2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
218 struct packet_skb_cb {
219 unsigned int origlen;
221 struct sockaddr_pkt pkt;
222 struct sockaddr_ll ll;
226 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
228 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
229 #define GET_PBLOCK_DESC(x, bid) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
233 #define GET_NEXT_PRB_BLK_NUM(x) \
234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 ((x)->kactive_blk_num+1) : 0)
237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
240 static int packet_direct_xmit(struct sk_buff *skb)
242 struct net_device *dev = skb->dev;
243 const struct net_device_ops *ops = dev->netdev_ops;
244 netdev_features_t features;
245 struct netdev_queue *txq;
249 if (unlikely(!netif_running(dev) ||
250 !netif_carrier_ok(dev))) {
252 return NET_XMIT_DROP;
255 features = netif_skb_features(skb);
256 if (skb_needs_linearize(skb, features) &&
257 __skb_linearize(skb)) {
259 return NET_XMIT_DROP;
262 queue_map = skb_get_queue_mapping(skb);
263 txq = netdev_get_tx_queue(dev, queue_map);
265 __netif_tx_lock_bh(txq);
266 if (unlikely(netif_xmit_frozen_or_stopped(txq))) {
267 ret = NETDEV_TX_BUSY;
272 ret = ops->ndo_start_xmit(skb, dev);
273 if (likely(dev_xmit_complete(ret)))
274 txq_trans_update(txq);
278 __netif_tx_unlock_bh(txq);
282 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
284 struct net_device *dev;
287 dev = rcu_dereference(po->cached_dev);
295 static void packet_cached_dev_assign(struct packet_sock *po,
296 struct net_device *dev)
298 rcu_assign_pointer(po->cached_dev, dev);
301 static void packet_cached_dev_reset(struct packet_sock *po)
303 RCU_INIT_POINTER(po->cached_dev, NULL);
306 static bool packet_use_direct_xmit(const struct packet_sock *po)
308 return po->xmit == packet_direct_xmit;
311 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
313 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
316 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
318 const struct net_device_ops *ops = dev->netdev_ops;
321 if (ops->ndo_select_queue) {
322 queue_index = ops->ndo_select_queue(dev, skb, NULL,
323 __packet_pick_tx_queue);
324 queue_index = netdev_cap_txqueue(dev, queue_index);
326 queue_index = __packet_pick_tx_queue(dev, skb);
329 skb_set_queue_mapping(skb, queue_index);
332 /* register_prot_hook must be invoked with the po->bind_lock held,
333 * or from a context in which asynchronous accesses to the packet
334 * socket is not possible (packet_create()).
336 static void register_prot_hook(struct sock *sk)
338 struct packet_sock *po = pkt_sk(sk);
342 __fanout_link(sk, po);
344 dev_add_pack(&po->prot_hook);
351 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
352 * held. If the sync parameter is true, we will temporarily drop
353 * the po->bind_lock and do a synchronize_net to make sure no
354 * asynchronous packet processing paths still refer to the elements
355 * of po->prot_hook. If the sync parameter is false, it is the
356 * callers responsibility to take care of this.
358 static void __unregister_prot_hook(struct sock *sk, bool sync)
360 struct packet_sock *po = pkt_sk(sk);
365 __fanout_unlink(sk, po);
367 __dev_remove_pack(&po->prot_hook);
372 spin_unlock(&po->bind_lock);
374 spin_lock(&po->bind_lock);
378 static void unregister_prot_hook(struct sock *sk, bool sync)
380 struct packet_sock *po = pkt_sk(sk);
383 __unregister_prot_hook(sk, sync);
386 static inline __pure struct page *pgv_to_page(void *addr)
388 if (is_vmalloc_addr(addr))
389 return vmalloc_to_page(addr);
390 return virt_to_page(addr);
393 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
395 union tpacket_uhdr h;
398 switch (po->tp_version) {
400 h.h1->tp_status = status;
401 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
404 h.h2->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
409 WARN(1, "TPACKET version not supported.\n");
416 static int __packet_get_status(struct packet_sock *po, void *frame)
418 union tpacket_uhdr h;
423 switch (po->tp_version) {
425 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
426 return h.h1->tp_status;
428 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
429 return h.h2->tp_status;
432 WARN(1, "TPACKET version not supported.\n");
438 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
441 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
444 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
445 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
446 return TP_STATUS_TS_SYS_HARDWARE;
447 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
448 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
449 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,
525 struct sk_buff_head *rb_queue)
527 struct tpacket_kbdq_core *pkc;
529 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
530 GET_PBDQC_FROM_RB(&po->rx_ring);
532 spin_lock_bh(&rb_queue->lock);
533 pkc->delete_blk_timer = 1;
534 spin_unlock_bh(&rb_queue->lock);
536 prb_del_retire_blk_timer(pkc);
539 static void prb_init_blk_timer(struct packet_sock *po,
540 struct tpacket_kbdq_core *pkc,
541 void (*func) (unsigned long))
543 init_timer(&pkc->retire_blk_timer);
544 pkc->retire_blk_timer.data = (long)po;
545 pkc->retire_blk_timer.function = func;
546 pkc->retire_blk_timer.expires = jiffies;
549 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
551 struct tpacket_kbdq_core *pkc;
556 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
557 GET_PBDQC_FROM_RB(&po->rx_ring);
558 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
561 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
562 int blk_size_in_bytes)
564 struct net_device *dev;
565 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
566 struct ethtool_cmd ecmd;
571 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
572 if (unlikely(!dev)) {
574 return DEFAULT_PRB_RETIRE_TOV;
576 err = __ethtool_get_settings(dev, &ecmd);
577 speed = ethtool_cmd_speed(&ecmd);
581 * If the link speed is so slow you don't really
582 * need to worry about perf anyways
584 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
585 return DEFAULT_PRB_RETIRE_TOV;
592 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
604 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
605 union tpacket_req_u *req_u)
607 p1->feature_req_word = req_u->req3.tp_feature_req_word;
610 static void init_prb_bdqc(struct packet_sock *po,
611 struct packet_ring_buffer *rb,
613 union tpacket_req_u *req_u, int tx_ring)
615 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
616 struct tpacket_block_desc *pbd;
618 memset(p1, 0x0, sizeof(*p1));
620 p1->knxt_seq_num = 1;
622 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
623 p1->pkblk_start = pg_vec[0].buffer;
624 p1->kblk_size = req_u->req3.tp_block_size;
625 p1->knum_blocks = req_u->req3.tp_block_nr;
626 p1->hdrlen = po->tp_hdrlen;
627 p1->version = po->tp_version;
628 p1->last_kactive_blk_num = 0;
629 po->stats.stats3.tp_freeze_q_cnt = 0;
630 if (req_u->req3.tp_retire_blk_tov)
631 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
633 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
634 req_u->req3.tp_block_size);
635 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
636 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
638 prb_init_ft_ops(p1, req_u);
639 prb_setup_retire_blk_timer(po, tx_ring);
640 prb_open_block(p1, pbd);
643 /* Do NOT update the last_blk_num first.
644 * Assumes sk_buff_head lock is held.
646 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
648 mod_timer(&pkc->retire_blk_timer,
649 jiffies + pkc->tov_in_jiffies);
650 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
655 * 1) We refresh the timer only when we open a block.
656 * By doing this we don't waste cycles refreshing the timer
657 * on packet-by-packet basis.
659 * With a 1MB block-size, on a 1Gbps line, it will take
660 * i) ~8 ms to fill a block + ii) memcpy etc.
661 * In this cut we are not accounting for the memcpy time.
663 * So, if the user sets the 'tmo' to 10ms then the timer
664 * will never fire while the block is still getting filled
665 * (which is what we want). However, the user could choose
666 * to close a block early and that's fine.
668 * But when the timer does fire, we check whether or not to refresh it.
669 * Since the tmo granularity is in msecs, it is not too expensive
670 * to refresh the timer, lets say every '8' msecs.
671 * Either the user can set the 'tmo' or we can derive it based on
672 * a) line-speed and b) block-size.
673 * prb_calc_retire_blk_tmo() calculates the tmo.
676 static void prb_retire_rx_blk_timer_expired(unsigned long data)
678 struct packet_sock *po = (struct packet_sock *)data;
679 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
681 struct tpacket_block_desc *pbd;
683 spin_lock(&po->sk.sk_receive_queue.lock);
685 frozen = prb_queue_frozen(pkc);
686 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
688 if (unlikely(pkc->delete_blk_timer))
691 /* We only need to plug the race when the block is partially filled.
693 * lock(); increment BLOCK_NUM_PKTS; unlock()
694 * copy_bits() is in progress ...
695 * timer fires on other cpu:
696 * we can't retire the current block because copy_bits
700 if (BLOCK_NUM_PKTS(pbd)) {
701 while (atomic_read(&pkc->blk_fill_in_prog)) {
702 /* Waiting for skb_copy_bits to finish... */
707 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
709 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
710 if (!prb_dispatch_next_block(pkc, po))
715 /* Case 1. Queue was frozen because user-space was
718 if (prb_curr_blk_in_use(pkc, pbd)) {
720 * Ok, user-space is still behind.
721 * So just refresh the timer.
725 /* Case 2. queue was frozen,user-space caught up,
726 * now the link went idle && the timer fired.
727 * We don't have a block to close.So we open this
728 * block and restart the timer.
729 * opening a block thaws the queue,restarts timer
730 * Thawing/timer-refresh is a side effect.
732 prb_open_block(pkc, pbd);
739 _prb_refresh_rx_retire_blk_timer(pkc);
742 spin_unlock(&po->sk.sk_receive_queue.lock);
745 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
746 struct tpacket_block_desc *pbd1, __u32 status)
748 /* Flush everything minus the block header */
750 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
755 /* Skip the block header(we know header WILL fit in 4K) */
758 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
759 for (; start < end; start += PAGE_SIZE)
760 flush_dcache_page(pgv_to_page(start));
765 /* Now update the block status. */
767 BLOCK_STATUS(pbd1) = status;
769 /* Flush the block header */
771 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
773 flush_dcache_page(pgv_to_page(start));
783 * 2) Increment active_blk_num
785 * Note:We DONT refresh the timer on purpose.
786 * Because almost always the next block will be opened.
788 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
789 struct tpacket_block_desc *pbd1,
790 struct packet_sock *po, unsigned int stat)
792 __u32 status = TP_STATUS_USER | stat;
794 struct tpacket3_hdr *last_pkt;
795 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
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 */
811 h1->ts_last_pkt.ts_sec = ts.tv_sec;
812 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
817 /* Flush the block */
818 prb_flush_block(pkc1, pbd1, status);
820 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
823 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
825 pkc->reset_pending_on_curr_blk = 0;
829 * Side effect of opening a block:
831 * 1) prb_queue is thawed.
832 * 2) retire_blk_timer is refreshed.
835 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
836 struct tpacket_block_desc *pbd1)
839 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
843 /* We could have just memset this but we will lose the
844 * flexibility of making the priv area sticky
847 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
848 BLOCK_NUM_PKTS(pbd1) = 0;
849 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
853 h1->ts_first_pkt.ts_sec = ts.tv_sec;
854 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
856 pkc1->pkblk_start = (char *)pbd1;
857 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
859 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
860 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
862 pbd1->version = pkc1->version;
863 pkc1->prev = pkc1->nxt_offset;
864 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
866 prb_thaw_queue(pkc1);
867 _prb_refresh_rx_retire_blk_timer(pkc1);
873 * Queue freeze logic:
874 * 1) Assume tp_block_nr = 8 blocks.
875 * 2) At time 't0', user opens Rx ring.
876 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
877 * 4) user-space is either sleeping or processing block '0'.
878 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
879 * it will close block-7,loop around and try to fill block '0'.
881 * __packet_lookup_frame_in_block
882 * prb_retire_current_block()
883 * prb_dispatch_next_block()
884 * |->(BLOCK_STATUS == USER) evaluates to true
885 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
886 * 6) Now there are two cases:
887 * 6.1) Link goes idle right after the queue is frozen.
888 * But remember, the last open_block() refreshed the timer.
889 * When this timer expires,it will refresh itself so that we can
890 * re-open block-0 in near future.
891 * 6.2) Link is busy and keeps on receiving packets. This is a simple
892 * case and __packet_lookup_frame_in_block will check if block-0
893 * is free and can now be re-used.
895 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
896 struct packet_sock *po)
898 pkc->reset_pending_on_curr_blk = 1;
899 po->stats.stats3.tp_freeze_q_cnt++;
902 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
905 * If the next block is free then we will dispatch it
906 * and return a good offset.
907 * Else, we will freeze the queue.
908 * So, caller must check the return value.
910 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
911 struct packet_sock *po)
913 struct tpacket_block_desc *pbd;
917 /* 1. Get current block num */
918 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
920 /* 2. If this block is currently in_use then freeze the queue */
921 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
922 prb_freeze_queue(pkc, po);
928 * open this block and return the offset where the first packet
929 * needs to get stored.
931 prb_open_block(pkc, pbd);
932 return (void *)pkc->nxt_offset;
935 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
936 struct packet_sock *po, unsigned int status)
938 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
940 /* retire/close the current block */
941 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
943 * Plug the case where copy_bits() is in progress on
944 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
945 * have space to copy the pkt in the current block and
946 * called prb_retire_current_block()
948 * We don't need to worry about the TMO case because
949 * the timer-handler already handled this case.
951 if (!(status & TP_STATUS_BLK_TMO)) {
952 while (atomic_read(&pkc->blk_fill_in_prog)) {
953 /* Waiting for skb_copy_bits to finish... */
957 prb_close_block(pkc, pbd, po, status);
962 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
963 struct tpacket_block_desc *pbd)
965 return TP_STATUS_USER & BLOCK_STATUS(pbd);
968 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
970 return pkc->reset_pending_on_curr_blk;
973 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
975 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
976 atomic_dec(&pkc->blk_fill_in_prog);
979 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
980 struct tpacket3_hdr *ppd)
982 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
985 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
986 struct tpacket3_hdr *ppd)
988 ppd->hv1.tp_rxhash = 0;
991 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
992 struct tpacket3_hdr *ppd)
994 if (vlan_tx_tag_present(pkc->skb)) {
995 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
996 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
997 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
999 ppd->hv1.tp_vlan_tci = 0;
1000 ppd->hv1.tp_vlan_tpid = 0;
1001 ppd->tp_status = TP_STATUS_AVAILABLE;
1005 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1006 struct tpacket3_hdr *ppd)
1008 ppd->hv1.tp_padding = 0;
1009 prb_fill_vlan_info(pkc, ppd);
1011 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1012 prb_fill_rxhash(pkc, ppd);
1014 prb_clear_rxhash(pkc, ppd);
1017 static void prb_fill_curr_block(char *curr,
1018 struct tpacket_kbdq_core *pkc,
1019 struct tpacket_block_desc *pbd,
1022 struct tpacket3_hdr *ppd;
1024 ppd = (struct tpacket3_hdr *)curr;
1025 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1027 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1028 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1029 BLOCK_NUM_PKTS(pbd) += 1;
1030 atomic_inc(&pkc->blk_fill_in_prog);
1031 prb_run_all_ft_ops(pkc, ppd);
1034 /* Assumes caller has the sk->rx_queue.lock */
1035 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1036 struct sk_buff *skb,
1041 struct tpacket_kbdq_core *pkc;
1042 struct tpacket_block_desc *pbd;
1045 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1046 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1048 /* Queue is frozen when user space is lagging behind */
1049 if (prb_queue_frozen(pkc)) {
1051 * Check if that last block which caused the queue to freeze,
1052 * is still in_use by user-space.
1054 if (prb_curr_blk_in_use(pkc, pbd)) {
1055 /* Can't record this packet */
1059 * Ok, the block was released by user-space.
1060 * Now let's open that block.
1061 * opening a block also thaws the queue.
1062 * Thawing is a side effect.
1064 prb_open_block(pkc, pbd);
1069 curr = pkc->nxt_offset;
1071 end = (char *)pbd + pkc->kblk_size;
1073 /* first try the current block */
1074 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1075 prb_fill_curr_block(curr, pkc, pbd, len);
1076 return (void *)curr;
1079 /* Ok, close the current block */
1080 prb_retire_current_block(pkc, po, 0);
1082 /* Now, try to dispatch the next block */
1083 curr = (char *)prb_dispatch_next_block(pkc, po);
1085 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1086 prb_fill_curr_block(curr, pkc, pbd, len);
1087 return (void *)curr;
1091 * No free blocks are available.user_space hasn't caught up yet.
1092 * Queue was just frozen and now this packet will get dropped.
1097 static void *packet_current_rx_frame(struct packet_sock *po,
1098 struct sk_buff *skb,
1099 int status, unsigned int len)
1102 switch (po->tp_version) {
1105 curr = packet_lookup_frame(po, &po->rx_ring,
1106 po->rx_ring.head, status);
1109 return __packet_lookup_frame_in_block(po, skb, status, len);
1111 WARN(1, "TPACKET version not supported\n");
1117 static void *prb_lookup_block(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1122 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1123 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1125 if (status != BLOCK_STATUS(pbd))
1130 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1133 if (rb->prb_bdqc.kactive_blk_num)
1134 prev = rb->prb_bdqc.kactive_blk_num-1;
1136 prev = rb->prb_bdqc.knum_blocks-1;
1140 /* Assumes caller has held the rx_queue.lock */
1141 static void *__prb_previous_block(struct packet_sock *po,
1142 struct packet_ring_buffer *rb,
1145 unsigned int previous = prb_previous_blk_num(rb);
1146 return prb_lookup_block(po, rb, previous, status);
1149 static void *packet_previous_rx_frame(struct packet_sock *po,
1150 struct packet_ring_buffer *rb,
1153 if (po->tp_version <= TPACKET_V2)
1154 return packet_previous_frame(po, rb, status);
1156 return __prb_previous_block(po, rb, status);
1159 static void packet_increment_rx_head(struct packet_sock *po,
1160 struct packet_ring_buffer *rb)
1162 switch (po->tp_version) {
1165 return packet_increment_head(rb);
1168 WARN(1, "TPACKET version not supported.\n");
1174 static void *packet_previous_frame(struct packet_sock *po,
1175 struct packet_ring_buffer *rb,
1178 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1179 return packet_lookup_frame(po, rb, previous, status);
1182 static void packet_increment_head(struct packet_ring_buffer *buff)
1184 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1187 static void packet_inc_pending(struct packet_ring_buffer *rb)
1189 this_cpu_inc(*rb->pending_refcnt);
1192 static void packet_dec_pending(struct packet_ring_buffer *rb)
1194 this_cpu_dec(*rb->pending_refcnt);
1197 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1199 unsigned int refcnt = 0;
1202 /* We don't use pending refcount in rx_ring. */
1203 if (rb->pending_refcnt == NULL)
1206 for_each_possible_cpu(cpu)
1207 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1212 static int packet_alloc_pending(struct packet_sock *po)
1214 po->rx_ring.pending_refcnt = NULL;
1216 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1217 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1223 static void packet_free_pending(struct packet_sock *po)
1225 free_percpu(po->tx_ring.pending_refcnt);
1228 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1230 struct sock *sk = &po->sk;
1233 if (po->prot_hook.func != tpacket_rcv)
1234 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1237 spin_lock(&sk->sk_receive_queue.lock);
1238 if (po->tp_version == TPACKET_V3)
1239 has_room = prb_lookup_block(po, &po->rx_ring,
1240 po->rx_ring.prb_bdqc.kactive_blk_num,
1243 has_room = packet_lookup_frame(po, &po->rx_ring,
1246 spin_unlock(&sk->sk_receive_queue.lock);
1251 static void packet_sock_destruct(struct sock *sk)
1253 skb_queue_purge(&sk->sk_error_queue);
1255 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1256 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1258 if (!sock_flag(sk, SOCK_DEAD)) {
1259 pr_err("Attempt to release alive packet socket: %p\n", sk);
1263 sk_refcnt_debug_dec(sk);
1266 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1268 int x = atomic_read(&f->rr_cur) + 1;
1276 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1277 struct sk_buff *skb,
1280 return reciprocal_scale(skb->rxhash, num);
1283 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1284 struct sk_buff *skb,
1289 cur = atomic_read(&f->rr_cur);
1290 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1291 fanout_rr_next(f, num))) != cur)
1296 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1297 struct sk_buff *skb,
1300 return smp_processor_id() % num;
1303 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1304 struct sk_buff *skb,
1307 return prandom_u32_max(num);
1310 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1311 struct sk_buff *skb,
1312 unsigned int idx, unsigned int skip,
1317 i = j = min_t(int, f->next[idx], num - 1);
1319 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1331 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1332 struct sk_buff *skb,
1335 return skb_get_queue_mapping(skb) % num;
1338 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1340 return f->flags & (flag >> 8);
1343 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1344 struct packet_type *pt, struct net_device *orig_dev)
1346 struct packet_fanout *f = pt->af_packet_priv;
1347 unsigned int num = f->num_members;
1348 struct packet_sock *po;
1351 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1358 case PACKET_FANOUT_HASH:
1360 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1361 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1366 idx = fanout_demux_hash(f, skb, num);
1368 case PACKET_FANOUT_LB:
1369 idx = fanout_demux_lb(f, skb, num);
1371 case PACKET_FANOUT_CPU:
1372 idx = fanout_demux_cpu(f, skb, num);
1374 case PACKET_FANOUT_RND:
1375 idx = fanout_demux_rnd(f, skb, num);
1377 case PACKET_FANOUT_QM:
1378 idx = fanout_demux_qm(f, skb, num);
1380 case PACKET_FANOUT_ROLLOVER:
1381 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1385 po = pkt_sk(f->arr[idx]);
1386 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1387 unlikely(!packet_rcv_has_room(po, skb))) {
1388 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1389 po = pkt_sk(f->arr[idx]);
1392 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1395 DEFINE_MUTEX(fanout_mutex);
1396 EXPORT_SYMBOL_GPL(fanout_mutex);
1397 static LIST_HEAD(fanout_list);
1399 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1401 struct packet_fanout *f = po->fanout;
1403 spin_lock(&f->lock);
1404 f->arr[f->num_members] = sk;
1407 spin_unlock(&f->lock);
1410 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1412 struct packet_fanout *f = po->fanout;
1415 spin_lock(&f->lock);
1416 for (i = 0; i < f->num_members; i++) {
1417 if (f->arr[i] == sk)
1420 BUG_ON(i >= f->num_members);
1421 f->arr[i] = f->arr[f->num_members - 1];
1423 spin_unlock(&f->lock);
1426 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1428 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1434 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1436 struct packet_sock *po = pkt_sk(sk);
1437 struct packet_fanout *f, *match;
1438 u8 type = type_flags & 0xff;
1439 u8 flags = type_flags >> 8;
1443 case PACKET_FANOUT_ROLLOVER:
1444 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1446 case PACKET_FANOUT_HASH:
1447 case PACKET_FANOUT_LB:
1448 case PACKET_FANOUT_CPU:
1449 case PACKET_FANOUT_RND:
1450 case PACKET_FANOUT_QM:
1462 mutex_lock(&fanout_mutex);
1464 list_for_each_entry(f, &fanout_list, list) {
1466 read_pnet(&f->net) == sock_net(sk)) {
1472 if (match && match->flags != flags)
1476 match = kzalloc(sizeof(*match), GFP_KERNEL);
1479 write_pnet(&match->net, sock_net(sk));
1482 match->flags = flags;
1483 atomic_set(&match->rr_cur, 0);
1484 INIT_LIST_HEAD(&match->list);
1485 spin_lock_init(&match->lock);
1486 atomic_set(&match->sk_ref, 0);
1487 match->prot_hook.type = po->prot_hook.type;
1488 match->prot_hook.dev = po->prot_hook.dev;
1489 match->prot_hook.func = packet_rcv_fanout;
1490 match->prot_hook.af_packet_priv = match;
1491 match->prot_hook.id_match = match_fanout_group;
1492 dev_add_pack(&match->prot_hook);
1493 list_add(&match->list, &fanout_list);
1496 if (match->type == type &&
1497 match->prot_hook.type == po->prot_hook.type &&
1498 match->prot_hook.dev == po->prot_hook.dev) {
1500 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1501 __dev_remove_pack(&po->prot_hook);
1503 atomic_inc(&match->sk_ref);
1504 __fanout_link(sk, po);
1509 mutex_unlock(&fanout_mutex);
1513 static void fanout_release(struct sock *sk)
1515 struct packet_sock *po = pkt_sk(sk);
1516 struct packet_fanout *f;
1522 mutex_lock(&fanout_mutex);
1525 if (atomic_dec_and_test(&f->sk_ref)) {
1527 dev_remove_pack(&f->prot_hook);
1530 mutex_unlock(&fanout_mutex);
1533 static const struct proto_ops packet_ops;
1535 static const struct proto_ops packet_ops_spkt;
1537 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1538 struct packet_type *pt, struct net_device *orig_dev)
1541 struct sockaddr_pkt *spkt;
1544 * When we registered the protocol we saved the socket in the data
1545 * field for just this event.
1548 sk = pt->af_packet_priv;
1551 * Yank back the headers [hope the device set this
1552 * right or kerboom...]
1554 * Incoming packets have ll header pulled,
1557 * For outgoing ones skb->data == skb_mac_header(skb)
1558 * so that this procedure is noop.
1561 if (skb->pkt_type == PACKET_LOOPBACK)
1564 if (!net_eq(dev_net(dev), sock_net(sk)))
1567 skb = skb_share_check(skb, GFP_ATOMIC);
1571 /* drop any routing info */
1574 /* drop conntrack reference */
1577 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1579 skb_push(skb, skb->data - skb_mac_header(skb));
1582 * The SOCK_PACKET socket receives _all_ frames.
1585 spkt->spkt_family = dev->type;
1586 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1587 spkt->spkt_protocol = skb->protocol;
1590 * Charge the memory to the socket. This is done specifically
1591 * to prevent sockets using all the memory up.
1594 if (sock_queue_rcv_skb(sk, skb) == 0)
1605 * Output a raw packet to a device layer. This bypasses all the other
1606 * protocol layers and you must therefore supply it with a complete frame
1609 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1610 struct msghdr *msg, size_t len)
1612 struct sock *sk = sock->sk;
1613 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1614 struct sk_buff *skb = NULL;
1615 struct net_device *dev;
1621 * Get and verify the address.
1625 if (msg->msg_namelen < sizeof(struct sockaddr))
1627 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1628 proto = saddr->spkt_protocol;
1630 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1633 * Find the device first to size check it
1636 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1639 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1645 if (!(dev->flags & IFF_UP))
1649 * You may not queue a frame bigger than the mtu. This is the lowest level
1650 * raw protocol and you must do your own fragmentation at this level.
1653 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1654 if (!netif_supports_nofcs(dev)) {
1655 err = -EPROTONOSUPPORT;
1658 extra_len = 4; /* We're doing our own CRC */
1662 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1666 size_t reserved = LL_RESERVED_SPACE(dev);
1667 int tlen = dev->needed_tailroom;
1668 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1671 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1674 /* FIXME: Save some space for broken drivers that write a hard
1675 * header at transmission time by themselves. PPP is the notable
1676 * one here. This should really be fixed at the driver level.
1678 skb_reserve(skb, reserved);
1679 skb_reset_network_header(skb);
1681 /* Try to align data part correctly */
1686 skb_reset_network_header(skb);
1688 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1694 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1695 /* Earlier code assumed this would be a VLAN pkt,
1696 * double-check this now that we have the actual
1699 struct ethhdr *ehdr;
1700 skb_reset_mac_header(skb);
1701 ehdr = eth_hdr(skb);
1702 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1708 skb->protocol = proto;
1710 skb->priority = sk->sk_priority;
1711 skb->mark = sk->sk_mark;
1713 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1715 if (unlikely(extra_len == 4))
1718 skb_probe_transport_header(skb, 0);
1720 dev_queue_xmit(skb);
1731 static unsigned int run_filter(const struct sk_buff *skb,
1732 const struct sock *sk,
1735 struct sk_filter *filter;
1738 filter = rcu_dereference(sk->sk_filter);
1740 res = SK_RUN_FILTER(filter, skb);
1747 * This function makes lazy skb cloning in hope that most of packets
1748 * are discarded by BPF.
1750 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1751 * and skb->cb are mangled. It works because (and until) packets
1752 * falling here are owned by current CPU. Output packets are cloned
1753 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1754 * sequencially, so that if we return skb to original state on exit,
1755 * we will not harm anyone.
1758 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1759 struct packet_type *pt, struct net_device *orig_dev)
1762 struct sockaddr_ll *sll;
1763 struct packet_sock *po;
1764 u8 *skb_head = skb->data;
1765 int skb_len = skb->len;
1766 unsigned int snaplen, res;
1768 if (skb->pkt_type == PACKET_LOOPBACK)
1771 sk = pt->af_packet_priv;
1774 if (!net_eq(dev_net(dev), sock_net(sk)))
1779 if (dev->header_ops) {
1780 /* The device has an explicit notion of ll header,
1781 * exported to higher levels.
1783 * Otherwise, the device hides details of its frame
1784 * structure, so that corresponding packet head is
1785 * never delivered to user.
1787 if (sk->sk_type != SOCK_DGRAM)
1788 skb_push(skb, skb->data - skb_mac_header(skb));
1789 else if (skb->pkt_type == PACKET_OUTGOING) {
1790 /* Special case: outgoing packets have ll header at head */
1791 skb_pull(skb, skb_network_offset(skb));
1797 res = run_filter(skb, sk, snaplen);
1799 goto drop_n_restore;
1803 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1806 if (skb_shared(skb)) {
1807 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1811 if (skb_head != skb->data) {
1812 skb->data = skb_head;
1819 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1822 sll = &PACKET_SKB_CB(skb)->sa.ll;
1823 sll->sll_family = AF_PACKET;
1824 sll->sll_hatype = dev->type;
1825 sll->sll_protocol = skb->protocol;
1826 sll->sll_pkttype = skb->pkt_type;
1827 if (unlikely(po->origdev))
1828 sll->sll_ifindex = orig_dev->ifindex;
1830 sll->sll_ifindex = dev->ifindex;
1832 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1834 PACKET_SKB_CB(skb)->origlen = skb->len;
1836 if (pskb_trim(skb, snaplen))
1839 skb_set_owner_r(skb, sk);
1843 /* drop conntrack reference */
1846 spin_lock(&sk->sk_receive_queue.lock);
1847 po->stats.stats1.tp_packets++;
1848 skb->dropcount = atomic_read(&sk->sk_drops);
1849 __skb_queue_tail(&sk->sk_receive_queue, skb);
1850 spin_unlock(&sk->sk_receive_queue.lock);
1851 sk->sk_data_ready(sk, skb->len);
1855 spin_lock(&sk->sk_receive_queue.lock);
1856 po->stats.stats1.tp_drops++;
1857 atomic_inc(&sk->sk_drops);
1858 spin_unlock(&sk->sk_receive_queue.lock);
1861 if (skb_head != skb->data && skb_shared(skb)) {
1862 skb->data = skb_head;
1870 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1871 struct packet_type *pt, struct net_device *orig_dev)
1874 struct packet_sock *po;
1875 struct sockaddr_ll *sll;
1876 union tpacket_uhdr h;
1877 u8 *skb_head = skb->data;
1878 int skb_len = skb->len;
1879 unsigned int snaplen, res;
1880 unsigned long status = TP_STATUS_USER;
1881 unsigned short macoff, netoff, hdrlen;
1882 struct sk_buff *copy_skb = NULL;
1886 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1887 * We may add members to them until current aligned size without forcing
1888 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1890 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1891 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1893 if (skb->pkt_type == PACKET_LOOPBACK)
1896 sk = pt->af_packet_priv;
1899 if (!net_eq(dev_net(dev), sock_net(sk)))
1902 if (dev->header_ops) {
1903 if (sk->sk_type != SOCK_DGRAM)
1904 skb_push(skb, skb->data - skb_mac_header(skb));
1905 else if (skb->pkt_type == PACKET_OUTGOING) {
1906 /* Special case: outgoing packets have ll header at head */
1907 skb_pull(skb, skb_network_offset(skb));
1911 if (skb->ip_summed == CHECKSUM_PARTIAL)
1912 status |= TP_STATUS_CSUMNOTREADY;
1916 res = run_filter(skb, sk, snaplen);
1918 goto drop_n_restore;
1922 if (sk->sk_type == SOCK_DGRAM) {
1923 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1926 unsigned int maclen = skb_network_offset(skb);
1927 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1928 (maclen < 16 ? 16 : maclen)) +
1930 macoff = netoff - maclen;
1932 if (po->tp_version <= TPACKET_V2) {
1933 if (macoff + snaplen > po->rx_ring.frame_size) {
1934 if (po->copy_thresh &&
1935 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1936 if (skb_shared(skb)) {
1937 copy_skb = skb_clone(skb, GFP_ATOMIC);
1939 copy_skb = skb_get(skb);
1940 skb_head = skb->data;
1943 skb_set_owner_r(copy_skb, sk);
1945 snaplen = po->rx_ring.frame_size - macoff;
1946 if ((int)snaplen < 0)
1950 spin_lock(&sk->sk_receive_queue.lock);
1951 h.raw = packet_current_rx_frame(po, skb,
1952 TP_STATUS_KERNEL, (macoff+snaplen));
1955 if (po->tp_version <= TPACKET_V2) {
1956 packet_increment_rx_head(po, &po->rx_ring);
1958 * LOSING will be reported till you read the stats,
1959 * because it's COR - Clear On Read.
1960 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1963 if (po->stats.stats1.tp_drops)
1964 status |= TP_STATUS_LOSING;
1966 po->stats.stats1.tp_packets++;
1968 status |= TP_STATUS_COPY;
1969 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1971 spin_unlock(&sk->sk_receive_queue.lock);
1973 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1975 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1976 getnstimeofday(&ts);
1978 status |= ts_status;
1980 switch (po->tp_version) {
1982 h.h1->tp_len = skb->len;
1983 h.h1->tp_snaplen = snaplen;
1984 h.h1->tp_mac = macoff;
1985 h.h1->tp_net = netoff;
1986 h.h1->tp_sec = ts.tv_sec;
1987 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1988 hdrlen = sizeof(*h.h1);
1991 h.h2->tp_len = skb->len;
1992 h.h2->tp_snaplen = snaplen;
1993 h.h2->tp_mac = macoff;
1994 h.h2->tp_net = netoff;
1995 h.h2->tp_sec = ts.tv_sec;
1996 h.h2->tp_nsec = ts.tv_nsec;
1997 if (vlan_tx_tag_present(skb)) {
1998 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1999 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2000 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2002 h.h2->tp_vlan_tci = 0;
2003 h.h2->tp_vlan_tpid = 0;
2005 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2006 hdrlen = sizeof(*h.h2);
2009 /* tp_nxt_offset,vlan are already populated above.
2010 * So DONT clear those fields here
2012 h.h3->tp_status |= status;
2013 h.h3->tp_len = skb->len;
2014 h.h3->tp_snaplen = snaplen;
2015 h.h3->tp_mac = macoff;
2016 h.h3->tp_net = netoff;
2017 h.h3->tp_sec = ts.tv_sec;
2018 h.h3->tp_nsec = ts.tv_nsec;
2019 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2020 hdrlen = sizeof(*h.h3);
2026 sll = h.raw + TPACKET_ALIGN(hdrlen);
2027 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2028 sll->sll_family = AF_PACKET;
2029 sll->sll_hatype = dev->type;
2030 sll->sll_protocol = skb->protocol;
2031 sll->sll_pkttype = skb->pkt_type;
2032 if (unlikely(po->origdev))
2033 sll->sll_ifindex = orig_dev->ifindex;
2035 sll->sll_ifindex = dev->ifindex;
2039 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2040 if (po->tp_version <= TPACKET_V2) {
2043 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2046 for (start = h.raw; start < end; start += PAGE_SIZE)
2047 flush_dcache_page(pgv_to_page(start));
2052 if (po->tp_version <= TPACKET_V2)
2053 __packet_set_status(po, h.raw, status);
2055 prb_clear_blk_fill_status(&po->rx_ring);
2057 sk->sk_data_ready(sk, 0);
2060 if (skb_head != skb->data && skb_shared(skb)) {
2061 skb->data = skb_head;
2069 po->stats.stats1.tp_drops++;
2070 spin_unlock(&sk->sk_receive_queue.lock);
2072 sk->sk_data_ready(sk, 0);
2073 kfree_skb(copy_skb);
2074 goto drop_n_restore;
2077 static void tpacket_destruct_skb(struct sk_buff *skb)
2079 struct packet_sock *po = pkt_sk(skb->sk);
2081 if (likely(po->tx_ring.pg_vec)) {
2085 ph = skb_shinfo(skb)->destructor_arg;
2086 packet_dec_pending(&po->tx_ring);
2088 ts = __packet_set_timestamp(po, ph, skb);
2089 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2095 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2096 void *frame, struct net_device *dev, int size_max,
2097 __be16 proto, unsigned char *addr, int hlen)
2099 union tpacket_uhdr ph;
2100 int to_write, offset, len, tp_len, nr_frags, len_max;
2101 struct socket *sock = po->sk.sk_socket;
2108 skb->protocol = proto;
2110 skb->priority = po->sk.sk_priority;
2111 skb->mark = po->sk.sk_mark;
2112 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2113 skb_shinfo(skb)->destructor_arg = ph.raw;
2115 switch (po->tp_version) {
2117 tp_len = ph.h2->tp_len;
2120 tp_len = ph.h1->tp_len;
2123 if (unlikely(tp_len > size_max)) {
2124 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2128 skb_reserve(skb, hlen);
2129 skb_reset_network_header(skb);
2131 if (!packet_use_direct_xmit(po))
2132 skb_probe_transport_header(skb, 0);
2133 if (unlikely(po->tp_tx_has_off)) {
2134 int off_min, off_max, off;
2135 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2136 off_max = po->tx_ring.frame_size - tp_len;
2137 if (sock->type == SOCK_DGRAM) {
2138 switch (po->tp_version) {
2140 off = ph.h2->tp_net;
2143 off = ph.h1->tp_net;
2147 switch (po->tp_version) {
2149 off = ph.h2->tp_mac;
2152 off = ph.h1->tp_mac;
2156 if (unlikely((off < off_min) || (off_max < off)))
2158 data = ph.raw + off;
2160 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2164 if (sock->type == SOCK_DGRAM) {
2165 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2167 if (unlikely(err < 0))
2169 } else if (dev->hard_header_len) {
2170 /* net device doesn't like empty head */
2171 if (unlikely(tp_len <= dev->hard_header_len)) {
2172 pr_err("packet size is too short (%d < %d)\n",
2173 tp_len, dev->hard_header_len);
2177 skb_push(skb, dev->hard_header_len);
2178 err = skb_store_bits(skb, 0, data,
2179 dev->hard_header_len);
2183 data += dev->hard_header_len;
2184 to_write -= dev->hard_header_len;
2187 offset = offset_in_page(data);
2188 len_max = PAGE_SIZE - offset;
2189 len = ((to_write > len_max) ? len_max : to_write);
2191 skb->data_len = to_write;
2192 skb->len += to_write;
2193 skb->truesize += to_write;
2194 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2196 while (likely(to_write)) {
2197 nr_frags = skb_shinfo(skb)->nr_frags;
2199 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2200 pr_err("Packet exceed the number of skb frags(%lu)\n",
2205 page = pgv_to_page(data);
2207 flush_dcache_page(page);
2209 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2212 len_max = PAGE_SIZE;
2213 len = ((to_write > len_max) ? len_max : to_write);
2219 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2221 struct sk_buff *skb;
2222 struct net_device *dev;
2224 int err, reserve = 0;
2226 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2227 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2228 int tp_len, size_max;
2229 unsigned char *addr;
2231 int status = TP_STATUS_AVAILABLE;
2234 mutex_lock(&po->pg_vec_lock);
2236 if (likely(saddr == NULL)) {
2237 dev = packet_cached_dev_get(po);
2242 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2244 if (msg->msg_namelen < (saddr->sll_halen
2245 + offsetof(struct sockaddr_ll,
2248 proto = saddr->sll_protocol;
2249 addr = saddr->sll_addr;
2250 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2254 if (unlikely(dev == NULL))
2257 if (unlikely(!(dev->flags & IFF_UP)))
2260 reserve = dev->hard_header_len;
2262 size_max = po->tx_ring.frame_size
2263 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2265 if (size_max > dev->mtu + reserve)
2266 size_max = dev->mtu + reserve;
2269 ph = packet_current_frame(po, &po->tx_ring,
2270 TP_STATUS_SEND_REQUEST);
2271 if (unlikely(ph == NULL)) {
2272 if (need_wait && need_resched())
2277 status = TP_STATUS_SEND_REQUEST;
2278 hlen = LL_RESERVED_SPACE(dev);
2279 tlen = dev->needed_tailroom;
2280 skb = sock_alloc_send_skb(&po->sk,
2281 hlen + tlen + sizeof(struct sockaddr_ll),
2284 if (unlikely(skb == NULL))
2287 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2290 if (unlikely(tp_len < 0)) {
2292 __packet_set_status(po, ph,
2293 TP_STATUS_AVAILABLE);
2294 packet_increment_head(&po->tx_ring);
2298 status = TP_STATUS_WRONG_FORMAT;
2304 packet_pick_tx_queue(dev, skb);
2306 skb->destructor = tpacket_destruct_skb;
2307 __packet_set_status(po, ph, TP_STATUS_SENDING);
2308 packet_inc_pending(&po->tx_ring);
2310 status = TP_STATUS_SEND_REQUEST;
2311 err = po->xmit(skb);
2312 if (unlikely(err > 0)) {
2313 err = net_xmit_errno(err);
2314 if (err && __packet_get_status(po, ph) ==
2315 TP_STATUS_AVAILABLE) {
2316 /* skb was destructed already */
2321 * skb was dropped but not destructed yet;
2322 * let's treat it like congestion or err < 0
2326 packet_increment_head(&po->tx_ring);
2328 } while (likely((ph != NULL) ||
2329 /* Note: packet_read_pending() might be slow if we have
2330 * to call it as it's per_cpu variable, but in fast-path
2331 * we already short-circuit the loop with the first
2332 * condition, and luckily don't have to go that path
2335 (need_wait && packet_read_pending(&po->tx_ring))));
2341 __packet_set_status(po, ph, status);
2346 mutex_unlock(&po->pg_vec_lock);
2350 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2351 size_t reserve, size_t len,
2352 size_t linear, int noblock,
2355 struct sk_buff *skb;
2357 /* Under a page? Don't bother with paged skb. */
2358 if (prepad + len < PAGE_SIZE || !linear)
2361 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2366 skb_reserve(skb, reserve);
2367 skb_put(skb, linear);
2368 skb->data_len = len - linear;
2369 skb->len += len - linear;
2374 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2376 struct sock *sk = sock->sk;
2377 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2378 struct sk_buff *skb;
2379 struct net_device *dev;
2381 unsigned char *addr;
2382 int err, reserve = 0;
2383 struct virtio_net_hdr vnet_hdr = { 0 };
2386 struct packet_sock *po = pkt_sk(sk);
2387 unsigned short gso_type = 0;
2392 * Get and verify the address.
2395 if (likely(saddr == NULL)) {
2396 dev = packet_cached_dev_get(po);
2401 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2403 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2405 proto = saddr->sll_protocol;
2406 addr = saddr->sll_addr;
2407 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2411 if (unlikely(dev == NULL))
2414 if (unlikely(!(dev->flags & IFF_UP)))
2417 if (sock->type == SOCK_RAW)
2418 reserve = dev->hard_header_len;
2419 if (po->has_vnet_hdr) {
2420 vnet_hdr_len = sizeof(vnet_hdr);
2423 if (len < vnet_hdr_len)
2426 len -= vnet_hdr_len;
2428 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2433 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2434 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2436 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2437 vnet_hdr.csum_offset + 2;
2440 if (vnet_hdr.hdr_len > len)
2443 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2444 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2445 case VIRTIO_NET_HDR_GSO_TCPV4:
2446 gso_type = SKB_GSO_TCPV4;
2448 case VIRTIO_NET_HDR_GSO_TCPV6:
2449 gso_type = SKB_GSO_TCPV6;
2451 case VIRTIO_NET_HDR_GSO_UDP:
2452 gso_type = SKB_GSO_UDP;
2458 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2459 gso_type |= SKB_GSO_TCP_ECN;
2461 if (vnet_hdr.gso_size == 0)
2467 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2468 if (!netif_supports_nofcs(dev)) {
2469 err = -EPROTONOSUPPORT;
2472 extra_len = 4; /* We're doing our own CRC */
2476 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2480 hlen = LL_RESERVED_SPACE(dev);
2481 tlen = dev->needed_tailroom;
2482 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2483 msg->msg_flags & MSG_DONTWAIT, &err);
2487 skb_set_network_header(skb, reserve);
2490 if (sock->type == SOCK_DGRAM &&
2491 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2494 /* Returns -EFAULT on error */
2495 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2499 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2501 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2502 /* Earlier code assumed this would be a VLAN pkt,
2503 * double-check this now that we have the actual
2506 struct ethhdr *ehdr;
2507 skb_reset_mac_header(skb);
2508 ehdr = eth_hdr(skb);
2509 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2515 skb->protocol = proto;
2517 skb->priority = sk->sk_priority;
2518 skb->mark = sk->sk_mark;
2520 packet_pick_tx_queue(dev, skb);
2522 if (po->has_vnet_hdr) {
2523 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2524 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2525 vnet_hdr.csum_offset)) {
2531 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2532 skb_shinfo(skb)->gso_type = gso_type;
2534 /* Header must be checked, and gso_segs computed. */
2535 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2536 skb_shinfo(skb)->gso_segs = 0;
2538 len += vnet_hdr_len;
2541 if (!packet_use_direct_xmit(po))
2542 skb_probe_transport_header(skb, reserve);
2543 if (unlikely(extra_len == 4))
2546 err = po->xmit(skb);
2547 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2563 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2564 struct msghdr *msg, size_t len)
2566 struct sock *sk = sock->sk;
2567 struct packet_sock *po = pkt_sk(sk);
2569 if (po->tx_ring.pg_vec)
2570 return tpacket_snd(po, msg);
2572 return packet_snd(sock, msg, len);
2576 * Close a PACKET socket. This is fairly simple. We immediately go
2577 * to 'closed' state and remove our protocol entry in the device list.
2580 static int packet_release(struct socket *sock)
2582 struct sock *sk = sock->sk;
2583 struct packet_sock *po;
2585 union tpacket_req_u req_u;
2593 mutex_lock(&net->packet.sklist_lock);
2594 sk_del_node_init_rcu(sk);
2595 mutex_unlock(&net->packet.sklist_lock);
2598 sock_prot_inuse_add(net, sk->sk_prot, -1);
2601 spin_lock(&po->bind_lock);
2602 unregister_prot_hook(sk, false);
2603 packet_cached_dev_reset(po);
2605 if (po->prot_hook.dev) {
2606 dev_put(po->prot_hook.dev);
2607 po->prot_hook.dev = NULL;
2609 spin_unlock(&po->bind_lock);
2611 packet_flush_mclist(sk);
2613 if (po->rx_ring.pg_vec) {
2614 memset(&req_u, 0, sizeof(req_u));
2615 packet_set_ring(sk, &req_u, 1, 0);
2618 if (po->tx_ring.pg_vec) {
2619 memset(&req_u, 0, sizeof(req_u));
2620 packet_set_ring(sk, &req_u, 1, 1);
2627 * Now the socket is dead. No more input will appear.
2634 skb_queue_purge(&sk->sk_receive_queue);
2635 packet_free_pending(po);
2636 sk_refcnt_debug_release(sk);
2643 * Attach a packet hook.
2646 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2648 struct packet_sock *po = pkt_sk(sk);
2649 const struct net_device *dev_curr;
2661 spin_lock(&po->bind_lock);
2663 proto_curr = po->prot_hook.type;
2664 dev_curr = po->prot_hook.dev;
2666 need_rehook = proto_curr != proto || dev_curr != dev;
2669 unregister_prot_hook(sk, true);
2672 po->prot_hook.type = proto;
2674 if (po->prot_hook.dev)
2675 dev_put(po->prot_hook.dev);
2677 po->prot_hook.dev = dev;
2679 po->ifindex = dev ? dev->ifindex : 0;
2680 packet_cached_dev_assign(po, dev);
2683 if (proto == 0 || !need_rehook)
2686 if (!dev || (dev->flags & IFF_UP)) {
2687 register_prot_hook(sk);
2689 sk->sk_err = ENETDOWN;
2690 if (!sock_flag(sk, SOCK_DEAD))
2691 sk->sk_error_report(sk);
2695 spin_unlock(&po->bind_lock);
2701 * Bind a packet socket to a device
2704 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2707 struct sock *sk = sock->sk;
2709 struct net_device *dev;
2716 if (addr_len != sizeof(struct sockaddr))
2718 strlcpy(name, uaddr->sa_data, sizeof(name));
2720 dev = dev_get_by_name(sock_net(sk), name);
2722 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2726 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2728 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2729 struct sock *sk = sock->sk;
2730 struct net_device *dev = NULL;
2738 if (addr_len < sizeof(struct sockaddr_ll))
2740 if (sll->sll_family != AF_PACKET)
2743 if (sll->sll_ifindex) {
2745 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2749 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2755 static struct proto packet_proto = {
2757 .owner = THIS_MODULE,
2758 .obj_size = sizeof(struct packet_sock),
2762 * Create a packet of type SOCK_PACKET.
2765 static int packet_create(struct net *net, struct socket *sock, int protocol,
2769 struct packet_sock *po;
2770 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2773 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2775 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2776 sock->type != SOCK_PACKET)
2777 return -ESOCKTNOSUPPORT;
2779 sock->state = SS_UNCONNECTED;
2782 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2786 sock->ops = &packet_ops;
2787 if (sock->type == SOCK_PACKET)
2788 sock->ops = &packet_ops_spkt;
2790 sock_init_data(sock, sk);
2793 sk->sk_family = PF_PACKET;
2795 po->xmit = dev_queue_xmit;
2797 err = packet_alloc_pending(po);
2801 packet_cached_dev_reset(po);
2803 sk->sk_destruct = packet_sock_destruct;
2804 sk_refcnt_debug_inc(sk);
2807 * Attach a protocol block
2810 spin_lock_init(&po->bind_lock);
2811 mutex_init(&po->pg_vec_lock);
2812 po->prot_hook.func = packet_rcv;
2814 if (sock->type == SOCK_PACKET)
2815 po->prot_hook.func = packet_rcv_spkt;
2817 po->prot_hook.af_packet_priv = sk;
2820 po->prot_hook.type = proto;
2821 register_prot_hook(sk);
2824 mutex_lock(&net->packet.sklist_lock);
2825 sk_add_node_rcu(sk, &net->packet.sklist);
2826 mutex_unlock(&net->packet.sklist_lock);
2829 sock_prot_inuse_add(net, &packet_proto, 1);
2840 * Pull a packet from our receive queue and hand it to the user.
2841 * If necessary we block.
2844 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2845 struct msghdr *msg, size_t len, int flags)
2847 struct sock *sk = sock->sk;
2848 struct sk_buff *skb;
2850 int vnet_hdr_len = 0;
2853 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2857 /* What error should we return now? EUNATTACH? */
2858 if (pkt_sk(sk)->ifindex < 0)
2862 if (flags & MSG_ERRQUEUE) {
2863 err = sock_recv_errqueue(sk, msg, len,
2864 SOL_PACKET, PACKET_TX_TIMESTAMP);
2869 * Call the generic datagram receiver. This handles all sorts
2870 * of horrible races and re-entrancy so we can forget about it
2871 * in the protocol layers.
2873 * Now it will return ENETDOWN, if device have just gone down,
2874 * but then it will block.
2877 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2880 * An error occurred so return it. Because skb_recv_datagram()
2881 * handles the blocking we don't see and worry about blocking
2888 if (pkt_sk(sk)->has_vnet_hdr) {
2889 struct virtio_net_hdr vnet_hdr = { 0 };
2892 vnet_hdr_len = sizeof(vnet_hdr);
2893 if (len < vnet_hdr_len)
2896 len -= vnet_hdr_len;
2898 if (skb_is_gso(skb)) {
2899 struct skb_shared_info *sinfo = skb_shinfo(skb);
2901 /* This is a hint as to how much should be linear. */
2902 vnet_hdr.hdr_len = skb_headlen(skb);
2903 vnet_hdr.gso_size = sinfo->gso_size;
2904 if (sinfo->gso_type & SKB_GSO_TCPV4)
2905 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2906 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2907 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2908 else if (sinfo->gso_type & SKB_GSO_UDP)
2909 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2910 else if (sinfo->gso_type & SKB_GSO_FCOE)
2914 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2915 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2917 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2919 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2920 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2921 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2922 vnet_hdr.csum_offset = skb->csum_offset;
2923 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2924 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2925 } /* else everything is zero */
2927 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2933 /* You lose any data beyond the buffer you gave. If it worries
2934 * a user program they can ask the device for its MTU
2940 msg->msg_flags |= MSG_TRUNC;
2943 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2947 sock_recv_ts_and_drops(msg, sk, skb);
2949 if (msg->msg_name) {
2950 /* If the address length field is there to be filled
2951 * in, we fill it in now.
2953 if (sock->type == SOCK_PACKET) {
2954 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
2955 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2957 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2958 msg->msg_namelen = sll->sll_halen +
2959 offsetof(struct sockaddr_ll, sll_addr);
2961 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2965 if (pkt_sk(sk)->auxdata) {
2966 struct tpacket_auxdata aux;
2968 aux.tp_status = TP_STATUS_USER;
2969 if (skb->ip_summed == CHECKSUM_PARTIAL)
2970 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2971 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2972 aux.tp_snaplen = skb->len;
2974 aux.tp_net = skb_network_offset(skb);
2975 if (vlan_tx_tag_present(skb)) {
2976 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2977 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2978 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2980 aux.tp_vlan_tci = 0;
2981 aux.tp_vlan_tpid = 0;
2983 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2987 * Free or return the buffer as appropriate. Again this
2988 * hides all the races and re-entrancy issues from us.
2990 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2993 skb_free_datagram(sk, skb);
2998 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2999 int *uaddr_len, int peer)
3001 struct net_device *dev;
3002 struct sock *sk = sock->sk;
3007 uaddr->sa_family = AF_PACKET;
3008 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3010 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3012 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3014 *uaddr_len = sizeof(*uaddr);
3019 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3020 int *uaddr_len, int peer)
3022 struct net_device *dev;
3023 struct sock *sk = sock->sk;
3024 struct packet_sock *po = pkt_sk(sk);
3025 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3030 sll->sll_family = AF_PACKET;
3031 sll->sll_ifindex = po->ifindex;
3032 sll->sll_protocol = po->num;
3033 sll->sll_pkttype = 0;
3035 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3037 sll->sll_hatype = dev->type;
3038 sll->sll_halen = dev->addr_len;
3039 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3041 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3045 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3050 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3054 case PACKET_MR_MULTICAST:
3055 if (i->alen != dev->addr_len)
3058 return dev_mc_add(dev, i->addr);
3060 return dev_mc_del(dev, i->addr);
3062 case PACKET_MR_PROMISC:
3063 return dev_set_promiscuity(dev, what);
3065 case PACKET_MR_ALLMULTI:
3066 return dev_set_allmulti(dev, what);
3068 case PACKET_MR_UNICAST:
3069 if (i->alen != dev->addr_len)
3072 return dev_uc_add(dev, i->addr);
3074 return dev_uc_del(dev, i->addr);
3082 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3084 for ( ; i; i = i->next) {
3085 if (i->ifindex == dev->ifindex)
3086 packet_dev_mc(dev, i, what);
3090 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3092 struct packet_sock *po = pkt_sk(sk);
3093 struct packet_mclist *ml, *i;
3094 struct net_device *dev;
3100 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3105 if (mreq->mr_alen > dev->addr_len)
3109 i = kmalloc(sizeof(*i), GFP_KERNEL);
3114 for (ml = po->mclist; ml; ml = ml->next) {
3115 if (ml->ifindex == mreq->mr_ifindex &&
3116 ml->type == mreq->mr_type &&
3117 ml->alen == mreq->mr_alen &&
3118 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3120 /* Free the new element ... */
3126 i->type = mreq->mr_type;
3127 i->ifindex = mreq->mr_ifindex;
3128 i->alen = mreq->mr_alen;
3129 memcpy(i->addr, mreq->mr_address, i->alen);
3131 i->next = po->mclist;
3133 err = packet_dev_mc(dev, i, 1);
3135 po->mclist = i->next;
3144 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3146 struct packet_mclist *ml, **mlp;
3150 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3151 if (ml->ifindex == mreq->mr_ifindex &&
3152 ml->type == mreq->mr_type &&
3153 ml->alen == mreq->mr_alen &&
3154 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3155 if (--ml->count == 0) {
3156 struct net_device *dev;
3158 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3160 packet_dev_mc(dev, ml, -1);
3168 return -EADDRNOTAVAIL;
3171 static void packet_flush_mclist(struct sock *sk)
3173 struct packet_sock *po = pkt_sk(sk);
3174 struct packet_mclist *ml;
3180 while ((ml = po->mclist) != NULL) {
3181 struct net_device *dev;
3183 po->mclist = ml->next;
3184 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3186 packet_dev_mc(dev, ml, -1);
3193 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3195 struct sock *sk = sock->sk;
3196 struct packet_sock *po = pkt_sk(sk);
3199 if (level != SOL_PACKET)
3200 return -ENOPROTOOPT;
3203 case PACKET_ADD_MEMBERSHIP:
3204 case PACKET_DROP_MEMBERSHIP:
3206 struct packet_mreq_max mreq;
3208 memset(&mreq, 0, sizeof(mreq));
3209 if (len < sizeof(struct packet_mreq))
3211 if (len > sizeof(mreq))
3213 if (copy_from_user(&mreq, optval, len))
3215 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3217 if (optname == PACKET_ADD_MEMBERSHIP)
3218 ret = packet_mc_add(sk, &mreq);
3220 ret = packet_mc_drop(sk, &mreq);
3224 case PACKET_RX_RING:
3225 case PACKET_TX_RING:
3227 union tpacket_req_u req_u;
3230 switch (po->tp_version) {
3233 len = sizeof(req_u.req);
3237 len = sizeof(req_u.req3);
3242 if (pkt_sk(sk)->has_vnet_hdr)
3244 if (copy_from_user(&req_u.req, optval, len))
3246 return packet_set_ring(sk, &req_u, 0,
3247 optname == PACKET_TX_RING);
3249 case PACKET_COPY_THRESH:
3253 if (optlen != sizeof(val))
3255 if (copy_from_user(&val, optval, sizeof(val)))
3258 pkt_sk(sk)->copy_thresh = val;
3261 case PACKET_VERSION:
3265 if (optlen != sizeof(val))
3267 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3269 if (copy_from_user(&val, optval, sizeof(val)))
3275 po->tp_version = val;
3281 case PACKET_RESERVE:
3285 if (optlen != sizeof(val))
3287 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3289 if (copy_from_user(&val, optval, sizeof(val)))
3291 po->tp_reserve = val;
3298 if (optlen != sizeof(val))
3300 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3302 if (copy_from_user(&val, optval, sizeof(val)))
3304 po->tp_loss = !!val;
3307 case PACKET_AUXDATA:
3311 if (optlen < sizeof(val))
3313 if (copy_from_user(&val, optval, sizeof(val)))
3316 po->auxdata = !!val;
3319 case PACKET_ORIGDEV:
3323 if (optlen < sizeof(val))
3325 if (copy_from_user(&val, optval, sizeof(val)))
3328 po->origdev = !!val;
3331 case PACKET_VNET_HDR:
3335 if (sock->type != SOCK_RAW)
3337 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3339 if (optlen < sizeof(val))
3341 if (copy_from_user(&val, optval, sizeof(val)))
3344 po->has_vnet_hdr = !!val;
3347 case PACKET_TIMESTAMP:
3351 if (optlen != sizeof(val))
3353 if (copy_from_user(&val, optval, sizeof(val)))
3356 po->tp_tstamp = val;
3363 if (optlen != sizeof(val))
3365 if (copy_from_user(&val, optval, sizeof(val)))
3368 return fanout_add(sk, val & 0xffff, val >> 16);
3370 case PACKET_TX_HAS_OFF:
3374 if (optlen != sizeof(val))
3376 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3378 if (copy_from_user(&val, optval, sizeof(val)))
3380 po->tp_tx_has_off = !!val;
3383 case PACKET_QDISC_BYPASS:
3387 if (optlen != sizeof(val))
3389 if (copy_from_user(&val, optval, sizeof(val)))
3392 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3396 return -ENOPROTOOPT;
3400 static int packet_getsockopt(struct socket *sock, int level, int optname,
3401 char __user *optval, int __user *optlen)
3404 int val, lv = sizeof(val);
3405 struct sock *sk = sock->sk;
3406 struct packet_sock *po = pkt_sk(sk);
3408 union tpacket_stats_u st;
3410 if (level != SOL_PACKET)
3411 return -ENOPROTOOPT;
3413 if (get_user(len, optlen))
3420 case PACKET_STATISTICS:
3421 spin_lock_bh(&sk->sk_receive_queue.lock);
3422 memcpy(&st, &po->stats, sizeof(st));
3423 memset(&po->stats, 0, sizeof(po->stats));
3424 spin_unlock_bh(&sk->sk_receive_queue.lock);
3426 if (po->tp_version == TPACKET_V3) {
3427 lv = sizeof(struct tpacket_stats_v3);
3428 st.stats3.tp_packets += st.stats3.tp_drops;
3431 lv = sizeof(struct tpacket_stats);
3432 st.stats1.tp_packets += st.stats1.tp_drops;
3437 case PACKET_AUXDATA:
3440 case PACKET_ORIGDEV:
3443 case PACKET_VNET_HDR:
3444 val = po->has_vnet_hdr;
3446 case PACKET_VERSION:
3447 val = po->tp_version;
3450 if (len > sizeof(int))
3452 if (copy_from_user(&val, optval, len))
3456 val = sizeof(struct tpacket_hdr);
3459 val = sizeof(struct tpacket2_hdr);
3462 val = sizeof(struct tpacket3_hdr);
3468 case PACKET_RESERVE:
3469 val = po->tp_reserve;
3474 case PACKET_TIMESTAMP:
3475 val = po->tp_tstamp;
3479 ((u32)po->fanout->id |
3480 ((u32)po->fanout->type << 16) |
3481 ((u32)po->fanout->flags << 24)) :
3484 case PACKET_TX_HAS_OFF:
3485 val = po->tp_tx_has_off;
3487 case PACKET_QDISC_BYPASS:
3488 val = packet_use_direct_xmit(po);
3491 return -ENOPROTOOPT;
3496 if (put_user(len, optlen))
3498 if (copy_to_user(optval, data, len))
3504 static int packet_notifier(struct notifier_block *this,
3505 unsigned long msg, void *ptr)
3508 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3509 struct net *net = dev_net(dev);
3512 sk_for_each_rcu(sk, &net->packet.sklist) {
3513 struct packet_sock *po = pkt_sk(sk);
3516 case NETDEV_UNREGISTER:
3518 packet_dev_mclist(dev, po->mclist, -1);
3522 if (dev->ifindex == po->ifindex) {
3523 spin_lock(&po->bind_lock);
3525 __unregister_prot_hook(sk, false);
3526 sk->sk_err = ENETDOWN;
3527 if (!sock_flag(sk, SOCK_DEAD))
3528 sk->sk_error_report(sk);
3530 if (msg == NETDEV_UNREGISTER) {
3531 packet_cached_dev_reset(po);
3533 if (po->prot_hook.dev)
3534 dev_put(po->prot_hook.dev);
3535 po->prot_hook.dev = NULL;
3537 spin_unlock(&po->bind_lock);
3541 if (dev->ifindex == po->ifindex) {
3542 spin_lock(&po->bind_lock);
3544 register_prot_hook(sk);
3545 spin_unlock(&po->bind_lock);
3555 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3558 struct sock *sk = sock->sk;
3563 int amount = sk_wmem_alloc_get(sk);
3565 return put_user(amount, (int __user *)arg);
3569 struct sk_buff *skb;
3572 spin_lock_bh(&sk->sk_receive_queue.lock);
3573 skb = skb_peek(&sk->sk_receive_queue);
3576 spin_unlock_bh(&sk->sk_receive_queue.lock);
3577 return put_user(amount, (int __user *)arg);
3580 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3582 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3592 case SIOCGIFBRDADDR:
3593 case SIOCSIFBRDADDR:
3594 case SIOCGIFNETMASK:
3595 case SIOCSIFNETMASK:
3596 case SIOCGIFDSTADDR:
3597 case SIOCSIFDSTADDR:
3599 return inet_dgram_ops.ioctl(sock, cmd, arg);
3603 return -ENOIOCTLCMD;
3608 static unsigned int packet_poll(struct file *file, struct socket *sock,
3611 struct sock *sk = sock->sk;
3612 struct packet_sock *po = pkt_sk(sk);
3613 unsigned int mask = datagram_poll(file, sock, wait);
3615 spin_lock_bh(&sk->sk_receive_queue.lock);
3616 if (po->rx_ring.pg_vec) {
3617 if (!packet_previous_rx_frame(po, &po->rx_ring,
3619 mask |= POLLIN | POLLRDNORM;
3621 spin_unlock_bh(&sk->sk_receive_queue.lock);
3622 spin_lock_bh(&sk->sk_write_queue.lock);
3623 if (po->tx_ring.pg_vec) {
3624 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3625 mask |= POLLOUT | POLLWRNORM;
3627 spin_unlock_bh(&sk->sk_write_queue.lock);
3632 /* Dirty? Well, I still did not learn better way to account
3636 static void packet_mm_open(struct vm_area_struct *vma)
3638 struct file *file = vma->vm_file;
3639 struct socket *sock = file->private_data;
3640 struct sock *sk = sock->sk;
3643 atomic_inc(&pkt_sk(sk)->mapped);
3646 static void packet_mm_close(struct vm_area_struct *vma)
3648 struct file *file = vma->vm_file;
3649 struct socket *sock = file->private_data;
3650 struct sock *sk = sock->sk;
3653 atomic_dec(&pkt_sk(sk)->mapped);
3656 static const struct vm_operations_struct packet_mmap_ops = {
3657 .open = packet_mm_open,
3658 .close = packet_mm_close,
3661 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3666 for (i = 0; i < len; i++) {
3667 if (likely(pg_vec[i].buffer)) {
3668 if (is_vmalloc_addr(pg_vec[i].buffer))
3669 vfree(pg_vec[i].buffer);
3671 free_pages((unsigned long)pg_vec[i].buffer,
3673 pg_vec[i].buffer = NULL;
3679 static char *alloc_one_pg_vec_page(unsigned long order)
3682 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3683 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3685 buffer = (char *) __get_free_pages(gfp_flags, order);
3689 /* __get_free_pages failed, fall back to vmalloc */
3690 buffer = vzalloc((1 << order) * PAGE_SIZE);
3694 /* vmalloc failed, lets dig into swap here */
3695 gfp_flags &= ~__GFP_NORETRY;
3696 buffer = (char *) __get_free_pages(gfp_flags, order);
3700 /* complete and utter failure */
3704 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3706 unsigned int block_nr = req->tp_block_nr;
3710 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3711 if (unlikely(!pg_vec))
3714 for (i = 0; i < block_nr; i++) {
3715 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3716 if (unlikely(!pg_vec[i].buffer))
3717 goto out_free_pgvec;
3724 free_pg_vec(pg_vec, order, block_nr);
3729 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3730 int closing, int tx_ring)
3732 struct pgv *pg_vec = NULL;
3733 struct packet_sock *po = pkt_sk(sk);
3734 int was_running, order = 0;
3735 struct packet_ring_buffer *rb;
3736 struct sk_buff_head *rb_queue;
3739 /* Added to avoid minimal code churn */
3740 struct tpacket_req *req = &req_u->req;
3742 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3743 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3744 WARN(1, "Tx-ring is not supported.\n");
3748 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3749 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3753 if (atomic_read(&po->mapped))
3755 if (packet_read_pending(rb))
3759 if (req->tp_block_nr) {
3760 /* Sanity tests and some calculations */
3762 if (unlikely(rb->pg_vec))
3765 switch (po->tp_version) {
3767 po->tp_hdrlen = TPACKET_HDRLEN;
3770 po->tp_hdrlen = TPACKET2_HDRLEN;
3773 po->tp_hdrlen = TPACKET3_HDRLEN;
3778 if (unlikely((int)req->tp_block_size <= 0))
3780 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3782 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3785 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3788 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3789 if (unlikely(rb->frames_per_block <= 0))
3791 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3796 order = get_order(req->tp_block_size);
3797 pg_vec = alloc_pg_vec(req, order);
3798 if (unlikely(!pg_vec))
3800 switch (po->tp_version) {
3802 /* Transmit path is not supported. We checked
3803 * it above but just being paranoid
3806 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3815 if (unlikely(req->tp_frame_nr))
3821 /* Detach socket from network */
3822 spin_lock(&po->bind_lock);
3823 was_running = po->running;
3827 __unregister_prot_hook(sk, false);
3829 spin_unlock(&po->bind_lock);
3834 mutex_lock(&po->pg_vec_lock);
3835 if (closing || atomic_read(&po->mapped) == 0) {
3837 spin_lock_bh(&rb_queue->lock);
3838 swap(rb->pg_vec, pg_vec);
3839 rb->frame_max = (req->tp_frame_nr - 1);
3841 rb->frame_size = req->tp_frame_size;
3842 spin_unlock_bh(&rb_queue->lock);
3844 swap(rb->pg_vec_order, order);
3845 swap(rb->pg_vec_len, req->tp_block_nr);
3847 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3848 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3849 tpacket_rcv : packet_rcv;
3850 skb_queue_purge(rb_queue);
3851 if (atomic_read(&po->mapped))
3852 pr_err("packet_mmap: vma is busy: %d\n",
3853 atomic_read(&po->mapped));
3855 mutex_unlock(&po->pg_vec_lock);
3857 spin_lock(&po->bind_lock);
3860 register_prot_hook(sk);
3862 spin_unlock(&po->bind_lock);
3863 if (closing && (po->tp_version > TPACKET_V2)) {
3864 /* Because we don't support block-based V3 on tx-ring */
3866 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3871 free_pg_vec(pg_vec, order, req->tp_block_nr);
3876 static int packet_mmap(struct file *file, struct socket *sock,
3877 struct vm_area_struct *vma)
3879 struct sock *sk = sock->sk;
3880 struct packet_sock *po = pkt_sk(sk);
3881 unsigned long size, expected_size;
3882 struct packet_ring_buffer *rb;
3883 unsigned long start;
3890 mutex_lock(&po->pg_vec_lock);
3893 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3895 expected_size += rb->pg_vec_len
3901 if (expected_size == 0)
3904 size = vma->vm_end - vma->vm_start;
3905 if (size != expected_size)
3908 start = vma->vm_start;
3909 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3910 if (rb->pg_vec == NULL)
3913 for (i = 0; i < rb->pg_vec_len; i++) {
3915 void *kaddr = rb->pg_vec[i].buffer;
3918 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3919 page = pgv_to_page(kaddr);
3920 err = vm_insert_page(vma, start, page);
3929 atomic_inc(&po->mapped);
3930 vma->vm_ops = &packet_mmap_ops;
3934 mutex_unlock(&po->pg_vec_lock);
3938 static const struct proto_ops packet_ops_spkt = {
3939 .family = PF_PACKET,
3940 .owner = THIS_MODULE,
3941 .release = packet_release,
3942 .bind = packet_bind_spkt,
3943 .connect = sock_no_connect,
3944 .socketpair = sock_no_socketpair,
3945 .accept = sock_no_accept,
3946 .getname = packet_getname_spkt,
3947 .poll = datagram_poll,
3948 .ioctl = packet_ioctl,
3949 .listen = sock_no_listen,
3950 .shutdown = sock_no_shutdown,
3951 .setsockopt = sock_no_setsockopt,
3952 .getsockopt = sock_no_getsockopt,
3953 .sendmsg = packet_sendmsg_spkt,
3954 .recvmsg = packet_recvmsg,
3955 .mmap = sock_no_mmap,
3956 .sendpage = sock_no_sendpage,
3959 static const struct proto_ops packet_ops = {
3960 .family = PF_PACKET,
3961 .owner = THIS_MODULE,
3962 .release = packet_release,
3963 .bind = packet_bind,
3964 .connect = sock_no_connect,
3965 .socketpair = sock_no_socketpair,
3966 .accept = sock_no_accept,
3967 .getname = packet_getname,
3968 .poll = packet_poll,
3969 .ioctl = packet_ioctl,
3970 .listen = sock_no_listen,
3971 .shutdown = sock_no_shutdown,
3972 .setsockopt = packet_setsockopt,
3973 .getsockopt = packet_getsockopt,
3974 .sendmsg = packet_sendmsg,
3975 .recvmsg = packet_recvmsg,
3976 .mmap = packet_mmap,
3977 .sendpage = sock_no_sendpage,
3980 static const struct net_proto_family packet_family_ops = {
3981 .family = PF_PACKET,
3982 .create = packet_create,
3983 .owner = THIS_MODULE,
3986 static struct notifier_block packet_netdev_notifier = {
3987 .notifier_call = packet_notifier,
3990 #ifdef CONFIG_PROC_FS
3992 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3995 struct net *net = seq_file_net(seq);
3998 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4001 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4003 struct net *net = seq_file_net(seq);
4004 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4007 static void packet_seq_stop(struct seq_file *seq, void *v)
4013 static int packet_seq_show(struct seq_file *seq, void *v)
4015 if (v == SEQ_START_TOKEN)
4016 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4018 struct sock *s = sk_entry(v);
4019 const struct packet_sock *po = pkt_sk(s);
4022 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4024 atomic_read(&s->sk_refcnt),
4029 atomic_read(&s->sk_rmem_alloc),
4030 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4037 static const struct seq_operations packet_seq_ops = {
4038 .start = packet_seq_start,
4039 .next = packet_seq_next,
4040 .stop = packet_seq_stop,
4041 .show = packet_seq_show,
4044 static int packet_seq_open(struct inode *inode, struct file *file)
4046 return seq_open_net(inode, file, &packet_seq_ops,
4047 sizeof(struct seq_net_private));
4050 static const struct file_operations packet_seq_fops = {
4051 .owner = THIS_MODULE,
4052 .open = packet_seq_open,
4054 .llseek = seq_lseek,
4055 .release = seq_release_net,
4060 static int __net_init packet_net_init(struct net *net)
4062 mutex_init(&net->packet.sklist_lock);
4063 INIT_HLIST_HEAD(&net->packet.sklist);
4065 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4071 static void __net_exit packet_net_exit(struct net *net)
4073 remove_proc_entry("packet", net->proc_net);
4076 static struct pernet_operations packet_net_ops = {
4077 .init = packet_net_init,
4078 .exit = packet_net_exit,
4082 static void __exit packet_exit(void)
4084 unregister_netdevice_notifier(&packet_netdev_notifier);
4085 unregister_pernet_subsys(&packet_net_ops);
4086 sock_unregister(PF_PACKET);
4087 proto_unregister(&packet_proto);
4090 static int __init packet_init(void)
4092 int rc = proto_register(&packet_proto, 0);
4097 sock_register(&packet_family_ops);
4098 register_pernet_subsys(&packet_net_ops);
4099 register_netdevice_notifier(&packet_netdev_notifier);
4104 module_init(packet_init);
4105 module_exit(packet_exit);
4106 MODULE_LICENSE("GPL");
4107 MODULE_ALIAS_NETPROTO(PF_PACKET);