdr))
1982 *len -= sizeof(vnet_hdr);
1984 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le()))
1987 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
1991 * This function makes lazy skb cloning in hope that most of packets
1992 * are discarded by BPF.
1994 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1995 * and skb->cb are mangled. It works because (and until) packets
1996 * falling here are owned by current CPU. Output packets are cloned
1997 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1998 * sequencially, so that if we return skb to original state on exit,
1999 * we will not harm anyone.
2002 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2003 struct packet_type *pt, struct net_device *orig_dev)
2006 struct sockaddr_ll *sll;
2007 struct packet_sock *po;
2008 u8 *skb_head = skb->data;
2009 int skb_len = skb->len;
2010 unsigned int snaplen, res;
2011 bool is_drop_n_account = false;
2013 if (skb->pkt_type == PACKET_LOOPBACK)
2016 sk = pt->af_packet_priv;
2019 if (!net_eq(dev_net(dev), sock_net(sk)))
2024 if (dev->header_ops) {
2025 /* The device has an explicit notion of ll header,
2026 * exported to higher levels.
2028 * Otherwise, the device hides details of its frame
2029 * structure, so that corresponding packet head is
2030 * never delivered to user.
2032 if (sk->sk_type != SOCK_DGRAM)
2033 skb_push(skb, skb->data - skb_mac_header(skb));
2034 else if (skb->pkt_type == PACKET_OUTGOING) {
2035 /* Special case: outgoing packets have ll header at head */
2036 skb_pull(skb, skb_network_offset(skb));
2042 res = run_filter(skb, sk, snaplen);
2044 goto drop_n_restore;
2048 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2051 if (skb_shared(skb)) {
2052 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2056 if (skb_head != skb->data) {
2057 skb->data = skb_head;
2064 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2066 sll = &PACKET_SKB_CB(skb)->sa.ll;
2067 sll->sll_hatype = dev->type;
2068 sll->sll_pkttype = skb->pkt_type;
2069 if (unlikely(po->origdev))
2070 sll->sll_ifindex = orig_dev->ifindex;
2072 sll->sll_ifindex = dev->ifindex;
2074 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2076 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2077 * Use their space for storing the original skb length.
2079 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2081 if (pskb_trim(skb, snaplen))
2084 skb_set_owner_r(skb, sk);
2088 /* drop conntrack reference */
2091 spin_lock(&sk->sk_receive_queue.lock);
2092 po->stats.stats1.tp_packets++;
2093 sock_skb_set_dropcount(sk, skb);
2094 __skb_queue_tail(&sk->sk_receive_queue, skb);
2095 spin_unlock(&sk->sk_receive_queue.lock);
2096 sk->sk_data_ready(sk);
2100 is_drop_n_account = true;
2101 spin_lock(&sk->sk_receive_queue.lock);
2102 po->stats.stats1.tp_drops++;
2103 atomic_inc(&sk->sk_drops);
2104 spin_unlock(&sk->sk_receive_queue.lock);
2107 if (skb_head != skb->data && skb_shared(skb)) {
2108 skb->data = skb_head;
2112 if (!is_drop_n_account)
2119 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2120 struct packet_type *pt, struct net_device *orig_dev)
2123 struct packet_sock *po;
2124 struct sockaddr_ll *sll;
2125 union tpacket_uhdr h;
2126 u8 *skb_head = skb->data;
2127 int skb_len = skb->len;
2128 unsigned int snaplen, res;
2129 unsigned long status = TP_STATUS_USER;
2130 unsigned short macoff, netoff, hdrlen;
2131 struct sk_buff *copy_skb = NULL;
2134 bool is_drop_n_account = false;
2136 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2137 * We may add members to them until current aligned size without forcing
2138 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2140 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2141 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2143 if (skb->pkt_type == PACKET_LOOPBACK)
2146 sk = pt->af_packet_priv;
2149 if (!net_eq(dev_net(dev), sock_net(sk)))
2152 if (dev->header_ops) {
2153 if (sk->sk_type != SOCK_DGRAM)
2154 skb_push(skb, skb->data - skb_mac_header(skb));
2155 else if (skb->pkt_type == PACKET_OUTGOING) {
2156 /* Special case: outgoing packets have ll header at head */
2157 skb_pull(skb, skb_network_offset(skb));
2163 res = run_filter(skb, sk, snaplen);
2165 goto drop_n_restore;
2167 if (skb->ip_summed == CHECKSUM_PARTIAL)
2168 status |= TP_STATUS_CSUMNOTREADY;
2169 else if (skb->pkt_type != PACKET_OUTGOING &&
2170 (skb->ip_summed == CHECKSUM_COMPLETE ||
2171 skb_csum_unnecessary(skb)))
2172 status |= TP_STATUS_CSUM_VALID;
2177 if (sk->sk_type == SOCK_DGRAM) {
2178 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2181 unsigned int maclen = skb_network_offset(skb);
2182 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2183 (maclen < 16 ? 16 : maclen)) +
2185 if (po->has_vnet_hdr)
2186 netoff += sizeof(struct virtio_net_hdr);
2187 macoff = netoff - maclen;
2189 if (po->tp_version <= TPACKET_V2) {
2190 if (macoff + snaplen > po->rx_ring.frame_size) {
2191 if (po->copy_thresh &&
2192 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2193 if (skb_shared(skb)) {
2194 copy_skb = skb_clone(skb, GFP_ATOMIC);
2196 copy_skb = skb_get(skb);
2197 skb_head = skb->data;
2200 skb_set_owner_r(copy_skb, sk);
2202 snaplen = po->rx_ring.frame_size - macoff;
2203 if ((int)snaplen < 0)
2206 } else if (unlikely(macoff + snaplen >
2207 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2210 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2211 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2212 snaplen, nval, macoff);
2214 if (unlikely((int)snaplen < 0)) {
2216 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2219 spin_lock(&sk->sk_receive_queue.lock);
2220 h.raw = packet_current_rx_frame(po, skb,
2221 TP_STATUS_KERNEL, (macoff+snaplen));
2223 goto drop_n_account;
2224 if (po->tp_version <= TPACKET_V2) {
2225 packet_increment_rx_head(po, &po->rx_ring);
2227 * LOSING will be reported till you read the stats,
2228 * because it's COR - Clear On Read.
2229 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2232 if (po->stats.stats1.tp_drops)
2233 status |= TP_STATUS_LOSING;
2235 po->stats.stats1.tp_packets++;
2237 status |= TP_STATUS_COPY;
2238 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2240 spin_unlock(&sk->sk_receive_queue.lock);
2242 if (po->has_vnet_hdr) {
2243 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2244 sizeof(struct virtio_net_hdr),
2246 spin_lock(&sk->sk_receive_queue.lock);
2247 goto drop_n_account;
2251 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2253 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2254 getnstimeofday(&ts);
2256 status |= ts_status;
2258 switch (po->tp_version) {
2260 h.h1->tp_len = skb->len;
2261 h.h1->tp_snaplen = snaplen;
2262 h.h1->tp_mac = macoff;
2263 h.h1->tp_net = netoff;
2264 h.h1->tp_sec = ts.tv_sec;
2265 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2266 hdrlen = sizeof(*h.h1);
2269 h.h2->tp_len = skb->len;
2270 h.h2->tp_snaplen = snaplen;
2271 h.h2->tp_mac = macoff;
2272 h.h2->tp_net = netoff;
2273 h.h2->tp_sec = ts.tv_sec;
2274 h.h2->tp_nsec = ts.tv_nsec;
2275 if (skb_vlan_tag_present(skb)) {
2276 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2277 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2278 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2280 h.h2->tp_vlan_tci = 0;
2281 h.h2->tp_vlan_tpid = 0;
2283 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2284 hdrlen = sizeof(*h.h2);
2287 /* tp_nxt_offset,vlan are already populated above.
2288 * So DONT clear those fields here
2290 h.h3->tp_status |= status;
2291 h.h3->tp_len = skb->len;
2292 h.h3->tp_snaplen = snaplen;
2293 h.h3->tp_mac = macoff;
2294 h.h3->tp_net = netoff;
2295 h.h3->tp_sec = ts.tv_sec;
2296 h.h3->tp_nsec = ts.tv_nsec;
2297 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2298 hdrlen = sizeof(*h.h3);
2304 sll = h.raw + TPACKET_ALIGN(hdrlen);
2305 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2306 sll->sll_family = AF_PACKET;
2307 sll->sll_hatype = dev->type;
2308 sll->sll_protocol = skb->protocol;
2309 sll->sll_pkttype = skb->pkt_type;
2310 if (unlikely(po->origdev))
2311 sll->sll_ifindex = orig_dev->ifindex;
2313 sll->sll_ifindex = dev->ifindex;
2317 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2318 if (po->tp_version <= TPACKET_V2) {
2321 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2324 for (start = h.raw; start < end; start += PAGE_SIZE)
2325 flush_dcache_page(pgv_to_page(start));
2330 if (po->tp_version <= TPACKET_V2) {
2331 __packet_set_status(po, h.raw, status);
2332 sk->sk_data_ready(sk);
2334 prb_clear_blk_fill_status(&po->rx_ring);
2338 if (skb_head != skb->data && skb_shared(skb)) {
2339 skb->data = skb_head;
2343 if (!is_drop_n_account)
2350 is_drop_n_account = true;
2351 po->stats.stats1.tp_drops++;
2352 spin_unlock(&sk->sk_receive_queue.lock);
2354 sk->sk_data_ready(sk);
2355 kfree_skb(copy_skb);
2356 goto drop_n_restore;
2359 static void tpacket_destruct_skb(struct sk_buff *skb)
2361 struct packet_sock *po = pkt_sk(skb->sk);
2363 if (likely(po->tx_ring.pg_vec)) {
2367 ph = skb_shinfo(skb)->destructor_arg;
2368 packet_dec_pending(&po->tx_ring);
2370 ts = __packet_set_timestamp(po, ph, skb);
2371 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2377 static void tpacket_set_protocol(const struct net_device *dev,
2378 struct sk_buff *skb)
2380 if (dev->type == ARPHRD_ETHER) {
2381 skb_reset_mac_header(skb);
2382 skb->protocol = eth_hdr(skb)->h_proto;
2386 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2388 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2389 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2390 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2391 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2392 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2393 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2394 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2396 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2402 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2403 struct virtio_net_hdr *vnet_hdr)
2405 if (*len < sizeof(*vnet_hdr))
2407 *len -= sizeof(*vnet_hdr);
2409 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2412 return __packet_snd_vnet_parse(vnet_hdr, *len);
2415 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2416 void *frame, struct net_device *dev, void *data, int tp_len,
2417 __be16 proto, unsigned char *addr, int hlen, int copylen,
2418 const struct sockcm_cookie *sockc)
2420 union tpacket_uhdr ph;
2421 int to_write, offset, len, nr_frags, len_max;
2422 struct socket *sock = po->sk.sk_socket;
2428 skb->protocol = proto;
2430 skb->priority = po->sk.sk_priority;
2431 skb->mark = po->sk.sk_mark;
2432 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2433 skb_shinfo(skb)->destructor_arg = ph.raw;
2435 skb_reserve(skb, hlen);
2436 skb_reset_network_header(skb);
2440 if (sock->type == SOCK_DGRAM) {
2441 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2443 if (unlikely(err < 0))
2445 } else if (copylen) {
2446 int hdrlen = min_t(int, copylen, tp_len);
2448 skb_push(skb, dev->hard_header_len);
2449 skb_put(skb, copylen - dev->hard_header_len);
2450 err = skb_store_bits(skb, 0, data, hdrlen);
2453 if (!dev_validate_header(dev, skb->data, hdrlen))
2456 tpacket_set_protocol(dev, skb);
2462 offset = offset_in_page(data);
2463 len_max = PAGE_SIZE - offset;
2464 len = ((to_write > len_max) ? len_max : to_write);
2466 skb->data_len = to_write;
2467 skb->len += to_write;
2468 skb->truesize += to_write;
2469 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2471 while (likely(to_write)) {
2472 nr_frags = skb_shinfo(skb)->nr_frags;
2474 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2475 pr_err("Packet exceed the number of skb frags(%lu)\n",
2480 page = pgv_to_page(data);
2482 flush_dcache_page(page);
2484 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2487 len_max = PAGE_SIZE;
2488 len = ((to_write > len_max) ? len_max : to_write);
2491 skb_probe_transport_header(skb, 0);
2496 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2497 int size_max, void **data)
2499 union tpacket_uhdr ph;
2504 switch (po->tp_version) {
2506 if (ph.h3->tp_next_offset != 0) {
2507 pr_warn_once("variable sized slot not supported");
2510 tp_len = ph.h3->tp_len;
2513 tp_len = ph.h2->tp_len;
2516 tp_len = ph.h1->tp_len;
2519 if (unlikely(tp_len > size_max)) {
2520 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2524 if (unlikely(po->tp_tx_has_off)) {
2525 int off_min, off_max;
2527 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2528 off_max = po->tx_ring.frame_size - tp_len;
2529 if (po->sk.sk_type == SOCK_DGRAM) {
2530 switch (po->tp_version) {
2532 off = ph.h3->tp_net;
2535 off = ph.h2->tp_net;
2538 off = ph.h1->tp_net;
2542 switch (po->tp_version) {
2544 off = ph.h3->tp_mac;
2547 off = ph.h2->tp_mac;
2550 off = ph.h1->tp_mac;
2554 if (unlikely((off < off_min) || (off_max < off)))
2557 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2560 *data = frame + off;
2564 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2566 struct sk_buff *skb;
2567 struct net_device *dev;
2568 struct virtio_net_hdr *vnet_hdr = NULL;
2569 struct sockcm_cookie sockc;
2571 int err, reserve = 0;
2573 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2574 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2575 int tp_len, size_max;
2576 unsigned char *addr;
2579 int status = TP_STATUS_AVAILABLE;
2580 int hlen, tlen, copylen = 0;
2582 mutex_lock(&po->pg_vec_lock);
2584 if (likely(saddr == NULL)) {
2585 dev = packet_cached_dev_get(po);
2590 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2592 if (msg->msg_namelen < (saddr->sll_halen
2593 + offsetof(struct sockaddr_ll,
2596 proto = saddr->sll_protocol;
2597 addr = saddr->sll_addr;
2598 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2601 sockc.tsflags = po->sk.sk_tsflags;
2602 if (msg->msg_controllen) {
2603 err = sock_cmsg_send(&po->sk, msg, &sockc);
2609 if (unlikely(dev == NULL))
2612 if (unlikely(!(dev->flags & IFF_UP)))
2615 if (po->sk.sk_socket->type == SOCK_RAW)
2616 reserve = dev->hard_header_len;
2617 size_max = po->tx_ring.frame_size
2618 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2620 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2621 size_max = dev->mtu + reserve + VLAN_HLEN;
2624 ph = packet_current_frame(po, &po->tx_ring,
2625 TP_STATUS_SEND_REQUEST);
2626 if (unlikely(ph == NULL)) {
2627 if (need_wait && need_resched())
2633 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2637 status = TP_STATUS_SEND_REQUEST;
2638 hlen = LL_RESERVED_SPACE(dev);
2639 tlen = dev->needed_tailroom;
2640 if (po->has_vnet_hdr) {
2642 data += sizeof(*vnet_hdr);
2643 tp_len -= sizeof(*vnet_hdr);
2645 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2649 copylen = __virtio16_to_cpu(vio_le(),
2652 copylen = max_t(int, copylen, dev->hard_header_len);
2653 skb = sock_alloc_send_skb(&po->sk,
2654 hlen + tlen + sizeof(struct sockaddr_ll) +
2655 (copylen - dev->hard_header_len),
2658 if (unlikely(skb == NULL)) {
2659 /* we assume the socket was initially writeable ... */
2660 if (likely(len_sum > 0))
2664 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2665 addr, hlen, copylen, &sockc);
2666 if (likely(tp_len >= 0) &&
2667 tp_len > dev->mtu + reserve &&
2668 !po->has_vnet_hdr &&
2669 !packet_extra_vlan_len_allowed(dev, skb))
2672 if (unlikely(tp_len < 0)) {
2675 __packet_set_status(po, ph,
2676 TP_STATUS_AVAILABLE);
2677 packet_increment_head(&po->tx_ring);
2681 status = TP_STATUS_WRONG_FORMAT;
2687 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2693 packet_pick_tx_queue(dev, skb);
2695 skb->destructor = tpacket_destruct_skb;
2696 __packet_set_status(po, ph, TP_STATUS_SENDING);
2697 packet_inc_pending(&po->tx_ring);
2699 status = TP_STATUS_SEND_REQUEST;
2700 err = po->xmit(skb);
2701 if (unlikely(err > 0)) {
2702 err = net_xmit_errno(err);
2703 if (err && __packet_get_status(po, ph) ==
2704 TP_STATUS_AVAILABLE) {
2705 /* skb was destructed already */
2710 * skb was dropped but not destructed yet;
2711 * let's treat it like congestion or err < 0
2715 packet_increment_head(&po->tx_ring);
2717 } while (likely((ph != NULL) ||
2718 /* Note: packet_read_pending() might be slow if we have
2719 * to call it as it's per_cpu variable, but in fast-path
2720 * we already short-circuit the loop with the first
2721 * condition, and luckily don't have to go that path
2724 (need_wait && packet_read_pending(&po->tx_ring))));
2730 __packet_set_status(po, ph, status);
2735 mutex_unlock(&po->pg_vec_lock);
2739 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2740 size_t reserve, size_t len,
2741 size_t linear, int noblock,
2744 struct sk_buff *skb;
2746 /* Under a page? Don't bother with paged skb. */
2747 if (prepad + len < PAGE_SIZE || !linear)
2750 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2755 skb_reserve(skb, reserve);
2756 skb_put(skb, linear);
2757 skb->data_len = len - linear;
2758 skb->len += len - linear;
2763 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2765 struct sock *sk = sock->sk;
2766 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2767 struct sk_buff *skb;
2768 struct net_device *dev;
2770 unsigned char *addr;
2771 int err, reserve = 0;
2772 struct sockcm_cookie sockc;
2773 struct virtio_net_hdr vnet_hdr = { 0 };
2775 struct packet_sock *po = pkt_sk(sk);
2780 * Get and verify the address.
2783 if (likely(saddr == NULL)) {
2784 dev = packet_cached_dev_get(po);
2789 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2791 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2793 proto = saddr->sll_protocol;
2794 addr = saddr->sll_addr;
2795 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2799 if (unlikely(dev == NULL))
2802 if (unlikely(!(dev->flags & IFF_UP)))
2805 sockc.tsflags = sk->sk_tsflags;
2806 sockc.mark = sk->sk_mark;
2807 if (msg->msg_controllen) {
2808 err = sock_cmsg_send(sk, msg, &sockc);
2813 if (sock->type == SOCK_RAW)
2814 reserve = dev->hard_header_len;
2815 if (po->has_vnet_hdr) {
2816 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2821 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2822 if (!netif_supports_nofcs(dev)) {
2823 err = -EPROTONOSUPPORT;
2826 extra_len = 4; /* We're doing our own CRC */
2830 if (!vnet_hdr.gso_type &&
2831 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2835 hlen = LL_RESERVED_SPACE(dev);
2836 tlen = dev->needed_tailroom;
2837 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2838 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2839 msg->msg_flags & MSG_DONTWAIT, &err);
2843 skb_set_network_header(skb, reserve);
2846 if (sock->type == SOCK_DGRAM) {
2847 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2848 if (unlikely(offset < 0))
2852 /* Returns -EFAULT on error */
2853 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2857 if (sock->type == SOCK_RAW &&
2858 !dev_validate_header(dev, skb->data, len)) {
2863 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2865 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2866 !packet_extra_vlan_len_allowed(dev, skb)) {
2871 skb->protocol = proto;
2873 skb->priority = sk->sk_priority;
2874 skb->mark = sockc.mark;
2876 packet_pick_tx_queue(dev, skb);
2878 if (po->has_vnet_hdr) {
2879 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2882 len += sizeof(vnet_hdr);
2885 skb_probe_transport_header(skb, reserve);
2887 if (unlikely(extra_len == 4))
2890 err = po->xmit(skb);
2891 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2907 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2909 struct sock *sk = sock->sk;
2910 struct packet_sock *po = pkt_sk(sk);
2912 if (po->tx_ring.pg_vec)
2913 return tpacket_snd(po, msg);
2915 return packet_snd(sock, msg, len);
2919 * Close a PACKET socket. This is fairly simple. We immediately go
2920 * to 'closed' state and remove our protocol entry in the device list.
2923 static int packet_release(struct socket *sock)
2925 struct sock *sk = sock->sk;
2926 struct packet_sock *po;
2928 union tpacket_req_u req_u;
2936 mutex_lock(&net->packet.sklist_lock);
2937 sk_del_node_init_rcu(sk);
2938 mutex_unlock(&net->packet.sklist_lock);
2941 sock_prot_inuse_add(net, sk->sk_prot, -1);
2944 spin_lock(&po->bind_lock);
2945 unregister_prot_hook(sk, false);
2946 packet_cached_dev_reset(po);
2948 if (po->prot_hook.dev) {
2949 dev_put(po->prot_hook.dev);
2950 po->prot_hook.dev = NULL;
2952 spin_unlock(&po->bind_lock);
2954 packet_flush_mclist(sk);
2956 if (po->rx_ring.pg_vec) {
2957 memset(&req_u, 0, sizeof(req_u));
2958 packet_set_ring(sk, &req_u, 1, 0);
2961 if (po->tx_ring.pg_vec) {
2962 memset(&req_u, 0, sizeof(req_u));
2963 packet_set_ring(sk, &req_u, 1, 1);
2970 * Now the socket is dead. No more input will appear.
2977 skb_queue_purge(&sk->sk_receive_queue);
2978 packet_free_pending(po);
2979 sk_refcnt_debug_release(sk);
2986 * Attach a packet hook.
2989 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2992 struct packet_sock *po = pkt_sk(sk);
2993 struct net_device *dev_curr;
2996 struct net_device *dev = NULL;
2998 bool unlisted = false;
3004 spin_lock(&po->bind_lock);
3008 dev = dev_get_by_name_rcu(sock_net(sk), name);
3013 } else if (ifindex) {
3014 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3024 proto_curr = po->prot_hook.type;
3025 dev_curr = po->prot_hook.dev;
3027 need_rehook = proto_curr != proto || dev_curr != dev;
3032 __unregister_prot_hook(sk, true);
3034 dev_curr = po->prot_hook.dev;
3036 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3041 po->prot_hook.type = proto;
3043 if (unlikely(unlisted)) {
3045 po->prot_hook.dev = NULL;
3047 packet_cached_dev_reset(po);
3049 po->prot_hook.dev = dev;
3050 po->ifindex = dev ? dev->ifindex : 0;
3051 packet_cached_dev_assign(po, dev);
3057 if (proto == 0 || !need_rehook)
3060 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3061 register_prot_hook(sk);
3063 sk->sk_err = ENETDOWN;
3064 if (!sock_flag(sk, SOCK_DEAD))
3065 sk->sk_error_report(sk);
3070 spin_unlock(&po->bind_lock);
3076 * Bind a packet socket to a device
3079 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3082 struct sock *sk = sock->sk;
3089 if (addr_len != sizeof(struct sockaddr))
3091 strlcpy(name, uaddr->sa_data, sizeof(name));
3093 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3096 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3098 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3099 struct sock *sk = sock->sk;
3105 if (addr_len < sizeof(struct sockaddr_ll))
3107 if (sll->sll_family != AF_PACKET)
3110 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3111 sll->sll_protocol ? : pkt_sk(sk)->num);
3114 static struct proto packet_proto = {
3116 .owner = THIS_MODULE,
3117 .obj_size = sizeof(struct packet_sock),
3121 * Create a packet of type SOCK_PACKET.
3124 static int packet_create(struct net *net, struct socket *sock, int protocol,
3128 struct packet_sock *po;
3129 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3132 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3134 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3135 sock->type != SOCK_PACKET)
3136 return -ESOCKTNOSUPPORT;
3138 sock->state = SS_UNCONNECTED;
3141 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3145 sock->ops = &packet_ops;
3146 if (sock->type == SOCK_PACKET)
3147 sock->ops = &packet_ops_spkt;
3149 sock_init_data(sock, sk);
3152 sk->sk_family = PF_PACKET;
3154 po->xmit = dev_queue_xmit;
3156 err = packet_alloc_pending(po);
3160 packet_cached_dev_reset(po);
3162 sk->sk_destruct = packet_sock_destruct;
3163 sk_refcnt_debug_inc(sk);
3166 * Attach a protocol block
3169 spin_lock_init(&po->bind_lock);
3170 mutex_init(&po->pg_vec_lock);
3171 po->rollover = NULL;
3172 po->prot_hook.func = packet_rcv;
3174 if (sock->type == SOCK_PACKET)
3175 po->prot_hook.func = packet_rcv_spkt;
3177 po->prot_hook.af_packet_priv = sk;
3180 po->prot_hook.type = proto;
3181 register_prot_hook(sk);
3184 mutex_lock(&net->packet.sklist_lock);
3185 sk_add_node_rcu(sk, &net->packet.sklist);
3186 mutex_unlock(&net->packet.sklist_lock);
3189 sock_prot_inuse_add(net, &packet_proto, 1);
3200 * Pull a packet from our receive queue and hand it to the user.
3201 * If necessary we block.
3204 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3207 struct sock *sk = sock->sk;
3208 struct sk_buff *skb;
3210 int vnet_hdr_len = 0;
3211 unsigned int origlen = 0;
3214 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3218 /* What error should we return now? EUNATTACH? */
3219 if (pkt_sk(sk)->ifindex < 0)
3223 if (flags & MSG_ERRQUEUE) {
3224 err = sock_recv_errqueue(sk, msg, len,
3225 SOL_PACKET, PACKET_TX_TIMESTAMP);
3230 * Call the generic datagram receiver. This handles all sorts
3231 * of horrible races and re-entrancy so we can forget about it
3232 * in the protocol layers.
3234 * Now it will return ENETDOWN, if device have just gone down,
3235 * but then it will block.
3238 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3241 * An error occurred so return it. Because skb_recv_datagram()
3242 * handles the blocking we don't see and worry about blocking
3249 if (pkt_sk(sk)->pressure)
3250 packet_rcv_has_room(pkt_sk(sk), NULL);
3252 if (pkt_sk(sk)->has_vnet_hdr) {
3253 err = packet_rcv_vnet(msg, skb, &len);
3256 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3259 /* You lose any data beyond the buffer you gave. If it worries
3260 * a user program they can ask the device for its MTU
3266 msg->msg_flags |= MSG_TRUNC;
3269 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3273 if (sock->type != SOCK_PACKET) {
3274 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3276 /* Original length was stored in sockaddr_ll fields */
3277 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3278 sll->sll_family = AF_PACKET;
3279 sll->sll_protocol = skb->protocol;
3282 sock_recv_ts_and_drops(msg, sk, skb);
3284 if (msg->msg_name) {
3285 /* If the address length field is there to be filled
3286 * in, we fill it in now.
3288 if (sock->type == SOCK_PACKET) {
3289 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3290 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3292 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3294 msg->msg_namelen = sll->sll_halen +
3295 offsetof(struct sockaddr_ll, sll_addr);
3297 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3301 if (pkt_sk(sk)->auxdata) {
3302 struct tpacket_auxdata aux;
3304 aux.tp_status = TP_STATUS_USER;
3305 if (skb->ip_summed == CHECKSUM_PARTIAL)
3306 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3307 else if (skb->pkt_type != PACKET_OUTGOING &&
3308 (skb->ip_summed == CHECKSUM_COMPLETE ||
3309 skb_csum_unnecessary(skb)))
3310 aux.tp_status |= TP_STATUS_CSUM_VALID;
3312 aux.tp_len = origlen;
3313 aux.tp_snaplen = skb->len;
3315 aux.tp_net = skb_network_offset(skb);
3316 if (skb_vlan_tag_present(skb)) {
3317 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3318 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3319 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3321 aux.tp_vlan_tci = 0;
3322 aux.tp_vlan_tpid = 0;
3324 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3328 * Free or return the buffer as appropriate. Again this
3329 * hides all the races and re-entrancy issues from us.
3331 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3334 skb_free_datagram(sk, skb);
3339 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3340 int *uaddr_len, int peer)
3342 struct net_device *dev;
3343 struct sock *sk = sock->sk;
3348 uaddr->sa_family = AF_PACKET;
3349 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3351 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3353 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3355 *uaddr_len = sizeof(*uaddr);
3360 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3361 int *uaddr_len, int peer)
3363 struct net_device *dev;
3364 struct sock *sk = sock->sk;
3365 struct packet_sock *po = pkt_sk(sk);
3366 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3371 sll->sll_family = AF_PACKET;
3372 sll->sll_ifindex = po->ifindex;
3373 sll->sll_protocol = po->num;
3374 sll->sll_pkttype = 0;
3376 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3378 sll->sll_hatype = dev->type;
3379 sll->sll_halen = dev->addr_len;
3380 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3382 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3386 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3391 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3395 case PACKET_MR_MULTICAST:
3396 if (i->alen != dev->addr_len)
3399 return dev_mc_add(dev, i->addr);
3401 return dev_mc_del(dev, i->addr);
3403 case PACKET_MR_PROMISC:
3404 return dev_set_promiscuity(dev, what);
3405 case PACKET_MR_ALLMULTI:
3406 return dev_set_allmulti(dev, what);
3407 case PACKET_MR_UNICAST:
3408 if (i->alen != dev->addr_len)
3411 return dev_uc_add(dev, i->addr);
3413 return dev_uc_del(dev, i->addr);
3421 static void packet_dev_mclist_delete(struct net_device *dev,
3422 struct packet_mclist **mlp)
3424 struct packet_mclist *ml;
3426 while ((ml = *mlp) != NULL) {
3427 if (ml->ifindex == dev->ifindex) {
3428 packet_dev_mc(dev, ml, -1);
3436 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3438 struct packet_sock *po = pkt_sk(sk);
3439 struct packet_mclist *ml, *i;
3440 struct net_device *dev;
3446 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3451 if (mreq->mr_alen > dev->addr_len)
3455 i = kmalloc(sizeof(*i), GFP_KERNEL);
3460 for (ml = po->mclist; ml; ml = ml->next) {
3461 if (ml->ifindex == mreq->mr_ifindex &&
3462 ml->type == mreq->mr_type &&
3463 ml->alen == mreq->mr_alen &&
3464 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3466 /* Free the new element ... */
3472 i->type = mreq->mr_type;
3473 i->ifindex = mreq->mr_ifindex;
3474 i->alen = mreq->mr_alen;
3475 memcpy(i->addr, mreq->mr_address, i->alen);
3476 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3478 i->next = po->mclist;
3480 err = packet_dev_mc(dev, i, 1);
3482 po->mclist = i->next;
3491 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3493 struct packet_mclist *ml, **mlp;
3497 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3498 if (ml->ifindex == mreq->mr_ifindex &&
3499 ml->type == mreq->mr_type &&
3500 ml->alen == mreq->mr_alen &&
3501 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3502 if (--ml->count == 0) {
3503 struct net_device *dev;
3505 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3507 packet_dev_mc(dev, ml, -1);
3517 static void packet_flush_mclist(struct sock *sk)
3519 struct packet_sock *po = pkt_sk(sk);
3520 struct packet_mclist *ml;
3526 while ((ml = po->mclist) != NULL) {
3527 struct net_device *dev;
3529 po->mclist = ml->next;
3530 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3532 packet_dev_mc(dev, ml, -1);
3539 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3541 struct sock *sk = sock->sk;
3542 struct packet_sock *po = pkt_sk(sk);
3545 if (level != SOL_PACKET)
3546 return -ENOPROTOOPT;
3549 case PACKET_ADD_MEMBERSHIP:
3550 case PACKET_DROP_MEMBERSHIP:
3552 struct packet_mreq_max mreq;
3554 memset(&mreq, 0, sizeof(mreq));
3555 if (len < sizeof(struct packet_mreq))
3557 if (len > sizeof(mreq))
3559 if (copy_from_user(&mreq, optval, len))
3561 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3563 if (optname == PACKET_ADD_MEMBERSHIP)
3564 ret = packet_mc_add(sk, &mreq);
3566 ret = packet_mc_drop(sk, &mreq);
3570 case PACKET_RX_RING:
3571 case PACKET_TX_RING:
3573 union tpacket_req_u req_u;
3576 switch (po->tp_version) {
3579 len = sizeof(req_u.req);
3583 len = sizeof(req_u.req3);
3588 if (copy_from_user(&req_u.req, optval, len))
3590 return packet_set_ring(sk, &req_u, 0,
3591 optname == PACKET_TX_RING);
3593 case PACKET_COPY_THRESH:
3597 if (optlen != sizeof(val))
3599 if (copy_from_user(&val, optval, sizeof(val)))
3602 pkt_sk(sk)->copy_thresh = val;
3605 case PACKET_VERSION:
3609 if (optlen != sizeof(val))
3611 if (copy_from_user(&val, optval, sizeof(val)))
3622 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3625 po->tp_version = val;
3631 case PACKET_RESERVE:
3635 if (optlen != sizeof(val))
3637 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3639 if (copy_from_user(&val, optval, sizeof(val)))
3641 po->tp_reserve = val;
3648 if (optlen != sizeof(val))
3650 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3652 if (copy_from_user(&val, optval, sizeof(val)))
3654 po->tp_loss = !!val;
3657 case PACKET_AUXDATA:
3661 if (optlen < sizeof(val))
3663 if (copy_from_user(&val, optval, sizeof(val)))
3666 po->auxdata = !!val;
3669 case PACKET_ORIGDEV:
3673 if (optlen < sizeof(val))
3675 if (copy_from_user(&val, optval, sizeof(val)))
3678 po->origdev = !!val;
3681 case PACKET_VNET_HDR:
3685 if (sock->type != SOCK_RAW)
3687 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3689 if (optlen < sizeof(val))
3691 if (copy_from_user(&val, optval, sizeof(val)))
3694 po->has_vnet_hdr = !!val;
3697 case PACKET_TIMESTAMP:
3701 if (optlen != sizeof(val))
3703 if (copy_from_user(&val, optval, sizeof(val)))
3706 po->tp_tstamp = val;
3713 if (optlen != sizeof(val))
3715 if (copy_from_user(&val, optval, sizeof(val)))
3718 return fanout_add(sk, val & 0xffff, val >> 16);
3720 case PACKET_FANOUT_DATA:
3725 return fanout_set_data(po, optval, optlen);
3727 case PACKET_TX_HAS_OFF:
3731 if (optlen != sizeof(val))
3733 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3735 if (copy_from_user(&val, optval, sizeof(val)))
3737 po->tp_tx_has_off = !!val;
3740 case PACKET_QDISC_BYPASS:
3744 if (optlen != sizeof(val))
3746 if (copy_from_user(&val, optval, sizeof(val)))
3749 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3753 return -ENOPROTOOPT;
3757 static int packet_getsockopt(struct socket *sock, int level, int optname,
3758 char __user *optval, int __user *optlen)
3761 int val, lv = sizeof(val);
3762 struct sock *sk = sock->sk;
3763 struct packet_sock *po = pkt_sk(sk);
3765 union tpacket_stats_u st;
3766 struct tpacket_rollover_stats rstats;
3768 if (level != SOL_PACKET)
3769 return -ENOPROTOOPT;
3771 if (get_user(len, optlen))
3778 case PACKET_STATISTICS:
3779 spin_lock_bh(&sk->sk_receive_queue.lock);
3780 memcpy(&st, &po->stats, sizeof(st));
3781 memset(&po->stats, 0, sizeof(po->stats));
3782 spin_unlock_bh(&sk->sk_receive_queue.lock);
3784 if (po->tp_version == TPACKET_V3) {
3785 lv = sizeof(struct tpacket_stats_v3);
3786 st.stats3.tp_packets += st.stats3.tp_drops;
3789 lv = sizeof(struct tpacket_stats);
3790 st.stats1.tp_packets += st.stats1.tp_drops;
3795 case PACKET_AUXDATA:
3798 case PACKET_ORIGDEV:
3801 case PACKET_VNET_HDR:
3802 val = po->has_vnet_hdr;
3804 case PACKET_VERSION:
3805 val = po->tp_version;
3808 if (len > sizeof(int))
3810 if (copy_from_user(&val, optval, len))
3814 val = sizeof(struct tpacket_hdr);
3817 val = sizeof(struct tpacket2_hdr);
3820 val = sizeof(struct tpacket3_hdr);
3826 case PACKET_RESERVE:
3827 val = po->tp_reserve;
3832 case PACKET_TIMESTAMP:
3833 val = po->tp_tstamp;
3837 ((u32)po->fanout->id |
3838 ((u32)po->fanout->type << 16) |
3839 ((u32)po->fanout->flags << 24)) :
3842 case PACKET_ROLLOVER_STATS:
3845 rstats.tp_all = atomic_long_read(&po->rollover->num);
3846 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3847 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3849 lv = sizeof(rstats);
3851 case PACKET_TX_HAS_OFF:
3852 val = po->tp_tx_has_off;
3854 case PACKET_QDISC_BYPASS:
3855 val = packet_use_direct_xmit(po);
3858 return -ENOPROTOOPT;
3863 if (put_user(len, optlen))
3865 if (copy_to_user(optval, data, len))
3871 #ifdef CONFIG_COMPAT
3872 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3873 char __user *optval, unsigned int optlen)
3875 struct packet_sock *po = pkt_sk(sock->sk);
3877 if (level != SOL_PACKET)
3878 return -ENOPROTOOPT;
3880 if (optname == PACKET_FANOUT_DATA &&
3881 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3882 optval = (char __user *)get_compat_bpf_fprog(optval);
3885 optlen = sizeof(struct sock_fprog);
3888 return packet_setsockopt(sock, level, optname, optval, optlen);
3892 static int packet_notifier(struct notifier_block *this,
3893 unsigned long msg, void *ptr)
3896 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3897 struct net *net = dev_net(dev);
3900 sk_for_each_rcu(sk, &net->packet.sklist) {
3901 struct packet_sock *po = pkt_sk(sk);
3904 case NETDEV_UNREGISTER:
3906 packet_dev_mclist_delete(dev, &po->mclist);
3910 if (dev->ifindex == po->ifindex) {
3911 spin_lock(&po->bind_lock);
3913 __unregister_prot_hook(sk, false);
3914 sk->sk_err = ENETDOWN;
3915 if (!sock_flag(sk, SOCK_DEAD))
3916 sk->sk_error_report(sk);
3918 if (msg == NETDEV_UNREGISTER) {
3919 packet_cached_dev_reset(po);
3922 if (po->prot_hook.dev)
3923 dev_put(po->prot_hook.dev);
3924 po->prot_hook.dev = NULL;
3926 spin_unlock(&po->bind_lock);
3930 if (dev->ifindex == po->ifindex) {
3931 spin_lock(&po->bind_lock);
3933 register_prot_hook(sk);
3934 spin_unlock(&po->bind_lock);
3944 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3947 struct sock *sk = sock->sk;
3952 int amount = sk_wmem_alloc_get(sk);
3954 return put_user(amount, (int __user *)arg);
3958 struct sk_buff *skb;
3961 spin_lock_bh(&sk->sk_receive_queue.lock);
3962 skb = skb_peek(&sk->sk_receive_queue);
3965 spin_unlock_bh(&sk->sk_receive_queue.lock);
3966 return put_user(amount, (int __user *)arg);
3969 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3971 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3981 case SIOCGIFBRDADDR:
3982 case SIOCSIFBRDADDR:
3983 case SIOCGIFNETMASK:
3984 case SIOCSIFNETMASK:
3985 case SIOCGIFDSTADDR:
3986 case SIOCSIFDSTADDR:
3988 return inet_dgram_ops.ioctl(sock, cmd, arg);
3992 return -ENOIOCTLCMD;
3997 static unsigned int packet_poll(struct file *file, struct socket *sock,
4000 struct sock *sk = sock->sk;
4001 struct packet_sock *po = pkt_sk(sk);
4002 unsigned int mask = datagram_poll(file, sock, wait);
4004 spin_lock_bh(&sk->sk_receive_queue.lock);
4005 if (po->rx_ring.pg_vec) {
4006 if (!packet_previous_rx_frame(po, &po->rx_ring,
4008 mask |= POLLIN | POLLRDNORM;
4010 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4012 spin_unlock_bh(&sk->sk_receive_queue.lock);
4013 spin_lock_bh(&sk->sk_write_queue.lock);
4014 if (po->tx_ring.pg_vec) {
4015 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4016 mask |= POLLOUT | POLLWRNORM;
4018 spin_unlock_bh(&sk->sk_write_queue.lock);
4023 /* Dirty? Well, I still did not learn better way to account
4027 static void packet_mm_open(struct vm_area_struct *vma)
4029 struct file *file = vma->vm_file;
4030 struct socket *sock = file->private_data;
4031 struct sock *sk = sock->sk;
4034 atomic_inc(&pkt_sk(sk)->mapped);
4037 static void packet_mm_close(struct vm_area_struct *vma)
4039 struct file *file = vma->vm_file;
4040 struct socket *sock = file->private_data;
4041 struct sock *sk = sock->sk;
4044 atomic_dec(&pkt_sk(sk)->mapped);
4047 static const struct vm_operations_struct packet_mmap_ops = {
4048 .open = packet_mm_open,
4049 .close = packet_mm_close,
4052 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4057 for (i = 0; i < len; i++) {
4058 if (likely(pg_vec[i].buffer)) {
4059 if (is_vmalloc_addr(pg_vec[i].buffer))
4060 vfree(pg_vec[i].buffer);
4062 free_pages((unsigned long)pg_vec[i].buffer,
4064 pg_vec[i].buffer = NULL;
4070 static char *alloc_one_pg_vec_page(unsigned long order)
4073 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4074 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4076 buffer = (char *) __get_free_pages(gfp_flags, order);
4080 /* __get_free_pages failed, fall back to vmalloc */
4081 buffer = vzalloc((1 << order) * PAGE_SIZE);
4085 /* vmalloc failed, lets dig into swap here */
4086 gfp_flags &= ~__GFP_NORETRY;
4087 buffer = (char *) __get_free_pages(gfp_flags, order);
4091 /* complete and utter failure */
4095 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4097 unsigned int block_nr = req->tp_block_nr;
4101 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4102 if (unlikely(!pg_vec))
4105 for (i = 0; i < block_nr; i++) {
4106 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4107 if (unlikely(!pg_vec[i].buffer))
4108 goto out_free_pgvec;
4115 free_pg_vec(pg_vec, order, block_nr);
4120 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4121 int closing, int tx_ring)
4123 struct pgv *pg_vec = NULL;
4124 struct packet_sock *po = pkt_sk(sk);
4125 int was_running, order = 0;
4126 struct packet_ring_buffer *rb;
4127 struct sk_buff_head *rb_queue;
4130 /* Added to avoid minimal code churn */
4131 struct tpacket_req *req = &req_u->req;
4135 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4136 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4140 if (atomic_read(&po->mapped))
4142 if (packet_read_pending(rb))
4146 if (req->tp_block_nr) {
4147 /* Sanity tests and some calculations */
4149 if (unlikely(rb->pg_vec))
4152 switch (po->tp_version) {
4154 po->tp_hdrlen = TPACKET_HDRLEN;
4157 po->tp_hdrlen = TPACKET2_HDRLEN;
4160 po->tp_hdrlen = TPACKET3_HDRLEN;
4165 if (unlikely((int)req->tp_block_size <= 0))
4167 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4169 if (po->tp_version >= TPACKET_V3 &&
4170 (int)(req->tp_block_size -
4171 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4173 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4176 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4179 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4180 if (unlikely(rb->frames_per_block == 0))
4182 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4187 order = get_order(req->tp_block_size);
4188 pg_vec = alloc_pg_vec(req, order);
4189 if (unlikely(!pg_vec))
4191 switch (po->tp_version) {
4193 /* Block transmit is not supported yet */
4195 init_prb_bdqc(po, rb, pg_vec, req_u);
4197 struct tpacket_req3 *req3 = &req_u->req3;
4199 if (req3->tp_retire_blk_tov ||
4200 req3->tp_sizeof_priv ||
4201 req3->tp_feature_req_word) {
4214 if (unlikely(req->tp_frame_nr))
4219 /* Detach socket from network */
4220 spin_lock(&po->bind_lock);
4221 was_running = po->running;
4225 __unregister_prot_hook(sk, false);
4227 spin_unlock(&po->bind_lock);
4232 mutex_lock(&po->pg_vec_lock);
4233 if (closing || atomic_read(&po->mapped) == 0) {
4235 spin_lock_bh(&rb_queue->lock);
4236 swap(rb->pg_vec, pg_vec);
4237 rb->frame_max = (req->tp_frame_nr - 1);
4239 rb->frame_size = req->tp_frame_size;
4240 spin_unlock_bh(&rb_queue->lock);
4242 swap(rb->pg_vec_order, order);
4243 swap(rb->pg_vec_len, req->tp_block_nr);
4245 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4246 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4247 tpacket_rcv : packet_rcv;
4248 skb_queue_purge(rb_queue);
4249 if (atomic_read(&po->mapped))
4250 pr_err("packet_mmap: vma is busy: %d\n",
4251 atomic_read(&po->mapped));
4253 mutex_unlock(&po->pg_vec_lock);
4255 spin_lock(&po->bind_lock);
4258 register_prot_hook(sk);
4260 spin_unlock(&po->bind_lock);
4261 if (closing && (po->tp_version > TPACKET_V2)) {
4262 /* Because we don't support block-based V3 on tx-ring */
4264 prb_shutdown_retire_blk_timer(po, rb_queue);
4268 free_pg_vec(pg_vec, order, req->tp_block_nr);
4274 static int packet_mmap(struct file *file, struct socket *sock,
4275 struct vm_area_struct *vma)
4277 struct sock *sk = sock->sk;
4278 struct packet_sock *po = pkt_sk(sk);
4279 unsigned long size, expected_size;
4280 struct packet_ring_buffer *rb;
4281 unsigned long start;
4288 mutex_lock(&po->pg_vec_lock);
4291 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4293 expected_size += rb->pg_vec_len
4299 if (expected_size == 0)
4302 size = vma->vm_end - vma->vm_start;
4303 if (size != expected_size)
4306 start = vma->vm_start;
4307 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4308 if (rb->pg_vec == NULL)
4311 for (i = 0; i < rb->pg_vec_len; i++) {
4313 void *kaddr = rb->pg_vec[i].buffer;
4316 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4317 page = pgv_to_page(kaddr);
4318 err = vm_insert_page(vma, start, page);
4327 atomic_inc(&po->mapped);
4328 vma->vm_ops = &packet_mmap_ops;
4332 mutex_unlock(&po->pg_vec_lock);
4336 static const struct proto_ops packet_ops_spkt = {
4337 .family = PF_PACKET,
4338 .owner = THIS_MODULE,
4339 .release = packet_release,
4340 .bind = packet_bind_spkt,
4341 .connect = sock_no_connect,
4342 .socketpair = sock_no_socketpair,
4343 .accept = sock_no_accept,
4344 .getname = packet_getname_spkt,
4345 .poll = datagram_poll,
4346 .ioctl = packet_ioctl,
4347 .listen = sock_no_listen,
4348 .shutdown = sock_no_shutdown,
4349 .setsockopt = sock_no_setsockopt,
4350 .getsockopt = sock_no_getsockopt,
4351 .sendmsg = packet_sendmsg_spkt,
4352 .recvmsg = packet_recvmsg,
4353 .mmap = sock_no_mmap,
4354 .sendpage = sock_no_sendpage,
4357 static const struct proto_ops packet_ops = {
4358 .family = PF_PACKET,
4359 .owner = THIS_MODULE,
4360 .release = packet_release,
4361 .bind = packet_bind,
4362 .connect = sock_no_connect,
4363 .socketpair = sock_no_socketpair,
4364 .accept = sock_no_accept,
4365 .getname = packet_getname,
4366 .poll = packet_poll,
4367 .ioctl = packet_ioctl,
4368 .listen = sock_no_listen,
4369 .shutdown = sock_no_shutdown,
4370 .setsockopt = packet_setsockopt,
4371 .getsockopt = packet_getsockopt,
4372 #ifdef CONFIG_COMPAT
4373 .compat_setsockopt = compat_packet_setsockopt,
4375 .sendmsg = packet_sendmsg,
4376 .recvmsg = packet_recvmsg,
4377 .mmap = packet_mmap,
4378 .sendpage = sock_no_sendpage,
4381 static const struct net_proto_family packet_family_ops = {
4382 .family = PF_PACKET,
4383 .create = packet_create,
4384 .owner = THIS_MODULE,
4387 static struct notifier_block packet_netdev_notifier = {
4388 .notifier_call = packet_notifier,
4391 #ifdef CONFIG_PROC_FS
4393 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4396 struct net *net = seq_file_net(seq);
4399 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4402 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4404 struct net *net = seq_file_net(seq);
4405 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4408 static void packet_seq_stop(struct seq_file *seq, void *v)
4414 static int packet_seq_show(struct seq_file *seq, void *v)
4416 if (v == SEQ_START_TOKEN)
4417 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4419 struct sock *s = sk_entry(v);
4420 const struct packet_sock *po = pkt_sk(s);
4423 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4425 atomic_read(&s->sk_refcnt),
4430 atomic_read(&s->sk_rmem_alloc),
4431 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4438 static const struct seq_operations packet_seq_ops = {
4439 .start = packet_seq_start,
4440 .next = packet_seq_next,
4441 .stop = packet_seq_stop,
4442 .show = packet_seq_show,
4445 static int packet_seq_open(struct inode *inode, struct file *file)
4447 return seq_open_net(inode, file, &packet_seq_ops,
4448 sizeof(struct seq_net_private));
4451 static const struct file_operations packet_seq_fops = {
4452 .owner = THIS_MODULE,
4453 .open = packet_seq_open,
4455 .llseek = seq_lseek,
4456 .release = seq_release_net,
4461 static int __net_init packet_net_init(struct net *net)
4463 mutex_init(&net->packet.sklist_lock);
4464 INIT_HLIST_HEAD(&net->packet.sklist);
4466 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4472 static void __net_exit packet_net_exit(struct net *net)
4474 remove_proc_entry("packet", net->proc_net);
4477 static struct pernet_operations packet_net_ops = {
4478 .init = packet_net_init,
4479 .exit = packet_net_exit,
4483 static void __exit packet_exit(void)
4485 unregister_netdevice_notifier(&packet_netdev_notifier);
4486 unregister_pernet_subsys(&packet_net_ops);
4487 sock_unregister(PF_PACKET);
4488 proto_unregister(&packet_proto);
4491 static int __init packet_init(void)
4493 int rc = proto_register(&packet_proto, 0);
4498 sock_register(&packet_family_ops);
4499 register_pernet_subsys(&packet_net_ops);
4500 register_netdevice_notifier(&packet_netdev_notifier);
4505 module_init(packet_init);
4506 module_exit(packet_exit);
4507 MODULE_LICENSE("GPL");
4508 MODULE_ALIAS_NETPROTO(PF_PACKET);