2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
113 Incoming, dev->hard_header!=NULL
114 mac_header -> ll header
117 Outgoing, dev->hard_header!=NULL
118 mac_header -> ll header
121 Incoming, dev->hard_header==NULL
122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
124 assymetry between rx and tx paths.
127 Outgoing, dev->hard_header==NULL
128 mac_header -> data. ll header is still not built!
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
138 dev->hard_header != NULL
139 mac_header -> ll header
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
150 /* Private packet socket structures. */
152 /* identical to struct packet_mreq except it has
153 * a longer address field.
155 struct packet_mreq_max {
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 #define PGV_FROM_VMALLOC 1
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
199 struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206 struct tpacket_block_desc *);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209 static void prb_init_blk_timer(struct packet_sock *,
210 struct tpacket_kbdq_core *,
211 void (*func) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 struct tpacket3_hdr *);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 struct tpacket3_hdr *);
217 static void packet_flush_mclist(struct sock *sk);
219 struct packet_skb_cb {
221 struct sockaddr_pkt pkt;
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
227 unsigned int origlen;
228 struct sockaddr_ll ll;
233 #define vio_le() virtio_legacy_is_little_endian()
235 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
237 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
249 static int packet_direct_xmit(struct sk_buff *skb)
251 struct net_device *dev = skb->dev;
252 netdev_features_t features;
253 struct netdev_queue *txq;
254 int ret = NETDEV_TX_BUSY;
256 if (unlikely(!netif_running(dev) ||
257 !netif_carrier_ok(dev)))
260 features = netif_skb_features(skb);
261 if (skb_needs_linearize(skb, features) &&
262 __skb_linearize(skb))
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
369 __fanout_unlink(sk, po);
371 __dev_remove_pack(&po->prot_hook);
376 spin_unlock(&po->bind_lock);
378 spin_lock(&po->bind_lock);
382 static void unregister_prot_hook(struct sock *sk, bool sync)
384 struct packet_sock *po = pkt_sk(sk);
387 __unregister_prot_hook(sk, sync);
390 static inline struct page * __pure pgv_to_page(void *addr)
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
399 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
413 WARN(1, "TPACKET version not supported.\n");
420 static int __packet_get_status(struct packet_sock *po, void *frame)
422 union tpacket_uhdr h;
427 switch (po->tp_version) {
429 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430 return h.h1->tp_status;
432 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433 return h.h2->tp_status;
436 WARN(1, "TPACKET version not supported.\n");
442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
445 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
448 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
452 if (ktime_to_timespec_cond(skb->tstamp, ts))
453 return TP_STATUS_TS_SOFTWARE;
458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
461 union tpacket_uhdr h;
465 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
469 switch (po->tp_version) {
471 h.h1->tp_sec = ts.tv_sec;
472 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
475 h.h2->tp_sec = ts.tv_sec;
476 h.h2->tp_nsec = ts.tv_nsec;
480 WARN(1, "TPACKET version not supported.\n");
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
491 static void *packet_lookup_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 unsigned int position,
496 unsigned int pg_vec_pos, frame_offset;
497 union tpacket_uhdr h;
499 pg_vec_pos = position / rb->frames_per_block;
500 frame_offset = position % rb->frames_per_block;
502 h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 (frame_offset * rb->frame_size);
505 if (status != __packet_get_status(po, h.raw))
511 static void *packet_current_frame(struct packet_sock *po,
512 struct packet_ring_buffer *rb,
515 return packet_lookup_frame(po, rb, rb->head, status);
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
520 del_timer_sync(&pkc->retire_blk_timer);
523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524 struct sk_buff_head *rb_queue)
526 struct tpacket_kbdq_core *pkc;
528 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
530 spin_lock_bh(&rb_queue->lock);
531 pkc->delete_blk_timer = 1;
532 spin_unlock_bh(&rb_queue->lock);
534 prb_del_retire_blk_timer(pkc);
537 static void prb_init_blk_timer(struct packet_sock *po,
538 struct tpacket_kbdq_core *pkc,
539 void (*func) (unsigned long))
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
549 struct tpacket_kbdq_core *pkc;
551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 int blk_size_in_bytes)
558 struct net_device *dev;
559 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560 struct ethtool_cmd ecmd;
565 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
566 if (unlikely(!dev)) {
568 return DEFAULT_PRB_RETIRE_TOV;
570 err = __ethtool_get_settings(dev, &ecmd);
571 speed = ethtool_cmd_speed(&ecmd);
575 * If the link speed is so slow you don't really
576 * need to worry about perf anyways
578 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
579 return DEFAULT_PRB_RETIRE_TOV;
586 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
598 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
599 union tpacket_req_u *req_u)
601 p1->feature_req_word = req_u->req3.tp_feature_req_word;
604 static void init_prb_bdqc(struct packet_sock *po,
605 struct packet_ring_buffer *rb,
607 union tpacket_req_u *req_u)
609 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
610 struct tpacket_block_desc *pbd;
612 memset(p1, 0x0, sizeof(*p1));
614 p1->knxt_seq_num = 1;
616 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
617 p1->pkblk_start = pg_vec[0].buffer;
618 p1->kblk_size = req_u->req3.tp_block_size;
619 p1->knum_blocks = req_u->req3.tp_block_nr;
620 p1->hdrlen = po->tp_hdrlen;
621 p1->version = po->tp_version;
622 p1->last_kactive_blk_num = 0;
623 po->stats.stats3.tp_freeze_q_cnt = 0;
624 if (req_u->req3.tp_retire_blk_tov)
625 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
627 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
628 req_u->req3.tp_block_size);
629 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
630 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
632 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
633 prb_init_ft_ops(p1, req_u);
634 prb_setup_retire_blk_timer(po);
635 prb_open_block(p1, pbd);
638 /* Do NOT update the last_blk_num first.
639 * Assumes sk_buff_head lock is held.
641 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
643 mod_timer(&pkc->retire_blk_timer,
644 jiffies + pkc->tov_in_jiffies);
645 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
650 * 1) We refresh the timer only when we open a block.
651 * By doing this we don't waste cycles refreshing the timer
652 * on packet-by-packet basis.
654 * With a 1MB block-size, on a 1Gbps line, it will take
655 * i) ~8 ms to fill a block + ii) memcpy etc.
656 * In this cut we are not accounting for the memcpy time.
658 * So, if the user sets the 'tmo' to 10ms then the timer
659 * will never fire while the block is still getting filled
660 * (which is what we want). However, the user could choose
661 * to close a block early and that's fine.
663 * But when the timer does fire, we check whether or not to refresh it.
664 * Since the tmo granularity is in msecs, it is not too expensive
665 * to refresh the timer, lets say every '8' msecs.
666 * Either the user can set the 'tmo' or we can derive it based on
667 * a) line-speed and b) block-size.
668 * prb_calc_retire_blk_tmo() calculates the tmo.
671 static void prb_retire_rx_blk_timer_expired(unsigned long data)
673 struct packet_sock *po = (struct packet_sock *)data;
674 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
676 struct tpacket_block_desc *pbd;
678 spin_lock(&po->sk.sk_receive_queue.lock);
680 frozen = prb_queue_frozen(pkc);
681 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
683 if (unlikely(pkc->delete_blk_timer))
686 /* We only need to plug the race when the block is partially filled.
688 * lock(); increment BLOCK_NUM_PKTS; unlock()
689 * copy_bits() is in progress ...
690 * timer fires on other cpu:
691 * we can't retire the current block because copy_bits
695 if (BLOCK_NUM_PKTS(pbd)) {
696 while (atomic_read(&pkc->blk_fill_in_prog)) {
697 /* Waiting for skb_copy_bits to finish... */
702 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
704 if (!BLOCK_NUM_PKTS(pbd)) {
705 /* An empty block. Just refresh the timer. */
708 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
709 if (!prb_dispatch_next_block(pkc, po))
714 /* Case 1. Queue was frozen because user-space was
717 if (prb_curr_blk_in_use(pkc, pbd)) {
719 * Ok, user-space is still behind.
720 * So just refresh the timer.
724 /* Case 2. queue was frozen,user-space caught up,
725 * now the link went idle && the timer fired.
726 * We don't have a block to close.So we open this
727 * block and restart the timer.
728 * opening a block thaws the queue,restarts timer
729 * Thawing/timer-refresh is a side effect.
731 prb_open_block(pkc, pbd);
738 _prb_refresh_rx_retire_blk_timer(pkc);
741 spin_unlock(&po->sk.sk_receive_queue.lock);
744 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
745 struct tpacket_block_desc *pbd1, __u32 status)
747 /* Flush everything minus the block header */
749 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
754 /* Skip the block header(we know header WILL fit in 4K) */
757 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
758 for (; start < end; start += PAGE_SIZE)
759 flush_dcache_page(pgv_to_page(start));
764 /* Now update the block status. */
766 BLOCK_STATUS(pbd1) = status;
768 /* Flush the block header */
770 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
772 flush_dcache_page(pgv_to_page(start));
782 * 2) Increment active_blk_num
784 * Note:We DONT refresh the timer on purpose.
785 * Because almost always the next block will be opened.
787 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
788 struct tpacket_block_desc *pbd1,
789 struct packet_sock *po, unsigned int stat)
791 __u32 status = TP_STATUS_USER | stat;
793 struct tpacket3_hdr *last_pkt;
794 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
795 struct sock *sk = &po->sk;
797 if (po->stats.stats3.tp_drops)
798 status |= TP_STATUS_LOSING;
800 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
801 last_pkt->tp_next_offset = 0;
803 /* Get the ts of the last pkt */
804 if (BLOCK_NUM_PKTS(pbd1)) {
805 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
806 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
808 /* Ok, we tmo'd - so get the current time.
810 * It shouldn't really happen as we don't close empty
811 * blocks. See prb_retire_rx_blk_timer_expired().
815 h1->ts_last_pkt.ts_sec = ts.tv_sec;
816 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
821 /* Flush the block */
822 prb_flush_block(pkc1, pbd1, status);
824 sk->sk_data_ready(sk);
826 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
829 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
831 pkc->reset_pending_on_curr_blk = 0;
835 * Side effect of opening a block:
837 * 1) prb_queue is thawed.
838 * 2) retire_blk_timer is refreshed.
841 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
842 struct tpacket_block_desc *pbd1)
845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
849 /* We could have just memset this but we will lose the
850 * flexibility of making the priv area sticky
853 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
854 BLOCK_NUM_PKTS(pbd1) = 0;
855 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
859 h1->ts_first_pkt.ts_sec = ts.tv_sec;
860 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
862 pkc1->pkblk_start = (char *)pbd1;
863 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
868 pbd1->version = pkc1->version;
869 pkc1->prev = pkc1->nxt_offset;
870 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
872 prb_thaw_queue(pkc1);
873 _prb_refresh_rx_retire_blk_timer(pkc1);
879 * Queue freeze logic:
880 * 1) Assume tp_block_nr = 8 blocks.
881 * 2) At time 't0', user opens Rx ring.
882 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
883 * 4) user-space is either sleeping or processing block '0'.
884 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
885 * it will close block-7,loop around and try to fill block '0'.
887 * __packet_lookup_frame_in_block
888 * prb_retire_current_block()
889 * prb_dispatch_next_block()
890 * |->(BLOCK_STATUS == USER) evaluates to true
891 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
892 * 6) Now there are two cases:
893 * 6.1) Link goes idle right after the queue is frozen.
894 * But remember, the last open_block() refreshed the timer.
895 * When this timer expires,it will refresh itself so that we can
896 * re-open block-0 in near future.
897 * 6.2) Link is busy and keeps on receiving packets. This is a simple
898 * case and __packet_lookup_frame_in_block will check if block-0
899 * is free and can now be re-used.
901 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
902 struct packet_sock *po)
904 pkc->reset_pending_on_curr_blk = 1;
905 po->stats.stats3.tp_freeze_q_cnt++;
908 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
911 * If the next block is free then we will dispatch it
912 * and return a good offset.
913 * Else, we will freeze the queue.
914 * So, caller must check the return value.
916 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
917 struct packet_sock *po)
919 struct tpacket_block_desc *pbd;
923 /* 1. Get current block num */
924 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
926 /* 2. If this block is currently in_use then freeze the queue */
927 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
928 prb_freeze_queue(pkc, po);
934 * open this block and return the offset where the first packet
935 * needs to get stored.
937 prb_open_block(pkc, pbd);
938 return (void *)pkc->nxt_offset;
941 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
942 struct packet_sock *po, unsigned int status)
944 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
946 /* retire/close the current block */
947 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
949 * Plug the case where copy_bits() is in progress on
950 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
951 * have space to copy the pkt in the current block and
952 * called prb_retire_current_block()
954 * We don't need to worry about the TMO case because
955 * the timer-handler already handled this case.
957 if (!(status & TP_STATUS_BLK_TMO)) {
958 while (atomic_read(&pkc->blk_fill_in_prog)) {
959 /* Waiting for skb_copy_bits to finish... */
963 prb_close_block(pkc, pbd, po, status);
968 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
969 struct tpacket_block_desc *pbd)
971 return TP_STATUS_USER & BLOCK_STATUS(pbd);
974 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
976 return pkc->reset_pending_on_curr_blk;
979 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
981 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
982 atomic_dec(&pkc->blk_fill_in_prog);
985 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
986 struct tpacket3_hdr *ppd)
988 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
991 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
992 struct tpacket3_hdr *ppd)
994 ppd->hv1.tp_rxhash = 0;
997 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
998 struct tpacket3_hdr *ppd)
1000 if (skb_vlan_tag_present(pkc->skb)) {
1001 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1002 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1003 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1005 ppd->hv1.tp_vlan_tci = 0;
1006 ppd->hv1.tp_vlan_tpid = 0;
1007 ppd->tp_status = TP_STATUS_AVAILABLE;
1011 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1012 struct tpacket3_hdr *ppd)
1014 ppd->hv1.tp_padding = 0;
1015 prb_fill_vlan_info(pkc, ppd);
1017 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1018 prb_fill_rxhash(pkc, ppd);
1020 prb_clear_rxhash(pkc, ppd);
1023 static void prb_fill_curr_block(char *curr,
1024 struct tpacket_kbdq_core *pkc,
1025 struct tpacket_block_desc *pbd,
1028 struct tpacket3_hdr *ppd;
1030 ppd = (struct tpacket3_hdr *)curr;
1031 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1035 BLOCK_NUM_PKTS(pbd) += 1;
1036 atomic_inc(&pkc->blk_fill_in_prog);
1037 prb_run_all_ft_ops(pkc, ppd);
1040 /* Assumes caller has the sk->rx_queue.lock */
1041 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1042 struct sk_buff *skb,
1047 struct tpacket_kbdq_core *pkc;
1048 struct tpacket_block_desc *pbd;
1051 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1052 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1054 /* Queue is frozen when user space is lagging behind */
1055 if (prb_queue_frozen(pkc)) {
1057 * Check if that last block which caused the queue to freeze,
1058 * is still in_use by user-space.
1060 if (prb_curr_blk_in_use(pkc, pbd)) {
1061 /* Can't record this packet */
1065 * Ok, the block was released by user-space.
1066 * Now let's open that block.
1067 * opening a block also thaws the queue.
1068 * Thawing is a side effect.
1070 prb_open_block(pkc, pbd);
1075 curr = pkc->nxt_offset;
1077 end = (char *)pbd + pkc->kblk_size;
1079 /* first try the current block */
1080 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1081 prb_fill_curr_block(curr, pkc, pbd, len);
1082 return (void *)curr;
1085 /* Ok, close the current block */
1086 prb_retire_current_block(pkc, po, 0);
1088 /* Now, try to dispatch the next block */
1089 curr = (char *)prb_dispatch_next_block(pkc, po);
1091 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1092 prb_fill_curr_block(curr, pkc, pbd, len);
1093 return (void *)curr;
1097 * No free blocks are available.user_space hasn't caught up yet.
1098 * Queue was just frozen and now this packet will get dropped.
1103 static void *packet_current_rx_frame(struct packet_sock *po,
1104 struct sk_buff *skb,
1105 int status, unsigned int len)
1108 switch (po->tp_version) {
1111 curr = packet_lookup_frame(po, &po->rx_ring,
1112 po->rx_ring.head, status);
1115 return __packet_lookup_frame_in_block(po, skb, status, len);
1117 WARN(1, "TPACKET version not supported\n");
1123 static void *prb_lookup_block(struct packet_sock *po,
1124 struct packet_ring_buffer *rb,
1128 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1129 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1131 if (status != BLOCK_STATUS(pbd))
1136 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1139 if (rb->prb_bdqc.kactive_blk_num)
1140 prev = rb->prb_bdqc.kactive_blk_num-1;
1142 prev = rb->prb_bdqc.knum_blocks-1;
1146 /* Assumes caller has held the rx_queue.lock */
1147 static void *__prb_previous_block(struct packet_sock *po,
1148 struct packet_ring_buffer *rb,
1151 unsigned int previous = prb_previous_blk_num(rb);
1152 return prb_lookup_block(po, rb, previous, status);
1155 static void *packet_previous_rx_frame(struct packet_sock *po,
1156 struct packet_ring_buffer *rb,
1159 if (po->tp_version <= TPACKET_V2)
1160 return packet_previous_frame(po, rb, status);
1162 return __prb_previous_block(po, rb, status);
1165 static void packet_increment_rx_head(struct packet_sock *po,
1166 struct packet_ring_buffer *rb)
1168 switch (po->tp_version) {
1171 return packet_increment_head(rb);
1174 WARN(1, "TPACKET version not supported.\n");
1180 static void *packet_previous_frame(struct packet_sock *po,
1181 struct packet_ring_buffer *rb,
1184 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1185 return packet_lookup_frame(po, rb, previous, status);
1188 static void packet_increment_head(struct packet_ring_buffer *buff)
1190 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1193 static void packet_inc_pending(struct packet_ring_buffer *rb)
1195 this_cpu_inc(*rb->pending_refcnt);
1198 static void packet_dec_pending(struct packet_ring_buffer *rb)
1200 this_cpu_dec(*rb->pending_refcnt);
1203 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1205 unsigned int refcnt = 0;
1208 /* We don't use pending refcount in rx_ring. */
1209 if (rb->pending_refcnt == NULL)
1212 for_each_possible_cpu(cpu)
1213 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1218 static int packet_alloc_pending(struct packet_sock *po)
1220 po->rx_ring.pending_refcnt = NULL;
1222 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1223 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1229 static void packet_free_pending(struct packet_sock *po)
1231 free_percpu(po->tx_ring.pending_refcnt);
1234 #define ROOM_POW_OFF 2
1235 #define ROOM_NONE 0x0
1236 #define ROOM_LOW 0x1
1237 #define ROOM_NORMAL 0x2
1239 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1243 len = po->rx_ring.frame_max + 1;
1244 idx = po->rx_ring.head;
1246 idx += len >> pow_off;
1249 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1252 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1256 len = po->rx_ring.prb_bdqc.knum_blocks;
1257 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1259 idx += len >> pow_off;
1262 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1265 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1267 struct sock *sk = &po->sk;
1268 int ret = ROOM_NONE;
1270 if (po->prot_hook.func != tpacket_rcv) {
1271 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1272 - (skb ? skb->truesize : 0);
1273 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1281 if (po->tp_version == TPACKET_V3) {
1282 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1284 else if (__tpacket_v3_has_room(po, 0))
1287 if (__tpacket_has_room(po, ROOM_POW_OFF))
1289 else if (__tpacket_has_room(po, 0))
1296 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1301 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1302 ret = __packet_rcv_has_room(po, skb);
1303 has_room = ret == ROOM_NORMAL;
1304 if (po->pressure == has_room)
1305 po->pressure = !has_room;
1306 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1311 static void packet_sock_destruct(struct sock *sk)
1313 skb_queue_purge(&sk->sk_error_queue);
1315 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1316 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1318 if (!sock_flag(sk, SOCK_DEAD)) {
1319 pr_err("Attempt to release alive packet socket: %p\n", sk);
1323 sk_refcnt_debug_dec(sk);
1326 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1331 rxhash = skb_get_hash(skb);
1332 for (i = 0; i < ROLLOVER_HLEN; i++)
1333 if (po->rollover->history[i] == rxhash)
1336 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1337 return count > (ROLLOVER_HLEN >> 1);
1340 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1341 struct sk_buff *skb,
1344 return reciprocal_scale(skb_get_hash(skb), num);
1347 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1348 struct sk_buff *skb,
1351 unsigned int val = atomic_inc_return(&f->rr_cur);
1356 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1357 struct sk_buff *skb,
1360 return smp_processor_id() % num;
1363 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1364 struct sk_buff *skb,
1367 return prandom_u32_max(num);
1370 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1371 struct sk_buff *skb,
1372 unsigned int idx, bool try_self,
1375 struct packet_sock *po, *po_next, *po_skip = NULL;
1376 unsigned int i, j, room = ROOM_NONE;
1378 po = pkt_sk(f->arr[idx]);
1381 room = packet_rcv_has_room(po, skb);
1382 if (room == ROOM_NORMAL ||
1383 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1388 i = j = min_t(int, po->rollover->sock, num - 1);
1390 po_next = pkt_sk(f->arr[i]);
1391 if (po_next != po_skip && !po_next->pressure &&
1392 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1394 po->rollover->sock = i;
1395 atomic_long_inc(&po->rollover->num);
1396 if (room == ROOM_LOW)
1397 atomic_long_inc(&po->rollover->num_huge);
1405 atomic_long_inc(&po->rollover->num_failed);
1409 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1410 struct sk_buff *skb,
1413 return skb_get_queue_mapping(skb) % num;
1416 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1417 struct sk_buff *skb,
1420 struct bpf_prog *prog;
1421 unsigned int ret = 0;
1424 prog = rcu_dereference(f->bpf_prog);
1426 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1432 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1434 return f->flags & (flag >> 8);
1437 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1438 struct packet_type *pt, struct net_device *orig_dev)
1440 struct packet_fanout *f = pt->af_packet_priv;
1441 unsigned int num = READ_ONCE(f->num_members);
1442 struct net *net = read_pnet(&f->net);
1443 struct packet_sock *po;
1446 if (!net_eq(dev_net(dev), net) || !num) {
1451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1452 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1457 case PACKET_FANOUT_HASH:
1459 idx = fanout_demux_hash(f, skb, num);
1461 case PACKET_FANOUT_LB:
1462 idx = fanout_demux_lb(f, skb, num);
1464 case PACKET_FANOUT_CPU:
1465 idx = fanout_demux_cpu(f, skb, num);
1467 case PACKET_FANOUT_RND:
1468 idx = fanout_demux_rnd(f, skb, num);
1470 case PACKET_FANOUT_QM:
1471 idx = fanout_demux_qm(f, skb, num);
1473 case PACKET_FANOUT_ROLLOVER:
1474 idx = fanout_demux_rollover(f, skb, 0, false, num);
1476 case PACKET_FANOUT_CBPF:
1477 case PACKET_FANOUT_EBPF:
1478 idx = fanout_demux_bpf(f, skb, num);
1482 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1483 idx = fanout_demux_rollover(f, skb, idx, true, num);
1485 po = pkt_sk(f->arr[idx]);
1486 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1489 DEFINE_MUTEX(fanout_mutex);
1490 EXPORT_SYMBOL_GPL(fanout_mutex);
1491 static LIST_HEAD(fanout_list);
1493 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1495 struct packet_fanout *f = po->fanout;
1497 spin_lock(&f->lock);
1498 f->arr[f->num_members] = sk;
1501 spin_unlock(&f->lock);
1504 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1506 struct packet_fanout *f = po->fanout;
1509 spin_lock(&f->lock);
1510 for (i = 0; i < f->num_members; i++) {
1511 if (f->arr[i] == sk)
1514 BUG_ON(i >= f->num_members);
1515 f->arr[i] = f->arr[f->num_members - 1];
1517 spin_unlock(&f->lock);
1520 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1522 if (sk->sk_family != PF_PACKET)
1525 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1528 static void fanout_init_data(struct packet_fanout *f)
1531 case PACKET_FANOUT_LB:
1532 atomic_set(&f->rr_cur, 0);
1534 case PACKET_FANOUT_CBPF:
1535 case PACKET_FANOUT_EBPF:
1536 RCU_INIT_POINTER(f->bpf_prog, NULL);
1541 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1543 struct bpf_prog *old;
1545 spin_lock(&f->lock);
1546 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1547 rcu_assign_pointer(f->bpf_prog, new);
1548 spin_unlock(&f->lock);
1552 bpf_prog_destroy(old);
1556 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1559 struct bpf_prog *new;
1560 struct sock_fprog fprog;
1563 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1565 if (len != sizeof(fprog))
1567 if (copy_from_user(&fprog, data, len))
1570 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1574 __fanout_set_data_bpf(po->fanout, new);
1578 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1581 struct bpf_prog *new;
1584 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1586 if (len != sizeof(fd))
1588 if (copy_from_user(&fd, data, len))
1591 new = bpf_prog_get(fd);
1593 return PTR_ERR(new);
1594 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
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;
1846 * Get and verify the address.
1850 if (msg->msg_namelen < sizeof(struct sockaddr))
1852 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1853 proto = saddr->spkt_protocol;
1855 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1858 * Find the device first to size check it
1861 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1864 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1870 if (!(dev->flags & IFF_UP))
1874 * You may not queue a frame bigger than the mtu. This is the lowest level
1875 * raw protocol and you must do your own fragmentation at this level.
1878 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1879 if (!netif_supports_nofcs(dev)) {
1880 err = -EPROTONOSUPPORT;
1883 extra_len = 4; /* We're doing our own CRC */
1887 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1891 size_t reserved = LL_RESERVED_SPACE(dev);
1892 int tlen = dev->needed_tailroom;
1893 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1896 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1899 /* FIXME: Save some space for broken drivers that write a hard
1900 * header at transmission time by themselves. PPP is the notable
1901 * one here. This should really be fixed at the driver level.
1903 skb_reserve(skb, reserved);
1904 skb_reset_network_header(skb);
1906 /* Try to align data part correctly */
1911 skb_reset_network_header(skb);
1913 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1919 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1920 !packet_extra_vlan_len_allowed(dev, skb)) {
1925 skb->protocol = proto;
1927 skb->priority = sk->sk_priority;
1928 skb->mark = sk->sk_mark;
1930 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1932 if (unlikely(extra_len == 4))
1935 skb_probe_transport_header(skb, 0);
1937 dev_queue_xmit(skb);
1948 static unsigned int run_filter(struct sk_buff *skb,
1949 const struct sock *sk,
1952 struct sk_filter *filter;
1955 filter = rcu_dereference(sk->sk_filter);
1957 res = bpf_prog_run_clear_cb(filter->prog, skb);
1964 * This function makes lazy skb cloning in hope that most of packets
1965 * are discarded by BPF.
1967 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1968 * and skb->cb are mangled. It works because (and until) packets
1969 * falling here are owned by current CPU. Output packets are cloned
1970 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1971 * sequencially, so that if we return skb to original state on exit,
1972 * we will not harm anyone.
1975 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1976 struct packet_type *pt, struct net_device *orig_dev)
1979 struct sockaddr_ll *sll;
1980 struct packet_sock *po;
1981 u8 *skb_head = skb->data;
1982 int skb_len = skb->len;
1983 unsigned int snaplen, res;
1985 if (skb->pkt_type == PACKET_LOOPBACK)
1988 sk = pt->af_packet_priv;
1991 if (!net_eq(dev_net(dev), sock_net(sk)))
1996 if (dev->header_ops) {
1997 /* The device has an explicit notion of ll header,
1998 * exported to higher levels.
2000 * Otherwise, the device hides details of its frame
2001 * structure, so that corresponding packet head is
2002 * never delivered to user.
2004 if (sk->sk_type != SOCK_DGRAM)
2005 skb_push(skb, skb->data - skb_mac_header(skb));
2006 else if (skb->pkt_type == PACKET_OUTGOING) {
2007 /* Special case: outgoing packets have ll header at head */
2008 skb_pull(skb, skb_network_offset(skb));
2014 res = run_filter(skb, sk, snaplen);
2016 goto drop_n_restore;
2020 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2023 if (skb_shared(skb)) {
2024 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2028 if (skb_head != skb->data) {
2029 skb->data = skb_head;
2036 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2038 sll = &PACKET_SKB_CB(skb)->sa.ll;
2039 sll->sll_hatype = dev->type;
2040 sll->sll_pkttype = skb->pkt_type;
2041 if (unlikely(po->origdev))
2042 sll->sll_ifindex = orig_dev->ifindex;
2044 sll->sll_ifindex = dev->ifindex;
2046 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2048 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2049 * Use their space for storing the original skb length.
2051 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2053 if (pskb_trim(skb, snaplen))
2056 skb_set_owner_r(skb, sk);
2060 /* drop conntrack reference */
2063 spin_lock(&sk->sk_receive_queue.lock);
2064 po->stats.stats1.tp_packets++;
2065 sock_skb_set_dropcount(sk, skb);
2066 __skb_queue_tail(&sk->sk_receive_queue, skb);
2067 spin_unlock(&sk->sk_receive_queue.lock);
2068 sk->sk_data_ready(sk);
2072 spin_lock(&sk->sk_receive_queue.lock);
2073 po->stats.stats1.tp_drops++;
2074 atomic_inc(&sk->sk_drops);
2075 spin_unlock(&sk->sk_receive_queue.lock);
2078 if (skb_head != skb->data && skb_shared(skb)) {
2079 skb->data = skb_head;
2087 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2088 struct packet_type *pt, struct net_device *orig_dev)
2091 struct packet_sock *po;
2092 struct sockaddr_ll *sll;
2093 union tpacket_uhdr h;
2094 u8 *skb_head = skb->data;
2095 int skb_len = skb->len;
2096 unsigned int snaplen, res;
2097 unsigned long status = TP_STATUS_USER;
2098 unsigned short macoff, netoff, hdrlen;
2099 struct sk_buff *copy_skb = NULL;
2103 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2104 * We may add members to them until current aligned size without forcing
2105 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2107 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2108 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2110 if (skb->pkt_type == PACKET_LOOPBACK)
2113 sk = pt->af_packet_priv;
2116 if (!net_eq(dev_net(dev), sock_net(sk)))
2119 if (dev->header_ops) {
2120 if (sk->sk_type != SOCK_DGRAM)
2121 skb_push(skb, skb->data - skb_mac_header(skb));
2122 else if (skb->pkt_type == PACKET_OUTGOING) {
2123 /* Special case: outgoing packets have ll header at head */
2124 skb_pull(skb, skb_network_offset(skb));
2130 res = run_filter(skb, sk, snaplen);
2132 goto drop_n_restore;
2134 if (skb->ip_summed == CHECKSUM_PARTIAL)
2135 status |= TP_STATUS_CSUMNOTREADY;
2136 else if (skb->pkt_type != PACKET_OUTGOING &&
2137 (skb->ip_summed == CHECKSUM_COMPLETE ||
2138 skb_csum_unnecessary(skb)))
2139 status |= TP_STATUS_CSUM_VALID;
2144 if (sk->sk_type == SOCK_DGRAM) {
2145 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2148 unsigned int maclen = skb_network_offset(skb);
2149 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2150 (maclen < 16 ? 16 : maclen)) +
2152 macoff = netoff - maclen;
2154 if (po->tp_version <= TPACKET_V2) {
2155 if (macoff + snaplen > po->rx_ring.frame_size) {
2156 if (po->copy_thresh &&
2157 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2158 if (skb_shared(skb)) {
2159 copy_skb = skb_clone(skb, GFP_ATOMIC);
2161 copy_skb = skb_get(skb);
2162 skb_head = skb->data;
2165 skb_set_owner_r(copy_skb, sk);
2167 snaplen = po->rx_ring.frame_size - macoff;
2168 if ((int)snaplen < 0)
2171 } else if (unlikely(macoff + snaplen >
2172 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2175 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2176 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2177 snaplen, nval, macoff);
2179 if (unlikely((int)snaplen < 0)) {
2181 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2184 spin_lock(&sk->sk_receive_queue.lock);
2185 h.raw = packet_current_rx_frame(po, skb,
2186 TP_STATUS_KERNEL, (macoff+snaplen));
2189 if (po->tp_version <= TPACKET_V2) {
2190 packet_increment_rx_head(po, &po->rx_ring);
2192 * LOSING will be reported till you read the stats,
2193 * because it's COR - Clear On Read.
2194 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2197 if (po->stats.stats1.tp_drops)
2198 status |= TP_STATUS_LOSING;
2200 po->stats.stats1.tp_packets++;
2202 status |= TP_STATUS_COPY;
2203 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2205 spin_unlock(&sk->sk_receive_queue.lock);
2207 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2209 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2210 getnstimeofday(&ts);
2212 status |= ts_status;
2214 switch (po->tp_version) {
2216 h.h1->tp_len = skb->len;
2217 h.h1->tp_snaplen = snaplen;
2218 h.h1->tp_mac = macoff;
2219 h.h1->tp_net = netoff;
2220 h.h1->tp_sec = ts.tv_sec;
2221 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2222 hdrlen = sizeof(*h.h1);
2225 h.h2->tp_len = skb->len;
2226 h.h2->tp_snaplen = snaplen;
2227 h.h2->tp_mac = macoff;
2228 h.h2->tp_net = netoff;
2229 h.h2->tp_sec = ts.tv_sec;
2230 h.h2->tp_nsec = ts.tv_nsec;
2231 if (skb_vlan_tag_present(skb)) {
2232 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2233 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2234 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2236 h.h2->tp_vlan_tci = 0;
2237 h.h2->tp_vlan_tpid = 0;
2239 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2240 hdrlen = sizeof(*h.h2);
2243 /* tp_nxt_offset,vlan are already populated above.
2244 * So DONT clear those fields here
2246 h.h3->tp_status |= status;
2247 h.h3->tp_len = skb->len;
2248 h.h3->tp_snaplen = snaplen;
2249 h.h3->tp_mac = macoff;
2250 h.h3->tp_net = netoff;
2251 h.h3->tp_sec = ts.tv_sec;
2252 h.h3->tp_nsec = ts.tv_nsec;
2253 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2254 hdrlen = sizeof(*h.h3);
2260 sll = h.raw + TPACKET_ALIGN(hdrlen);
2261 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2262 sll->sll_family = AF_PACKET;
2263 sll->sll_hatype = dev->type;
2264 sll->sll_protocol = skb->protocol;
2265 sll->sll_pkttype = skb->pkt_type;
2266 if (unlikely(po->origdev))
2267 sll->sll_ifindex = orig_dev->ifindex;
2269 sll->sll_ifindex = dev->ifindex;
2273 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2274 if (po->tp_version <= TPACKET_V2) {
2277 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2280 for (start = h.raw; start < end; start += PAGE_SIZE)
2281 flush_dcache_page(pgv_to_page(start));
2286 if (po->tp_version <= TPACKET_V2) {
2287 __packet_set_status(po, h.raw, status);
2288 sk->sk_data_ready(sk);
2290 prb_clear_blk_fill_status(&po->rx_ring);
2294 if (skb_head != skb->data && skb_shared(skb)) {
2295 skb->data = skb_head;
2303 po->stats.stats1.tp_drops++;
2304 spin_unlock(&sk->sk_receive_queue.lock);
2306 sk->sk_data_ready(sk);
2307 kfree_skb(copy_skb);
2308 goto drop_n_restore;
2311 static void tpacket_destruct_skb(struct sk_buff *skb)
2313 struct packet_sock *po = pkt_sk(skb->sk);
2315 if (likely(po->tx_ring.pg_vec)) {
2319 ph = skb_shinfo(skb)->destructor_arg;
2320 packet_dec_pending(&po->tx_ring);
2322 ts = __packet_set_timestamp(po, ph, skb);
2323 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2329 static bool ll_header_truncated(const struct net_device *dev, int len)
2331 /* net device doesn't like empty head */
2332 if (unlikely(len <= dev->hard_header_len)) {
2333 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2334 current->comm, len, dev->hard_header_len);
2341 static void tpacket_set_protocol(const struct net_device *dev,
2342 struct sk_buff *skb)
2344 if (dev->type == ARPHRD_ETHER) {
2345 skb_reset_mac_header(skb);
2346 skb->protocol = eth_hdr(skb)->h_proto;
2350 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2351 void *frame, struct net_device *dev, int size_max,
2352 __be16 proto, unsigned char *addr, int hlen)
2354 union tpacket_uhdr ph;
2355 int to_write, offset, len, tp_len, nr_frags, len_max;
2356 struct socket *sock = po->sk.sk_socket;
2363 skb->protocol = proto;
2365 skb->priority = po->sk.sk_priority;
2366 skb->mark = po->sk.sk_mark;
2367 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2368 skb_shinfo(skb)->destructor_arg = ph.raw;
2370 switch (po->tp_version) {
2372 tp_len = ph.h2->tp_len;
2375 tp_len = ph.h1->tp_len;
2378 if (unlikely(tp_len > size_max)) {
2379 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2383 skb_reserve(skb, hlen);
2384 skb_reset_network_header(skb);
2386 if (unlikely(po->tp_tx_has_off)) {
2387 int off_min, off_max, off;
2388 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2389 off_max = po->tx_ring.frame_size - tp_len;
2390 if (sock->type == SOCK_DGRAM) {
2391 switch (po->tp_version) {
2393 off = ph.h2->tp_net;
2396 off = ph.h1->tp_net;
2400 switch (po->tp_version) {
2402 off = ph.h2->tp_mac;
2405 off = ph.h1->tp_mac;
2409 if (unlikely((off < off_min) || (off_max < off)))
2411 data = ph.raw + off;
2413 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2417 if (sock->type == SOCK_DGRAM) {
2418 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2420 if (unlikely(err < 0))
2422 } else if (dev->hard_header_len) {
2423 if (ll_header_truncated(dev, tp_len))
2426 skb_push(skb, dev->hard_header_len);
2427 err = skb_store_bits(skb, 0, data,
2428 dev->hard_header_len);
2432 tpacket_set_protocol(dev, skb);
2434 data += dev->hard_header_len;
2435 to_write -= dev->hard_header_len;
2438 offset = offset_in_page(data);
2439 len_max = PAGE_SIZE - offset;
2440 len = ((to_write > len_max) ? len_max : to_write);
2442 skb->data_len = to_write;
2443 skb->len += to_write;
2444 skb->truesize += to_write;
2445 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2447 while (likely(to_write)) {
2448 nr_frags = skb_shinfo(skb)->nr_frags;
2450 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2451 pr_err("Packet exceed the number of skb frags(%lu)\n",
2456 page = pgv_to_page(data);
2458 flush_dcache_page(page);
2460 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2463 len_max = PAGE_SIZE;
2464 len = ((to_write > len_max) ? len_max : to_write);
2467 skb_probe_transport_header(skb, 0);
2472 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2474 struct sk_buff *skb;
2475 struct net_device *dev;
2477 int err, reserve = 0;
2479 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2480 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2481 int tp_len, size_max;
2482 unsigned char *addr;
2484 int status = TP_STATUS_AVAILABLE;
2487 mutex_lock(&po->pg_vec_lock);
2489 if (likely(saddr == NULL)) {
2490 dev = packet_cached_dev_get(po);
2495 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2497 if (msg->msg_namelen < (saddr->sll_halen
2498 + offsetof(struct sockaddr_ll,
2501 proto = saddr->sll_protocol;
2502 addr = saddr->sll_addr;
2503 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2507 if (unlikely(dev == NULL))
2510 if (unlikely(!(dev->flags & IFF_UP)))
2513 reserve = dev->hard_header_len + VLAN_HLEN;
2514 size_max = po->tx_ring.frame_size
2515 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2517 if (size_max > dev->mtu + reserve)
2518 size_max = dev->mtu + reserve;
2521 ph = packet_current_frame(po, &po->tx_ring,
2522 TP_STATUS_SEND_REQUEST);
2523 if (unlikely(ph == NULL)) {
2524 if (need_wait && need_resched())
2529 status = TP_STATUS_SEND_REQUEST;
2530 hlen = LL_RESERVED_SPACE(dev);
2531 tlen = dev->needed_tailroom;
2532 skb = sock_alloc_send_skb(&po->sk,
2533 hlen + tlen + sizeof(struct sockaddr_ll),
2536 if (unlikely(skb == NULL)) {
2537 /* we assume the socket was initially writeable ... */
2538 if (likely(len_sum > 0))
2542 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2544 if (likely(tp_len >= 0) &&
2545 tp_len > dev->mtu + dev->hard_header_len &&
2546 !packet_extra_vlan_len_allowed(dev, skb))
2549 if (unlikely(tp_len < 0)) {
2551 __packet_set_status(po, ph,
2552 TP_STATUS_AVAILABLE);
2553 packet_increment_head(&po->tx_ring);
2557 status = TP_STATUS_WRONG_FORMAT;
2563 packet_pick_tx_queue(dev, skb);
2565 skb->destructor = tpacket_destruct_skb;
2566 __packet_set_status(po, ph, TP_STATUS_SENDING);
2567 packet_inc_pending(&po->tx_ring);
2569 status = TP_STATUS_SEND_REQUEST;
2570 err = po->xmit(skb);
2571 if (unlikely(err > 0)) {
2572 err = net_xmit_errno(err);
2573 if (err && __packet_get_status(po, ph) ==
2574 TP_STATUS_AVAILABLE) {
2575 /* skb was destructed already */
2580 * skb was dropped but not destructed yet;
2581 * let's treat it like congestion or err < 0
2585 packet_increment_head(&po->tx_ring);
2587 } while (likely((ph != NULL) ||
2588 /* Note: packet_read_pending() might be slow if we have
2589 * to call it as it's per_cpu variable, but in fast-path
2590 * we already short-circuit the loop with the first
2591 * condition, and luckily don't have to go that path
2594 (need_wait && packet_read_pending(&po->tx_ring))));
2600 __packet_set_status(po, ph, status);
2605 mutex_unlock(&po->pg_vec_lock);
2609 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2610 size_t reserve, size_t len,
2611 size_t linear, int noblock,
2614 struct sk_buff *skb;
2616 /* Under a page? Don't bother with paged skb. */
2617 if (prepad + len < PAGE_SIZE || !linear)
2620 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2625 skb_reserve(skb, reserve);
2626 skb_put(skb, linear);
2627 skb->data_len = len - linear;
2628 skb->len += len - linear;
2633 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2635 struct sock *sk = sock->sk;
2636 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2637 struct sk_buff *skb;
2638 struct net_device *dev;
2640 unsigned char *addr;
2641 int err, reserve = 0;
2642 struct sockcm_cookie sockc;
2643 struct virtio_net_hdr vnet_hdr = { 0 };
2646 struct packet_sock *po = pkt_sk(sk);
2647 unsigned short gso_type = 0;
2653 * Get and verify the address.
2656 if (likely(saddr == NULL)) {
2657 dev = packet_cached_dev_get(po);
2662 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2664 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2666 proto = saddr->sll_protocol;
2667 addr = saddr->sll_addr;
2668 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2672 if (unlikely(dev == NULL))
2675 if (unlikely(!(dev->flags & IFF_UP)))
2678 sockc.mark = sk->sk_mark;
2679 if (msg->msg_controllen) {
2680 err = sock_cmsg_send(sk, msg, &sockc);
2685 if (sock->type == SOCK_RAW)
2686 reserve = dev->hard_header_len;
2687 if (po->has_vnet_hdr) {
2688 vnet_hdr_len = sizeof(vnet_hdr);
2691 if (len < vnet_hdr_len)
2694 len -= vnet_hdr_len;
2697 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2698 if (n != vnet_hdr_len)
2701 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2702 (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2703 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2704 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2705 vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2706 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2707 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2710 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2713 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2714 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2715 case VIRTIO_NET_HDR_GSO_TCPV4:
2716 gso_type = SKB_GSO_TCPV4;
2718 case VIRTIO_NET_HDR_GSO_TCPV6:
2719 gso_type = SKB_GSO_TCPV6;
2721 case VIRTIO_NET_HDR_GSO_UDP:
2722 gso_type = SKB_GSO_UDP;
2728 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2729 gso_type |= SKB_GSO_TCP_ECN;
2731 if (vnet_hdr.gso_size == 0)
2737 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2738 if (!netif_supports_nofcs(dev)) {
2739 err = -EPROTONOSUPPORT;
2742 extra_len = 4; /* We're doing our own CRC */
2746 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2750 hlen = LL_RESERVED_SPACE(dev);
2751 tlen = dev->needed_tailroom;
2752 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2753 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2754 msg->msg_flags & MSG_DONTWAIT, &err);
2758 skb_set_network_header(skb, reserve);
2761 if (sock->type == SOCK_DGRAM) {
2762 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2763 if (unlikely(offset < 0))
2766 if (ll_header_truncated(dev, len))
2770 /* Returns -EFAULT on error */
2771 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2775 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2777 if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2778 !packet_extra_vlan_len_allowed(dev, skb)) {
2783 skb->protocol = proto;
2785 skb->priority = sk->sk_priority;
2786 skb->mark = sockc.mark;
2788 packet_pick_tx_queue(dev, skb);
2790 if (po->has_vnet_hdr) {
2791 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2792 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2793 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2794 if (!skb_partial_csum_set(skb, s, o)) {
2800 skb_shinfo(skb)->gso_size =
2801 __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2802 skb_shinfo(skb)->gso_type = gso_type;
2804 /* Header must be checked, and gso_segs computed. */
2805 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2806 skb_shinfo(skb)->gso_segs = 0;
2808 len += vnet_hdr_len;
2811 skb_probe_transport_header(skb, reserve);
2813 if (unlikely(extra_len == 4))
2816 err = po->xmit(skb);
2817 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2833 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2835 struct sock *sk = sock->sk;
2836 struct packet_sock *po = pkt_sk(sk);
2838 if (po->tx_ring.pg_vec)
2839 return tpacket_snd(po, msg);
2841 return packet_snd(sock, msg, len);
2845 * Close a PACKET socket. This is fairly simple. We immediately go
2846 * to 'closed' state and remove our protocol entry in the device list.
2849 static int packet_release(struct socket *sock)
2851 struct sock *sk = sock->sk;
2852 struct packet_sock *po;
2854 union tpacket_req_u req_u;
2862 mutex_lock(&net->packet.sklist_lock);
2863 sk_del_node_init_rcu(sk);
2864 mutex_unlock(&net->packet.sklist_lock);
2867 sock_prot_inuse_add(net, sk->sk_prot, -1);
2870 spin_lock(&po->bind_lock);
2871 unregister_prot_hook(sk, false);
2872 packet_cached_dev_reset(po);
2874 if (po->prot_hook.dev) {
2875 dev_put(po->prot_hook.dev);
2876 po->prot_hook.dev = NULL;
2878 spin_unlock(&po->bind_lock);
2880 packet_flush_mclist(sk);
2882 if (po->rx_ring.pg_vec) {
2883 memset(&req_u, 0, sizeof(req_u));
2884 packet_set_ring(sk, &req_u, 1, 0);
2887 if (po->tx_ring.pg_vec) {
2888 memset(&req_u, 0, sizeof(req_u));
2889 packet_set_ring(sk, &req_u, 1, 1);
2896 * Now the socket is dead. No more input will appear.
2903 skb_queue_purge(&sk->sk_receive_queue);
2904 packet_free_pending(po);
2905 sk_refcnt_debug_release(sk);
2912 * Attach a packet hook.
2915 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2918 struct packet_sock *po = pkt_sk(sk);
2919 struct net_device *dev_curr;
2922 struct net_device *dev = NULL;
2924 bool unlisted = false;
2930 spin_lock(&po->bind_lock);
2934 dev = dev_get_by_name_rcu(sock_net(sk), name);
2939 } else if (ifindex) {
2940 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2950 proto_curr = po->prot_hook.type;
2951 dev_curr = po->prot_hook.dev;
2953 need_rehook = proto_curr != proto || dev_curr != dev;
2958 __unregister_prot_hook(sk, true);
2960 dev_curr = po->prot_hook.dev;
2962 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2967 po->prot_hook.type = proto;
2969 if (unlikely(unlisted)) {
2971 po->prot_hook.dev = NULL;
2973 packet_cached_dev_reset(po);
2975 po->prot_hook.dev = dev;
2976 po->ifindex = dev ? dev->ifindex : 0;
2977 packet_cached_dev_assign(po, dev);
2983 if (proto == 0 || !need_rehook)
2986 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2987 register_prot_hook(sk);
2989 sk->sk_err = ENETDOWN;
2990 if (!sock_flag(sk, SOCK_DEAD))
2991 sk->sk_error_report(sk);
2996 spin_unlock(&po->bind_lock);
3002 * Bind a packet socket to a device
3005 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3008 struct sock *sk = sock->sk;
3015 if (addr_len != sizeof(struct sockaddr))
3017 strlcpy(name, uaddr->sa_data, sizeof(name));
3019 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3022 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3024 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3025 struct sock *sk = sock->sk;
3031 if (addr_len < sizeof(struct sockaddr_ll))
3033 if (sll->sll_family != AF_PACKET)
3036 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3037 sll->sll_protocol ? : pkt_sk(sk)->num);
3040 static struct proto packet_proto = {
3042 .owner = THIS_MODULE,
3043 .obj_size = sizeof(struct packet_sock),
3047 * Create a packet of type SOCK_PACKET.
3050 static int packet_create(struct net *net, struct socket *sock, int protocol,
3054 struct packet_sock *po;
3055 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3058 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3060 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3061 sock->type != SOCK_PACKET)
3062 return -ESOCKTNOSUPPORT;
3064 sock->state = SS_UNCONNECTED;
3067 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3071 sock->ops = &packet_ops;
3072 if (sock->type == SOCK_PACKET)
3073 sock->ops = &packet_ops_spkt;
3075 sock_init_data(sock, sk);
3078 sk->sk_family = PF_PACKET;
3080 po->xmit = dev_queue_xmit;
3082 err = packet_alloc_pending(po);
3086 packet_cached_dev_reset(po);
3088 sk->sk_destruct = packet_sock_destruct;
3089 sk_refcnt_debug_inc(sk);
3092 * Attach a protocol block
3095 spin_lock_init(&po->bind_lock);
3096 mutex_init(&po->pg_vec_lock);
3097 po->rollover = NULL;
3098 po->prot_hook.func = packet_rcv;
3100 if (sock->type == SOCK_PACKET)
3101 po->prot_hook.func = packet_rcv_spkt;
3103 po->prot_hook.af_packet_priv = sk;
3106 po->prot_hook.type = proto;
3107 register_prot_hook(sk);
3110 mutex_lock(&net->packet.sklist_lock);
3111 sk_add_node_rcu(sk, &net->packet.sklist);
3112 mutex_unlock(&net->packet.sklist_lock);
3115 sock_prot_inuse_add(net, &packet_proto, 1);
3126 * Pull a packet from our receive queue and hand it to the user.
3127 * If necessary we block.
3130 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3133 struct sock *sk = sock->sk;
3134 struct sk_buff *skb;
3136 int vnet_hdr_len = 0;
3137 unsigned int origlen = 0;
3140 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3144 /* What error should we return now? EUNATTACH? */
3145 if (pkt_sk(sk)->ifindex < 0)
3149 if (flags & MSG_ERRQUEUE) {
3150 err = sock_recv_errqueue(sk, msg, len,
3151 SOL_PACKET, PACKET_TX_TIMESTAMP);
3156 * Call the generic datagram receiver. This handles all sorts
3157 * of horrible races and re-entrancy so we can forget about it
3158 * in the protocol layers.
3160 * Now it will return ENETDOWN, if device have just gone down,
3161 * but then it will block.
3164 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3167 * An error occurred so return it. Because skb_recv_datagram()
3168 * handles the blocking we don't see and worry about blocking
3175 if (pkt_sk(sk)->pressure)
3176 packet_rcv_has_room(pkt_sk(sk), NULL);
3178 if (pkt_sk(sk)->has_vnet_hdr) {
3179 struct virtio_net_hdr vnet_hdr = { 0 };
3182 vnet_hdr_len = sizeof(vnet_hdr);
3183 if (len < vnet_hdr_len)
3186 len -= vnet_hdr_len;
3188 if (skb_is_gso(skb)) {
3189 struct skb_shared_info *sinfo = skb_shinfo(skb);
3191 /* This is a hint as to how much should be linear. */
3193 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3195 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3196 if (sinfo->gso_type & SKB_GSO_TCPV4)
3197 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3198 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3199 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3200 else if (sinfo->gso_type & SKB_GSO_UDP)
3201 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3202 else if (sinfo->gso_type & SKB_GSO_FCOE)
3206 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3207 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3209 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3211 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3212 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3213 vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3214 skb_checksum_start_offset(skb));
3215 vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3217 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3218 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3219 } /* else everything is zero */
3221 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3226 /* You lose any data beyond the buffer you gave. If it worries
3227 * a user program they can ask the device for its MTU
3233 msg->msg_flags |= MSG_TRUNC;
3236 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3240 if (sock->type != SOCK_PACKET) {
3241 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3243 /* Original length was stored in sockaddr_ll fields */
3244 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3245 sll->sll_family = AF_PACKET;
3246 sll->sll_protocol = skb->protocol;
3249 sock_recv_ts_and_drops(msg, sk, skb);
3251 if (msg->msg_name) {
3252 /* If the address length field is there to be filled
3253 * in, we fill it in now.
3255 if (sock->type == SOCK_PACKET) {
3256 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3257 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3259 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3261 msg->msg_namelen = sll->sll_halen +
3262 offsetof(struct sockaddr_ll, sll_addr);
3264 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3268 if (pkt_sk(sk)->auxdata) {
3269 struct tpacket_auxdata aux;
3271 aux.tp_status = TP_STATUS_USER;
3272 if (skb->ip_summed == CHECKSUM_PARTIAL)
3273 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3274 else if (skb->pkt_type != PACKET_OUTGOING &&
3275 (skb->ip_summed == CHECKSUM_COMPLETE ||
3276 skb_csum_unnecessary(skb)))
3277 aux.tp_status |= TP_STATUS_CSUM_VALID;
3279 aux.tp_len = origlen;
3280 aux.tp_snaplen = skb->len;
3282 aux.tp_net = skb_network_offset(skb);
3283 if (skb_vlan_tag_present(skb)) {
3284 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3285 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3286 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3288 aux.tp_vlan_tci = 0;
3289 aux.tp_vlan_tpid = 0;
3291 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3295 * Free or return the buffer as appropriate. Again this
3296 * hides all the races and re-entrancy issues from us.
3298 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3301 skb_free_datagram(sk, skb);
3306 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3307 int *uaddr_len, int peer)
3309 struct net_device *dev;
3310 struct sock *sk = sock->sk;
3315 uaddr->sa_family = AF_PACKET;
3316 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3318 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3320 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3322 *uaddr_len = sizeof(*uaddr);
3327 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3328 int *uaddr_len, int peer)
3330 struct net_device *dev;
3331 struct sock *sk = sock->sk;
3332 struct packet_sock *po = pkt_sk(sk);
3333 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3338 sll->sll_family = AF_PACKET;
3339 sll->sll_ifindex = po->ifindex;
3340 sll->sll_protocol = po->num;
3341 sll->sll_pkttype = 0;
3343 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3345 sll->sll_hatype = dev->type;
3346 sll->sll_halen = dev->addr_len;
3347 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3349 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3353 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3358 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3362 case PACKET_MR_MULTICAST:
3363 if (i->alen != dev->addr_len)
3366 return dev_mc_add(dev, i->addr);
3368 return dev_mc_del(dev, i->addr);
3370 case PACKET_MR_PROMISC:
3371 return dev_set_promiscuity(dev, what);
3372 case PACKET_MR_ALLMULTI:
3373 return dev_set_allmulti(dev, what);
3374 case PACKET_MR_UNICAST:
3375 if (i->alen != dev->addr_len)
3378 return dev_uc_add(dev, i->addr);
3380 return dev_uc_del(dev, i->addr);
3388 static void packet_dev_mclist_delete(struct net_device *dev,
3389 struct packet_mclist **mlp)
3391 struct packet_mclist *ml;
3393 while ((ml = *mlp) != NULL) {
3394 if (ml->ifindex == dev->ifindex) {
3395 packet_dev_mc(dev, ml, -1);
3403 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3405 struct packet_sock *po = pkt_sk(sk);
3406 struct packet_mclist *ml, *i;
3407 struct net_device *dev;
3413 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3418 if (mreq->mr_alen > dev->addr_len)
3422 i = kmalloc(sizeof(*i), GFP_KERNEL);
3427 for (ml = po->mclist; ml; ml = ml->next) {
3428 if (ml->ifindex == mreq->mr_ifindex &&
3429 ml->type == mreq->mr_type &&
3430 ml->alen == mreq->mr_alen &&
3431 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3433 /* Free the new element ... */
3439 i->type = mreq->mr_type;
3440 i->ifindex = mreq->mr_ifindex;
3441 i->alen = mreq->mr_alen;
3442 memcpy(i->addr, mreq->mr_address, i->alen);
3444 i->next = po->mclist;
3446 err = packet_dev_mc(dev, i, 1);
3448 po->mclist = i->next;
3457 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3459 struct packet_mclist *ml, **mlp;
3463 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3464 if (ml->ifindex == mreq->mr_ifindex &&
3465 ml->type == mreq->mr_type &&
3466 ml->alen == mreq->mr_alen &&
3467 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3468 if (--ml->count == 0) {
3469 struct net_device *dev;
3471 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3473 packet_dev_mc(dev, ml, -1);
3483 static void packet_flush_mclist(struct sock *sk)
3485 struct packet_sock *po = pkt_sk(sk);
3486 struct packet_mclist *ml;
3492 while ((ml = po->mclist) != NULL) {
3493 struct net_device *dev;
3495 po->mclist = ml->next;
3496 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3498 packet_dev_mc(dev, ml, -1);
3505 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3507 struct sock *sk = sock->sk;
3508 struct packet_sock *po = pkt_sk(sk);
3511 if (level != SOL_PACKET)
3512 return -ENOPROTOOPT;
3515 case PACKET_ADD_MEMBERSHIP:
3516 case PACKET_DROP_MEMBERSHIP:
3518 struct packet_mreq_max mreq;
3520 memset(&mreq, 0, sizeof(mreq));
3521 if (len < sizeof(struct packet_mreq))
3523 if (len > sizeof(mreq))
3525 if (copy_from_user(&mreq, optval, len))
3527 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3529 if (optname == PACKET_ADD_MEMBERSHIP)
3530 ret = packet_mc_add(sk, &mreq);
3532 ret = packet_mc_drop(sk, &mreq);
3536 case PACKET_RX_RING:
3537 case PACKET_TX_RING:
3539 union tpacket_req_u req_u;
3542 switch (po->tp_version) {
3545 len = sizeof(req_u.req);
3549 len = sizeof(req_u.req3);
3554 if (pkt_sk(sk)->has_vnet_hdr)
3556 if (copy_from_user(&req_u.req, optval, len))
3558 return packet_set_ring(sk, &req_u, 0,
3559 optname == PACKET_TX_RING);
3561 case PACKET_COPY_THRESH:
3565 if (optlen != sizeof(val))
3567 if (copy_from_user(&val, optval, sizeof(val)))
3570 pkt_sk(sk)->copy_thresh = val;
3573 case PACKET_VERSION:
3577 if (optlen != sizeof(val))
3579 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3581 if (copy_from_user(&val, optval, sizeof(val)))
3587 po->tp_version = val;
3593 case PACKET_RESERVE:
3597 if (optlen != sizeof(val))
3599 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3601 if (copy_from_user(&val, optval, sizeof(val)))
3603 po->tp_reserve = val;
3610 if (optlen != sizeof(val))
3612 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3614 if (copy_from_user(&val, optval, sizeof(val)))
3616 po->tp_loss = !!val;
3619 case PACKET_AUXDATA:
3623 if (optlen < sizeof(val))
3625 if (copy_from_user(&val, optval, sizeof(val)))
3628 po->auxdata = !!val;
3631 case PACKET_ORIGDEV:
3635 if (optlen < sizeof(val))
3637 if (copy_from_user(&val, optval, sizeof(val)))
3640 po->origdev = !!val;
3643 case PACKET_VNET_HDR:
3647 if (sock->type != SOCK_RAW)
3649 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3651 if (optlen < sizeof(val))
3653 if (copy_from_user(&val, optval, sizeof(val)))
3656 po->has_vnet_hdr = !!val;
3659 case PACKET_TIMESTAMP:
3663 if (optlen != sizeof(val))
3665 if (copy_from_user(&val, optval, sizeof(val)))
3668 po->tp_tstamp = val;
3675 if (optlen != sizeof(val))
3677 if (copy_from_user(&val, optval, sizeof(val)))
3680 return fanout_add(sk, val & 0xffff, val >> 16);
3682 case PACKET_FANOUT_DATA:
3687 return fanout_set_data(po, optval, optlen);
3689 case PACKET_TX_HAS_OFF:
3693 if (optlen != sizeof(val))
3695 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3697 if (copy_from_user(&val, optval, sizeof(val)))
3699 po->tp_tx_has_off = !!val;
3702 case PACKET_QDISC_BYPASS:
3706 if (optlen != sizeof(val))
3708 if (copy_from_user(&val, optval, sizeof(val)))
3711 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3715 return -ENOPROTOOPT;
3719 static int packet_getsockopt(struct socket *sock, int level, int optname,
3720 char __user *optval, int __user *optlen)
3723 int val, lv = sizeof(val);
3724 struct sock *sk = sock->sk;
3725 struct packet_sock *po = pkt_sk(sk);
3727 union tpacket_stats_u st;
3728 struct tpacket_rollover_stats rstats;
3730 if (level != SOL_PACKET)
3731 return -ENOPROTOOPT;
3733 if (get_user(len, optlen))
3740 case PACKET_STATISTICS:
3741 spin_lock_bh(&sk->sk_receive_queue.lock);
3742 memcpy(&st, &po->stats, sizeof(st));
3743 memset(&po->stats, 0, sizeof(po->stats));
3744 spin_unlock_bh(&sk->sk_receive_queue.lock);
3746 if (po->tp_version == TPACKET_V3) {
3747 lv = sizeof(struct tpacket_stats_v3);
3748 st.stats3.tp_packets += st.stats3.tp_drops;
3751 lv = sizeof(struct tpacket_stats);
3752 st.stats1.tp_packets += st.stats1.tp_drops;
3757 case PACKET_AUXDATA:
3760 case PACKET_ORIGDEV:
3763 case PACKET_VNET_HDR:
3764 val = po->has_vnet_hdr;
3766 case PACKET_VERSION:
3767 val = po->tp_version;
3770 if (len > sizeof(int))
3772 if (copy_from_user(&val, optval, len))
3776 val = sizeof(struct tpacket_hdr);
3779 val = sizeof(struct tpacket2_hdr);
3782 val = sizeof(struct tpacket3_hdr);
3788 case PACKET_RESERVE:
3789 val = po->tp_reserve;
3794 case PACKET_TIMESTAMP:
3795 val = po->tp_tstamp;
3799 ((u32)po->fanout->id |
3800 ((u32)po->fanout->type << 16) |
3801 ((u32)po->fanout->flags << 24)) :
3804 case PACKET_ROLLOVER_STATS:
3807 rstats.tp_all = atomic_long_read(&po->rollover->num);
3808 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3809 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3811 lv = sizeof(rstats);
3813 case PACKET_TX_HAS_OFF:
3814 val = po->tp_tx_has_off;
3816 case PACKET_QDISC_BYPASS:
3817 val = packet_use_direct_xmit(po);
3820 return -ENOPROTOOPT;
3825 if (put_user(len, optlen))
3827 if (copy_to_user(optval, data, len))
3833 static int packet_notifier(struct notifier_block *this,
3834 unsigned long msg, void *ptr)
3837 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3838 struct net *net = dev_net(dev);
3841 sk_for_each_rcu(sk, &net->packet.sklist) {
3842 struct packet_sock *po = pkt_sk(sk);
3845 case NETDEV_UNREGISTER:
3847 packet_dev_mclist_delete(dev, &po->mclist);
3851 if (dev->ifindex == po->ifindex) {
3852 spin_lock(&po->bind_lock);
3854 __unregister_prot_hook(sk, false);
3855 sk->sk_err = ENETDOWN;
3856 if (!sock_flag(sk, SOCK_DEAD))
3857 sk->sk_error_report(sk);
3859 if (msg == NETDEV_UNREGISTER) {
3860 packet_cached_dev_reset(po);
3862 if (po->prot_hook.dev)
3863 dev_put(po->prot_hook.dev);
3864 po->prot_hook.dev = NULL;
3866 spin_unlock(&po->bind_lock);
3870 if (dev->ifindex == po->ifindex) {
3871 spin_lock(&po->bind_lock);
3873 register_prot_hook(sk);
3874 spin_unlock(&po->bind_lock);
3884 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3887 struct sock *sk = sock->sk;
3892 int amount = sk_wmem_alloc_get(sk);
3894 return put_user(amount, (int __user *)arg);
3898 struct sk_buff *skb;
3901 spin_lock_bh(&sk->sk_receive_queue.lock);
3902 skb = skb_peek(&sk->sk_receive_queue);
3905 spin_unlock_bh(&sk->sk_receive_queue.lock);
3906 return put_user(amount, (int __user *)arg);
3909 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3911 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3921 case SIOCGIFBRDADDR:
3922 case SIOCSIFBRDADDR:
3923 case SIOCGIFNETMASK:
3924 case SIOCSIFNETMASK:
3925 case SIOCGIFDSTADDR:
3926 case SIOCSIFDSTADDR:
3928 return inet_dgram_ops.ioctl(sock, cmd, arg);
3932 return -ENOIOCTLCMD;
3937 static unsigned int packet_poll(struct file *file, struct socket *sock,
3940 struct sock *sk = sock->sk;
3941 struct packet_sock *po = pkt_sk(sk);
3942 unsigned int mask = datagram_poll(file, sock, wait);
3944 spin_lock_bh(&sk->sk_receive_queue.lock);
3945 if (po->rx_ring.pg_vec) {
3946 if (!packet_previous_rx_frame(po, &po->rx_ring,
3948 mask |= POLLIN | POLLRDNORM;
3950 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3952 spin_unlock_bh(&sk->sk_receive_queue.lock);
3953 spin_lock_bh(&sk->sk_write_queue.lock);
3954 if (po->tx_ring.pg_vec) {
3955 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3956 mask |= POLLOUT | POLLWRNORM;
3958 spin_unlock_bh(&sk->sk_write_queue.lock);
3963 /* Dirty? Well, I still did not learn better way to account
3967 static void packet_mm_open(struct vm_area_struct *vma)
3969 struct file *file = vma->vm_file;
3970 struct socket *sock = file->private_data;
3971 struct sock *sk = sock->sk;
3974 atomic_inc(&pkt_sk(sk)->mapped);
3977 static void packet_mm_close(struct vm_area_struct *vma)
3979 struct file *file = vma->vm_file;
3980 struct socket *sock = file->private_data;
3981 struct sock *sk = sock->sk;
3984 atomic_dec(&pkt_sk(sk)->mapped);
3987 static const struct vm_operations_struct packet_mmap_ops = {
3988 .open = packet_mm_open,
3989 .close = packet_mm_close,
3992 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3997 for (i = 0; i < len; i++) {
3998 if (likely(pg_vec[i].buffer)) {
3999 if (is_vmalloc_addr(pg_vec[i].buffer))
4000 vfree(pg_vec[i].buffer);
4002 free_pages((unsigned long)pg_vec[i].buffer,
4004 pg_vec[i].buffer = NULL;
4010 static char *alloc_one_pg_vec_page(unsigned long order)
4013 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4014 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4016 buffer = (char *) __get_free_pages(gfp_flags, order);
4020 /* __get_free_pages failed, fall back to vmalloc */
4021 buffer = vzalloc((1 << order) * PAGE_SIZE);
4025 /* vmalloc failed, lets dig into swap here */
4026 gfp_flags &= ~__GFP_NORETRY;
4027 buffer = (char *) __get_free_pages(gfp_flags, order);
4031 /* complete and utter failure */
4035 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4037 unsigned int block_nr = req->tp_block_nr;
4041 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4042 if (unlikely(!pg_vec))
4045 for (i = 0; i < block_nr; i++) {
4046 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4047 if (unlikely(!pg_vec[i].buffer))
4048 goto out_free_pgvec;
4055 free_pg_vec(pg_vec, order, block_nr);
4060 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4061 int closing, int tx_ring)
4063 struct pgv *pg_vec = NULL;
4064 struct packet_sock *po = pkt_sk(sk);
4065 int was_running, order = 0;
4066 struct packet_ring_buffer *rb;
4067 struct sk_buff_head *rb_queue;
4070 /* Added to avoid minimal code churn */
4071 struct tpacket_req *req = &req_u->req;
4073 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4074 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4075 WARN(1, "Tx-ring is not supported.\n");
4079 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4080 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4084 if (atomic_read(&po->mapped))
4086 if (packet_read_pending(rb))
4090 if (req->tp_block_nr) {
4091 /* Sanity tests and some calculations */
4093 if (unlikely(rb->pg_vec))
4096 switch (po->tp_version) {
4098 po->tp_hdrlen = TPACKET_HDRLEN;
4101 po->tp_hdrlen = TPACKET2_HDRLEN;
4104 po->tp_hdrlen = TPACKET3_HDRLEN;
4109 if (unlikely((int)req->tp_block_size <= 0))
4111 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
4113 if (po->tp_version >= TPACKET_V3 &&
4114 (int)(req->tp_block_size -
4115 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4117 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4120 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4123 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
4124 if (unlikely(rb->frames_per_block <= 0))
4126 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4131 order = get_order(req->tp_block_size);
4132 pg_vec = alloc_pg_vec(req, order);
4133 if (unlikely(!pg_vec))
4135 switch (po->tp_version) {
4137 /* Transmit path is not supported. We checked
4138 * it above but just being paranoid
4141 init_prb_bdqc(po, rb, pg_vec, req_u);
4150 if (unlikely(req->tp_frame_nr))
4156 /* Detach socket from network */
4157 spin_lock(&po->bind_lock);
4158 was_running = po->running;
4162 __unregister_prot_hook(sk, false);
4164 spin_unlock(&po->bind_lock);
4169 mutex_lock(&po->pg_vec_lock);
4170 if (closing || atomic_read(&po->mapped) == 0) {
4172 spin_lock_bh(&rb_queue->lock);
4173 swap(rb->pg_vec, pg_vec);
4174 rb->frame_max = (req->tp_frame_nr - 1);
4176 rb->frame_size = req->tp_frame_size;
4177 spin_unlock_bh(&rb_queue->lock);
4179 swap(rb->pg_vec_order, order);
4180 swap(rb->pg_vec_len, req->tp_block_nr);
4182 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4183 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4184 tpacket_rcv : packet_rcv;
4185 skb_queue_purge(rb_queue);
4186 if (atomic_read(&po->mapped))
4187 pr_err("packet_mmap: vma is busy: %d\n",
4188 atomic_read(&po->mapped));
4190 mutex_unlock(&po->pg_vec_lock);
4192 spin_lock(&po->bind_lock);
4195 register_prot_hook(sk);
4197 spin_unlock(&po->bind_lock);
4198 if (closing && (po->tp_version > TPACKET_V2)) {
4199 /* Because we don't support block-based V3 on tx-ring */
4201 prb_shutdown_retire_blk_timer(po, rb_queue);
4206 free_pg_vec(pg_vec, order, req->tp_block_nr);
4211 static int packet_mmap(struct file *file, struct socket *sock,
4212 struct vm_area_struct *vma)
4214 struct sock *sk = sock->sk;
4215 struct packet_sock *po = pkt_sk(sk);
4216 unsigned long size, expected_size;
4217 struct packet_ring_buffer *rb;
4218 unsigned long start;
4225 mutex_lock(&po->pg_vec_lock);
4228 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4230 expected_size += rb->pg_vec_len
4236 if (expected_size == 0)
4239 size = vma->vm_end - vma->vm_start;
4240 if (size != expected_size)
4243 start = vma->vm_start;
4244 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4245 if (rb->pg_vec == NULL)
4248 for (i = 0; i < rb->pg_vec_len; i++) {
4250 void *kaddr = rb->pg_vec[i].buffer;
4253 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4254 page = pgv_to_page(kaddr);
4255 err = vm_insert_page(vma, start, page);
4264 atomic_inc(&po->mapped);
4265 vma->vm_ops = &packet_mmap_ops;
4269 mutex_unlock(&po->pg_vec_lock);
4273 static const struct proto_ops packet_ops_spkt = {
4274 .family = PF_PACKET,
4275 .owner = THIS_MODULE,
4276 .release = packet_release,
4277 .bind = packet_bind_spkt,
4278 .connect = sock_no_connect,
4279 .socketpair = sock_no_socketpair,
4280 .accept = sock_no_accept,
4281 .getname = packet_getname_spkt,
4282 .poll = datagram_poll,
4283 .ioctl = packet_ioctl,
4284 .listen = sock_no_listen,
4285 .shutdown = sock_no_shutdown,
4286 .setsockopt = sock_no_setsockopt,
4287 .getsockopt = sock_no_getsockopt,
4288 .sendmsg = packet_sendmsg_spkt,
4289 .recvmsg = packet_recvmsg,
4290 .mmap = sock_no_mmap,
4291 .sendpage = sock_no_sendpage,
4294 static const struct proto_ops packet_ops = {
4295 .family = PF_PACKET,
4296 .owner = THIS_MODULE,
4297 .release = packet_release,
4298 .bind = packet_bind,
4299 .connect = sock_no_connect,
4300 .socketpair = sock_no_socketpair,
4301 .accept = sock_no_accept,
4302 .getname = packet_getname,
4303 .poll = packet_poll,
4304 .ioctl = packet_ioctl,
4305 .listen = sock_no_listen,
4306 .shutdown = sock_no_shutdown,
4307 .setsockopt = packet_setsockopt,
4308 .getsockopt = packet_getsockopt,
4309 .sendmsg = packet_sendmsg,
4310 .recvmsg = packet_recvmsg,
4311 .mmap = packet_mmap,
4312 .sendpage = sock_no_sendpage,
4315 static const struct net_proto_family packet_family_ops = {
4316 .family = PF_PACKET,
4317 .create = packet_create,
4318 .owner = THIS_MODULE,
4321 static struct notifier_block packet_netdev_notifier = {
4322 .notifier_call = packet_notifier,
4325 #ifdef CONFIG_PROC_FS
4327 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4330 struct net *net = seq_file_net(seq);
4333 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4336 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4338 struct net *net = seq_file_net(seq);
4339 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4342 static void packet_seq_stop(struct seq_file *seq, void *v)
4348 static int packet_seq_show(struct seq_file *seq, void *v)
4350 if (v == SEQ_START_TOKEN)
4351 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4353 struct sock *s = sk_entry(v);
4354 const struct packet_sock *po = pkt_sk(s);
4357 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4359 atomic_read(&s->sk_refcnt),
4364 atomic_read(&s->sk_rmem_alloc),
4365 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4372 static const struct seq_operations packet_seq_ops = {
4373 .start = packet_seq_start,
4374 .next = packet_seq_next,
4375 .stop = packet_seq_stop,
4376 .show = packet_seq_show,
4379 static int packet_seq_open(struct inode *inode, struct file *file)
4381 return seq_open_net(inode, file, &packet_seq_ops,
4382 sizeof(struct seq_net_private));
4385 static const struct file_operations packet_seq_fops = {
4386 .owner = THIS_MODULE,
4387 .open = packet_seq_open,
4389 .llseek = seq_lseek,
4390 .release = seq_release_net,
4395 static int __net_init packet_net_init(struct net *net)
4397 mutex_init(&net->packet.sklist_lock);
4398 INIT_HLIST_HEAD(&net->packet.sklist);
4400 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4406 static void __net_exit packet_net_exit(struct net *net)
4408 remove_proc_entry("packet", net->proc_net);
4411 static struct pernet_operations packet_net_ops = {
4412 .init = packet_net_init,
4413 .exit = packet_net_exit,
4417 static void __exit packet_exit(void)
4419 unregister_netdevice_notifier(&packet_netdev_notifier);
4420 unregister_pernet_subsys(&packet_net_ops);
4421 sock_unregister(PF_PACKET);
4422 proto_unregister(&packet_proto);
4425 static int __init packet_init(void)
4427 int rc = proto_register(&packet_proto, 0);
4432 sock_register(&packet_family_ops);
4433 register_pernet_subsys(&packet_net_ops);
4434 register_netdevice_notifier(&packet_netdev_notifier);
4439 module_init(packet_init);
4440 module_exit(packet_exit);
4441 MODULE_LICENSE("GPL");
4442 MODULE_ALIAS_NETPROTO(PF_PACKET);