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(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 packet_sock *po;
1445 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1452 skb = ip_check_defrag(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 (ptype->af_packet_priv == (void *)((struct packet_sock *)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);
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 const struct proto_ops packet_ops;
1746 static const struct proto_ops packet_ops_spkt;
1748 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1749 struct packet_type *pt, struct net_device *orig_dev)
1752 struct sockaddr_pkt *spkt;
1755 * When we registered the protocol we saved the socket in the data
1756 * field for just this event.
1759 sk = pt->af_packet_priv;
1762 * Yank back the headers [hope the device set this
1763 * right or kerboom...]
1765 * Incoming packets have ll header pulled,
1768 * For outgoing ones skb->data == skb_mac_header(skb)
1769 * so that this procedure is noop.
1772 if (skb->pkt_type == PACKET_LOOPBACK)
1775 if (!net_eq(dev_net(dev), sock_net(sk)))
1778 skb = skb_share_check(skb, GFP_ATOMIC);
1782 /* drop any routing info */
1785 /* drop conntrack reference */
1788 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1790 skb_push(skb, skb->data - skb_mac_header(skb));
1793 * The SOCK_PACKET socket receives _all_ frames.
1796 spkt->spkt_family = dev->type;
1797 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1798 spkt->spkt_protocol = skb->protocol;
1801 * Charge the memory to the socket. This is done specifically
1802 * to prevent sockets using all the memory up.
1805 if (sock_queue_rcv_skb(sk, skb) == 0)
1816 * Output a raw packet to a device layer. This bypasses all the other
1817 * protocol layers and you must therefore supply it with a complete frame
1820 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1823 struct sock *sk = sock->sk;
1824 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1825 struct sk_buff *skb = NULL;
1826 struct net_device *dev;
1832 * Get and verify the address.
1836 if (msg->msg_namelen < sizeof(struct sockaddr))
1838 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1839 proto = saddr->spkt_protocol;
1841 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1844 * Find the device first to size check it
1847 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1850 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1856 if (!(dev->flags & IFF_UP))
1860 * You may not queue a frame bigger than the mtu. This is the lowest level
1861 * raw protocol and you must do your own fragmentation at this level.
1864 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1865 if (!netif_supports_nofcs(dev)) {
1866 err = -EPROTONOSUPPORT;
1869 extra_len = 4; /* We're doing our own CRC */
1873 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1877 size_t reserved = LL_RESERVED_SPACE(dev);
1878 int tlen = dev->needed_tailroom;
1879 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1882 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1885 /* FIXME: Save some space for broken drivers that write a hard
1886 * header at transmission time by themselves. PPP is the notable
1887 * one here. This should really be fixed at the driver level.
1889 skb_reserve(skb, reserved);
1890 skb_reset_network_header(skb);
1892 /* Try to align data part correctly */
1897 skb_reset_network_header(skb);
1899 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1905 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1906 /* Earlier code assumed this would be a VLAN pkt,
1907 * double-check this now that we have the actual
1910 struct ethhdr *ehdr;
1911 skb_reset_mac_header(skb);
1912 ehdr = eth_hdr(skb);
1913 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1919 skb->protocol = proto;
1921 skb->priority = sk->sk_priority;
1922 skb->mark = sk->sk_mark;
1924 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1926 if (unlikely(extra_len == 4))
1929 skb_probe_transport_header(skb, 0);
1931 dev_queue_xmit(skb);
1942 static unsigned int run_filter(const struct sk_buff *skb,
1943 const struct sock *sk,
1946 struct sk_filter *filter;
1949 filter = rcu_dereference(sk->sk_filter);
1951 res = SK_RUN_FILTER(filter, skb);
1958 * This function makes lazy skb cloning in hope that most of packets
1959 * are discarded by BPF.
1961 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1962 * and skb->cb are mangled. It works because (and until) packets
1963 * falling here are owned by current CPU. Output packets are cloned
1964 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1965 * sequencially, so that if we return skb to original state on exit,
1966 * we will not harm anyone.
1969 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1970 struct packet_type *pt, struct net_device *orig_dev)
1973 struct sockaddr_ll *sll;
1974 struct packet_sock *po;
1975 u8 *skb_head = skb->data;
1976 int skb_len = skb->len;
1977 unsigned int snaplen, res;
1979 if (skb->pkt_type == PACKET_LOOPBACK)
1982 sk = pt->af_packet_priv;
1985 if (!net_eq(dev_net(dev), sock_net(sk)))
1990 if (dev->header_ops) {
1991 /* The device has an explicit notion of ll header,
1992 * exported to higher levels.
1994 * Otherwise, the device hides details of its frame
1995 * structure, so that corresponding packet head is
1996 * never delivered to user.
1998 if (sk->sk_type != SOCK_DGRAM)
1999 skb_push(skb, skb->data - skb_mac_header(skb));
2000 else if (skb->pkt_type == PACKET_OUTGOING) {
2001 /* Special case: outgoing packets have ll header at head */
2002 skb_pull(skb, skb_network_offset(skb));
2008 res = run_filter(skb, sk, snaplen);
2010 goto drop_n_restore;
2014 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2017 if (skb_shared(skb)) {
2018 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2022 if (skb_head != skb->data) {
2023 skb->data = skb_head;
2030 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2032 sll = &PACKET_SKB_CB(skb)->sa.ll;
2033 sll->sll_hatype = dev->type;
2034 sll->sll_pkttype = skb->pkt_type;
2035 if (unlikely(po->origdev))
2036 sll->sll_ifindex = orig_dev->ifindex;
2038 sll->sll_ifindex = dev->ifindex;
2040 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2042 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2043 * Use their space for storing the original skb length.
2045 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2047 if (pskb_trim(skb, snaplen))
2050 skb_set_owner_r(skb, sk);
2054 /* drop conntrack reference */
2057 spin_lock(&sk->sk_receive_queue.lock);
2058 po->stats.stats1.tp_packets++;
2059 sock_skb_set_dropcount(sk, skb);
2060 __skb_queue_tail(&sk->sk_receive_queue, skb);
2061 spin_unlock(&sk->sk_receive_queue.lock);
2062 sk->sk_data_ready(sk);
2066 spin_lock(&sk->sk_receive_queue.lock);
2067 po->stats.stats1.tp_drops++;
2068 atomic_inc(&sk->sk_drops);
2069 spin_unlock(&sk->sk_receive_queue.lock);
2072 if (skb_head != skb->data && skb_shared(skb)) {
2073 skb->data = skb_head;
2081 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2082 struct packet_type *pt, struct net_device *orig_dev)
2085 struct packet_sock *po;
2086 struct sockaddr_ll *sll;
2087 union tpacket_uhdr h;
2088 u8 *skb_head = skb->data;
2089 int skb_len = skb->len;
2090 unsigned int snaplen, res;
2091 unsigned long status = TP_STATUS_USER;
2092 unsigned short macoff, netoff, hdrlen;
2093 struct sk_buff *copy_skb = NULL;
2097 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2098 * We may add members to them until current aligned size without forcing
2099 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2101 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2102 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2104 if (skb->pkt_type == PACKET_LOOPBACK)
2107 sk = pt->af_packet_priv;
2110 if (!net_eq(dev_net(dev), sock_net(sk)))
2113 if (dev->header_ops) {
2114 if (sk->sk_type != SOCK_DGRAM)
2115 skb_push(skb, skb->data - skb_mac_header(skb));
2116 else if (skb->pkt_type == PACKET_OUTGOING) {
2117 /* Special case: outgoing packets have ll header at head */
2118 skb_pull(skb, skb_network_offset(skb));
2124 res = run_filter(skb, sk, snaplen);
2126 goto drop_n_restore;
2128 if (skb->ip_summed == CHECKSUM_PARTIAL)
2129 status |= TP_STATUS_CSUMNOTREADY;
2130 else if (skb->pkt_type != PACKET_OUTGOING &&
2131 (skb->ip_summed == CHECKSUM_COMPLETE ||
2132 skb_csum_unnecessary(skb)))
2133 status |= TP_STATUS_CSUM_VALID;
2138 if (sk->sk_type == SOCK_DGRAM) {
2139 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2142 unsigned int maclen = skb_network_offset(skb);
2143 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2144 (maclen < 16 ? 16 : maclen)) +
2146 macoff = netoff - maclen;
2148 if (po->tp_version <= TPACKET_V2) {
2149 if (macoff + snaplen > po->rx_ring.frame_size) {
2150 if (po->copy_thresh &&
2151 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2152 if (skb_shared(skb)) {
2153 copy_skb = skb_clone(skb, GFP_ATOMIC);
2155 copy_skb = skb_get(skb);
2156 skb_head = skb->data;
2159 skb_set_owner_r(copy_skb, sk);
2161 snaplen = po->rx_ring.frame_size - macoff;
2162 if ((int)snaplen < 0)
2165 } else if (unlikely(macoff + snaplen >
2166 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2169 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2170 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2171 snaplen, nval, macoff);
2173 if (unlikely((int)snaplen < 0)) {
2175 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2178 spin_lock(&sk->sk_receive_queue.lock);
2179 h.raw = packet_current_rx_frame(po, skb,
2180 TP_STATUS_KERNEL, (macoff+snaplen));
2183 if (po->tp_version <= TPACKET_V2) {
2184 packet_increment_rx_head(po, &po->rx_ring);
2186 * LOSING will be reported till you read the stats,
2187 * because it's COR - Clear On Read.
2188 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2191 if (po->stats.stats1.tp_drops)
2192 status |= TP_STATUS_LOSING;
2194 po->stats.stats1.tp_packets++;
2196 status |= TP_STATUS_COPY;
2197 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2199 spin_unlock(&sk->sk_receive_queue.lock);
2201 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2203 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2204 getnstimeofday(&ts);
2206 status |= ts_status;
2208 switch (po->tp_version) {
2210 h.h1->tp_len = skb->len;
2211 h.h1->tp_snaplen = snaplen;
2212 h.h1->tp_mac = macoff;
2213 h.h1->tp_net = netoff;
2214 h.h1->tp_sec = ts.tv_sec;
2215 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2216 hdrlen = sizeof(*h.h1);
2219 h.h2->tp_len = skb->len;
2220 h.h2->tp_snaplen = snaplen;
2221 h.h2->tp_mac = macoff;
2222 h.h2->tp_net = netoff;
2223 h.h2->tp_sec = ts.tv_sec;
2224 h.h2->tp_nsec = ts.tv_nsec;
2225 if (skb_vlan_tag_present(skb)) {
2226 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2227 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2228 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2230 h.h2->tp_vlan_tci = 0;
2231 h.h2->tp_vlan_tpid = 0;
2233 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2234 hdrlen = sizeof(*h.h2);
2237 /* tp_nxt_offset,vlan are already populated above.
2238 * So DONT clear those fields here
2240 h.h3->tp_status |= status;
2241 h.h3->tp_len = skb->len;
2242 h.h3->tp_snaplen = snaplen;
2243 h.h3->tp_mac = macoff;
2244 h.h3->tp_net = netoff;
2245 h.h3->tp_sec = ts.tv_sec;
2246 h.h3->tp_nsec = ts.tv_nsec;
2247 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2248 hdrlen = sizeof(*h.h3);
2254 sll = h.raw + TPACKET_ALIGN(hdrlen);
2255 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2256 sll->sll_family = AF_PACKET;
2257 sll->sll_hatype = dev->type;
2258 sll->sll_protocol = skb->protocol;
2259 sll->sll_pkttype = skb->pkt_type;
2260 if (unlikely(po->origdev))
2261 sll->sll_ifindex = orig_dev->ifindex;
2263 sll->sll_ifindex = dev->ifindex;
2267 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2268 if (po->tp_version <= TPACKET_V2) {
2271 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2274 for (start = h.raw; start < end; start += PAGE_SIZE)
2275 flush_dcache_page(pgv_to_page(start));
2280 if (po->tp_version <= TPACKET_V2) {
2281 __packet_set_status(po, h.raw, status);
2282 sk->sk_data_ready(sk);
2284 prb_clear_blk_fill_status(&po->rx_ring);
2288 if (skb_head != skb->data && skb_shared(skb)) {
2289 skb->data = skb_head;
2297 po->stats.stats1.tp_drops++;
2298 spin_unlock(&sk->sk_receive_queue.lock);
2300 sk->sk_data_ready(sk);
2301 kfree_skb(copy_skb);
2302 goto drop_n_restore;
2305 static void tpacket_destruct_skb(struct sk_buff *skb)
2307 struct packet_sock *po = pkt_sk(skb->sk);
2309 if (likely(po->tx_ring.pg_vec)) {
2313 ph = skb_shinfo(skb)->destructor_arg;
2314 packet_dec_pending(&po->tx_ring);
2316 ts = __packet_set_timestamp(po, ph, skb);
2317 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2323 static bool ll_header_truncated(const struct net_device *dev, int len)
2325 /* net device doesn't like empty head */
2326 if (unlikely(len <= dev->hard_header_len)) {
2327 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2328 current->comm, len, dev->hard_header_len);
2335 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2336 void *frame, struct net_device *dev, int size_max,
2337 __be16 proto, unsigned char *addr, int hlen)
2339 union tpacket_uhdr ph;
2340 int to_write, offset, len, tp_len, nr_frags, len_max;
2341 struct socket *sock = po->sk.sk_socket;
2348 skb->protocol = proto;
2350 skb->priority = po->sk.sk_priority;
2351 skb->mark = po->sk.sk_mark;
2352 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2353 skb_shinfo(skb)->destructor_arg = ph.raw;
2355 switch (po->tp_version) {
2357 tp_len = ph.h2->tp_len;
2360 tp_len = ph.h1->tp_len;
2363 if (unlikely(tp_len > size_max)) {
2364 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2368 skb_reserve(skb, hlen);
2369 skb_reset_network_header(skb);
2371 if (!packet_use_direct_xmit(po))
2372 skb_probe_transport_header(skb, 0);
2373 if (unlikely(po->tp_tx_has_off)) {
2374 int off_min, off_max, off;
2375 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2376 off_max = po->tx_ring.frame_size - tp_len;
2377 if (sock->type == SOCK_DGRAM) {
2378 switch (po->tp_version) {
2380 off = ph.h2->tp_net;
2383 off = ph.h1->tp_net;
2387 switch (po->tp_version) {
2389 off = ph.h2->tp_mac;
2392 off = ph.h1->tp_mac;
2396 if (unlikely((off < off_min) || (off_max < off)))
2398 data = ph.raw + off;
2400 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2404 if (sock->type == SOCK_DGRAM) {
2405 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2407 if (unlikely(err < 0))
2409 } else if (dev->hard_header_len) {
2410 if (ll_header_truncated(dev, tp_len))
2413 skb_push(skb, dev->hard_header_len);
2414 err = skb_store_bits(skb, 0, data,
2415 dev->hard_header_len);
2419 data += dev->hard_header_len;
2420 to_write -= dev->hard_header_len;
2423 offset = offset_in_page(data);
2424 len_max = PAGE_SIZE - offset;
2425 len = ((to_write > len_max) ? len_max : to_write);
2427 skb->data_len = to_write;
2428 skb->len += to_write;
2429 skb->truesize += to_write;
2430 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2432 while (likely(to_write)) {
2433 nr_frags = skb_shinfo(skb)->nr_frags;
2435 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2436 pr_err("Packet exceed the number of skb frags(%lu)\n",
2441 page = pgv_to_page(data);
2443 flush_dcache_page(page);
2445 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2448 len_max = PAGE_SIZE;
2449 len = ((to_write > len_max) ? len_max : to_write);
2455 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2457 struct sk_buff *skb;
2458 struct net_device *dev;
2460 int err, reserve = 0;
2462 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2463 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2464 int tp_len, size_max;
2465 unsigned char *addr;
2467 int status = TP_STATUS_AVAILABLE;
2470 mutex_lock(&po->pg_vec_lock);
2472 if (likely(saddr == NULL)) {
2473 dev = packet_cached_dev_get(po);
2478 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2480 if (msg->msg_namelen < (saddr->sll_halen
2481 + offsetof(struct sockaddr_ll,
2484 proto = saddr->sll_protocol;
2485 addr = saddr->sll_addr;
2486 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2490 if (unlikely(dev == NULL))
2493 if (unlikely(!(dev->flags & IFF_UP)))
2496 reserve = dev->hard_header_len + VLAN_HLEN;
2497 size_max = po->tx_ring.frame_size
2498 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2500 if (size_max > dev->mtu + reserve)
2501 size_max = dev->mtu + reserve;
2504 ph = packet_current_frame(po, &po->tx_ring,
2505 TP_STATUS_SEND_REQUEST);
2506 if (unlikely(ph == NULL)) {
2507 if (need_wait && need_resched())
2512 status = TP_STATUS_SEND_REQUEST;
2513 hlen = LL_RESERVED_SPACE(dev);
2514 tlen = dev->needed_tailroom;
2515 skb = sock_alloc_send_skb(&po->sk,
2516 hlen + tlen + sizeof(struct sockaddr_ll),
2519 if (unlikely(skb == NULL)) {
2520 /* we assume the socket was initially writeable ... */
2521 if (likely(len_sum > 0))
2525 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2527 if (likely(tp_len >= 0) &&
2528 tp_len > dev->mtu + dev->hard_header_len) {
2529 struct ethhdr *ehdr;
2530 /* Earlier code assumed this would be a VLAN pkt,
2531 * double-check this now that we have the actual
2535 skb_reset_mac_header(skb);
2536 ehdr = eth_hdr(skb);
2537 if (ehdr->h_proto != htons(ETH_P_8021Q))
2540 if (unlikely(tp_len < 0)) {
2542 __packet_set_status(po, ph,
2543 TP_STATUS_AVAILABLE);
2544 packet_increment_head(&po->tx_ring);
2548 status = TP_STATUS_WRONG_FORMAT;
2554 packet_pick_tx_queue(dev, skb);
2556 skb->destructor = tpacket_destruct_skb;
2557 __packet_set_status(po, ph, TP_STATUS_SENDING);
2558 packet_inc_pending(&po->tx_ring);
2560 status = TP_STATUS_SEND_REQUEST;
2561 err = po->xmit(skb);
2562 if (unlikely(err > 0)) {
2563 err = net_xmit_errno(err);
2564 if (err && __packet_get_status(po, ph) ==
2565 TP_STATUS_AVAILABLE) {
2566 /* skb was destructed already */
2571 * skb was dropped but not destructed yet;
2572 * let's treat it like congestion or err < 0
2576 packet_increment_head(&po->tx_ring);
2578 } while (likely((ph != NULL) ||
2579 /* Note: packet_read_pending() might be slow if we have
2580 * to call it as it's per_cpu variable, but in fast-path
2581 * we already short-circuit the loop with the first
2582 * condition, and luckily don't have to go that path
2585 (need_wait && packet_read_pending(&po->tx_ring))));
2591 __packet_set_status(po, ph, status);
2596 mutex_unlock(&po->pg_vec_lock);
2600 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2601 size_t reserve, size_t len,
2602 size_t linear, int noblock,
2605 struct sk_buff *skb;
2607 /* Under a page? Don't bother with paged skb. */
2608 if (prepad + len < PAGE_SIZE || !linear)
2611 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2616 skb_reserve(skb, reserve);
2617 skb_put(skb, linear);
2618 skb->data_len = len - linear;
2619 skb->len += len - linear;
2624 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2626 struct sock *sk = sock->sk;
2627 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2628 struct sk_buff *skb;
2629 struct net_device *dev;
2631 unsigned char *addr;
2632 int err, reserve = 0;
2633 struct virtio_net_hdr vnet_hdr = { 0 };
2636 struct packet_sock *po = pkt_sk(sk);
2637 unsigned short gso_type = 0;
2643 * Get and verify the address.
2646 if (likely(saddr == NULL)) {
2647 dev = packet_cached_dev_get(po);
2652 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2654 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2656 proto = saddr->sll_protocol;
2657 addr = saddr->sll_addr;
2658 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2662 if (unlikely(dev == NULL))
2665 if (unlikely(!(dev->flags & IFF_UP)))
2668 if (sock->type == SOCK_RAW)
2669 reserve = dev->hard_header_len;
2670 if (po->has_vnet_hdr) {
2671 vnet_hdr_len = sizeof(vnet_hdr);
2674 if (len < vnet_hdr_len)
2677 len -= vnet_hdr_len;
2680 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2681 if (n != vnet_hdr_len)
2684 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2685 (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2686 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2687 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2688 vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2689 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2690 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2693 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2696 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2697 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2698 case VIRTIO_NET_HDR_GSO_TCPV4:
2699 gso_type = SKB_GSO_TCPV4;
2701 case VIRTIO_NET_HDR_GSO_TCPV6:
2702 gso_type = SKB_GSO_TCPV6;
2704 case VIRTIO_NET_HDR_GSO_UDP:
2705 gso_type = SKB_GSO_UDP;
2711 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2712 gso_type |= SKB_GSO_TCP_ECN;
2714 if (vnet_hdr.gso_size == 0)
2720 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2721 if (!netif_supports_nofcs(dev)) {
2722 err = -EPROTONOSUPPORT;
2725 extra_len = 4; /* We're doing our own CRC */
2729 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2733 hlen = LL_RESERVED_SPACE(dev);
2734 tlen = dev->needed_tailroom;
2735 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2736 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2737 msg->msg_flags & MSG_DONTWAIT, &err);
2741 skb_set_network_header(skb, reserve);
2744 if (sock->type == SOCK_DGRAM) {
2745 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2746 if (unlikely(offset < 0))
2749 if (ll_header_truncated(dev, len))
2753 /* Returns -EFAULT on error */
2754 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2758 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2760 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2761 /* Earlier code assumed this would be a VLAN pkt,
2762 * double-check this now that we have the actual
2765 struct ethhdr *ehdr;
2766 skb_reset_mac_header(skb);
2767 ehdr = eth_hdr(skb);
2768 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2774 skb->protocol = proto;
2776 skb->priority = sk->sk_priority;
2777 skb->mark = sk->sk_mark;
2779 packet_pick_tx_queue(dev, skb);
2781 if (po->has_vnet_hdr) {
2782 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2783 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2784 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2785 if (!skb_partial_csum_set(skb, s, o)) {
2791 skb_shinfo(skb)->gso_size =
2792 __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2793 skb_shinfo(skb)->gso_type = gso_type;
2795 /* Header must be checked, and gso_segs computed. */
2796 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2797 skb_shinfo(skb)->gso_segs = 0;
2799 len += vnet_hdr_len;
2802 if (!packet_use_direct_xmit(po))
2803 skb_probe_transport_header(skb, reserve);
2804 if (unlikely(extra_len == 4))
2807 err = po->xmit(skb);
2808 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2824 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2826 struct sock *sk = sock->sk;
2827 struct packet_sock *po = pkt_sk(sk);
2829 if (po->tx_ring.pg_vec)
2830 return tpacket_snd(po, msg);
2832 return packet_snd(sock, msg, len);
2836 * Close a PACKET socket. This is fairly simple. We immediately go
2837 * to 'closed' state and remove our protocol entry in the device list.
2840 static int packet_release(struct socket *sock)
2842 struct sock *sk = sock->sk;
2843 struct packet_sock *po;
2845 union tpacket_req_u req_u;
2853 mutex_lock(&net->packet.sklist_lock);
2854 sk_del_node_init_rcu(sk);
2855 mutex_unlock(&net->packet.sklist_lock);
2858 sock_prot_inuse_add(net, sk->sk_prot, -1);
2861 spin_lock(&po->bind_lock);
2862 unregister_prot_hook(sk, false);
2863 packet_cached_dev_reset(po);
2865 if (po->prot_hook.dev) {
2866 dev_put(po->prot_hook.dev);
2867 po->prot_hook.dev = NULL;
2869 spin_unlock(&po->bind_lock);
2871 packet_flush_mclist(sk);
2873 if (po->rx_ring.pg_vec) {
2874 memset(&req_u, 0, sizeof(req_u));
2875 packet_set_ring(sk, &req_u, 1, 0);
2878 if (po->tx_ring.pg_vec) {
2879 memset(&req_u, 0, sizeof(req_u));
2880 packet_set_ring(sk, &req_u, 1, 1);
2887 * Now the socket is dead. No more input will appear.
2894 skb_queue_purge(&sk->sk_receive_queue);
2895 packet_free_pending(po);
2896 sk_refcnt_debug_release(sk);
2903 * Attach a packet hook.
2906 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2908 struct packet_sock *po = pkt_sk(sk);
2909 struct net_device *dev_curr;
2921 spin_lock(&po->bind_lock);
2923 proto_curr = po->prot_hook.type;
2924 dev_curr = po->prot_hook.dev;
2926 need_rehook = proto_curr != proto || dev_curr != dev;
2929 unregister_prot_hook(sk, true);
2932 po->prot_hook.type = proto;
2933 po->prot_hook.dev = dev;
2935 po->ifindex = dev ? dev->ifindex : 0;
2936 packet_cached_dev_assign(po, dev);
2941 if (proto == 0 || !need_rehook)
2944 if (!dev || (dev->flags & IFF_UP)) {
2945 register_prot_hook(sk);
2947 sk->sk_err = ENETDOWN;
2948 if (!sock_flag(sk, SOCK_DEAD))
2949 sk->sk_error_report(sk);
2953 spin_unlock(&po->bind_lock);
2959 * Bind a packet socket to a device
2962 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2965 struct sock *sk = sock->sk;
2967 struct net_device *dev;
2974 if (addr_len != sizeof(struct sockaddr))
2976 strlcpy(name, uaddr->sa_data, sizeof(name));
2978 dev = dev_get_by_name(sock_net(sk), name);
2980 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2984 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2986 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2987 struct sock *sk = sock->sk;
2988 struct net_device *dev = NULL;
2996 if (addr_len < sizeof(struct sockaddr_ll))
2998 if (sll->sll_family != AF_PACKET)
3001 if (sll->sll_ifindex) {
3003 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
3007 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
3013 static struct proto packet_proto = {
3015 .owner = THIS_MODULE,
3016 .obj_size = sizeof(struct packet_sock),
3020 * Create a packet of type SOCK_PACKET.
3023 static int packet_create(struct net *net, struct socket *sock, int protocol,
3027 struct packet_sock *po;
3028 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3031 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3033 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3034 sock->type != SOCK_PACKET)
3035 return -ESOCKTNOSUPPORT;
3037 sock->state = SS_UNCONNECTED;
3040 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3044 sock->ops = &packet_ops;
3045 if (sock->type == SOCK_PACKET)
3046 sock->ops = &packet_ops_spkt;
3048 sock_init_data(sock, sk);
3051 sk->sk_family = PF_PACKET;
3053 po->xmit = dev_queue_xmit;
3055 err = packet_alloc_pending(po);
3059 packet_cached_dev_reset(po);
3061 sk->sk_destruct = packet_sock_destruct;
3062 sk_refcnt_debug_inc(sk);
3065 * Attach a protocol block
3068 spin_lock_init(&po->bind_lock);
3069 mutex_init(&po->pg_vec_lock);
3070 po->rollover = NULL;
3071 po->prot_hook.func = packet_rcv;
3073 if (sock->type == SOCK_PACKET)
3074 po->prot_hook.func = packet_rcv_spkt;
3076 po->prot_hook.af_packet_priv = sk;
3079 po->prot_hook.type = proto;
3080 register_prot_hook(sk);
3083 mutex_lock(&net->packet.sklist_lock);
3084 sk_add_node_rcu(sk, &net->packet.sklist);
3085 mutex_unlock(&net->packet.sklist_lock);
3088 sock_prot_inuse_add(net, &packet_proto, 1);
3099 * Pull a packet from our receive queue and hand it to the user.
3100 * If necessary we block.
3103 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3106 struct sock *sk = sock->sk;
3107 struct sk_buff *skb;
3109 int vnet_hdr_len = 0;
3110 unsigned int origlen = 0;
3113 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3117 /* What error should we return now? EUNATTACH? */
3118 if (pkt_sk(sk)->ifindex < 0)
3122 if (flags & MSG_ERRQUEUE) {
3123 err = sock_recv_errqueue(sk, msg, len,
3124 SOL_PACKET, PACKET_TX_TIMESTAMP);
3129 * Call the generic datagram receiver. This handles all sorts
3130 * of horrible races and re-entrancy so we can forget about it
3131 * in the protocol layers.
3133 * Now it will return ENETDOWN, if device have just gone down,
3134 * but then it will block.
3137 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3140 * An error occurred so return it. Because skb_recv_datagram()
3141 * handles the blocking we don't see and worry about blocking
3148 if (pkt_sk(sk)->pressure)
3149 packet_rcv_has_room(pkt_sk(sk), NULL);
3151 if (pkt_sk(sk)->has_vnet_hdr) {
3152 struct virtio_net_hdr vnet_hdr = { 0 };
3155 vnet_hdr_len = sizeof(vnet_hdr);
3156 if (len < vnet_hdr_len)
3159 len -= vnet_hdr_len;
3161 if (skb_is_gso(skb)) {
3162 struct skb_shared_info *sinfo = skb_shinfo(skb);
3164 /* This is a hint as to how much should be linear. */
3166 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3168 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3169 if (sinfo->gso_type & SKB_GSO_TCPV4)
3170 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3171 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3172 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3173 else if (sinfo->gso_type & SKB_GSO_UDP)
3174 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3175 else if (sinfo->gso_type & SKB_GSO_FCOE)
3179 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3180 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3182 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3184 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3185 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3186 vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3187 skb_checksum_start_offset(skb));
3188 vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3190 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3191 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3192 } /* else everything is zero */
3194 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3199 /* You lose any data beyond the buffer you gave. If it worries
3200 * a user program they can ask the device for its MTU
3206 msg->msg_flags |= MSG_TRUNC;
3209 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3213 if (sock->type != SOCK_PACKET) {
3214 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3216 /* Original length was stored in sockaddr_ll fields */
3217 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3218 sll->sll_family = AF_PACKET;
3219 sll->sll_protocol = skb->protocol;
3222 sock_recv_ts_and_drops(msg, sk, skb);
3224 if (msg->msg_name) {
3225 /* If the address length field is there to be filled
3226 * in, we fill it in now.
3228 if (sock->type == SOCK_PACKET) {
3229 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3230 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3232 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3234 msg->msg_namelen = sll->sll_halen +
3235 offsetof(struct sockaddr_ll, sll_addr);
3237 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3241 if (pkt_sk(sk)->auxdata) {
3242 struct tpacket_auxdata aux;
3244 aux.tp_status = TP_STATUS_USER;
3245 if (skb->ip_summed == CHECKSUM_PARTIAL)
3246 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3247 else if (skb->pkt_type != PACKET_OUTGOING &&
3248 (skb->ip_summed == CHECKSUM_COMPLETE ||
3249 skb_csum_unnecessary(skb)))
3250 aux.tp_status |= TP_STATUS_CSUM_VALID;
3252 aux.tp_len = origlen;
3253 aux.tp_snaplen = skb->len;
3255 aux.tp_net = skb_network_offset(skb);
3256 if (skb_vlan_tag_present(skb)) {
3257 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3258 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3259 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3261 aux.tp_vlan_tci = 0;
3262 aux.tp_vlan_tpid = 0;
3264 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3268 * Free or return the buffer as appropriate. Again this
3269 * hides all the races and re-entrancy issues from us.
3271 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3274 skb_free_datagram(sk, skb);
3279 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3280 int *uaddr_len, int peer)
3282 struct net_device *dev;
3283 struct sock *sk = sock->sk;
3288 uaddr->sa_family = AF_PACKET;
3289 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3291 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3293 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3295 *uaddr_len = sizeof(*uaddr);
3300 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3301 int *uaddr_len, int peer)
3303 struct net_device *dev;
3304 struct sock *sk = sock->sk;
3305 struct packet_sock *po = pkt_sk(sk);
3306 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3311 sll->sll_family = AF_PACKET;
3312 sll->sll_ifindex = po->ifindex;
3313 sll->sll_protocol = po->num;
3314 sll->sll_pkttype = 0;
3316 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3318 sll->sll_hatype = dev->type;
3319 sll->sll_halen = dev->addr_len;
3320 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3322 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3326 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3331 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3335 case PACKET_MR_MULTICAST:
3336 if (i->alen != dev->addr_len)
3339 return dev_mc_add(dev, i->addr);
3341 return dev_mc_del(dev, i->addr);
3343 case PACKET_MR_PROMISC:
3344 return dev_set_promiscuity(dev, what);
3345 case PACKET_MR_ALLMULTI:
3346 return dev_set_allmulti(dev, what);
3347 case PACKET_MR_UNICAST:
3348 if (i->alen != dev->addr_len)
3351 return dev_uc_add(dev, i->addr);
3353 return dev_uc_del(dev, i->addr);
3361 static void packet_dev_mclist_delete(struct net_device *dev,
3362 struct packet_mclist **mlp)
3364 struct packet_mclist *ml;
3366 while ((ml = *mlp) != NULL) {
3367 if (ml->ifindex == dev->ifindex) {
3368 packet_dev_mc(dev, ml, -1);
3376 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3378 struct packet_sock *po = pkt_sk(sk);
3379 struct packet_mclist *ml, *i;
3380 struct net_device *dev;
3386 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3391 if (mreq->mr_alen > dev->addr_len)
3395 i = kmalloc(sizeof(*i), GFP_KERNEL);
3400 for (ml = po->mclist; ml; ml = ml->next) {
3401 if (ml->ifindex == mreq->mr_ifindex &&
3402 ml->type == mreq->mr_type &&
3403 ml->alen == mreq->mr_alen &&
3404 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3406 /* Free the new element ... */
3412 i->type = mreq->mr_type;
3413 i->ifindex = mreq->mr_ifindex;
3414 i->alen = mreq->mr_alen;
3415 memcpy(i->addr, mreq->mr_address, i->alen);
3417 i->next = po->mclist;
3419 err = packet_dev_mc(dev, i, 1);
3421 po->mclist = i->next;
3430 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3432 struct packet_mclist *ml, **mlp;
3436 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3437 if (ml->ifindex == mreq->mr_ifindex &&
3438 ml->type == mreq->mr_type &&
3439 ml->alen == mreq->mr_alen &&
3440 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3441 if (--ml->count == 0) {
3442 struct net_device *dev;
3444 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3446 packet_dev_mc(dev, ml, -1);
3456 static void packet_flush_mclist(struct sock *sk)
3458 struct packet_sock *po = pkt_sk(sk);
3459 struct packet_mclist *ml;
3465 while ((ml = po->mclist) != NULL) {
3466 struct net_device *dev;
3468 po->mclist = ml->next;
3469 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3471 packet_dev_mc(dev, ml, -1);
3478 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3480 struct sock *sk = sock->sk;
3481 struct packet_sock *po = pkt_sk(sk);
3484 if (level != SOL_PACKET)
3485 return -ENOPROTOOPT;
3488 case PACKET_ADD_MEMBERSHIP:
3489 case PACKET_DROP_MEMBERSHIP:
3491 struct packet_mreq_max mreq;
3493 memset(&mreq, 0, sizeof(mreq));
3494 if (len < sizeof(struct packet_mreq))
3496 if (len > sizeof(mreq))
3498 if (copy_from_user(&mreq, optval, len))
3500 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3502 if (optname == PACKET_ADD_MEMBERSHIP)
3503 ret = packet_mc_add(sk, &mreq);
3505 ret = packet_mc_drop(sk, &mreq);
3509 case PACKET_RX_RING:
3510 case PACKET_TX_RING:
3512 union tpacket_req_u req_u;
3515 switch (po->tp_version) {
3518 len = sizeof(req_u.req);
3522 len = sizeof(req_u.req3);
3527 if (pkt_sk(sk)->has_vnet_hdr)
3529 if (copy_from_user(&req_u.req, optval, len))
3531 return packet_set_ring(sk, &req_u, 0,
3532 optname == PACKET_TX_RING);
3534 case PACKET_COPY_THRESH:
3538 if (optlen != sizeof(val))
3540 if (copy_from_user(&val, optval, sizeof(val)))
3543 pkt_sk(sk)->copy_thresh = val;
3546 case PACKET_VERSION:
3550 if (optlen != sizeof(val))
3552 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3554 if (copy_from_user(&val, optval, sizeof(val)))
3560 po->tp_version = val;
3566 case PACKET_RESERVE:
3570 if (optlen != sizeof(val))
3572 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3574 if (copy_from_user(&val, optval, sizeof(val)))
3576 po->tp_reserve = val;
3583 if (optlen != sizeof(val))
3585 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3587 if (copy_from_user(&val, optval, sizeof(val)))
3589 po->tp_loss = !!val;
3592 case PACKET_AUXDATA:
3596 if (optlen < sizeof(val))
3598 if (copy_from_user(&val, optval, sizeof(val)))
3601 po->auxdata = !!val;
3604 case PACKET_ORIGDEV:
3608 if (optlen < sizeof(val))
3610 if (copy_from_user(&val, optval, sizeof(val)))
3613 po->origdev = !!val;
3616 case PACKET_VNET_HDR:
3620 if (sock->type != SOCK_RAW)
3622 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3624 if (optlen < sizeof(val))
3626 if (copy_from_user(&val, optval, sizeof(val)))
3629 po->has_vnet_hdr = !!val;
3632 case PACKET_TIMESTAMP:
3636 if (optlen != sizeof(val))
3638 if (copy_from_user(&val, optval, sizeof(val)))
3641 po->tp_tstamp = val;
3648 if (optlen != sizeof(val))
3650 if (copy_from_user(&val, optval, sizeof(val)))
3653 return fanout_add(sk, val & 0xffff, val >> 16);
3655 case PACKET_FANOUT_DATA:
3660 return fanout_set_data(po, optval, optlen);
3662 case PACKET_TX_HAS_OFF:
3666 if (optlen != sizeof(val))
3668 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3670 if (copy_from_user(&val, optval, sizeof(val)))
3672 po->tp_tx_has_off = !!val;
3675 case PACKET_QDISC_BYPASS:
3679 if (optlen != sizeof(val))
3681 if (copy_from_user(&val, optval, sizeof(val)))
3684 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3688 return -ENOPROTOOPT;
3692 static int packet_getsockopt(struct socket *sock, int level, int optname,
3693 char __user *optval, int __user *optlen)
3696 int val, lv = sizeof(val);
3697 struct sock *sk = sock->sk;
3698 struct packet_sock *po = pkt_sk(sk);
3700 union tpacket_stats_u st;
3701 struct tpacket_rollover_stats rstats;
3703 if (level != SOL_PACKET)
3704 return -ENOPROTOOPT;
3706 if (get_user(len, optlen))
3713 case PACKET_STATISTICS:
3714 spin_lock_bh(&sk->sk_receive_queue.lock);
3715 memcpy(&st, &po->stats, sizeof(st));
3716 memset(&po->stats, 0, sizeof(po->stats));
3717 spin_unlock_bh(&sk->sk_receive_queue.lock);
3719 if (po->tp_version == TPACKET_V3) {
3720 lv = sizeof(struct tpacket_stats_v3);
3721 st.stats3.tp_packets += st.stats3.tp_drops;
3724 lv = sizeof(struct tpacket_stats);
3725 st.stats1.tp_packets += st.stats1.tp_drops;
3730 case PACKET_AUXDATA:
3733 case PACKET_ORIGDEV:
3736 case PACKET_VNET_HDR:
3737 val = po->has_vnet_hdr;
3739 case PACKET_VERSION:
3740 val = po->tp_version;
3743 if (len > sizeof(int))
3745 if (copy_from_user(&val, optval, len))
3749 val = sizeof(struct tpacket_hdr);
3752 val = sizeof(struct tpacket2_hdr);
3755 val = sizeof(struct tpacket3_hdr);
3761 case PACKET_RESERVE:
3762 val = po->tp_reserve;
3767 case PACKET_TIMESTAMP:
3768 val = po->tp_tstamp;
3772 ((u32)po->fanout->id |
3773 ((u32)po->fanout->type << 16) |
3774 ((u32)po->fanout->flags << 24)) :
3777 case PACKET_ROLLOVER_STATS:
3780 rstats.tp_all = atomic_long_read(&po->rollover->num);
3781 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3782 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3784 lv = sizeof(rstats);
3786 case PACKET_TX_HAS_OFF:
3787 val = po->tp_tx_has_off;
3789 case PACKET_QDISC_BYPASS:
3790 val = packet_use_direct_xmit(po);
3793 return -ENOPROTOOPT;
3798 if (put_user(len, optlen))
3800 if (copy_to_user(optval, data, len))
3806 static int packet_notifier(struct notifier_block *this,
3807 unsigned long msg, void *ptr)
3810 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3811 struct net *net = dev_net(dev);
3814 sk_for_each_rcu(sk, &net->packet.sklist) {
3815 struct packet_sock *po = pkt_sk(sk);
3818 case NETDEV_UNREGISTER:
3820 packet_dev_mclist_delete(dev, &po->mclist);
3824 if (dev->ifindex == po->ifindex) {
3825 spin_lock(&po->bind_lock);
3827 __unregister_prot_hook(sk, false);
3828 sk->sk_err = ENETDOWN;
3829 if (!sock_flag(sk, SOCK_DEAD))
3830 sk->sk_error_report(sk);
3832 if (msg == NETDEV_UNREGISTER) {
3833 packet_cached_dev_reset(po);
3835 if (po->prot_hook.dev)
3836 dev_put(po->prot_hook.dev);
3837 po->prot_hook.dev = NULL;
3839 spin_unlock(&po->bind_lock);
3843 if (dev->ifindex == po->ifindex) {
3844 spin_lock(&po->bind_lock);
3846 register_prot_hook(sk);
3847 spin_unlock(&po->bind_lock);
3857 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3860 struct sock *sk = sock->sk;
3865 int amount = sk_wmem_alloc_get(sk);
3867 return put_user(amount, (int __user *)arg);
3871 struct sk_buff *skb;
3874 spin_lock_bh(&sk->sk_receive_queue.lock);
3875 skb = skb_peek(&sk->sk_receive_queue);
3878 spin_unlock_bh(&sk->sk_receive_queue.lock);
3879 return put_user(amount, (int __user *)arg);
3882 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3884 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3894 case SIOCGIFBRDADDR:
3895 case SIOCSIFBRDADDR:
3896 case SIOCGIFNETMASK:
3897 case SIOCSIFNETMASK:
3898 case SIOCGIFDSTADDR:
3899 case SIOCSIFDSTADDR:
3901 return inet_dgram_ops.ioctl(sock, cmd, arg);
3905 return -ENOIOCTLCMD;
3910 static unsigned int packet_poll(struct file *file, struct socket *sock,
3913 struct sock *sk = sock->sk;
3914 struct packet_sock *po = pkt_sk(sk);
3915 unsigned int mask = datagram_poll(file, sock, wait);
3917 spin_lock_bh(&sk->sk_receive_queue.lock);
3918 if (po->rx_ring.pg_vec) {
3919 if (!packet_previous_rx_frame(po, &po->rx_ring,
3921 mask |= POLLIN | POLLRDNORM;
3923 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3925 spin_unlock_bh(&sk->sk_receive_queue.lock);
3926 spin_lock_bh(&sk->sk_write_queue.lock);
3927 if (po->tx_ring.pg_vec) {
3928 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3929 mask |= POLLOUT | POLLWRNORM;
3931 spin_unlock_bh(&sk->sk_write_queue.lock);
3936 /* Dirty? Well, I still did not learn better way to account
3940 static void packet_mm_open(struct vm_area_struct *vma)
3942 struct file *file = vma->vm_file;
3943 struct socket *sock = file->private_data;
3944 struct sock *sk = sock->sk;
3947 atomic_inc(&pkt_sk(sk)->mapped);
3950 static void packet_mm_close(struct vm_area_struct *vma)
3952 struct file *file = vma->vm_file;
3953 struct socket *sock = file->private_data;
3954 struct sock *sk = sock->sk;
3957 atomic_dec(&pkt_sk(sk)->mapped);
3960 static const struct vm_operations_struct packet_mmap_ops = {
3961 .open = packet_mm_open,
3962 .close = packet_mm_close,
3965 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3970 for (i = 0; i < len; i++) {
3971 if (likely(pg_vec[i].buffer)) {
3972 if (is_vmalloc_addr(pg_vec[i].buffer))
3973 vfree(pg_vec[i].buffer);
3975 free_pages((unsigned long)pg_vec[i].buffer,
3977 pg_vec[i].buffer = NULL;
3983 static char *alloc_one_pg_vec_page(unsigned long order)
3986 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3987 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3989 buffer = (char *) __get_free_pages(gfp_flags, order);
3993 /* __get_free_pages failed, fall back to vmalloc */
3994 buffer = vzalloc((1 << order) * PAGE_SIZE);
3998 /* vmalloc failed, lets dig into swap here */
3999 gfp_flags &= ~__GFP_NORETRY;
4000 buffer = (char *) __get_free_pages(gfp_flags, order);
4004 /* complete and utter failure */
4008 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4010 unsigned int block_nr = req->tp_block_nr;
4014 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4015 if (unlikely(!pg_vec))
4018 for (i = 0; i < block_nr; i++) {
4019 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4020 if (unlikely(!pg_vec[i].buffer))
4021 goto out_free_pgvec;
4028 free_pg_vec(pg_vec, order, block_nr);
4033 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4034 int closing, int tx_ring)
4036 struct pgv *pg_vec = NULL;
4037 struct packet_sock *po = pkt_sk(sk);
4038 int was_running, order = 0;
4039 struct packet_ring_buffer *rb;
4040 struct sk_buff_head *rb_queue;
4043 /* Added to avoid minimal code churn */
4044 struct tpacket_req *req = &req_u->req;
4046 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4047 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4048 WARN(1, "Tx-ring is not supported.\n");
4052 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4053 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4057 if (atomic_read(&po->mapped))
4059 if (packet_read_pending(rb))
4063 if (req->tp_block_nr) {
4064 /* Sanity tests and some calculations */
4066 if (unlikely(rb->pg_vec))
4069 switch (po->tp_version) {
4071 po->tp_hdrlen = TPACKET_HDRLEN;
4074 po->tp_hdrlen = TPACKET2_HDRLEN;
4077 po->tp_hdrlen = TPACKET3_HDRLEN;
4082 if (unlikely((int)req->tp_block_size <= 0))
4084 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
4086 if (po->tp_version >= TPACKET_V3 &&
4087 (int)(req->tp_block_size -
4088 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4090 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4093 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4096 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
4097 if (unlikely(rb->frames_per_block <= 0))
4099 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4104 order = get_order(req->tp_block_size);
4105 pg_vec = alloc_pg_vec(req, order);
4106 if (unlikely(!pg_vec))
4108 switch (po->tp_version) {
4110 /* Transmit path is not supported. We checked
4111 * it above but just being paranoid
4114 init_prb_bdqc(po, rb, pg_vec, req_u);
4123 if (unlikely(req->tp_frame_nr))
4129 /* Detach socket from network */
4130 spin_lock(&po->bind_lock);
4131 was_running = po->running;
4135 __unregister_prot_hook(sk, false);
4137 spin_unlock(&po->bind_lock);
4142 mutex_lock(&po->pg_vec_lock);
4143 if (closing || atomic_read(&po->mapped) == 0) {
4145 spin_lock_bh(&rb_queue->lock);
4146 swap(rb->pg_vec, pg_vec);
4147 rb->frame_max = (req->tp_frame_nr - 1);
4149 rb->frame_size = req->tp_frame_size;
4150 spin_unlock_bh(&rb_queue->lock);
4152 swap(rb->pg_vec_order, order);
4153 swap(rb->pg_vec_len, req->tp_block_nr);
4155 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4156 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4157 tpacket_rcv : packet_rcv;
4158 skb_queue_purge(rb_queue);
4159 if (atomic_read(&po->mapped))
4160 pr_err("packet_mmap: vma is busy: %d\n",
4161 atomic_read(&po->mapped));
4163 mutex_unlock(&po->pg_vec_lock);
4165 spin_lock(&po->bind_lock);
4168 register_prot_hook(sk);
4170 spin_unlock(&po->bind_lock);
4171 if (closing && (po->tp_version > TPACKET_V2)) {
4172 /* Because we don't support block-based V3 on tx-ring */
4174 prb_shutdown_retire_blk_timer(po, rb_queue);
4179 free_pg_vec(pg_vec, order, req->tp_block_nr);
4184 static int packet_mmap(struct file *file, struct socket *sock,
4185 struct vm_area_struct *vma)
4187 struct sock *sk = sock->sk;
4188 struct packet_sock *po = pkt_sk(sk);
4189 unsigned long size, expected_size;
4190 struct packet_ring_buffer *rb;
4191 unsigned long start;
4198 mutex_lock(&po->pg_vec_lock);
4201 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4203 expected_size += rb->pg_vec_len
4209 if (expected_size == 0)
4212 size = vma->vm_end - vma->vm_start;
4213 if (size != expected_size)
4216 start = vma->vm_start;
4217 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4218 if (rb->pg_vec == NULL)
4221 for (i = 0; i < rb->pg_vec_len; i++) {
4223 void *kaddr = rb->pg_vec[i].buffer;
4226 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4227 page = pgv_to_page(kaddr);
4228 err = vm_insert_page(vma, start, page);
4237 atomic_inc(&po->mapped);
4238 vma->vm_ops = &packet_mmap_ops;
4242 mutex_unlock(&po->pg_vec_lock);
4246 static const struct proto_ops packet_ops_spkt = {
4247 .family = PF_PACKET,
4248 .owner = THIS_MODULE,
4249 .release = packet_release,
4250 .bind = packet_bind_spkt,
4251 .connect = sock_no_connect,
4252 .socketpair = sock_no_socketpair,
4253 .accept = sock_no_accept,
4254 .getname = packet_getname_spkt,
4255 .poll = datagram_poll,
4256 .ioctl = packet_ioctl,
4257 .listen = sock_no_listen,
4258 .shutdown = sock_no_shutdown,
4259 .setsockopt = sock_no_setsockopt,
4260 .getsockopt = sock_no_getsockopt,
4261 .sendmsg = packet_sendmsg_spkt,
4262 .recvmsg = packet_recvmsg,
4263 .mmap = sock_no_mmap,
4264 .sendpage = sock_no_sendpage,
4267 static const struct proto_ops packet_ops = {
4268 .family = PF_PACKET,
4269 .owner = THIS_MODULE,
4270 .release = packet_release,
4271 .bind = packet_bind,
4272 .connect = sock_no_connect,
4273 .socketpair = sock_no_socketpair,
4274 .accept = sock_no_accept,
4275 .getname = packet_getname,
4276 .poll = packet_poll,
4277 .ioctl = packet_ioctl,
4278 .listen = sock_no_listen,
4279 .shutdown = sock_no_shutdown,
4280 .setsockopt = packet_setsockopt,
4281 .getsockopt = packet_getsockopt,
4282 .sendmsg = packet_sendmsg,
4283 .recvmsg = packet_recvmsg,
4284 .mmap = packet_mmap,
4285 .sendpage = sock_no_sendpage,
4288 static const struct net_proto_family packet_family_ops = {
4289 .family = PF_PACKET,
4290 .create = packet_create,
4291 .owner = THIS_MODULE,
4294 static struct notifier_block packet_netdev_notifier = {
4295 .notifier_call = packet_notifier,
4298 #ifdef CONFIG_PROC_FS
4300 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4303 struct net *net = seq_file_net(seq);
4306 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4309 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4311 struct net *net = seq_file_net(seq);
4312 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4315 static void packet_seq_stop(struct seq_file *seq, void *v)
4321 static int packet_seq_show(struct seq_file *seq, void *v)
4323 if (v == SEQ_START_TOKEN)
4324 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4326 struct sock *s = sk_entry(v);
4327 const struct packet_sock *po = pkt_sk(s);
4330 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4332 atomic_read(&s->sk_refcnt),
4337 atomic_read(&s->sk_rmem_alloc),
4338 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4345 static const struct seq_operations packet_seq_ops = {
4346 .start = packet_seq_start,
4347 .next = packet_seq_next,
4348 .stop = packet_seq_stop,
4349 .show = packet_seq_show,
4352 static int packet_seq_open(struct inode *inode, struct file *file)
4354 return seq_open_net(inode, file, &packet_seq_ops,
4355 sizeof(struct seq_net_private));
4358 static const struct file_operations packet_seq_fops = {
4359 .owner = THIS_MODULE,
4360 .open = packet_seq_open,
4362 .llseek = seq_lseek,
4363 .release = seq_release_net,
4368 static int __net_init packet_net_init(struct net *net)
4370 mutex_init(&net->packet.sklist_lock);
4371 INIT_HLIST_HEAD(&net->packet.sklist);
4373 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4379 static void __net_exit packet_net_exit(struct net *net)
4381 remove_proc_entry("packet", net->proc_net);
4384 static struct pernet_operations packet_net_ops = {
4385 .init = packet_net_init,
4386 .exit = packet_net_exit,
4390 static void __exit packet_exit(void)
4392 unregister_netdevice_notifier(&packet_netdev_notifier);
4393 unregister_pernet_subsys(&packet_net_ops);
4394 sock_unregister(PF_PACKET);
4395 proto_unregister(&packet_proto);
4398 static int __init packet_init(void)
4400 int rc = proto_register(&packet_proto, 0);
4405 sock_register(&packet_family_ops);
4406 register_pernet_subsys(&packet_net_ops);
4407 register_netdevice_notifier(&packet_netdev_notifier);
4412 module_init(packet_init);
4413 module_exit(packet_exit);
4414 MODULE_LICENSE("GPL");
4415 MODULE_ALIAS_NETPROTO(PF_PACKET);