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/reciprocal_div.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)
313 return (u16) smp_processor_id() % dev->real_num_tx_queues;
316 /* register_prot_hook must be invoked with the po->bind_lock held,
317 * or from a context in which asynchronous accesses to the packet
318 * socket is not possible (packet_create()).
320 static void register_prot_hook(struct sock *sk)
322 struct packet_sock *po = pkt_sk(sk);
326 __fanout_link(sk, po);
328 dev_add_pack(&po->prot_hook);
335 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
336 * held. If the sync parameter is true, we will temporarily drop
337 * the po->bind_lock and do a synchronize_net to make sure no
338 * asynchronous packet processing paths still refer to the elements
339 * of po->prot_hook. If the sync parameter is false, it is the
340 * callers responsibility to take care of this.
342 static void __unregister_prot_hook(struct sock *sk, bool sync)
344 struct packet_sock *po = pkt_sk(sk);
349 __fanout_unlink(sk, po);
351 __dev_remove_pack(&po->prot_hook);
356 spin_unlock(&po->bind_lock);
358 spin_lock(&po->bind_lock);
362 static void unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
367 __unregister_prot_hook(sk, sync);
370 static inline __pure struct page *pgv_to_page(void *addr)
372 if (is_vmalloc_addr(addr))
373 return vmalloc_to_page(addr);
374 return virt_to_page(addr);
377 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
379 union tpacket_uhdr h;
382 switch (po->tp_version) {
384 h.h1->tp_status = status;
385 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
388 h.h2->tp_status = status;
389 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
393 WARN(1, "TPACKET version not supported.\n");
400 static int __packet_get_status(struct packet_sock *po, void *frame)
402 union tpacket_uhdr h;
407 switch (po->tp_version) {
409 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
410 return h.h1->tp_status;
412 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
413 return h.h2->tp_status;
416 WARN(1, "TPACKET version not supported.\n");
422 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
425 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
428 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
429 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
430 return TP_STATUS_TS_SYS_HARDWARE;
431 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
432 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
433 return TP_STATUS_TS_RAW_HARDWARE;
436 if (ktime_to_timespec_cond(skb->tstamp, ts))
437 return TP_STATUS_TS_SOFTWARE;
442 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
445 union tpacket_uhdr h;
449 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
453 switch (po->tp_version) {
455 h.h1->tp_sec = ts.tv_sec;
456 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
459 h.h2->tp_sec = ts.tv_sec;
460 h.h2->tp_nsec = ts.tv_nsec;
464 WARN(1, "TPACKET version not supported.\n");
468 /* one flush is safe, as both fields always lie on the same cacheline */
469 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
475 static void *packet_lookup_frame(struct packet_sock *po,
476 struct packet_ring_buffer *rb,
477 unsigned int position,
480 unsigned int pg_vec_pos, frame_offset;
481 union tpacket_uhdr h;
483 pg_vec_pos = position / rb->frames_per_block;
484 frame_offset = position % rb->frames_per_block;
486 h.raw = rb->pg_vec[pg_vec_pos].buffer +
487 (frame_offset * rb->frame_size);
489 if (status != __packet_get_status(po, h.raw))
495 static void *packet_current_frame(struct packet_sock *po,
496 struct packet_ring_buffer *rb,
499 return packet_lookup_frame(po, rb, rb->head, status);
502 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
504 del_timer_sync(&pkc->retire_blk_timer);
507 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
509 struct sk_buff_head *rb_queue)
511 struct tpacket_kbdq_core *pkc;
513 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
514 GET_PBDQC_FROM_RB(&po->rx_ring);
516 spin_lock_bh(&rb_queue->lock);
517 pkc->delete_blk_timer = 1;
518 spin_unlock_bh(&rb_queue->lock);
520 prb_del_retire_blk_timer(pkc);
523 static void prb_init_blk_timer(struct packet_sock *po,
524 struct tpacket_kbdq_core *pkc,
525 void (*func) (unsigned long))
527 init_timer(&pkc->retire_blk_timer);
528 pkc->retire_blk_timer.data = (long)po;
529 pkc->retire_blk_timer.function = func;
530 pkc->retire_blk_timer.expires = jiffies;
533 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
535 struct tpacket_kbdq_core *pkc;
540 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
541 GET_PBDQC_FROM_RB(&po->rx_ring);
542 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
545 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
546 int blk_size_in_bytes)
548 struct net_device *dev;
549 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
550 struct ethtool_cmd ecmd;
555 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
556 if (unlikely(!dev)) {
558 return DEFAULT_PRB_RETIRE_TOV;
560 err = __ethtool_get_settings(dev, &ecmd);
561 speed = ethtool_cmd_speed(&ecmd);
565 * If the link speed is so slow you don't really
566 * need to worry about perf anyways
568 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
569 return DEFAULT_PRB_RETIRE_TOV;
576 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
588 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
589 union tpacket_req_u *req_u)
591 p1->feature_req_word = req_u->req3.tp_feature_req_word;
594 static void init_prb_bdqc(struct packet_sock *po,
595 struct packet_ring_buffer *rb,
597 union tpacket_req_u *req_u, int tx_ring)
599 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
600 struct tpacket_block_desc *pbd;
602 memset(p1, 0x0, sizeof(*p1));
604 p1->knxt_seq_num = 1;
606 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
607 p1->pkblk_start = pg_vec[0].buffer;
608 p1->kblk_size = req_u->req3.tp_block_size;
609 p1->knum_blocks = req_u->req3.tp_block_nr;
610 p1->hdrlen = po->tp_hdrlen;
611 p1->version = po->tp_version;
612 p1->last_kactive_blk_num = 0;
613 po->stats.stats3.tp_freeze_q_cnt = 0;
614 if (req_u->req3.tp_retire_blk_tov)
615 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
617 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
618 req_u->req3.tp_block_size);
619 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
620 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
622 prb_init_ft_ops(p1, req_u);
623 prb_setup_retire_blk_timer(po, tx_ring);
624 prb_open_block(p1, pbd);
627 /* Do NOT update the last_blk_num first.
628 * Assumes sk_buff_head lock is held.
630 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
632 mod_timer(&pkc->retire_blk_timer,
633 jiffies + pkc->tov_in_jiffies);
634 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
639 * 1) We refresh the timer only when we open a block.
640 * By doing this we don't waste cycles refreshing the timer
641 * on packet-by-packet basis.
643 * With a 1MB block-size, on a 1Gbps line, it will take
644 * i) ~8 ms to fill a block + ii) memcpy etc.
645 * In this cut we are not accounting for the memcpy time.
647 * So, if the user sets the 'tmo' to 10ms then the timer
648 * will never fire while the block is still getting filled
649 * (which is what we want). However, the user could choose
650 * to close a block early and that's fine.
652 * But when the timer does fire, we check whether or not to refresh it.
653 * Since the tmo granularity is in msecs, it is not too expensive
654 * to refresh the timer, lets say every '8' msecs.
655 * Either the user can set the 'tmo' or we can derive it based on
656 * a) line-speed and b) block-size.
657 * prb_calc_retire_blk_tmo() calculates the tmo.
660 static void prb_retire_rx_blk_timer_expired(unsigned long data)
662 struct packet_sock *po = (struct packet_sock *)data;
663 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
665 struct tpacket_block_desc *pbd;
667 spin_lock(&po->sk.sk_receive_queue.lock);
669 frozen = prb_queue_frozen(pkc);
670 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
672 if (unlikely(pkc->delete_blk_timer))
675 /* We only need to plug the race when the block is partially filled.
677 * lock(); increment BLOCK_NUM_PKTS; unlock()
678 * copy_bits() is in progress ...
679 * timer fires on other cpu:
680 * we can't retire the current block because copy_bits
684 if (BLOCK_NUM_PKTS(pbd)) {
685 while (atomic_read(&pkc->blk_fill_in_prog)) {
686 /* Waiting for skb_copy_bits to finish... */
691 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
693 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
694 if (!prb_dispatch_next_block(pkc, po))
699 /* Case 1. Queue was frozen because user-space was
702 if (prb_curr_blk_in_use(pkc, pbd)) {
704 * Ok, user-space is still behind.
705 * So just refresh the timer.
709 /* Case 2. queue was frozen,user-space caught up,
710 * now the link went idle && the timer fired.
711 * We don't have a block to close.So we open this
712 * block and restart the timer.
713 * opening a block thaws the queue,restarts timer
714 * Thawing/timer-refresh is a side effect.
716 prb_open_block(pkc, pbd);
723 _prb_refresh_rx_retire_blk_timer(pkc);
726 spin_unlock(&po->sk.sk_receive_queue.lock);
729 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
730 struct tpacket_block_desc *pbd1, __u32 status)
732 /* Flush everything minus the block header */
734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
739 /* Skip the block header(we know header WILL fit in 4K) */
742 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
743 for (; start < end; start += PAGE_SIZE)
744 flush_dcache_page(pgv_to_page(start));
749 /* Now update the block status. */
751 BLOCK_STATUS(pbd1) = status;
753 /* Flush the block header */
755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
757 flush_dcache_page(pgv_to_page(start));
767 * 2) Increment active_blk_num
769 * Note:We DONT refresh the timer on purpose.
770 * Because almost always the next block will be opened.
772 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
773 struct tpacket_block_desc *pbd1,
774 struct packet_sock *po, unsigned int stat)
776 __u32 status = TP_STATUS_USER | stat;
778 struct tpacket3_hdr *last_pkt;
779 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
781 if (po->stats.stats3.tp_drops)
782 status |= TP_STATUS_LOSING;
784 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
785 last_pkt->tp_next_offset = 0;
787 /* Get the ts of the last pkt */
788 if (BLOCK_NUM_PKTS(pbd1)) {
789 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
790 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
792 /* Ok, we tmo'd - so get the current time */
795 h1->ts_last_pkt.ts_sec = ts.tv_sec;
796 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
801 /* Flush the block */
802 prb_flush_block(pkc1, pbd1, status);
804 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
807 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
809 pkc->reset_pending_on_curr_blk = 0;
813 * Side effect of opening a block:
815 * 1) prb_queue is thawed.
816 * 2) retire_blk_timer is refreshed.
819 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
820 struct tpacket_block_desc *pbd1)
823 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
827 /* We could have just memset this but we will lose the
828 * flexibility of making the priv area sticky
831 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
832 BLOCK_NUM_PKTS(pbd1) = 0;
833 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 h1->ts_first_pkt.ts_sec = ts.tv_sec;
838 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
840 pkc1->pkblk_start = (char *)pbd1;
841 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
843 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
844 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
846 pbd1->version = pkc1->version;
847 pkc1->prev = pkc1->nxt_offset;
848 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
850 prb_thaw_queue(pkc1);
851 _prb_refresh_rx_retire_blk_timer(pkc1);
857 * Queue freeze logic:
858 * 1) Assume tp_block_nr = 8 blocks.
859 * 2) At time 't0', user opens Rx ring.
860 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
861 * 4) user-space is either sleeping or processing block '0'.
862 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
863 * it will close block-7,loop around and try to fill block '0'.
865 * __packet_lookup_frame_in_block
866 * prb_retire_current_block()
867 * prb_dispatch_next_block()
868 * |->(BLOCK_STATUS == USER) evaluates to true
869 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
870 * 6) Now there are two cases:
871 * 6.1) Link goes idle right after the queue is frozen.
872 * But remember, the last open_block() refreshed the timer.
873 * When this timer expires,it will refresh itself so that we can
874 * re-open block-0 in near future.
875 * 6.2) Link is busy and keeps on receiving packets. This is a simple
876 * case and __packet_lookup_frame_in_block will check if block-0
877 * is free and can now be re-used.
879 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
880 struct packet_sock *po)
882 pkc->reset_pending_on_curr_blk = 1;
883 po->stats.stats3.tp_freeze_q_cnt++;
886 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
889 * If the next block is free then we will dispatch it
890 * and return a good offset.
891 * Else, we will freeze the queue.
892 * So, caller must check the return value.
894 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
895 struct packet_sock *po)
897 struct tpacket_block_desc *pbd;
901 /* 1. Get current block num */
902 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
904 /* 2. If this block is currently in_use then freeze the queue */
905 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
906 prb_freeze_queue(pkc, po);
912 * open this block and return the offset where the first packet
913 * needs to get stored.
915 prb_open_block(pkc, pbd);
916 return (void *)pkc->nxt_offset;
919 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
920 struct packet_sock *po, unsigned int status)
922 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
924 /* retire/close the current block */
925 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
927 * Plug the case where copy_bits() is in progress on
928 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
929 * have space to copy the pkt in the current block and
930 * called prb_retire_current_block()
932 * We don't need to worry about the TMO case because
933 * the timer-handler already handled this case.
935 if (!(status & TP_STATUS_BLK_TMO)) {
936 while (atomic_read(&pkc->blk_fill_in_prog)) {
937 /* Waiting for skb_copy_bits to finish... */
941 prb_close_block(pkc, pbd, po, status);
946 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
947 struct tpacket_block_desc *pbd)
949 return TP_STATUS_USER & BLOCK_STATUS(pbd);
952 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
954 return pkc->reset_pending_on_curr_blk;
957 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
959 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
960 atomic_dec(&pkc->blk_fill_in_prog);
963 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
964 struct tpacket3_hdr *ppd)
966 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
969 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
970 struct tpacket3_hdr *ppd)
972 ppd->hv1.tp_rxhash = 0;
975 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
976 struct tpacket3_hdr *ppd)
978 if (vlan_tx_tag_present(pkc->skb)) {
979 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
980 ppd->tp_status = TP_STATUS_VLAN_VALID;
982 ppd->hv1.tp_vlan_tci = 0;
983 ppd->tp_status = TP_STATUS_AVAILABLE;
987 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
988 struct tpacket3_hdr *ppd)
990 prb_fill_vlan_info(pkc, ppd);
992 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
993 prb_fill_rxhash(pkc, ppd);
995 prb_clear_rxhash(pkc, ppd);
998 static void prb_fill_curr_block(char *curr,
999 struct tpacket_kbdq_core *pkc,
1000 struct tpacket_block_desc *pbd,
1003 struct tpacket3_hdr *ppd;
1005 ppd = (struct tpacket3_hdr *)curr;
1006 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1008 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1009 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1010 BLOCK_NUM_PKTS(pbd) += 1;
1011 atomic_inc(&pkc->blk_fill_in_prog);
1012 prb_run_all_ft_ops(pkc, ppd);
1015 /* Assumes caller has the sk->rx_queue.lock */
1016 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1017 struct sk_buff *skb,
1022 struct tpacket_kbdq_core *pkc;
1023 struct tpacket_block_desc *pbd;
1026 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1027 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1029 /* Queue is frozen when user space is lagging behind */
1030 if (prb_queue_frozen(pkc)) {
1032 * Check if that last block which caused the queue to freeze,
1033 * is still in_use by user-space.
1035 if (prb_curr_blk_in_use(pkc, pbd)) {
1036 /* Can't record this packet */
1040 * Ok, the block was released by user-space.
1041 * Now let's open that block.
1042 * opening a block also thaws the queue.
1043 * Thawing is a side effect.
1045 prb_open_block(pkc, pbd);
1050 curr = pkc->nxt_offset;
1052 end = (char *)pbd + pkc->kblk_size;
1054 /* first try the current block */
1055 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1056 prb_fill_curr_block(curr, pkc, pbd, len);
1057 return (void *)curr;
1060 /* Ok, close the current block */
1061 prb_retire_current_block(pkc, po, 0);
1063 /* Now, try to dispatch the next block */
1064 curr = (char *)prb_dispatch_next_block(pkc, po);
1066 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1067 prb_fill_curr_block(curr, pkc, pbd, len);
1068 return (void *)curr;
1072 * No free blocks are available.user_space hasn't caught up yet.
1073 * Queue was just frozen and now this packet will get dropped.
1078 static void *packet_current_rx_frame(struct packet_sock *po,
1079 struct sk_buff *skb,
1080 int status, unsigned int len)
1083 switch (po->tp_version) {
1086 curr = packet_lookup_frame(po, &po->rx_ring,
1087 po->rx_ring.head, status);
1090 return __packet_lookup_frame_in_block(po, skb, status, len);
1092 WARN(1, "TPACKET version not supported\n");
1098 static void *prb_lookup_block(struct packet_sock *po,
1099 struct packet_ring_buffer *rb,
1103 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1104 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1106 if (status != BLOCK_STATUS(pbd))
1111 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1114 if (rb->prb_bdqc.kactive_blk_num)
1115 prev = rb->prb_bdqc.kactive_blk_num-1;
1117 prev = rb->prb_bdqc.knum_blocks-1;
1121 /* Assumes caller has held the rx_queue.lock */
1122 static void *__prb_previous_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1126 unsigned int previous = prb_previous_blk_num(rb);
1127 return prb_lookup_block(po, rb, previous, status);
1130 static void *packet_previous_rx_frame(struct packet_sock *po,
1131 struct packet_ring_buffer *rb,
1134 if (po->tp_version <= TPACKET_V2)
1135 return packet_previous_frame(po, rb, status);
1137 return __prb_previous_block(po, rb, status);
1140 static void packet_increment_rx_head(struct packet_sock *po,
1141 struct packet_ring_buffer *rb)
1143 switch (po->tp_version) {
1146 return packet_increment_head(rb);
1149 WARN(1, "TPACKET version not supported.\n");
1155 static void *packet_previous_frame(struct packet_sock *po,
1156 struct packet_ring_buffer *rb,
1159 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1160 return packet_lookup_frame(po, rb, previous, status);
1163 static void packet_increment_head(struct packet_ring_buffer *buff)
1165 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1168 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1170 struct sock *sk = &po->sk;
1173 if (po->prot_hook.func != tpacket_rcv)
1174 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1177 spin_lock(&sk->sk_receive_queue.lock);
1178 if (po->tp_version == TPACKET_V3)
1179 has_room = prb_lookup_block(po, &po->rx_ring,
1180 po->rx_ring.prb_bdqc.kactive_blk_num,
1183 has_room = packet_lookup_frame(po, &po->rx_ring,
1186 spin_unlock(&sk->sk_receive_queue.lock);
1191 static void packet_sock_destruct(struct sock *sk)
1193 skb_queue_purge(&sk->sk_error_queue);
1195 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1196 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1198 if (!sock_flag(sk, SOCK_DEAD)) {
1199 pr_err("Attempt to release alive packet socket: %p\n", sk);
1203 sk_refcnt_debug_dec(sk);
1206 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1208 int x = atomic_read(&f->rr_cur) + 1;
1216 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1217 struct sk_buff *skb,
1220 return reciprocal_divide(skb->rxhash, num);
1223 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1224 struct sk_buff *skb,
1229 cur = atomic_read(&f->rr_cur);
1230 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1231 fanout_rr_next(f, num))) != cur)
1236 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1237 struct sk_buff *skb,
1240 return smp_processor_id() % num;
1243 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1244 struct sk_buff *skb,
1247 return reciprocal_divide(prandom_u32(), num);
1250 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1251 struct sk_buff *skb,
1252 unsigned int idx, unsigned int skip,
1257 i = j = min_t(int, f->next[idx], num - 1);
1259 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1271 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1273 return f->flags & (flag >> 8);
1276 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1277 struct packet_type *pt, struct net_device *orig_dev)
1279 struct packet_fanout *f = pt->af_packet_priv;
1280 unsigned int num = f->num_members;
1281 struct packet_sock *po;
1284 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1291 case PACKET_FANOUT_HASH:
1293 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1294 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1298 skb_get_rxhash(skb);
1299 idx = fanout_demux_hash(f, skb, num);
1301 case PACKET_FANOUT_LB:
1302 idx = fanout_demux_lb(f, skb, num);
1304 case PACKET_FANOUT_CPU:
1305 idx = fanout_demux_cpu(f, skb, num);
1307 case PACKET_FANOUT_RND:
1308 idx = fanout_demux_rnd(f, skb, num);
1310 case PACKET_FANOUT_ROLLOVER:
1311 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1315 po = pkt_sk(f->arr[idx]);
1316 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1317 unlikely(!packet_rcv_has_room(po, skb))) {
1318 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1319 po = pkt_sk(f->arr[idx]);
1322 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1325 DEFINE_MUTEX(fanout_mutex);
1326 EXPORT_SYMBOL_GPL(fanout_mutex);
1327 static LIST_HEAD(fanout_list);
1329 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1331 struct packet_fanout *f = po->fanout;
1333 spin_lock(&f->lock);
1334 f->arr[f->num_members] = sk;
1337 spin_unlock(&f->lock);
1340 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1342 struct packet_fanout *f = po->fanout;
1345 spin_lock(&f->lock);
1346 for (i = 0; i < f->num_members; i++) {
1347 if (f->arr[i] == sk)
1350 BUG_ON(i >= f->num_members);
1351 f->arr[i] = f->arr[f->num_members - 1];
1353 spin_unlock(&f->lock);
1356 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1358 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1364 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1366 struct packet_sock *po = pkt_sk(sk);
1367 struct packet_fanout *f, *match;
1368 u8 type = type_flags & 0xff;
1369 u8 flags = type_flags >> 8;
1373 case PACKET_FANOUT_ROLLOVER:
1374 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1376 case PACKET_FANOUT_HASH:
1377 case PACKET_FANOUT_LB:
1378 case PACKET_FANOUT_CPU:
1379 case PACKET_FANOUT_RND:
1391 mutex_lock(&fanout_mutex);
1393 list_for_each_entry(f, &fanout_list, list) {
1395 read_pnet(&f->net) == sock_net(sk)) {
1401 if (match && match->flags != flags)
1405 match = kzalloc(sizeof(*match), GFP_KERNEL);
1408 write_pnet(&match->net, sock_net(sk));
1411 match->flags = flags;
1412 atomic_set(&match->rr_cur, 0);
1413 INIT_LIST_HEAD(&match->list);
1414 spin_lock_init(&match->lock);
1415 atomic_set(&match->sk_ref, 0);
1416 match->prot_hook.type = po->prot_hook.type;
1417 match->prot_hook.dev = po->prot_hook.dev;
1418 match->prot_hook.func = packet_rcv_fanout;
1419 match->prot_hook.af_packet_priv = match;
1420 match->prot_hook.id_match = match_fanout_group;
1421 dev_add_pack(&match->prot_hook);
1422 list_add(&match->list, &fanout_list);
1425 if (match->type == type &&
1426 match->prot_hook.type == po->prot_hook.type &&
1427 match->prot_hook.dev == po->prot_hook.dev) {
1429 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1430 __dev_remove_pack(&po->prot_hook);
1432 atomic_inc(&match->sk_ref);
1433 __fanout_link(sk, po);
1438 mutex_unlock(&fanout_mutex);
1442 static void fanout_release(struct sock *sk)
1444 struct packet_sock *po = pkt_sk(sk);
1445 struct packet_fanout *f;
1451 mutex_lock(&fanout_mutex);
1454 if (atomic_dec_and_test(&f->sk_ref)) {
1456 dev_remove_pack(&f->prot_hook);
1459 mutex_unlock(&fanout_mutex);
1462 static const struct proto_ops packet_ops;
1464 static const struct proto_ops packet_ops_spkt;
1466 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1467 struct packet_type *pt, struct net_device *orig_dev)
1470 struct sockaddr_pkt *spkt;
1473 * When we registered the protocol we saved the socket in the data
1474 * field for just this event.
1477 sk = pt->af_packet_priv;
1480 * Yank back the headers [hope the device set this
1481 * right or kerboom...]
1483 * Incoming packets have ll header pulled,
1486 * For outgoing ones skb->data == skb_mac_header(skb)
1487 * so that this procedure is noop.
1490 if (skb->pkt_type == PACKET_LOOPBACK)
1493 if (!net_eq(dev_net(dev), sock_net(sk)))
1496 skb = skb_share_check(skb, GFP_ATOMIC);
1500 /* drop any routing info */
1503 /* drop conntrack reference */
1506 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1508 skb_push(skb, skb->data - skb_mac_header(skb));
1511 * The SOCK_PACKET socket receives _all_ frames.
1514 spkt->spkt_family = dev->type;
1515 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1516 spkt->spkt_protocol = skb->protocol;
1519 * Charge the memory to the socket. This is done specifically
1520 * to prevent sockets using all the memory up.
1523 if (sock_queue_rcv_skb(sk, skb) == 0)
1534 * Output a raw packet to a device layer. This bypasses all the other
1535 * protocol layers and you must therefore supply it with a complete frame
1538 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1539 struct msghdr *msg, size_t len)
1541 struct sock *sk = sock->sk;
1542 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1543 struct sk_buff *skb = NULL;
1544 struct net_device *dev;
1550 * Get and verify the address.
1554 if (msg->msg_namelen < sizeof(struct sockaddr))
1556 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1557 proto = saddr->spkt_protocol;
1559 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1562 * Find the device first to size check it
1565 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1568 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1574 if (!(dev->flags & IFF_UP))
1578 * You may not queue a frame bigger than the mtu. This is the lowest level
1579 * raw protocol and you must do your own fragmentation at this level.
1582 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1583 if (!netif_supports_nofcs(dev)) {
1584 err = -EPROTONOSUPPORT;
1587 extra_len = 4; /* We're doing our own CRC */
1591 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1595 size_t reserved = LL_RESERVED_SPACE(dev);
1596 int tlen = dev->needed_tailroom;
1597 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1600 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1603 /* FIXME: Save some space for broken drivers that write a hard
1604 * header at transmission time by themselves. PPP is the notable
1605 * one here. This should really be fixed at the driver level.
1607 skb_reserve(skb, reserved);
1608 skb_reset_network_header(skb);
1610 /* Try to align data part correctly */
1615 skb_reset_network_header(skb);
1617 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1623 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1624 /* Earlier code assumed this would be a VLAN pkt,
1625 * double-check this now that we have the actual
1628 struct ethhdr *ehdr;
1629 skb_reset_mac_header(skb);
1630 ehdr = eth_hdr(skb);
1631 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1637 skb->protocol = proto;
1639 skb->priority = sk->sk_priority;
1640 skb->mark = sk->sk_mark;
1642 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1644 if (unlikely(extra_len == 4))
1647 skb_probe_transport_header(skb, 0);
1649 dev_queue_xmit(skb);
1660 static unsigned int run_filter(const struct sk_buff *skb,
1661 const struct sock *sk,
1664 struct sk_filter *filter;
1667 filter = rcu_dereference(sk->sk_filter);
1669 res = SK_RUN_FILTER(filter, skb);
1676 * This function makes lazy skb cloning in hope that most of packets
1677 * are discarded by BPF.
1679 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1680 * and skb->cb are mangled. It works because (and until) packets
1681 * falling here are owned by current CPU. Output packets are cloned
1682 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1683 * sequencially, so that if we return skb to original state on exit,
1684 * we will not harm anyone.
1687 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1688 struct packet_type *pt, struct net_device *orig_dev)
1691 struct sockaddr_ll *sll;
1692 struct packet_sock *po;
1693 u8 *skb_head = skb->data;
1694 int skb_len = skb->len;
1695 unsigned int snaplen, res;
1697 if (skb->pkt_type == PACKET_LOOPBACK)
1700 sk = pt->af_packet_priv;
1703 if (!net_eq(dev_net(dev), sock_net(sk)))
1708 if (dev->header_ops) {
1709 /* The device has an explicit notion of ll header,
1710 * exported to higher levels.
1712 * Otherwise, the device hides details of its frame
1713 * structure, so that corresponding packet head is
1714 * never delivered to user.
1716 if (sk->sk_type != SOCK_DGRAM)
1717 skb_push(skb, skb->data - skb_mac_header(skb));
1718 else if (skb->pkt_type == PACKET_OUTGOING) {
1719 /* Special case: outgoing packets have ll header at head */
1720 skb_pull(skb, skb_network_offset(skb));
1726 res = run_filter(skb, sk, snaplen);
1728 goto drop_n_restore;
1732 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1735 if (skb_shared(skb)) {
1736 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1740 if (skb_head != skb->data) {
1741 skb->data = skb_head;
1748 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1751 sll = &PACKET_SKB_CB(skb)->sa.ll;
1752 sll->sll_family = AF_PACKET;
1753 sll->sll_hatype = dev->type;
1754 sll->sll_protocol = skb->protocol;
1755 sll->sll_pkttype = skb->pkt_type;
1756 if (unlikely(po->origdev))
1757 sll->sll_ifindex = orig_dev->ifindex;
1759 sll->sll_ifindex = dev->ifindex;
1761 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1763 PACKET_SKB_CB(skb)->origlen = skb->len;
1765 if (pskb_trim(skb, snaplen))
1768 skb_set_owner_r(skb, sk);
1772 /* drop conntrack reference */
1775 spin_lock(&sk->sk_receive_queue.lock);
1776 po->stats.stats1.tp_packets++;
1777 skb->dropcount = atomic_read(&sk->sk_drops);
1778 __skb_queue_tail(&sk->sk_receive_queue, skb);
1779 spin_unlock(&sk->sk_receive_queue.lock);
1780 sk->sk_data_ready(sk, skb->len);
1784 spin_lock(&sk->sk_receive_queue.lock);
1785 po->stats.stats1.tp_drops++;
1786 atomic_inc(&sk->sk_drops);
1787 spin_unlock(&sk->sk_receive_queue.lock);
1790 if (skb_head != skb->data && skb_shared(skb)) {
1791 skb->data = skb_head;
1799 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1800 struct packet_type *pt, struct net_device *orig_dev)
1803 struct packet_sock *po;
1804 struct sockaddr_ll *sll;
1805 union tpacket_uhdr h;
1806 u8 *skb_head = skb->data;
1807 int skb_len = skb->len;
1808 unsigned int snaplen, res;
1809 unsigned long status = TP_STATUS_USER;
1810 unsigned short macoff, netoff, hdrlen;
1811 struct sk_buff *copy_skb = NULL;
1815 if (skb->pkt_type == PACKET_LOOPBACK)
1818 sk = pt->af_packet_priv;
1821 if (!net_eq(dev_net(dev), sock_net(sk)))
1824 if (dev->header_ops) {
1825 if (sk->sk_type != SOCK_DGRAM)
1826 skb_push(skb, skb->data - skb_mac_header(skb));
1827 else if (skb->pkt_type == PACKET_OUTGOING) {
1828 /* Special case: outgoing packets have ll header at head */
1829 skb_pull(skb, skb_network_offset(skb));
1833 if (skb->ip_summed == CHECKSUM_PARTIAL)
1834 status |= TP_STATUS_CSUMNOTREADY;
1838 res = run_filter(skb, sk, snaplen);
1840 goto drop_n_restore;
1844 if (sk->sk_type == SOCK_DGRAM) {
1845 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1848 unsigned int maclen = skb_network_offset(skb);
1849 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1850 (maclen < 16 ? 16 : maclen)) +
1852 macoff = netoff - maclen;
1854 if (po->tp_version <= TPACKET_V2) {
1855 if (macoff + snaplen > po->rx_ring.frame_size) {
1856 if (po->copy_thresh &&
1857 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1858 if (skb_shared(skb)) {
1859 copy_skb = skb_clone(skb, GFP_ATOMIC);
1861 copy_skb = skb_get(skb);
1862 skb_head = skb->data;
1865 skb_set_owner_r(copy_skb, sk);
1867 snaplen = po->rx_ring.frame_size - macoff;
1868 if ((int)snaplen < 0)
1872 spin_lock(&sk->sk_receive_queue.lock);
1873 h.raw = packet_current_rx_frame(po, skb,
1874 TP_STATUS_KERNEL, (macoff+snaplen));
1877 if (po->tp_version <= TPACKET_V2) {
1878 packet_increment_rx_head(po, &po->rx_ring);
1880 * LOSING will be reported till you read the stats,
1881 * because it's COR - Clear On Read.
1882 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1885 if (po->stats.stats1.tp_drops)
1886 status |= TP_STATUS_LOSING;
1888 po->stats.stats1.tp_packets++;
1890 status |= TP_STATUS_COPY;
1891 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1893 spin_unlock(&sk->sk_receive_queue.lock);
1895 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1897 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1898 getnstimeofday(&ts);
1900 status |= ts_status;
1902 switch (po->tp_version) {
1904 h.h1->tp_len = skb->len;
1905 h.h1->tp_snaplen = snaplen;
1906 h.h1->tp_mac = macoff;
1907 h.h1->tp_net = netoff;
1908 h.h1->tp_sec = ts.tv_sec;
1909 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1910 hdrlen = sizeof(*h.h1);
1913 h.h2->tp_len = skb->len;
1914 h.h2->tp_snaplen = snaplen;
1915 h.h2->tp_mac = macoff;
1916 h.h2->tp_net = netoff;
1917 h.h2->tp_sec = ts.tv_sec;
1918 h.h2->tp_nsec = ts.tv_nsec;
1919 if (vlan_tx_tag_present(skb)) {
1920 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1921 status |= TP_STATUS_VLAN_VALID;
1923 h.h2->tp_vlan_tci = 0;
1925 h.h2->tp_padding = 0;
1926 hdrlen = sizeof(*h.h2);
1929 /* tp_nxt_offset,vlan are already populated above.
1930 * So DONT clear those fields here
1932 h.h3->tp_status |= status;
1933 h.h3->tp_len = skb->len;
1934 h.h3->tp_snaplen = snaplen;
1935 h.h3->tp_mac = macoff;
1936 h.h3->tp_net = netoff;
1937 h.h3->tp_sec = ts.tv_sec;
1938 h.h3->tp_nsec = ts.tv_nsec;
1939 hdrlen = sizeof(*h.h3);
1945 sll = h.raw + TPACKET_ALIGN(hdrlen);
1946 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1947 sll->sll_family = AF_PACKET;
1948 sll->sll_hatype = dev->type;
1949 sll->sll_protocol = skb->protocol;
1950 sll->sll_pkttype = skb->pkt_type;
1951 if (unlikely(po->origdev))
1952 sll->sll_ifindex = orig_dev->ifindex;
1954 sll->sll_ifindex = dev->ifindex;
1957 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1961 if (po->tp_version <= TPACKET_V2) {
1962 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1963 + macoff + snaplen);
1964 for (start = h.raw; start < end; start += PAGE_SIZE)
1965 flush_dcache_page(pgv_to_page(start));
1970 if (po->tp_version <= TPACKET_V2)
1971 __packet_set_status(po, h.raw, status);
1973 prb_clear_blk_fill_status(&po->rx_ring);
1975 sk->sk_data_ready(sk, 0);
1978 if (skb_head != skb->data && skb_shared(skb)) {
1979 skb->data = skb_head;
1987 po->stats.stats1.tp_drops++;
1988 spin_unlock(&sk->sk_receive_queue.lock);
1990 sk->sk_data_ready(sk, 0);
1991 kfree_skb(copy_skb);
1992 goto drop_n_restore;
1995 static void tpacket_destruct_skb(struct sk_buff *skb)
1997 struct packet_sock *po = pkt_sk(skb->sk);
2000 if (likely(po->tx_ring.pg_vec)) {
2003 ph = skb_shinfo(skb)->destructor_arg;
2004 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
2005 atomic_dec(&po->tx_ring.pending);
2007 ts = __packet_set_timestamp(po, ph, skb);
2008 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2014 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2015 void *frame, struct net_device *dev, int size_max,
2016 __be16 proto, unsigned char *addr, int hlen)
2018 union tpacket_uhdr ph;
2019 int to_write, offset, len, tp_len, nr_frags, len_max;
2020 struct socket *sock = po->sk.sk_socket;
2027 skb->protocol = proto;
2029 skb->priority = po->sk.sk_priority;
2030 skb->mark = po->sk.sk_mark;
2031 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2032 skb_shinfo(skb)->destructor_arg = ph.raw;
2034 switch (po->tp_version) {
2036 tp_len = ph.h2->tp_len;
2039 tp_len = ph.h1->tp_len;
2042 if (unlikely(tp_len > size_max)) {
2043 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2047 skb_reserve(skb, hlen);
2048 skb_reset_network_header(skb);
2050 if (!packet_use_direct_xmit(po))
2051 skb_probe_transport_header(skb, 0);
2052 if (unlikely(po->tp_tx_has_off)) {
2053 int off_min, off_max, off;
2054 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2055 off_max = po->tx_ring.frame_size - tp_len;
2056 if (sock->type == SOCK_DGRAM) {
2057 switch (po->tp_version) {
2059 off = ph.h2->tp_net;
2062 off = ph.h1->tp_net;
2066 switch (po->tp_version) {
2068 off = ph.h2->tp_mac;
2071 off = ph.h1->tp_mac;
2075 if (unlikely((off < off_min) || (off_max < off)))
2077 data = ph.raw + off;
2079 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2083 if (sock->type == SOCK_DGRAM) {
2084 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2086 if (unlikely(err < 0))
2088 } else if (dev->hard_header_len) {
2089 /* net device doesn't like empty head */
2090 if (unlikely(tp_len <= dev->hard_header_len)) {
2091 pr_err("packet size is too short (%d < %d)\n",
2092 tp_len, dev->hard_header_len);
2096 skb_push(skb, dev->hard_header_len);
2097 err = skb_store_bits(skb, 0, data,
2098 dev->hard_header_len);
2102 data += dev->hard_header_len;
2103 to_write -= dev->hard_header_len;
2106 offset = offset_in_page(data);
2107 len_max = PAGE_SIZE - offset;
2108 len = ((to_write > len_max) ? len_max : to_write);
2110 skb->data_len = to_write;
2111 skb->len += to_write;
2112 skb->truesize += to_write;
2113 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2115 while (likely(to_write)) {
2116 nr_frags = skb_shinfo(skb)->nr_frags;
2118 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2119 pr_err("Packet exceed the number of skb frags(%lu)\n",
2124 page = pgv_to_page(data);
2126 flush_dcache_page(page);
2128 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2131 len_max = PAGE_SIZE;
2132 len = ((to_write > len_max) ? len_max : to_write);
2138 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2140 struct sk_buff *skb;
2141 struct net_device *dev;
2143 int err, reserve = 0;
2145 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2146 int tp_len, size_max;
2147 unsigned char *addr;
2149 int status = TP_STATUS_AVAILABLE;
2152 mutex_lock(&po->pg_vec_lock);
2154 if (likely(saddr == NULL)) {
2155 dev = packet_cached_dev_get(po);
2160 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2162 if (msg->msg_namelen < (saddr->sll_halen
2163 + offsetof(struct sockaddr_ll,
2166 proto = saddr->sll_protocol;
2167 addr = saddr->sll_addr;
2168 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2172 if (unlikely(dev == NULL))
2175 if (unlikely(!(dev->flags & IFF_UP)))
2178 reserve = dev->hard_header_len;
2180 size_max = po->tx_ring.frame_size
2181 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2183 if (size_max > dev->mtu + reserve)
2184 size_max = dev->mtu + reserve;
2187 ph = packet_current_frame(po, &po->tx_ring,
2188 TP_STATUS_SEND_REQUEST);
2190 if (unlikely(ph == NULL)) {
2195 status = TP_STATUS_SEND_REQUEST;
2196 hlen = LL_RESERVED_SPACE(dev);
2197 tlen = dev->needed_tailroom;
2198 skb = sock_alloc_send_skb(&po->sk,
2199 hlen + tlen + sizeof(struct sockaddr_ll),
2202 if (unlikely(skb == NULL))
2205 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2208 if (unlikely(tp_len < 0)) {
2210 __packet_set_status(po, ph,
2211 TP_STATUS_AVAILABLE);
2212 packet_increment_head(&po->tx_ring);
2216 status = TP_STATUS_WRONG_FORMAT;
2222 skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2223 skb->destructor = tpacket_destruct_skb;
2224 __packet_set_status(po, ph, TP_STATUS_SENDING);
2225 atomic_inc(&po->tx_ring.pending);
2227 status = TP_STATUS_SEND_REQUEST;
2228 err = po->xmit(skb);
2229 if (unlikely(err > 0)) {
2230 err = net_xmit_errno(err);
2231 if (err && __packet_get_status(po, ph) ==
2232 TP_STATUS_AVAILABLE) {
2233 /* skb was destructed already */
2238 * skb was dropped but not destructed yet;
2239 * let's treat it like congestion or err < 0
2243 packet_increment_head(&po->tx_ring);
2245 } while (likely((ph != NULL) ||
2246 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2247 (atomic_read(&po->tx_ring.pending))))
2254 __packet_set_status(po, ph, status);
2259 mutex_unlock(&po->pg_vec_lock);
2263 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2264 size_t reserve, size_t len,
2265 size_t linear, int noblock,
2268 struct sk_buff *skb;
2270 /* Under a page? Don't bother with paged skb. */
2271 if (prepad + len < PAGE_SIZE || !linear)
2274 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2279 skb_reserve(skb, reserve);
2280 skb_put(skb, linear);
2281 skb->data_len = len - linear;
2282 skb->len += len - linear;
2287 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2289 struct sock *sk = sock->sk;
2290 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2291 struct sk_buff *skb;
2292 struct net_device *dev;
2294 unsigned char *addr;
2295 int err, reserve = 0;
2296 struct virtio_net_hdr vnet_hdr = { 0 };
2299 struct packet_sock *po = pkt_sk(sk);
2300 unsigned short gso_type = 0;
2305 * Get and verify the address.
2308 if (likely(saddr == NULL)) {
2309 dev = packet_cached_dev_get(po);
2314 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2316 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2318 proto = saddr->sll_protocol;
2319 addr = saddr->sll_addr;
2320 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2324 if (unlikely(dev == NULL))
2327 if (unlikely(!(dev->flags & IFF_UP)))
2330 if (sock->type == SOCK_RAW)
2331 reserve = dev->hard_header_len;
2332 if (po->has_vnet_hdr) {
2333 vnet_hdr_len = sizeof(vnet_hdr);
2336 if (len < vnet_hdr_len)
2339 len -= vnet_hdr_len;
2341 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2346 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2347 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2349 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2350 vnet_hdr.csum_offset + 2;
2353 if (vnet_hdr.hdr_len > len)
2356 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2357 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2358 case VIRTIO_NET_HDR_GSO_TCPV4:
2359 gso_type = SKB_GSO_TCPV4;
2361 case VIRTIO_NET_HDR_GSO_TCPV6:
2362 gso_type = SKB_GSO_TCPV6;
2364 case VIRTIO_NET_HDR_GSO_UDP:
2365 gso_type = SKB_GSO_UDP;
2371 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2372 gso_type |= SKB_GSO_TCP_ECN;
2374 if (vnet_hdr.gso_size == 0)
2380 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2381 if (!netif_supports_nofcs(dev)) {
2382 err = -EPROTONOSUPPORT;
2385 extra_len = 4; /* We're doing our own CRC */
2389 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2393 hlen = LL_RESERVED_SPACE(dev);
2394 tlen = dev->needed_tailroom;
2395 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2396 msg->msg_flags & MSG_DONTWAIT, &err);
2400 skb_set_network_header(skb, reserve);
2403 if (sock->type == SOCK_DGRAM &&
2404 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2407 /* Returns -EFAULT on error */
2408 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2412 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2414 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2415 /* Earlier code assumed this would be a VLAN pkt,
2416 * double-check this now that we have the actual
2419 struct ethhdr *ehdr;
2420 skb_reset_mac_header(skb);
2421 ehdr = eth_hdr(skb);
2422 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2428 skb->protocol = proto;
2430 skb->priority = sk->sk_priority;
2431 skb->mark = sk->sk_mark;
2432 skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2434 if (po->has_vnet_hdr) {
2435 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2436 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2437 vnet_hdr.csum_offset)) {
2443 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2444 skb_shinfo(skb)->gso_type = gso_type;
2446 /* Header must be checked, and gso_segs computed. */
2447 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2448 skb_shinfo(skb)->gso_segs = 0;
2450 len += vnet_hdr_len;
2453 if (!packet_use_direct_xmit(po))
2454 skb_probe_transport_header(skb, reserve);
2455 if (unlikely(extra_len == 4))
2458 err = po->xmit(skb);
2459 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2475 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2476 struct msghdr *msg, size_t len)
2478 struct sock *sk = sock->sk;
2479 struct packet_sock *po = pkt_sk(sk);
2481 if (po->tx_ring.pg_vec)
2482 return tpacket_snd(po, msg);
2484 return packet_snd(sock, msg, len);
2488 * Close a PACKET socket. This is fairly simple. We immediately go
2489 * to 'closed' state and remove our protocol entry in the device list.
2492 static int packet_release(struct socket *sock)
2494 struct sock *sk = sock->sk;
2495 struct packet_sock *po;
2497 union tpacket_req_u req_u;
2505 mutex_lock(&net->packet.sklist_lock);
2506 sk_del_node_init_rcu(sk);
2507 mutex_unlock(&net->packet.sklist_lock);
2510 sock_prot_inuse_add(net, sk->sk_prot, -1);
2513 spin_lock(&po->bind_lock);
2514 unregister_prot_hook(sk, false);
2515 packet_cached_dev_reset(po);
2517 if (po->prot_hook.dev) {
2518 dev_put(po->prot_hook.dev);
2519 po->prot_hook.dev = NULL;
2521 spin_unlock(&po->bind_lock);
2523 packet_flush_mclist(sk);
2525 if (po->rx_ring.pg_vec) {
2526 memset(&req_u, 0, sizeof(req_u));
2527 packet_set_ring(sk, &req_u, 1, 0);
2530 if (po->tx_ring.pg_vec) {
2531 memset(&req_u, 0, sizeof(req_u));
2532 packet_set_ring(sk, &req_u, 1, 1);
2539 * Now the socket is dead. No more input will appear.
2546 skb_queue_purge(&sk->sk_receive_queue);
2547 sk_refcnt_debug_release(sk);
2554 * Attach a packet hook.
2557 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2559 struct packet_sock *po = pkt_sk(sk);
2570 spin_lock(&po->bind_lock);
2571 unregister_prot_hook(sk, true);
2574 po->prot_hook.type = protocol;
2575 if (po->prot_hook.dev)
2576 dev_put(po->prot_hook.dev);
2578 po->prot_hook.dev = dev;
2579 po->ifindex = dev ? dev->ifindex : 0;
2581 packet_cached_dev_assign(po, dev);
2586 if (!dev || (dev->flags & IFF_UP)) {
2587 register_prot_hook(sk);
2589 sk->sk_err = ENETDOWN;
2590 if (!sock_flag(sk, SOCK_DEAD))
2591 sk->sk_error_report(sk);
2595 spin_unlock(&po->bind_lock);
2601 * Bind a packet socket to a device
2604 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2607 struct sock *sk = sock->sk;
2609 struct net_device *dev;
2616 if (addr_len != sizeof(struct sockaddr))
2618 strlcpy(name, uaddr->sa_data, sizeof(name));
2620 dev = dev_get_by_name(sock_net(sk), name);
2622 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2626 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2628 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2629 struct sock *sk = sock->sk;
2630 struct net_device *dev = NULL;
2638 if (addr_len < sizeof(struct sockaddr_ll))
2640 if (sll->sll_family != AF_PACKET)
2643 if (sll->sll_ifindex) {
2645 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2649 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2655 static struct proto packet_proto = {
2657 .owner = THIS_MODULE,
2658 .obj_size = sizeof(struct packet_sock),
2662 * Create a packet of type SOCK_PACKET.
2665 static int packet_create(struct net *net, struct socket *sock, int protocol,
2669 struct packet_sock *po;
2670 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2673 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2675 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2676 sock->type != SOCK_PACKET)
2677 return -ESOCKTNOSUPPORT;
2679 sock->state = SS_UNCONNECTED;
2682 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2686 sock->ops = &packet_ops;
2687 if (sock->type == SOCK_PACKET)
2688 sock->ops = &packet_ops_spkt;
2690 sock_init_data(sock, sk);
2693 sk->sk_family = PF_PACKET;
2695 po->xmit = dev_queue_xmit;
2697 packet_cached_dev_reset(po);
2699 sk->sk_destruct = packet_sock_destruct;
2700 sk_refcnt_debug_inc(sk);
2703 * Attach a protocol block
2706 spin_lock_init(&po->bind_lock);
2707 mutex_init(&po->pg_vec_lock);
2708 po->prot_hook.func = packet_rcv;
2710 if (sock->type == SOCK_PACKET)
2711 po->prot_hook.func = packet_rcv_spkt;
2713 po->prot_hook.af_packet_priv = sk;
2716 po->prot_hook.type = proto;
2717 register_prot_hook(sk);
2720 mutex_lock(&net->packet.sklist_lock);
2721 sk_add_node_rcu(sk, &net->packet.sklist);
2722 mutex_unlock(&net->packet.sklist_lock);
2725 sock_prot_inuse_add(net, &packet_proto, 1);
2734 * Pull a packet from our receive queue and hand it to the user.
2735 * If necessary we block.
2738 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2739 struct msghdr *msg, size_t len, int flags)
2741 struct sock *sk = sock->sk;
2742 struct sk_buff *skb;
2744 int vnet_hdr_len = 0;
2747 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2751 /* What error should we return now? EUNATTACH? */
2752 if (pkt_sk(sk)->ifindex < 0)
2756 if (flags & MSG_ERRQUEUE) {
2757 err = sock_recv_errqueue(sk, msg, len,
2758 SOL_PACKET, PACKET_TX_TIMESTAMP);
2763 * Call the generic datagram receiver. This handles all sorts
2764 * of horrible races and re-entrancy so we can forget about it
2765 * in the protocol layers.
2767 * Now it will return ENETDOWN, if device have just gone down,
2768 * but then it will block.
2771 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2774 * An error occurred so return it. Because skb_recv_datagram()
2775 * handles the blocking we don't see and worry about blocking
2782 if (pkt_sk(sk)->has_vnet_hdr) {
2783 struct virtio_net_hdr vnet_hdr = { 0 };
2786 vnet_hdr_len = sizeof(vnet_hdr);
2787 if (len < vnet_hdr_len)
2790 len -= vnet_hdr_len;
2792 if (skb_is_gso(skb)) {
2793 struct skb_shared_info *sinfo = skb_shinfo(skb);
2795 /* This is a hint as to how much should be linear. */
2796 vnet_hdr.hdr_len = skb_headlen(skb);
2797 vnet_hdr.gso_size = sinfo->gso_size;
2798 if (sinfo->gso_type & SKB_GSO_TCPV4)
2799 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2800 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2801 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2802 else if (sinfo->gso_type & SKB_GSO_UDP)
2803 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2804 else if (sinfo->gso_type & SKB_GSO_FCOE)
2808 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2809 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2811 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2813 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2814 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2815 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2816 vnet_hdr.csum_offset = skb->csum_offset;
2817 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2818 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2819 } /* else everything is zero */
2821 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2827 /* You lose any data beyond the buffer you gave. If it worries
2828 * a user program they can ask the device for its MTU
2834 msg->msg_flags |= MSG_TRUNC;
2837 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2841 sock_recv_ts_and_drops(msg, sk, skb);
2843 if (msg->msg_name) {
2844 /* If the address length field is there to be filled
2845 * in, we fill it in now.
2847 if (sock->type == SOCK_PACKET) {
2848 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2850 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2851 msg->msg_namelen = sll->sll_halen +
2852 offsetof(struct sockaddr_ll, sll_addr);
2854 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2858 if (pkt_sk(sk)->auxdata) {
2859 struct tpacket_auxdata aux;
2861 aux.tp_status = TP_STATUS_USER;
2862 if (skb->ip_summed == CHECKSUM_PARTIAL)
2863 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2864 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2865 aux.tp_snaplen = skb->len;
2867 aux.tp_net = skb_network_offset(skb);
2868 if (vlan_tx_tag_present(skb)) {
2869 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2870 aux.tp_status |= TP_STATUS_VLAN_VALID;
2872 aux.tp_vlan_tci = 0;
2875 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2879 * Free or return the buffer as appropriate. Again this
2880 * hides all the races and re-entrancy issues from us.
2882 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2885 skb_free_datagram(sk, skb);
2890 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2891 int *uaddr_len, int peer)
2893 struct net_device *dev;
2894 struct sock *sk = sock->sk;
2899 uaddr->sa_family = AF_PACKET;
2900 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2902 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2904 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2906 *uaddr_len = sizeof(*uaddr);
2911 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2912 int *uaddr_len, int peer)
2914 struct net_device *dev;
2915 struct sock *sk = sock->sk;
2916 struct packet_sock *po = pkt_sk(sk);
2917 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2922 sll->sll_family = AF_PACKET;
2923 sll->sll_ifindex = po->ifindex;
2924 sll->sll_protocol = po->num;
2925 sll->sll_pkttype = 0;
2927 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2929 sll->sll_hatype = dev->type;
2930 sll->sll_halen = dev->addr_len;
2931 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2933 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2937 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2942 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2946 case PACKET_MR_MULTICAST:
2947 if (i->alen != dev->addr_len)
2950 return dev_mc_add(dev, i->addr);
2952 return dev_mc_del(dev, i->addr);
2954 case PACKET_MR_PROMISC:
2955 return dev_set_promiscuity(dev, what);
2957 case PACKET_MR_ALLMULTI:
2958 return dev_set_allmulti(dev, what);
2960 case PACKET_MR_UNICAST:
2961 if (i->alen != dev->addr_len)
2964 return dev_uc_add(dev, i->addr);
2966 return dev_uc_del(dev, i->addr);
2974 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2976 for ( ; i; i = i->next) {
2977 if (i->ifindex == dev->ifindex)
2978 packet_dev_mc(dev, i, what);
2982 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2984 struct packet_sock *po = pkt_sk(sk);
2985 struct packet_mclist *ml, *i;
2986 struct net_device *dev;
2992 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2997 if (mreq->mr_alen > dev->addr_len)
3001 i = kmalloc(sizeof(*i), GFP_KERNEL);
3006 for (ml = po->mclist; ml; ml = ml->next) {
3007 if (ml->ifindex == mreq->mr_ifindex &&
3008 ml->type == mreq->mr_type &&
3009 ml->alen == mreq->mr_alen &&
3010 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3012 /* Free the new element ... */
3018 i->type = mreq->mr_type;
3019 i->ifindex = mreq->mr_ifindex;
3020 i->alen = mreq->mr_alen;
3021 memcpy(i->addr, mreq->mr_address, i->alen);
3023 i->next = po->mclist;
3025 err = packet_dev_mc(dev, i, 1);
3027 po->mclist = i->next;
3036 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3038 struct packet_mclist *ml, **mlp;
3042 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3043 if (ml->ifindex == mreq->mr_ifindex &&
3044 ml->type == mreq->mr_type &&
3045 ml->alen == mreq->mr_alen &&
3046 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3047 if (--ml->count == 0) {
3048 struct net_device *dev;
3050 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3052 packet_dev_mc(dev, ml, -1);
3060 return -EADDRNOTAVAIL;
3063 static void packet_flush_mclist(struct sock *sk)
3065 struct packet_sock *po = pkt_sk(sk);
3066 struct packet_mclist *ml;
3072 while ((ml = po->mclist) != NULL) {
3073 struct net_device *dev;
3075 po->mclist = ml->next;
3076 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3078 packet_dev_mc(dev, ml, -1);
3085 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3087 struct sock *sk = sock->sk;
3088 struct packet_sock *po = pkt_sk(sk);
3091 if (level != SOL_PACKET)
3092 return -ENOPROTOOPT;
3095 case PACKET_ADD_MEMBERSHIP:
3096 case PACKET_DROP_MEMBERSHIP:
3098 struct packet_mreq_max mreq;
3100 memset(&mreq, 0, sizeof(mreq));
3101 if (len < sizeof(struct packet_mreq))
3103 if (len > sizeof(mreq))
3105 if (copy_from_user(&mreq, optval, len))
3107 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3109 if (optname == PACKET_ADD_MEMBERSHIP)
3110 ret = packet_mc_add(sk, &mreq);
3112 ret = packet_mc_drop(sk, &mreq);
3116 case PACKET_RX_RING:
3117 case PACKET_TX_RING:
3119 union tpacket_req_u req_u;
3122 switch (po->tp_version) {
3125 len = sizeof(req_u.req);
3129 len = sizeof(req_u.req3);
3134 if (pkt_sk(sk)->has_vnet_hdr)
3136 if (copy_from_user(&req_u.req, optval, len))
3138 return packet_set_ring(sk, &req_u, 0,
3139 optname == PACKET_TX_RING);
3141 case PACKET_COPY_THRESH:
3145 if (optlen != sizeof(val))
3147 if (copy_from_user(&val, optval, sizeof(val)))
3150 pkt_sk(sk)->copy_thresh = val;
3153 case PACKET_VERSION:
3157 if (optlen != sizeof(val))
3159 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3161 if (copy_from_user(&val, optval, sizeof(val)))
3167 po->tp_version = val;
3173 case PACKET_RESERVE:
3177 if (optlen != sizeof(val))
3179 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3181 if (copy_from_user(&val, optval, sizeof(val)))
3183 po->tp_reserve = val;
3190 if (optlen != sizeof(val))
3192 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3194 if (copy_from_user(&val, optval, sizeof(val)))
3196 po->tp_loss = !!val;
3199 case PACKET_AUXDATA:
3203 if (optlen < sizeof(val))
3205 if (copy_from_user(&val, optval, sizeof(val)))
3208 po->auxdata = !!val;
3211 case PACKET_ORIGDEV:
3215 if (optlen < sizeof(val))
3217 if (copy_from_user(&val, optval, sizeof(val)))
3220 po->origdev = !!val;
3223 case PACKET_VNET_HDR:
3227 if (sock->type != SOCK_RAW)
3229 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3231 if (optlen < sizeof(val))
3233 if (copy_from_user(&val, optval, sizeof(val)))
3236 po->has_vnet_hdr = !!val;
3239 case PACKET_TIMESTAMP:
3243 if (optlen != sizeof(val))
3245 if (copy_from_user(&val, optval, sizeof(val)))
3248 po->tp_tstamp = val;
3255 if (optlen != sizeof(val))
3257 if (copy_from_user(&val, optval, sizeof(val)))
3260 return fanout_add(sk, val & 0xffff, val >> 16);
3262 case PACKET_TX_HAS_OFF:
3266 if (optlen != sizeof(val))
3268 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3270 if (copy_from_user(&val, optval, sizeof(val)))
3272 po->tp_tx_has_off = !!val;
3275 case PACKET_QDISC_BYPASS:
3279 if (optlen != sizeof(val))
3281 if (copy_from_user(&val, optval, sizeof(val)))
3284 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3288 return -ENOPROTOOPT;
3292 static int packet_getsockopt(struct socket *sock, int level, int optname,
3293 char __user *optval, int __user *optlen)
3296 int val, lv = sizeof(val);
3297 struct sock *sk = sock->sk;
3298 struct packet_sock *po = pkt_sk(sk);
3300 union tpacket_stats_u st;
3302 if (level != SOL_PACKET)
3303 return -ENOPROTOOPT;
3305 if (get_user(len, optlen))
3312 case PACKET_STATISTICS:
3313 spin_lock_bh(&sk->sk_receive_queue.lock);
3314 memcpy(&st, &po->stats, sizeof(st));
3315 memset(&po->stats, 0, sizeof(po->stats));
3316 spin_unlock_bh(&sk->sk_receive_queue.lock);
3318 if (po->tp_version == TPACKET_V3) {
3319 lv = sizeof(struct tpacket_stats_v3);
3320 st.stats3.tp_packets += st.stats3.tp_drops;
3323 lv = sizeof(struct tpacket_stats);
3324 st.stats1.tp_packets += st.stats1.tp_drops;
3329 case PACKET_AUXDATA:
3332 case PACKET_ORIGDEV:
3335 case PACKET_VNET_HDR:
3336 val = po->has_vnet_hdr;
3338 case PACKET_VERSION:
3339 val = po->tp_version;
3342 if (len > sizeof(int))
3344 if (copy_from_user(&val, optval, len))
3348 val = sizeof(struct tpacket_hdr);
3351 val = sizeof(struct tpacket2_hdr);
3354 val = sizeof(struct tpacket3_hdr);
3360 case PACKET_RESERVE:
3361 val = po->tp_reserve;
3366 case PACKET_TIMESTAMP:
3367 val = po->tp_tstamp;
3371 ((u32)po->fanout->id |
3372 ((u32)po->fanout->type << 16) |
3373 ((u32)po->fanout->flags << 24)) :
3376 case PACKET_TX_HAS_OFF:
3377 val = po->tp_tx_has_off;
3379 case PACKET_QDISC_BYPASS:
3380 val = packet_use_direct_xmit(po);
3383 return -ENOPROTOOPT;
3388 if (put_user(len, optlen))
3390 if (copy_to_user(optval, data, len))
3396 static int packet_notifier(struct notifier_block *this,
3397 unsigned long msg, void *ptr)
3400 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3401 struct net *net = dev_net(dev);
3404 sk_for_each_rcu(sk, &net->packet.sklist) {
3405 struct packet_sock *po = pkt_sk(sk);
3408 case NETDEV_UNREGISTER:
3410 packet_dev_mclist(dev, po->mclist, -1);
3414 if (dev->ifindex == po->ifindex) {
3415 spin_lock(&po->bind_lock);
3417 __unregister_prot_hook(sk, false);
3418 sk->sk_err = ENETDOWN;
3419 if (!sock_flag(sk, SOCK_DEAD))
3420 sk->sk_error_report(sk);
3422 if (msg == NETDEV_UNREGISTER) {
3423 packet_cached_dev_reset(po);
3425 if (po->prot_hook.dev)
3426 dev_put(po->prot_hook.dev);
3427 po->prot_hook.dev = NULL;
3429 spin_unlock(&po->bind_lock);
3433 if (dev->ifindex == po->ifindex) {
3434 spin_lock(&po->bind_lock);
3436 register_prot_hook(sk);
3437 spin_unlock(&po->bind_lock);
3447 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3450 struct sock *sk = sock->sk;
3455 int amount = sk_wmem_alloc_get(sk);
3457 return put_user(amount, (int __user *)arg);
3461 struct sk_buff *skb;
3464 spin_lock_bh(&sk->sk_receive_queue.lock);
3465 skb = skb_peek(&sk->sk_receive_queue);
3468 spin_unlock_bh(&sk->sk_receive_queue.lock);
3469 return put_user(amount, (int __user *)arg);
3472 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3474 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3484 case SIOCGIFBRDADDR:
3485 case SIOCSIFBRDADDR:
3486 case SIOCGIFNETMASK:
3487 case SIOCSIFNETMASK:
3488 case SIOCGIFDSTADDR:
3489 case SIOCSIFDSTADDR:
3491 return inet_dgram_ops.ioctl(sock, cmd, arg);
3495 return -ENOIOCTLCMD;
3500 static unsigned int packet_poll(struct file *file, struct socket *sock,
3503 struct sock *sk = sock->sk;
3504 struct packet_sock *po = pkt_sk(sk);
3505 unsigned int mask = datagram_poll(file, sock, wait);
3507 spin_lock_bh(&sk->sk_receive_queue.lock);
3508 if (po->rx_ring.pg_vec) {
3509 if (!packet_previous_rx_frame(po, &po->rx_ring,
3511 mask |= POLLIN | POLLRDNORM;
3513 spin_unlock_bh(&sk->sk_receive_queue.lock);
3514 spin_lock_bh(&sk->sk_write_queue.lock);
3515 if (po->tx_ring.pg_vec) {
3516 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3517 mask |= POLLOUT | POLLWRNORM;
3519 spin_unlock_bh(&sk->sk_write_queue.lock);
3524 /* Dirty? Well, I still did not learn better way to account
3528 static void packet_mm_open(struct vm_area_struct *vma)
3530 struct file *file = vma->vm_file;
3531 struct socket *sock = file->private_data;
3532 struct sock *sk = sock->sk;
3535 atomic_inc(&pkt_sk(sk)->mapped);
3538 static void packet_mm_close(struct vm_area_struct *vma)
3540 struct file *file = vma->vm_file;
3541 struct socket *sock = file->private_data;
3542 struct sock *sk = sock->sk;
3545 atomic_dec(&pkt_sk(sk)->mapped);
3548 static const struct vm_operations_struct packet_mmap_ops = {
3549 .open = packet_mm_open,
3550 .close = packet_mm_close,
3553 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3558 for (i = 0; i < len; i++) {
3559 if (likely(pg_vec[i].buffer)) {
3560 if (is_vmalloc_addr(pg_vec[i].buffer))
3561 vfree(pg_vec[i].buffer);
3563 free_pages((unsigned long)pg_vec[i].buffer,
3565 pg_vec[i].buffer = NULL;
3571 static char *alloc_one_pg_vec_page(unsigned long order)
3573 char *buffer = NULL;
3574 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3575 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3577 buffer = (char *) __get_free_pages(gfp_flags, order);
3583 * __get_free_pages failed, fall back to vmalloc
3585 buffer = vzalloc((1 << order) * PAGE_SIZE);
3591 * vmalloc failed, lets dig into swap here
3593 gfp_flags &= ~__GFP_NORETRY;
3594 buffer = (char *)__get_free_pages(gfp_flags, order);
3599 * complete and utter failure
3604 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3606 unsigned int block_nr = req->tp_block_nr;
3610 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3611 if (unlikely(!pg_vec))
3614 for (i = 0; i < block_nr; i++) {
3615 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3616 if (unlikely(!pg_vec[i].buffer))
3617 goto out_free_pgvec;
3624 free_pg_vec(pg_vec, order, block_nr);
3629 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3630 int closing, int tx_ring)
3632 struct pgv *pg_vec = NULL;
3633 struct packet_sock *po = pkt_sk(sk);
3634 int was_running, order = 0;
3635 struct packet_ring_buffer *rb;
3636 struct sk_buff_head *rb_queue;
3639 /* Added to avoid minimal code churn */
3640 struct tpacket_req *req = &req_u->req;
3642 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3643 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3644 WARN(1, "Tx-ring is not supported.\n");
3648 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3649 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3653 if (atomic_read(&po->mapped))
3655 if (atomic_read(&rb->pending))
3659 if (req->tp_block_nr) {
3660 /* Sanity tests and some calculations */
3662 if (unlikely(rb->pg_vec))
3665 switch (po->tp_version) {
3667 po->tp_hdrlen = TPACKET_HDRLEN;
3670 po->tp_hdrlen = TPACKET2_HDRLEN;
3673 po->tp_hdrlen = TPACKET3_HDRLEN;
3678 if (unlikely((int)req->tp_block_size <= 0))
3680 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3682 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3685 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3688 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3689 if (unlikely(rb->frames_per_block <= 0))
3691 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3696 order = get_order(req->tp_block_size);
3697 pg_vec = alloc_pg_vec(req, order);
3698 if (unlikely(!pg_vec))
3700 switch (po->tp_version) {
3702 /* Transmit path is not supported. We checked
3703 * it above but just being paranoid
3706 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3715 if (unlikely(req->tp_frame_nr))
3721 /* Detach socket from network */
3722 spin_lock(&po->bind_lock);
3723 was_running = po->running;
3727 __unregister_prot_hook(sk, false);
3729 spin_unlock(&po->bind_lock);
3734 mutex_lock(&po->pg_vec_lock);
3735 if (closing || atomic_read(&po->mapped) == 0) {
3737 spin_lock_bh(&rb_queue->lock);
3738 swap(rb->pg_vec, pg_vec);
3739 rb->frame_max = (req->tp_frame_nr - 1);
3741 rb->frame_size = req->tp_frame_size;
3742 spin_unlock_bh(&rb_queue->lock);
3744 swap(rb->pg_vec_order, order);
3745 swap(rb->pg_vec_len, req->tp_block_nr);
3747 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3748 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3749 tpacket_rcv : packet_rcv;
3750 skb_queue_purge(rb_queue);
3751 if (atomic_read(&po->mapped))
3752 pr_err("packet_mmap: vma is busy: %d\n",
3753 atomic_read(&po->mapped));
3755 mutex_unlock(&po->pg_vec_lock);
3757 spin_lock(&po->bind_lock);
3760 register_prot_hook(sk);
3762 spin_unlock(&po->bind_lock);
3763 if (closing && (po->tp_version > TPACKET_V2)) {
3764 /* Because we don't support block-based V3 on tx-ring */
3766 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3771 free_pg_vec(pg_vec, order, req->tp_block_nr);
3776 static int packet_mmap(struct file *file, struct socket *sock,
3777 struct vm_area_struct *vma)
3779 struct sock *sk = sock->sk;
3780 struct packet_sock *po = pkt_sk(sk);
3781 unsigned long size, expected_size;
3782 struct packet_ring_buffer *rb;
3783 unsigned long start;
3790 mutex_lock(&po->pg_vec_lock);
3793 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3795 expected_size += rb->pg_vec_len
3801 if (expected_size == 0)
3804 size = vma->vm_end - vma->vm_start;
3805 if (size != expected_size)
3808 start = vma->vm_start;
3809 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3810 if (rb->pg_vec == NULL)
3813 for (i = 0; i < rb->pg_vec_len; i++) {
3815 void *kaddr = rb->pg_vec[i].buffer;
3818 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3819 page = pgv_to_page(kaddr);
3820 err = vm_insert_page(vma, start, page);
3829 atomic_inc(&po->mapped);
3830 vma->vm_ops = &packet_mmap_ops;
3834 mutex_unlock(&po->pg_vec_lock);
3838 static const struct proto_ops packet_ops_spkt = {
3839 .family = PF_PACKET,
3840 .owner = THIS_MODULE,
3841 .release = packet_release,
3842 .bind = packet_bind_spkt,
3843 .connect = sock_no_connect,
3844 .socketpair = sock_no_socketpair,
3845 .accept = sock_no_accept,
3846 .getname = packet_getname_spkt,
3847 .poll = datagram_poll,
3848 .ioctl = packet_ioctl,
3849 .listen = sock_no_listen,
3850 .shutdown = sock_no_shutdown,
3851 .setsockopt = sock_no_setsockopt,
3852 .getsockopt = sock_no_getsockopt,
3853 .sendmsg = packet_sendmsg_spkt,
3854 .recvmsg = packet_recvmsg,
3855 .mmap = sock_no_mmap,
3856 .sendpage = sock_no_sendpage,
3859 static const struct proto_ops packet_ops = {
3860 .family = PF_PACKET,
3861 .owner = THIS_MODULE,
3862 .release = packet_release,
3863 .bind = packet_bind,
3864 .connect = sock_no_connect,
3865 .socketpair = sock_no_socketpair,
3866 .accept = sock_no_accept,
3867 .getname = packet_getname,
3868 .poll = packet_poll,
3869 .ioctl = packet_ioctl,
3870 .listen = sock_no_listen,
3871 .shutdown = sock_no_shutdown,
3872 .setsockopt = packet_setsockopt,
3873 .getsockopt = packet_getsockopt,
3874 .sendmsg = packet_sendmsg,
3875 .recvmsg = packet_recvmsg,
3876 .mmap = packet_mmap,
3877 .sendpage = sock_no_sendpage,
3880 static const struct net_proto_family packet_family_ops = {
3881 .family = PF_PACKET,
3882 .create = packet_create,
3883 .owner = THIS_MODULE,
3886 static struct notifier_block packet_netdev_notifier = {
3887 .notifier_call = packet_notifier,
3890 #ifdef CONFIG_PROC_FS
3892 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3895 struct net *net = seq_file_net(seq);
3898 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3901 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3903 struct net *net = seq_file_net(seq);
3904 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3907 static void packet_seq_stop(struct seq_file *seq, void *v)
3913 static int packet_seq_show(struct seq_file *seq, void *v)
3915 if (v == SEQ_START_TOKEN)
3916 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3918 struct sock *s = sk_entry(v);
3919 const struct packet_sock *po = pkt_sk(s);
3922 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3924 atomic_read(&s->sk_refcnt),
3929 atomic_read(&s->sk_rmem_alloc),
3930 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3937 static const struct seq_operations packet_seq_ops = {
3938 .start = packet_seq_start,
3939 .next = packet_seq_next,
3940 .stop = packet_seq_stop,
3941 .show = packet_seq_show,
3944 static int packet_seq_open(struct inode *inode, struct file *file)
3946 return seq_open_net(inode, file, &packet_seq_ops,
3947 sizeof(struct seq_net_private));
3950 static const struct file_operations packet_seq_fops = {
3951 .owner = THIS_MODULE,
3952 .open = packet_seq_open,
3954 .llseek = seq_lseek,
3955 .release = seq_release_net,
3960 static int __net_init packet_net_init(struct net *net)
3962 mutex_init(&net->packet.sklist_lock);
3963 INIT_HLIST_HEAD(&net->packet.sklist);
3965 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3971 static void __net_exit packet_net_exit(struct net *net)
3973 remove_proc_entry("packet", net->proc_net);
3976 static struct pernet_operations packet_net_ops = {
3977 .init = packet_net_init,
3978 .exit = packet_net_exit,
3982 static void __exit packet_exit(void)
3984 unregister_netdevice_notifier(&packet_netdev_notifier);
3985 unregister_pernet_subsys(&packet_net_ops);
3986 sock_unregister(PF_PACKET);
3987 proto_unregister(&packet_proto);
3990 static int __init packet_init(void)
3992 int rc = proto_register(&packet_proto, 0);
3997 sock_register(&packet_family_ops);
3998 register_pernet_subsys(&packet_net_ops);
3999 register_netdevice_notifier(&packet_netdev_notifier);
4004 module_init(packet_init);
4005 module_exit(packet_exit);
4006 MODULE_LICENSE("GPL");
4007 MODULE_ALIAS_NETPROTO(PF_PACKET);