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>
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];
161 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
162 int closing, int tx_ring);
165 #define V3_ALIGNMENT (8)
167 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
169 #define BLK_PLUS_PRIV(sz_of_priv) \
170 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
172 #define PGV_FROM_VMALLOC 1
174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
183 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
184 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
185 struct packet_type *pt, struct net_device *orig_dev);
187 static void *packet_previous_frame(struct packet_sock *po,
188 struct packet_ring_buffer *rb,
190 static void packet_increment_head(struct packet_ring_buffer *buff);
191 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
192 struct tpacket_block_desc *);
193 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
194 struct packet_sock *);
195 static void prb_retire_current_block(struct tpacket_kbdq_core *,
196 struct packet_sock *, unsigned int status);
197 static int prb_queue_frozen(struct tpacket_kbdq_core *);
198 static void prb_open_block(struct tpacket_kbdq_core *,
199 struct tpacket_block_desc *);
200 static void prb_retire_rx_blk_timer_expired(unsigned long);
201 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
202 static void prb_init_blk_timer(struct packet_sock *,
203 struct tpacket_kbdq_core *,
204 void (*func) (unsigned long));
205 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
206 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
207 struct tpacket3_hdr *);
208 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
209 struct tpacket3_hdr *);
210 static void packet_flush_mclist(struct sock *sk);
212 struct packet_skb_cb {
213 unsigned int origlen;
215 struct sockaddr_pkt pkt;
216 struct sockaddr_ll ll;
220 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
222 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
223 #define GET_PBLOCK_DESC(x, bid) \
224 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
225 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
226 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
227 #define GET_NEXT_PRB_BLK_NUM(x) \
228 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
229 ((x)->kactive_blk_num+1) : 0)
231 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
232 static void __fanout_link(struct sock *sk, struct packet_sock *po);
234 /* register_prot_hook must be invoked with the po->bind_lock held,
235 * or from a context in which asynchronous accesses to the packet
236 * socket is not possible (packet_create()).
238 static void register_prot_hook(struct sock *sk)
240 struct packet_sock *po = pkt_sk(sk);
243 __fanout_link(sk, po);
245 dev_add_pack(&po->prot_hook);
251 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
252 * held. If the sync parameter is true, we will temporarily drop
253 * the po->bind_lock and do a synchronize_net to make sure no
254 * asynchronous packet processing paths still refer to the elements
255 * of po->prot_hook. If the sync parameter is false, it is the
256 * callers responsibility to take care of this.
258 static void __unregister_prot_hook(struct sock *sk, bool sync)
260 struct packet_sock *po = pkt_sk(sk);
264 __fanout_unlink(sk, po);
266 __dev_remove_pack(&po->prot_hook);
270 spin_unlock(&po->bind_lock);
272 spin_lock(&po->bind_lock);
276 static void unregister_prot_hook(struct sock *sk, bool sync)
278 struct packet_sock *po = pkt_sk(sk);
281 __unregister_prot_hook(sk, sync);
284 static inline __pure struct page *pgv_to_page(void *addr)
286 if (is_vmalloc_addr(addr))
287 return vmalloc_to_page(addr);
288 return virt_to_page(addr);
291 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
294 struct tpacket_hdr *h1;
295 struct tpacket2_hdr *h2;
300 switch (po->tp_version) {
302 h.h1->tp_status = status;
303 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
306 h.h2->tp_status = status;
307 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
311 WARN(1, "TPACKET version not supported.\n");
318 static int __packet_get_status(struct packet_sock *po, void *frame)
321 struct tpacket_hdr *h1;
322 struct tpacket2_hdr *h2;
329 switch (po->tp_version) {
331 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
332 return h.h1->tp_status;
334 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
335 return h.h2->tp_status;
338 WARN(1, "TPACKET version not supported.\n");
344 static void *packet_lookup_frame(struct packet_sock *po,
345 struct packet_ring_buffer *rb,
346 unsigned int position,
349 unsigned int pg_vec_pos, frame_offset;
351 struct tpacket_hdr *h1;
352 struct tpacket2_hdr *h2;
356 pg_vec_pos = position / rb->frames_per_block;
357 frame_offset = position % rb->frames_per_block;
359 h.raw = rb->pg_vec[pg_vec_pos].buffer +
360 (frame_offset * rb->frame_size);
362 if (status != __packet_get_status(po, h.raw))
368 static void *packet_current_frame(struct packet_sock *po,
369 struct packet_ring_buffer *rb,
372 return packet_lookup_frame(po, rb, rb->head, status);
375 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
377 del_timer_sync(&pkc->retire_blk_timer);
380 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
382 struct sk_buff_head *rb_queue)
384 struct tpacket_kbdq_core *pkc;
386 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
388 spin_lock(&rb_queue->lock);
389 pkc->delete_blk_timer = 1;
390 spin_unlock(&rb_queue->lock);
392 prb_del_retire_blk_timer(pkc);
395 static void prb_init_blk_timer(struct packet_sock *po,
396 struct tpacket_kbdq_core *pkc,
397 void (*func) (unsigned long))
399 init_timer(&pkc->retire_blk_timer);
400 pkc->retire_blk_timer.data = (long)po;
401 pkc->retire_blk_timer.function = func;
402 pkc->retire_blk_timer.expires = jiffies;
405 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
407 struct tpacket_kbdq_core *pkc;
412 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
413 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
416 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
417 int blk_size_in_bytes)
419 struct net_device *dev;
420 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
421 struct ethtool_cmd ecmd;
426 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
427 if (unlikely(!dev)) {
429 return DEFAULT_PRB_RETIRE_TOV;
431 err = __ethtool_get_settings(dev, &ecmd);
432 speed = ethtool_cmd_speed(&ecmd);
436 * If the link speed is so slow you don't really
437 * need to worry about perf anyways
439 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
440 return DEFAULT_PRB_RETIRE_TOV;
447 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
459 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
460 union tpacket_req_u *req_u)
462 p1->feature_req_word = req_u->req3.tp_feature_req_word;
465 static void init_prb_bdqc(struct packet_sock *po,
466 struct packet_ring_buffer *rb,
468 union tpacket_req_u *req_u, int tx_ring)
470 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
471 struct tpacket_block_desc *pbd;
473 memset(p1, 0x0, sizeof(*p1));
475 p1->knxt_seq_num = 1;
477 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
478 p1->pkblk_start = pg_vec[0].buffer;
479 p1->kblk_size = req_u->req3.tp_block_size;
480 p1->knum_blocks = req_u->req3.tp_block_nr;
481 p1->hdrlen = po->tp_hdrlen;
482 p1->version = po->tp_version;
483 p1->last_kactive_blk_num = 0;
484 po->stats_u.stats3.tp_freeze_q_cnt = 0;
485 if (req_u->req3.tp_retire_blk_tov)
486 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
488 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
489 req_u->req3.tp_block_size);
490 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
491 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
493 prb_init_ft_ops(p1, req_u);
494 prb_setup_retire_blk_timer(po, tx_ring);
495 prb_open_block(p1, pbd);
498 /* Do NOT update the last_blk_num first.
499 * Assumes sk_buff_head lock is held.
501 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
503 mod_timer(&pkc->retire_blk_timer,
504 jiffies + pkc->tov_in_jiffies);
505 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
510 * 1) We refresh the timer only when we open a block.
511 * By doing this we don't waste cycles refreshing the timer
512 * on packet-by-packet basis.
514 * With a 1MB block-size, on a 1Gbps line, it will take
515 * i) ~8 ms to fill a block + ii) memcpy etc.
516 * In this cut we are not accounting for the memcpy time.
518 * So, if the user sets the 'tmo' to 10ms then the timer
519 * will never fire while the block is still getting filled
520 * (which is what we want). However, the user could choose
521 * to close a block early and that's fine.
523 * But when the timer does fire, we check whether or not to refresh it.
524 * Since the tmo granularity is in msecs, it is not too expensive
525 * to refresh the timer, lets say every '8' msecs.
526 * Either the user can set the 'tmo' or we can derive it based on
527 * a) line-speed and b) block-size.
528 * prb_calc_retire_blk_tmo() calculates the tmo.
531 static void prb_retire_rx_blk_timer_expired(unsigned long data)
533 struct packet_sock *po = (struct packet_sock *)data;
534 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
536 struct tpacket_block_desc *pbd;
538 spin_lock(&po->sk.sk_receive_queue.lock);
540 frozen = prb_queue_frozen(pkc);
541 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
543 if (unlikely(pkc->delete_blk_timer))
546 /* We only need to plug the race when the block is partially filled.
548 * lock(); increment BLOCK_NUM_PKTS; unlock()
549 * copy_bits() is in progress ...
550 * timer fires on other cpu:
551 * we can't retire the current block because copy_bits
555 if (BLOCK_NUM_PKTS(pbd)) {
556 while (atomic_read(&pkc->blk_fill_in_prog)) {
557 /* Waiting for skb_copy_bits to finish... */
562 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
564 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
565 if (!prb_dispatch_next_block(pkc, po))
570 /* Case 1. Queue was frozen because user-space was
573 if (prb_curr_blk_in_use(pkc, pbd)) {
575 * Ok, user-space is still behind.
576 * So just refresh the timer.
580 /* Case 2. queue was frozen,user-space caught up,
581 * now the link went idle && the timer fired.
582 * We don't have a block to close.So we open this
583 * block and restart the timer.
584 * opening a block thaws the queue,restarts timer
585 * Thawing/timer-refresh is a side effect.
587 prb_open_block(pkc, pbd);
594 _prb_refresh_rx_retire_blk_timer(pkc);
597 spin_unlock(&po->sk.sk_receive_queue.lock);
600 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
601 struct tpacket_block_desc *pbd1, __u32 status)
603 /* Flush everything minus the block header */
605 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
610 /* Skip the block header(we know header WILL fit in 4K) */
613 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
614 for (; start < end; start += PAGE_SIZE)
615 flush_dcache_page(pgv_to_page(start));
620 /* Now update the block status. */
622 BLOCK_STATUS(pbd1) = status;
624 /* Flush the block header */
626 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
628 flush_dcache_page(pgv_to_page(start));
638 * 2) Increment active_blk_num
640 * Note:We DONT refresh the timer on purpose.
641 * Because almost always the next block will be opened.
643 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
644 struct tpacket_block_desc *pbd1,
645 struct packet_sock *po, unsigned int stat)
647 __u32 status = TP_STATUS_USER | stat;
649 struct tpacket3_hdr *last_pkt;
650 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
652 if (po->stats.tp_drops)
653 status |= TP_STATUS_LOSING;
655 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
656 last_pkt->tp_next_offset = 0;
658 /* Get the ts of the last pkt */
659 if (BLOCK_NUM_PKTS(pbd1)) {
660 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
661 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
663 /* Ok, we tmo'd - so get the current time */
666 h1->ts_last_pkt.ts_sec = ts.tv_sec;
667 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
672 /* Flush the block */
673 prb_flush_block(pkc1, pbd1, status);
675 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
678 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
680 pkc->reset_pending_on_curr_blk = 0;
684 * Side effect of opening a block:
686 * 1) prb_queue is thawed.
687 * 2) retire_blk_timer is refreshed.
690 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
691 struct tpacket_block_desc *pbd1)
694 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
698 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
700 /* We could have just memset this but we will lose the
701 * flexibility of making the priv area sticky
703 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
704 BLOCK_NUM_PKTS(pbd1) = 0;
705 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
707 h1->ts_first_pkt.ts_sec = ts.tv_sec;
708 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
709 pkc1->pkblk_start = (char *)pbd1;
710 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
711 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
712 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
713 pbd1->version = pkc1->version;
714 pkc1->prev = pkc1->nxt_offset;
715 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
716 prb_thaw_queue(pkc1);
717 _prb_refresh_rx_retire_blk_timer(pkc1);
724 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
725 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
731 * Queue freeze logic:
732 * 1) Assume tp_block_nr = 8 blocks.
733 * 2) At time 't0', user opens Rx ring.
734 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
735 * 4) user-space is either sleeping or processing block '0'.
736 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
737 * it will close block-7,loop around and try to fill block '0'.
739 * __packet_lookup_frame_in_block
740 * prb_retire_current_block()
741 * prb_dispatch_next_block()
742 * |->(BLOCK_STATUS == USER) evaluates to true
743 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
744 * 6) Now there are two cases:
745 * 6.1) Link goes idle right after the queue is frozen.
746 * But remember, the last open_block() refreshed the timer.
747 * When this timer expires,it will refresh itself so that we can
748 * re-open block-0 in near future.
749 * 6.2) Link is busy and keeps on receiving packets. This is a simple
750 * case and __packet_lookup_frame_in_block will check if block-0
751 * is free and can now be re-used.
753 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
754 struct packet_sock *po)
756 pkc->reset_pending_on_curr_blk = 1;
757 po->stats_u.stats3.tp_freeze_q_cnt++;
760 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
763 * If the next block is free then we will dispatch it
764 * and return a good offset.
765 * Else, we will freeze the queue.
766 * So, caller must check the return value.
768 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
769 struct packet_sock *po)
771 struct tpacket_block_desc *pbd;
775 /* 1. Get current block num */
776 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
778 /* 2. If this block is currently in_use then freeze the queue */
779 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
780 prb_freeze_queue(pkc, po);
786 * open this block and return the offset where the first packet
787 * needs to get stored.
789 prb_open_block(pkc, pbd);
790 return (void *)pkc->nxt_offset;
793 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
794 struct packet_sock *po, unsigned int status)
796 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
798 /* retire/close the current block */
799 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
801 * Plug the case where copy_bits() is in progress on
802 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
803 * have space to copy the pkt in the current block and
804 * called prb_retire_current_block()
806 * We don't need to worry about the TMO case because
807 * the timer-handler already handled this case.
809 if (!(status & TP_STATUS_BLK_TMO)) {
810 while (atomic_read(&pkc->blk_fill_in_prog)) {
811 /* Waiting for skb_copy_bits to finish... */
815 prb_close_block(pkc, pbd, po, status);
819 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
824 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
825 struct tpacket_block_desc *pbd)
827 return TP_STATUS_USER & BLOCK_STATUS(pbd);
830 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
832 return pkc->reset_pending_on_curr_blk;
835 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
837 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
838 atomic_dec(&pkc->blk_fill_in_prog);
841 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
842 struct tpacket3_hdr *ppd)
844 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
847 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
848 struct tpacket3_hdr *ppd)
850 ppd->hv1.tp_rxhash = 0;
853 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
854 struct tpacket3_hdr *ppd)
856 if (vlan_tx_tag_present(pkc->skb)) {
857 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
858 ppd->tp_status = TP_STATUS_VLAN_VALID;
860 ppd->hv1.tp_vlan_tci = 0;
861 ppd->tp_status = TP_STATUS_AVAILABLE;
865 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
866 struct tpacket3_hdr *ppd)
868 prb_fill_vlan_info(pkc, ppd);
870 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
871 prb_fill_rxhash(pkc, ppd);
873 prb_clear_rxhash(pkc, ppd);
876 static void prb_fill_curr_block(char *curr,
877 struct tpacket_kbdq_core *pkc,
878 struct tpacket_block_desc *pbd,
881 struct tpacket3_hdr *ppd;
883 ppd = (struct tpacket3_hdr *)curr;
884 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
886 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
887 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
888 BLOCK_NUM_PKTS(pbd) += 1;
889 atomic_inc(&pkc->blk_fill_in_prog);
890 prb_run_all_ft_ops(pkc, ppd);
893 /* Assumes caller has the sk->rx_queue.lock */
894 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
900 struct tpacket_kbdq_core *pkc;
901 struct tpacket_block_desc *pbd;
904 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
905 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
907 /* Queue is frozen when user space is lagging behind */
908 if (prb_queue_frozen(pkc)) {
910 * Check if that last block which caused the queue to freeze,
911 * is still in_use by user-space.
913 if (prb_curr_blk_in_use(pkc, pbd)) {
914 /* Can't record this packet */
918 * Ok, the block was released by user-space.
919 * Now let's open that block.
920 * opening a block also thaws the queue.
921 * Thawing is a side effect.
923 prb_open_block(pkc, pbd);
928 curr = pkc->nxt_offset;
930 end = (char *)pbd + pkc->kblk_size;
932 /* first try the current block */
933 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
934 prb_fill_curr_block(curr, pkc, pbd, len);
938 /* Ok, close the current block */
939 prb_retire_current_block(pkc, po, 0);
941 /* Now, try to dispatch the next block */
942 curr = (char *)prb_dispatch_next_block(pkc, po);
944 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 prb_fill_curr_block(curr, pkc, pbd, len);
950 * No free blocks are available.user_space hasn't caught up yet.
951 * Queue was just frozen and now this packet will get dropped.
956 static void *packet_current_rx_frame(struct packet_sock *po,
958 int status, unsigned int len)
961 switch (po->tp_version) {
964 curr = packet_lookup_frame(po, &po->rx_ring,
965 po->rx_ring.head, status);
968 return __packet_lookup_frame_in_block(po, skb, status, len);
970 WARN(1, "TPACKET version not supported\n");
976 static void *prb_lookup_block(struct packet_sock *po,
977 struct packet_ring_buffer *rb,
981 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
982 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
984 if (status != BLOCK_STATUS(pbd))
989 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
992 if (rb->prb_bdqc.kactive_blk_num)
993 prev = rb->prb_bdqc.kactive_blk_num-1;
995 prev = rb->prb_bdqc.knum_blocks-1;
999 /* Assumes caller has held the rx_queue.lock */
1000 static void *__prb_previous_block(struct packet_sock *po,
1001 struct packet_ring_buffer *rb,
1004 unsigned int previous = prb_previous_blk_num(rb);
1005 return prb_lookup_block(po, rb, previous, status);
1008 static void *packet_previous_rx_frame(struct packet_sock *po,
1009 struct packet_ring_buffer *rb,
1012 if (po->tp_version <= TPACKET_V2)
1013 return packet_previous_frame(po, rb, status);
1015 return __prb_previous_block(po, rb, status);
1018 static void packet_increment_rx_head(struct packet_sock *po,
1019 struct packet_ring_buffer *rb)
1021 switch (po->tp_version) {
1024 return packet_increment_head(rb);
1027 WARN(1, "TPACKET version not supported.\n");
1033 static void *packet_previous_frame(struct packet_sock *po,
1034 struct packet_ring_buffer *rb,
1037 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1038 return packet_lookup_frame(po, rb, previous, status);
1041 static void packet_increment_head(struct packet_ring_buffer *buff)
1043 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1046 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1048 struct sock *sk = &po->sk;
1051 if (po->prot_hook.func != tpacket_rcv)
1052 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1055 spin_lock(&sk->sk_receive_queue.lock);
1056 if (po->tp_version == TPACKET_V3)
1057 has_room = prb_lookup_block(po, &po->rx_ring,
1058 po->rx_ring.prb_bdqc.kactive_blk_num,
1061 has_room = packet_lookup_frame(po, &po->rx_ring,
1064 spin_unlock(&sk->sk_receive_queue.lock);
1069 static void packet_sock_destruct(struct sock *sk)
1071 skb_queue_purge(&sk->sk_error_queue);
1073 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1074 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1076 if (!sock_flag(sk, SOCK_DEAD)) {
1077 pr_err("Attempt to release alive packet socket: %p\n", sk);
1081 sk_refcnt_debug_dec(sk);
1084 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1086 int x = atomic_read(&f->rr_cur) + 1;
1094 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1095 struct sk_buff *skb,
1098 return (((u64)skb->rxhash) * num) >> 32;
1101 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1102 struct sk_buff *skb,
1107 cur = atomic_read(&f->rr_cur);
1108 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1109 fanout_rr_next(f, num))) != cur)
1114 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1115 struct sk_buff *skb,
1118 return smp_processor_id() % num;
1121 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1122 struct sk_buff *skb,
1123 unsigned int idx, unsigned int skip,
1128 i = j = min_t(int, f->next[idx], num - 1);
1130 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1142 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1144 return f->flags & (flag >> 8);
1147 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1148 struct packet_type *pt, struct net_device *orig_dev)
1150 struct packet_fanout *f = pt->af_packet_priv;
1151 unsigned int num = f->num_members;
1152 struct packet_sock *po;
1155 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1162 case PACKET_FANOUT_HASH:
1164 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1165 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1169 skb_get_rxhash(skb);
1170 idx = fanout_demux_hash(f, skb, num);
1172 case PACKET_FANOUT_LB:
1173 idx = fanout_demux_lb(f, skb, num);
1175 case PACKET_FANOUT_CPU:
1176 idx = fanout_demux_cpu(f, skb, num);
1178 case PACKET_FANOUT_ROLLOVER:
1179 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1183 po = pkt_sk(f->arr[idx]);
1184 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1185 unlikely(!packet_rcv_has_room(po, skb))) {
1186 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1187 po = pkt_sk(f->arr[idx]);
1190 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1193 DEFINE_MUTEX(fanout_mutex);
1194 EXPORT_SYMBOL_GPL(fanout_mutex);
1195 static LIST_HEAD(fanout_list);
1197 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1199 struct packet_fanout *f = po->fanout;
1201 spin_lock(&f->lock);
1202 f->arr[f->num_members] = sk;
1205 spin_unlock(&f->lock);
1208 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1210 struct packet_fanout *f = po->fanout;
1213 spin_lock(&f->lock);
1214 for (i = 0; i < f->num_members; i++) {
1215 if (f->arr[i] == sk)
1218 BUG_ON(i >= f->num_members);
1219 f->arr[i] = f->arr[f->num_members - 1];
1221 spin_unlock(&f->lock);
1224 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1226 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1232 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1234 struct packet_sock *po = pkt_sk(sk);
1235 struct packet_fanout *f, *match;
1236 u8 type = type_flags & 0xff;
1237 u8 flags = type_flags >> 8;
1241 case PACKET_FANOUT_ROLLOVER:
1242 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1244 case PACKET_FANOUT_HASH:
1245 case PACKET_FANOUT_LB:
1246 case PACKET_FANOUT_CPU:
1258 mutex_lock(&fanout_mutex);
1260 list_for_each_entry(f, &fanout_list, list) {
1262 read_pnet(&f->net) == sock_net(sk)) {
1268 if (match && match->flags != flags)
1272 match = kzalloc(sizeof(*match), GFP_KERNEL);
1275 write_pnet(&match->net, sock_net(sk));
1278 match->flags = flags;
1279 atomic_set(&match->rr_cur, 0);
1280 INIT_LIST_HEAD(&match->list);
1281 spin_lock_init(&match->lock);
1282 atomic_set(&match->sk_ref, 0);
1283 match->prot_hook.type = po->prot_hook.type;
1284 match->prot_hook.dev = po->prot_hook.dev;
1285 match->prot_hook.func = packet_rcv_fanout;
1286 match->prot_hook.af_packet_priv = match;
1287 match->prot_hook.id_match = match_fanout_group;
1288 dev_add_pack(&match->prot_hook);
1289 list_add(&match->list, &fanout_list);
1292 if (match->type == type &&
1293 match->prot_hook.type == po->prot_hook.type &&
1294 match->prot_hook.dev == po->prot_hook.dev) {
1296 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1297 __dev_remove_pack(&po->prot_hook);
1299 atomic_inc(&match->sk_ref);
1300 __fanout_link(sk, po);
1305 mutex_unlock(&fanout_mutex);
1309 static void fanout_release(struct sock *sk)
1311 struct packet_sock *po = pkt_sk(sk);
1312 struct packet_fanout *f;
1318 mutex_lock(&fanout_mutex);
1321 if (atomic_dec_and_test(&f->sk_ref)) {
1323 dev_remove_pack(&f->prot_hook);
1326 mutex_unlock(&fanout_mutex);
1329 static const struct proto_ops packet_ops;
1331 static const struct proto_ops packet_ops_spkt;
1333 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1334 struct packet_type *pt, struct net_device *orig_dev)
1337 struct sockaddr_pkt *spkt;
1340 * When we registered the protocol we saved the socket in the data
1341 * field for just this event.
1344 sk = pt->af_packet_priv;
1347 * Yank back the headers [hope the device set this
1348 * right or kerboom...]
1350 * Incoming packets have ll header pulled,
1353 * For outgoing ones skb->data == skb_mac_header(skb)
1354 * so that this procedure is noop.
1357 if (skb->pkt_type == PACKET_LOOPBACK)
1360 if (!net_eq(dev_net(dev), sock_net(sk)))
1363 skb = skb_share_check(skb, GFP_ATOMIC);
1367 /* drop any routing info */
1370 /* drop conntrack reference */
1373 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1375 skb_push(skb, skb->data - skb_mac_header(skb));
1378 * The SOCK_PACKET socket receives _all_ frames.
1381 spkt->spkt_family = dev->type;
1382 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1383 spkt->spkt_protocol = skb->protocol;
1386 * Charge the memory to the socket. This is done specifically
1387 * to prevent sockets using all the memory up.
1390 if (sock_queue_rcv_skb(sk, skb) == 0)
1401 * Output a raw packet to a device layer. This bypasses all the other
1402 * protocol layers and you must therefore supply it with a complete frame
1405 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1406 struct msghdr *msg, size_t len)
1408 struct sock *sk = sock->sk;
1409 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1410 struct sk_buff *skb = NULL;
1411 struct net_device *dev;
1417 * Get and verify the address.
1421 if (msg->msg_namelen < sizeof(struct sockaddr))
1423 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1424 proto = saddr->spkt_protocol;
1426 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1429 * Find the device first to size check it
1432 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1435 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1441 if (!(dev->flags & IFF_UP))
1445 * You may not queue a frame bigger than the mtu. This is the lowest level
1446 * raw protocol and you must do your own fragmentation at this level.
1449 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1450 if (!netif_supports_nofcs(dev)) {
1451 err = -EPROTONOSUPPORT;
1454 extra_len = 4; /* We're doing our own CRC */
1458 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1462 size_t reserved = LL_RESERVED_SPACE(dev);
1463 int tlen = dev->needed_tailroom;
1464 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1467 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1470 /* FIXME: Save some space for broken drivers that write a hard
1471 * header at transmission time by themselves. PPP is the notable
1472 * one here. This should really be fixed at the driver level.
1474 skb_reserve(skb, reserved);
1475 skb_reset_network_header(skb);
1477 /* Try to align data part correctly */
1482 skb_reset_network_header(skb);
1484 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1490 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1491 /* Earlier code assumed this would be a VLAN pkt,
1492 * double-check this now that we have the actual
1495 struct ethhdr *ehdr;
1496 skb_reset_mac_header(skb);
1497 ehdr = eth_hdr(skb);
1498 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1504 skb->protocol = proto;
1506 skb->priority = sk->sk_priority;
1507 skb->mark = sk->sk_mark;
1509 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1511 if (unlikely(extra_len == 4))
1514 skb_probe_transport_header(skb, 0);
1516 dev_queue_xmit(skb);
1527 static unsigned int run_filter(const struct sk_buff *skb,
1528 const struct sock *sk,
1531 struct sk_filter *filter;
1534 filter = rcu_dereference(sk->sk_filter);
1536 res = SK_RUN_FILTER(filter, skb);
1543 * This function makes lazy skb cloning in hope that most of packets
1544 * are discarded by BPF.
1546 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1547 * and skb->cb are mangled. It works because (and until) packets
1548 * falling here are owned by current CPU. Output packets are cloned
1549 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1550 * sequencially, so that if we return skb to original state on exit,
1551 * we will not harm anyone.
1554 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1555 struct packet_type *pt, struct net_device *orig_dev)
1558 struct sockaddr_ll *sll;
1559 struct packet_sock *po;
1560 u8 *skb_head = skb->data;
1561 int skb_len = skb->len;
1562 unsigned int snaplen, res;
1564 if (skb->pkt_type == PACKET_LOOPBACK)
1567 sk = pt->af_packet_priv;
1570 if (!net_eq(dev_net(dev), sock_net(sk)))
1575 if (dev->header_ops) {
1576 /* The device has an explicit notion of ll header,
1577 * exported to higher levels.
1579 * Otherwise, the device hides details of its frame
1580 * structure, so that corresponding packet head is
1581 * never delivered to user.
1583 if (sk->sk_type != SOCK_DGRAM)
1584 skb_push(skb, skb->data - skb_mac_header(skb));
1585 else if (skb->pkt_type == PACKET_OUTGOING) {
1586 /* Special case: outgoing packets have ll header at head */
1587 skb_pull(skb, skb_network_offset(skb));
1593 res = run_filter(skb, sk, snaplen);
1595 goto drop_n_restore;
1599 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1602 if (skb_shared(skb)) {
1603 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1607 if (skb_head != skb->data) {
1608 skb->data = skb_head;
1615 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1618 sll = &PACKET_SKB_CB(skb)->sa.ll;
1619 sll->sll_family = AF_PACKET;
1620 sll->sll_hatype = dev->type;
1621 sll->sll_protocol = skb->protocol;
1622 sll->sll_pkttype = skb->pkt_type;
1623 if (unlikely(po->origdev))
1624 sll->sll_ifindex = orig_dev->ifindex;
1626 sll->sll_ifindex = dev->ifindex;
1628 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1630 PACKET_SKB_CB(skb)->origlen = skb->len;
1632 if (pskb_trim(skb, snaplen))
1635 skb_set_owner_r(skb, sk);
1639 /* drop conntrack reference */
1642 spin_lock(&sk->sk_receive_queue.lock);
1643 po->stats.tp_packets++;
1644 skb->dropcount = atomic_read(&sk->sk_drops);
1645 __skb_queue_tail(&sk->sk_receive_queue, skb);
1646 spin_unlock(&sk->sk_receive_queue.lock);
1647 sk->sk_data_ready(sk, skb->len);
1651 spin_lock(&sk->sk_receive_queue.lock);
1652 po->stats.tp_drops++;
1653 atomic_inc(&sk->sk_drops);
1654 spin_unlock(&sk->sk_receive_queue.lock);
1657 if (skb_head != skb->data && skb_shared(skb)) {
1658 skb->data = skb_head;
1666 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1667 struct packet_type *pt, struct net_device *orig_dev)
1670 struct packet_sock *po;
1671 struct sockaddr_ll *sll;
1673 struct tpacket_hdr *h1;
1674 struct tpacket2_hdr *h2;
1675 struct tpacket3_hdr *h3;
1678 u8 *skb_head = skb->data;
1679 int skb_len = skb->len;
1680 unsigned int snaplen, res;
1681 unsigned long status = TP_STATUS_USER;
1682 unsigned short macoff, netoff, hdrlen;
1683 struct sk_buff *copy_skb = NULL;
1686 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1688 if (skb->pkt_type == PACKET_LOOPBACK)
1691 sk = pt->af_packet_priv;
1694 if (!net_eq(dev_net(dev), sock_net(sk)))
1697 if (dev->header_ops) {
1698 if (sk->sk_type != SOCK_DGRAM)
1699 skb_push(skb, skb->data - skb_mac_header(skb));
1700 else if (skb->pkt_type == PACKET_OUTGOING) {
1701 /* Special case: outgoing packets have ll header at head */
1702 skb_pull(skb, skb_network_offset(skb));
1706 if (skb->ip_summed == CHECKSUM_PARTIAL)
1707 status |= TP_STATUS_CSUMNOTREADY;
1711 res = run_filter(skb, sk, snaplen);
1713 goto drop_n_restore;
1717 if (sk->sk_type == SOCK_DGRAM) {
1718 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1721 unsigned int maclen = skb_network_offset(skb);
1722 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1723 (maclen < 16 ? 16 : maclen)) +
1725 macoff = netoff - maclen;
1727 if (po->tp_version <= TPACKET_V2) {
1728 if (macoff + snaplen > po->rx_ring.frame_size) {
1729 if (po->copy_thresh &&
1730 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1731 if (skb_shared(skb)) {
1732 copy_skb = skb_clone(skb, GFP_ATOMIC);
1734 copy_skb = skb_get(skb);
1735 skb_head = skb->data;
1738 skb_set_owner_r(copy_skb, sk);
1740 snaplen = po->rx_ring.frame_size - macoff;
1741 if ((int)snaplen < 0)
1745 spin_lock(&sk->sk_receive_queue.lock);
1746 h.raw = packet_current_rx_frame(po, skb,
1747 TP_STATUS_KERNEL, (macoff+snaplen));
1750 if (po->tp_version <= TPACKET_V2) {
1751 packet_increment_rx_head(po, &po->rx_ring);
1753 * LOSING will be reported till you read the stats,
1754 * because it's COR - Clear On Read.
1755 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1758 if (po->stats.tp_drops)
1759 status |= TP_STATUS_LOSING;
1761 po->stats.tp_packets++;
1763 status |= TP_STATUS_COPY;
1764 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1766 spin_unlock(&sk->sk_receive_queue.lock);
1768 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1770 switch (po->tp_version) {
1772 h.h1->tp_len = skb->len;
1773 h.h1->tp_snaplen = snaplen;
1774 h.h1->tp_mac = macoff;
1775 h.h1->tp_net = netoff;
1776 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1777 && shhwtstamps->syststamp.tv64)
1778 tv = ktime_to_timeval(shhwtstamps->syststamp);
1779 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1780 && shhwtstamps->hwtstamp.tv64)
1781 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1782 else if (skb->tstamp.tv64)
1783 tv = ktime_to_timeval(skb->tstamp);
1785 do_gettimeofday(&tv);
1786 h.h1->tp_sec = tv.tv_sec;
1787 h.h1->tp_usec = tv.tv_usec;
1788 hdrlen = sizeof(*h.h1);
1791 h.h2->tp_len = skb->len;
1792 h.h2->tp_snaplen = snaplen;
1793 h.h2->tp_mac = macoff;
1794 h.h2->tp_net = netoff;
1795 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1796 && shhwtstamps->syststamp.tv64)
1797 ts = ktime_to_timespec(shhwtstamps->syststamp);
1798 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1799 && shhwtstamps->hwtstamp.tv64)
1800 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1801 else if (skb->tstamp.tv64)
1802 ts = ktime_to_timespec(skb->tstamp);
1804 getnstimeofday(&ts);
1805 h.h2->tp_sec = ts.tv_sec;
1806 h.h2->tp_nsec = ts.tv_nsec;
1807 if (vlan_tx_tag_present(skb)) {
1808 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1809 status |= TP_STATUS_VLAN_VALID;
1811 h.h2->tp_vlan_tci = 0;
1813 h.h2->tp_padding = 0;
1814 hdrlen = sizeof(*h.h2);
1817 /* tp_nxt_offset,vlan are already populated above.
1818 * So DONT clear those fields here
1820 h.h3->tp_status |= status;
1821 h.h3->tp_len = skb->len;
1822 h.h3->tp_snaplen = snaplen;
1823 h.h3->tp_mac = macoff;
1824 h.h3->tp_net = netoff;
1825 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1826 && shhwtstamps->syststamp.tv64)
1827 ts = ktime_to_timespec(shhwtstamps->syststamp);
1828 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1829 && shhwtstamps->hwtstamp.tv64)
1830 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1831 else if (skb->tstamp.tv64)
1832 ts = ktime_to_timespec(skb->tstamp);
1834 getnstimeofday(&ts);
1835 h.h3->tp_sec = ts.tv_sec;
1836 h.h3->tp_nsec = ts.tv_nsec;
1837 hdrlen = sizeof(*h.h3);
1843 sll = h.raw + TPACKET_ALIGN(hdrlen);
1844 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1845 sll->sll_family = AF_PACKET;
1846 sll->sll_hatype = dev->type;
1847 sll->sll_protocol = skb->protocol;
1848 sll->sll_pkttype = skb->pkt_type;
1849 if (unlikely(po->origdev))
1850 sll->sll_ifindex = orig_dev->ifindex;
1852 sll->sll_ifindex = dev->ifindex;
1855 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1859 if (po->tp_version <= TPACKET_V2) {
1860 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1861 + macoff + snaplen);
1862 for (start = h.raw; start < end; start += PAGE_SIZE)
1863 flush_dcache_page(pgv_to_page(start));
1868 if (po->tp_version <= TPACKET_V2)
1869 __packet_set_status(po, h.raw, status);
1871 prb_clear_blk_fill_status(&po->rx_ring);
1873 sk->sk_data_ready(sk, 0);
1876 if (skb_head != skb->data && skb_shared(skb)) {
1877 skb->data = skb_head;
1885 po->stats.tp_drops++;
1886 spin_unlock(&sk->sk_receive_queue.lock);
1888 sk->sk_data_ready(sk, 0);
1889 kfree_skb(copy_skb);
1890 goto drop_n_restore;
1893 static void tpacket_destruct_skb(struct sk_buff *skb)
1895 struct packet_sock *po = pkt_sk(skb->sk);
1898 if (likely(po->tx_ring.pg_vec)) {
1899 ph = skb_shinfo(skb)->destructor_arg;
1900 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1901 atomic_dec(&po->tx_ring.pending);
1902 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1908 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1909 void *frame, struct net_device *dev, int size_max,
1910 __be16 proto, unsigned char *addr, int hlen)
1913 struct tpacket_hdr *h1;
1914 struct tpacket2_hdr *h2;
1917 int to_write, offset, len, tp_len, nr_frags, len_max;
1918 struct socket *sock = po->sk.sk_socket;
1925 skb->protocol = proto;
1927 skb->priority = po->sk.sk_priority;
1928 skb->mark = po->sk.sk_mark;
1929 skb_shinfo(skb)->destructor_arg = ph.raw;
1931 switch (po->tp_version) {
1933 tp_len = ph.h2->tp_len;
1936 tp_len = ph.h1->tp_len;
1939 if (unlikely(tp_len > size_max)) {
1940 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1944 skb_reserve(skb, hlen);
1945 skb_reset_network_header(skb);
1946 skb_probe_transport_header(skb, 0);
1948 if (po->tp_tx_has_off) {
1949 int off_min, off_max, off;
1950 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1951 off_max = po->tx_ring.frame_size - tp_len;
1952 if (sock->type == SOCK_DGRAM) {
1953 switch (po->tp_version) {
1955 off = ph.h2->tp_net;
1958 off = ph.h1->tp_net;
1962 switch (po->tp_version) {
1964 off = ph.h2->tp_mac;
1967 off = ph.h1->tp_mac;
1971 if (unlikely((off < off_min) || (off_max < off)))
1973 data = ph.raw + off;
1975 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1979 if (sock->type == SOCK_DGRAM) {
1980 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1982 if (unlikely(err < 0))
1984 } else if (dev->hard_header_len) {
1985 /* net device doesn't like empty head */
1986 if (unlikely(tp_len <= dev->hard_header_len)) {
1987 pr_err("packet size is too short (%d < %d)\n",
1988 tp_len, dev->hard_header_len);
1992 skb_push(skb, dev->hard_header_len);
1993 err = skb_store_bits(skb, 0, data,
1994 dev->hard_header_len);
1998 data += dev->hard_header_len;
1999 to_write -= dev->hard_header_len;
2002 offset = offset_in_page(data);
2003 len_max = PAGE_SIZE - offset;
2004 len = ((to_write > len_max) ? len_max : to_write);
2006 skb->data_len = to_write;
2007 skb->len += to_write;
2008 skb->truesize += to_write;
2009 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2011 while (likely(to_write)) {
2012 nr_frags = skb_shinfo(skb)->nr_frags;
2014 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2015 pr_err("Packet exceed the number of skb frags(%lu)\n",
2020 page = pgv_to_page(data);
2022 flush_dcache_page(page);
2024 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2027 len_max = PAGE_SIZE;
2028 len = ((to_write > len_max) ? len_max : to_write);
2034 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2036 struct sk_buff *skb;
2037 struct net_device *dev;
2039 bool need_rls_dev = false;
2040 int err, reserve = 0;
2042 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2043 int tp_len, size_max;
2044 unsigned char *addr;
2046 int status = TP_STATUS_AVAILABLE;
2049 mutex_lock(&po->pg_vec_lock);
2051 if (saddr == NULL) {
2052 dev = po->prot_hook.dev;
2057 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2059 if (msg->msg_namelen < (saddr->sll_halen
2060 + offsetof(struct sockaddr_ll,
2063 proto = saddr->sll_protocol;
2064 addr = saddr->sll_addr;
2065 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2066 need_rls_dev = true;
2070 if (unlikely(dev == NULL))
2073 reserve = dev->hard_header_len;
2076 if (unlikely(!(dev->flags & IFF_UP)))
2079 size_max = po->tx_ring.frame_size
2080 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2082 if (size_max > dev->mtu + reserve)
2083 size_max = dev->mtu + reserve;
2086 ph = packet_current_frame(po, &po->tx_ring,
2087 TP_STATUS_SEND_REQUEST);
2089 if (unlikely(ph == NULL)) {
2094 status = TP_STATUS_SEND_REQUEST;
2095 hlen = LL_RESERVED_SPACE(dev);
2096 tlen = dev->needed_tailroom;
2097 skb = sock_alloc_send_skb(&po->sk,
2098 hlen + tlen + sizeof(struct sockaddr_ll),
2101 if (unlikely(skb == NULL))
2104 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2107 if (unlikely(tp_len < 0)) {
2109 __packet_set_status(po, ph,
2110 TP_STATUS_AVAILABLE);
2111 packet_increment_head(&po->tx_ring);
2115 status = TP_STATUS_WRONG_FORMAT;
2121 skb->destructor = tpacket_destruct_skb;
2122 __packet_set_status(po, ph, TP_STATUS_SENDING);
2123 atomic_inc(&po->tx_ring.pending);
2125 status = TP_STATUS_SEND_REQUEST;
2126 err = dev_queue_xmit(skb);
2127 if (unlikely(err > 0)) {
2128 err = net_xmit_errno(err);
2129 if (err && __packet_get_status(po, ph) ==
2130 TP_STATUS_AVAILABLE) {
2131 /* skb was destructed already */
2136 * skb was dropped but not destructed yet;
2137 * let's treat it like congestion or err < 0
2141 packet_increment_head(&po->tx_ring);
2143 } while (likely((ph != NULL) ||
2144 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2145 (atomic_read(&po->tx_ring.pending))))
2152 __packet_set_status(po, ph, status);
2158 mutex_unlock(&po->pg_vec_lock);
2162 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2163 size_t reserve, size_t len,
2164 size_t linear, int noblock,
2167 struct sk_buff *skb;
2169 /* Under a page? Don't bother with paged skb. */
2170 if (prepad + len < PAGE_SIZE || !linear)
2173 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2178 skb_reserve(skb, reserve);
2179 skb_put(skb, linear);
2180 skb->data_len = len - linear;
2181 skb->len += len - linear;
2186 static int packet_snd(struct socket *sock,
2187 struct msghdr *msg, size_t len)
2189 struct sock *sk = sock->sk;
2190 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2191 struct sk_buff *skb;
2192 struct net_device *dev;
2194 bool need_rls_dev = false;
2195 unsigned char *addr;
2196 int err, reserve = 0;
2197 struct virtio_net_hdr vnet_hdr = { 0 };
2200 struct packet_sock *po = pkt_sk(sk);
2201 unsigned short gso_type = 0;
2206 * Get and verify the address.
2209 if (saddr == NULL) {
2210 dev = po->prot_hook.dev;
2215 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2217 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2219 proto = saddr->sll_protocol;
2220 addr = saddr->sll_addr;
2221 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2222 need_rls_dev = true;
2228 if (sock->type == SOCK_RAW)
2229 reserve = dev->hard_header_len;
2232 if (!(dev->flags & IFF_UP))
2235 if (po->has_vnet_hdr) {
2236 vnet_hdr_len = sizeof(vnet_hdr);
2239 if (len < vnet_hdr_len)
2242 len -= vnet_hdr_len;
2244 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2249 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2250 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2252 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2253 vnet_hdr.csum_offset + 2;
2256 if (vnet_hdr.hdr_len > len)
2259 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2260 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2261 case VIRTIO_NET_HDR_GSO_TCPV4:
2262 gso_type = SKB_GSO_TCPV4;
2264 case VIRTIO_NET_HDR_GSO_TCPV6:
2265 gso_type = SKB_GSO_TCPV6;
2267 case VIRTIO_NET_HDR_GSO_UDP:
2268 gso_type = SKB_GSO_UDP;
2274 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2275 gso_type |= SKB_GSO_TCP_ECN;
2277 if (vnet_hdr.gso_size == 0)
2283 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2284 if (!netif_supports_nofcs(dev)) {
2285 err = -EPROTONOSUPPORT;
2288 extra_len = 4; /* We're doing our own CRC */
2292 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2296 hlen = LL_RESERVED_SPACE(dev);
2297 tlen = dev->needed_tailroom;
2298 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2299 msg->msg_flags & MSG_DONTWAIT, &err);
2303 skb_set_network_header(skb, reserve);
2306 if (sock->type == SOCK_DGRAM &&
2307 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2310 /* Returns -EFAULT on error */
2311 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2315 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2317 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2318 /* Earlier code assumed this would be a VLAN pkt,
2319 * double-check this now that we have the actual
2322 struct ethhdr *ehdr;
2323 skb_reset_mac_header(skb);
2324 ehdr = eth_hdr(skb);
2325 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2331 skb->protocol = proto;
2333 skb->priority = sk->sk_priority;
2334 skb->mark = sk->sk_mark;
2336 if (po->has_vnet_hdr) {
2337 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2338 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2339 vnet_hdr.csum_offset)) {
2345 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2346 skb_shinfo(skb)->gso_type = gso_type;
2348 /* Header must be checked, and gso_segs computed. */
2349 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2350 skb_shinfo(skb)->gso_segs = 0;
2352 len += vnet_hdr_len;
2355 skb_probe_transport_header(skb, reserve);
2357 if (unlikely(extra_len == 4))
2364 err = dev_queue_xmit(skb);
2365 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2376 if (dev && need_rls_dev)
2382 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2383 struct msghdr *msg, size_t len)
2385 struct sock *sk = sock->sk;
2386 struct packet_sock *po = pkt_sk(sk);
2387 if (po->tx_ring.pg_vec)
2388 return tpacket_snd(po, msg);
2390 return packet_snd(sock, msg, len);
2394 * Close a PACKET socket. This is fairly simple. We immediately go
2395 * to 'closed' state and remove our protocol entry in the device list.
2398 static int packet_release(struct socket *sock)
2400 struct sock *sk = sock->sk;
2401 struct packet_sock *po;
2403 union tpacket_req_u req_u;
2411 mutex_lock(&net->packet.sklist_lock);
2412 sk_del_node_init_rcu(sk);
2413 mutex_unlock(&net->packet.sklist_lock);
2416 sock_prot_inuse_add(net, sk->sk_prot, -1);
2419 spin_lock(&po->bind_lock);
2420 unregister_prot_hook(sk, false);
2421 if (po->prot_hook.dev) {
2422 dev_put(po->prot_hook.dev);
2423 po->prot_hook.dev = NULL;
2425 spin_unlock(&po->bind_lock);
2427 packet_flush_mclist(sk);
2429 if (po->rx_ring.pg_vec) {
2430 memset(&req_u, 0, sizeof(req_u));
2431 packet_set_ring(sk, &req_u, 1, 0);
2434 if (po->tx_ring.pg_vec) {
2435 memset(&req_u, 0, sizeof(req_u));
2436 packet_set_ring(sk, &req_u, 1, 1);
2443 * Now the socket is dead. No more input will appear.
2450 skb_queue_purge(&sk->sk_receive_queue);
2451 sk_refcnt_debug_release(sk);
2458 * Attach a packet hook.
2461 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2463 struct packet_sock *po = pkt_sk(sk);
2474 spin_lock(&po->bind_lock);
2475 unregister_prot_hook(sk, true);
2477 po->prot_hook.type = protocol;
2478 if (po->prot_hook.dev)
2479 dev_put(po->prot_hook.dev);
2480 po->prot_hook.dev = dev;
2482 po->ifindex = dev ? dev->ifindex : 0;
2487 if (!dev || (dev->flags & IFF_UP)) {
2488 register_prot_hook(sk);
2490 sk->sk_err = ENETDOWN;
2491 if (!sock_flag(sk, SOCK_DEAD))
2492 sk->sk_error_report(sk);
2496 spin_unlock(&po->bind_lock);
2502 * Bind a packet socket to a device
2505 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2508 struct sock *sk = sock->sk;
2510 struct net_device *dev;
2517 if (addr_len != sizeof(struct sockaddr))
2519 strlcpy(name, uaddr->sa_data, sizeof(name));
2521 dev = dev_get_by_name(sock_net(sk), name);
2523 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2527 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2529 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2530 struct sock *sk = sock->sk;
2531 struct net_device *dev = NULL;
2539 if (addr_len < sizeof(struct sockaddr_ll))
2541 if (sll->sll_family != AF_PACKET)
2544 if (sll->sll_ifindex) {
2546 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2550 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2556 static struct proto packet_proto = {
2558 .owner = THIS_MODULE,
2559 .obj_size = sizeof(struct packet_sock),
2563 * Create a packet of type SOCK_PACKET.
2566 static int packet_create(struct net *net, struct socket *sock, int protocol,
2570 struct packet_sock *po;
2571 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2574 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2576 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2577 sock->type != SOCK_PACKET)
2578 return -ESOCKTNOSUPPORT;
2580 sock->state = SS_UNCONNECTED;
2583 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2587 sock->ops = &packet_ops;
2588 if (sock->type == SOCK_PACKET)
2589 sock->ops = &packet_ops_spkt;
2591 sock_init_data(sock, sk);
2594 sk->sk_family = PF_PACKET;
2597 sk->sk_destruct = packet_sock_destruct;
2598 sk_refcnt_debug_inc(sk);
2601 * Attach a protocol block
2604 spin_lock_init(&po->bind_lock);
2605 mutex_init(&po->pg_vec_lock);
2606 po->prot_hook.func = packet_rcv;
2608 if (sock->type == SOCK_PACKET)
2609 po->prot_hook.func = packet_rcv_spkt;
2611 po->prot_hook.af_packet_priv = sk;
2614 po->prot_hook.type = proto;
2615 register_prot_hook(sk);
2618 mutex_lock(&net->packet.sklist_lock);
2619 sk_add_node_rcu(sk, &net->packet.sklist);
2620 mutex_unlock(&net->packet.sklist_lock);
2623 sock_prot_inuse_add(net, &packet_proto, 1);
2631 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2633 struct sock_exterr_skb *serr;
2634 struct sk_buff *skb, *skb2;
2638 skb = skb_dequeue(&sk->sk_error_queue);
2644 msg->msg_flags |= MSG_TRUNC;
2647 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2651 sock_recv_timestamp(msg, sk, skb);
2653 serr = SKB_EXT_ERR(skb);
2654 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2655 sizeof(serr->ee), &serr->ee);
2657 msg->msg_flags |= MSG_ERRQUEUE;
2660 /* Reset and regenerate socket error */
2661 spin_lock_bh(&sk->sk_error_queue.lock);
2663 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2664 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2665 spin_unlock_bh(&sk->sk_error_queue.lock);
2666 sk->sk_error_report(sk);
2668 spin_unlock_bh(&sk->sk_error_queue.lock);
2677 * Pull a packet from our receive queue and hand it to the user.
2678 * If necessary we block.
2681 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2682 struct msghdr *msg, size_t len, int flags)
2684 struct sock *sk = sock->sk;
2685 struct sk_buff *skb;
2687 struct sockaddr_ll *sll;
2688 int vnet_hdr_len = 0;
2691 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2695 /* What error should we return now? EUNATTACH? */
2696 if (pkt_sk(sk)->ifindex < 0)
2700 if (flags & MSG_ERRQUEUE) {
2701 err = packet_recv_error(sk, msg, len);
2706 * Call the generic datagram receiver. This handles all sorts
2707 * of horrible races and re-entrancy so we can forget about it
2708 * in the protocol layers.
2710 * Now it will return ENETDOWN, if device have just gone down,
2711 * but then it will block.
2714 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2717 * An error occurred so return it. Because skb_recv_datagram()
2718 * handles the blocking we don't see and worry about blocking
2725 if (pkt_sk(sk)->has_vnet_hdr) {
2726 struct virtio_net_hdr vnet_hdr = { 0 };
2729 vnet_hdr_len = sizeof(vnet_hdr);
2730 if (len < vnet_hdr_len)
2733 len -= vnet_hdr_len;
2735 if (skb_is_gso(skb)) {
2736 struct skb_shared_info *sinfo = skb_shinfo(skb);
2738 /* This is a hint as to how much should be linear. */
2739 vnet_hdr.hdr_len = skb_headlen(skb);
2740 vnet_hdr.gso_size = sinfo->gso_size;
2741 if (sinfo->gso_type & SKB_GSO_TCPV4)
2742 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2743 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2744 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2745 else if (sinfo->gso_type & SKB_GSO_UDP)
2746 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2747 else if (sinfo->gso_type & SKB_GSO_FCOE)
2751 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2752 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2754 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2756 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2757 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2758 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2759 vnet_hdr.csum_offset = skb->csum_offset;
2760 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2761 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2762 } /* else everything is zero */
2764 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2771 * If the address length field is there to be filled in, we fill
2775 sll = &PACKET_SKB_CB(skb)->sa.ll;
2776 if (sock->type == SOCK_PACKET)
2777 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2779 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2782 * You lose any data beyond the buffer you gave. If it worries a
2783 * user program they can ask the device for its MTU anyway.
2789 msg->msg_flags |= MSG_TRUNC;
2792 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2796 sock_recv_ts_and_drops(msg, sk, skb);
2799 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2802 if (pkt_sk(sk)->auxdata) {
2803 struct tpacket_auxdata aux;
2805 aux.tp_status = TP_STATUS_USER;
2806 if (skb->ip_summed == CHECKSUM_PARTIAL)
2807 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2808 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2809 aux.tp_snaplen = skb->len;
2811 aux.tp_net = skb_network_offset(skb);
2812 if (vlan_tx_tag_present(skb)) {
2813 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2814 aux.tp_status |= TP_STATUS_VLAN_VALID;
2816 aux.tp_vlan_tci = 0;
2819 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2823 * Free or return the buffer as appropriate. Again this
2824 * hides all the races and re-entrancy issues from us.
2826 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2829 skb_free_datagram(sk, skb);
2834 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2835 int *uaddr_len, int peer)
2837 struct net_device *dev;
2838 struct sock *sk = sock->sk;
2843 uaddr->sa_family = AF_PACKET;
2845 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2847 strncpy(uaddr->sa_data, dev->name, 14);
2849 memset(uaddr->sa_data, 0, 14);
2851 *uaddr_len = sizeof(*uaddr);
2856 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2857 int *uaddr_len, int peer)
2859 struct net_device *dev;
2860 struct sock *sk = sock->sk;
2861 struct packet_sock *po = pkt_sk(sk);
2862 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2867 sll->sll_family = AF_PACKET;
2868 sll->sll_ifindex = po->ifindex;
2869 sll->sll_protocol = po->num;
2870 sll->sll_pkttype = 0;
2872 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2874 sll->sll_hatype = dev->type;
2875 sll->sll_halen = dev->addr_len;
2876 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2878 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2882 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2887 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2891 case PACKET_MR_MULTICAST:
2892 if (i->alen != dev->addr_len)
2895 return dev_mc_add(dev, i->addr);
2897 return dev_mc_del(dev, i->addr);
2899 case PACKET_MR_PROMISC:
2900 return dev_set_promiscuity(dev, what);
2902 case PACKET_MR_ALLMULTI:
2903 return dev_set_allmulti(dev, what);
2905 case PACKET_MR_UNICAST:
2906 if (i->alen != dev->addr_len)
2909 return dev_uc_add(dev, i->addr);
2911 return dev_uc_del(dev, i->addr);
2919 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2921 for ( ; i; i = i->next) {
2922 if (i->ifindex == dev->ifindex)
2923 packet_dev_mc(dev, i, what);
2927 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2929 struct packet_sock *po = pkt_sk(sk);
2930 struct packet_mclist *ml, *i;
2931 struct net_device *dev;
2937 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2942 if (mreq->mr_alen > dev->addr_len)
2946 i = kmalloc(sizeof(*i), GFP_KERNEL);
2951 for (ml = po->mclist; ml; ml = ml->next) {
2952 if (ml->ifindex == mreq->mr_ifindex &&
2953 ml->type == mreq->mr_type &&
2954 ml->alen == mreq->mr_alen &&
2955 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2957 /* Free the new element ... */
2963 i->type = mreq->mr_type;
2964 i->ifindex = mreq->mr_ifindex;
2965 i->alen = mreq->mr_alen;
2966 memcpy(i->addr, mreq->mr_address, i->alen);
2968 i->next = po->mclist;
2970 err = packet_dev_mc(dev, i, 1);
2972 po->mclist = i->next;
2981 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2983 struct packet_mclist *ml, **mlp;
2987 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2988 if (ml->ifindex == mreq->mr_ifindex &&
2989 ml->type == mreq->mr_type &&
2990 ml->alen == mreq->mr_alen &&
2991 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2992 if (--ml->count == 0) {
2993 struct net_device *dev;
2995 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2997 packet_dev_mc(dev, ml, -1);
3005 return -EADDRNOTAVAIL;
3008 static void packet_flush_mclist(struct sock *sk)
3010 struct packet_sock *po = pkt_sk(sk);
3011 struct packet_mclist *ml;
3017 while ((ml = po->mclist) != NULL) {
3018 struct net_device *dev;
3020 po->mclist = ml->next;
3021 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3023 packet_dev_mc(dev, ml, -1);
3030 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3032 struct sock *sk = sock->sk;
3033 struct packet_sock *po = pkt_sk(sk);
3036 if (level != SOL_PACKET)
3037 return -ENOPROTOOPT;
3040 case PACKET_ADD_MEMBERSHIP:
3041 case PACKET_DROP_MEMBERSHIP:
3043 struct packet_mreq_max mreq;
3045 memset(&mreq, 0, sizeof(mreq));
3046 if (len < sizeof(struct packet_mreq))
3048 if (len > sizeof(mreq))
3050 if (copy_from_user(&mreq, optval, len))
3052 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3054 if (optname == PACKET_ADD_MEMBERSHIP)
3055 ret = packet_mc_add(sk, &mreq);
3057 ret = packet_mc_drop(sk, &mreq);
3061 case PACKET_RX_RING:
3062 case PACKET_TX_RING:
3064 union tpacket_req_u req_u;
3067 switch (po->tp_version) {
3070 len = sizeof(req_u.req);
3074 len = sizeof(req_u.req3);
3079 if (pkt_sk(sk)->has_vnet_hdr)
3081 if (copy_from_user(&req_u.req, optval, len))
3083 return packet_set_ring(sk, &req_u, 0,
3084 optname == PACKET_TX_RING);
3086 case PACKET_COPY_THRESH:
3090 if (optlen != sizeof(val))
3092 if (copy_from_user(&val, optval, sizeof(val)))
3095 pkt_sk(sk)->copy_thresh = val;
3098 case PACKET_VERSION:
3102 if (optlen != sizeof(val))
3104 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3106 if (copy_from_user(&val, optval, sizeof(val)))
3112 po->tp_version = val;
3118 case PACKET_RESERVE:
3122 if (optlen != sizeof(val))
3124 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3126 if (copy_from_user(&val, optval, sizeof(val)))
3128 po->tp_reserve = val;
3135 if (optlen != sizeof(val))
3137 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3139 if (copy_from_user(&val, optval, sizeof(val)))
3141 po->tp_loss = !!val;
3144 case PACKET_AUXDATA:
3148 if (optlen < sizeof(val))
3150 if (copy_from_user(&val, optval, sizeof(val)))
3153 po->auxdata = !!val;
3156 case PACKET_ORIGDEV:
3160 if (optlen < sizeof(val))
3162 if (copy_from_user(&val, optval, sizeof(val)))
3165 po->origdev = !!val;
3168 case PACKET_VNET_HDR:
3172 if (sock->type != SOCK_RAW)
3174 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3176 if (optlen < sizeof(val))
3178 if (copy_from_user(&val, optval, sizeof(val)))
3181 po->has_vnet_hdr = !!val;
3184 case PACKET_TIMESTAMP:
3188 if (optlen != sizeof(val))
3190 if (copy_from_user(&val, optval, sizeof(val)))
3193 po->tp_tstamp = val;
3200 if (optlen != sizeof(val))
3202 if (copy_from_user(&val, optval, sizeof(val)))
3205 return fanout_add(sk, val & 0xffff, val >> 16);
3207 case PACKET_TX_HAS_OFF:
3211 if (optlen != sizeof(val))
3213 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3215 if (copy_from_user(&val, optval, sizeof(val)))
3217 po->tp_tx_has_off = !!val;
3221 return -ENOPROTOOPT;
3225 static int packet_getsockopt(struct socket *sock, int level, int optname,
3226 char __user *optval, int __user *optlen)
3229 int val, lv = sizeof(val);
3230 struct sock *sk = sock->sk;
3231 struct packet_sock *po = pkt_sk(sk);
3233 struct tpacket_stats st;
3234 union tpacket_stats_u st_u;
3236 if (level != SOL_PACKET)
3237 return -ENOPROTOOPT;
3239 if (get_user(len, optlen))
3246 case PACKET_STATISTICS:
3247 spin_lock_bh(&sk->sk_receive_queue.lock);
3248 if (po->tp_version == TPACKET_V3) {
3249 lv = sizeof(struct tpacket_stats_v3);
3250 memcpy(&st_u.stats3, &po->stats,
3251 sizeof(struct tpacket_stats));
3252 st_u.stats3.tp_freeze_q_cnt =
3253 po->stats_u.stats3.tp_freeze_q_cnt;
3254 st_u.stats3.tp_packets += po->stats.tp_drops;
3255 data = &st_u.stats3;
3257 lv = sizeof(struct tpacket_stats);
3259 st.tp_packets += st.tp_drops;
3262 memset(&po->stats, 0, sizeof(st));
3263 spin_unlock_bh(&sk->sk_receive_queue.lock);
3265 case PACKET_AUXDATA:
3268 case PACKET_ORIGDEV:
3271 case PACKET_VNET_HDR:
3272 val = po->has_vnet_hdr;
3274 case PACKET_VERSION:
3275 val = po->tp_version;
3278 if (len > sizeof(int))
3280 if (copy_from_user(&val, optval, len))
3284 val = sizeof(struct tpacket_hdr);
3287 val = sizeof(struct tpacket2_hdr);
3290 val = sizeof(struct tpacket3_hdr);
3296 case PACKET_RESERVE:
3297 val = po->tp_reserve;
3302 case PACKET_TIMESTAMP:
3303 val = po->tp_tstamp;
3307 ((u32)po->fanout->id |
3308 ((u32)po->fanout->type << 16) |
3309 ((u32)po->fanout->flags << 24)) :
3312 case PACKET_TX_HAS_OFF:
3313 val = po->tp_tx_has_off;
3316 return -ENOPROTOOPT;
3321 if (put_user(len, optlen))
3323 if (copy_to_user(optval, data, len))
3329 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3332 struct net_device *dev = data;
3333 struct net *net = dev_net(dev);
3336 sk_for_each_rcu(sk, &net->packet.sklist) {
3337 struct packet_sock *po = pkt_sk(sk);
3340 case NETDEV_UNREGISTER:
3342 packet_dev_mclist(dev, po->mclist, -1);
3346 if (dev->ifindex == po->ifindex) {
3347 spin_lock(&po->bind_lock);
3349 __unregister_prot_hook(sk, false);
3350 sk->sk_err = ENETDOWN;
3351 if (!sock_flag(sk, SOCK_DEAD))
3352 sk->sk_error_report(sk);
3354 if (msg == NETDEV_UNREGISTER) {
3356 if (po->prot_hook.dev)
3357 dev_put(po->prot_hook.dev);
3358 po->prot_hook.dev = NULL;
3360 spin_unlock(&po->bind_lock);
3364 if (dev->ifindex == po->ifindex) {
3365 spin_lock(&po->bind_lock);
3367 register_prot_hook(sk);
3368 spin_unlock(&po->bind_lock);
3378 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3381 struct sock *sk = sock->sk;
3386 int amount = sk_wmem_alloc_get(sk);
3388 return put_user(amount, (int __user *)arg);
3392 struct sk_buff *skb;
3395 spin_lock_bh(&sk->sk_receive_queue.lock);
3396 skb = skb_peek(&sk->sk_receive_queue);
3399 spin_unlock_bh(&sk->sk_receive_queue.lock);
3400 return put_user(amount, (int __user *)arg);
3403 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3405 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3415 case SIOCGIFBRDADDR:
3416 case SIOCSIFBRDADDR:
3417 case SIOCGIFNETMASK:
3418 case SIOCSIFNETMASK:
3419 case SIOCGIFDSTADDR:
3420 case SIOCSIFDSTADDR:
3422 return inet_dgram_ops.ioctl(sock, cmd, arg);
3426 return -ENOIOCTLCMD;
3431 static unsigned int packet_poll(struct file *file, struct socket *sock,
3434 struct sock *sk = sock->sk;
3435 struct packet_sock *po = pkt_sk(sk);
3436 unsigned int mask = datagram_poll(file, sock, wait);
3438 spin_lock_bh(&sk->sk_receive_queue.lock);
3439 if (po->rx_ring.pg_vec) {
3440 if (!packet_previous_rx_frame(po, &po->rx_ring,
3442 mask |= POLLIN | POLLRDNORM;
3444 spin_unlock_bh(&sk->sk_receive_queue.lock);
3445 spin_lock_bh(&sk->sk_write_queue.lock);
3446 if (po->tx_ring.pg_vec) {
3447 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3448 mask |= POLLOUT | POLLWRNORM;
3450 spin_unlock_bh(&sk->sk_write_queue.lock);
3455 /* Dirty? Well, I still did not learn better way to account
3459 static void packet_mm_open(struct vm_area_struct *vma)
3461 struct file *file = vma->vm_file;
3462 struct socket *sock = file->private_data;
3463 struct sock *sk = sock->sk;
3466 atomic_inc(&pkt_sk(sk)->mapped);
3469 static void packet_mm_close(struct vm_area_struct *vma)
3471 struct file *file = vma->vm_file;
3472 struct socket *sock = file->private_data;
3473 struct sock *sk = sock->sk;
3476 atomic_dec(&pkt_sk(sk)->mapped);
3479 static const struct vm_operations_struct packet_mmap_ops = {
3480 .open = packet_mm_open,
3481 .close = packet_mm_close,
3484 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3489 for (i = 0; i < len; i++) {
3490 if (likely(pg_vec[i].buffer)) {
3491 if (is_vmalloc_addr(pg_vec[i].buffer))
3492 vfree(pg_vec[i].buffer);
3494 free_pages((unsigned long)pg_vec[i].buffer,
3496 pg_vec[i].buffer = NULL;
3502 static char *alloc_one_pg_vec_page(unsigned long order)
3504 char *buffer = NULL;
3505 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3506 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3508 buffer = (char *) __get_free_pages(gfp_flags, order);
3514 * __get_free_pages failed, fall back to vmalloc
3516 buffer = vzalloc((1 << order) * PAGE_SIZE);
3522 * vmalloc failed, lets dig into swap here
3524 gfp_flags &= ~__GFP_NORETRY;
3525 buffer = (char *)__get_free_pages(gfp_flags, order);
3530 * complete and utter failure
3535 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3537 unsigned int block_nr = req->tp_block_nr;
3541 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3542 if (unlikely(!pg_vec))
3545 for (i = 0; i < block_nr; i++) {
3546 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3547 if (unlikely(!pg_vec[i].buffer))
3548 goto out_free_pgvec;
3555 free_pg_vec(pg_vec, order, block_nr);
3560 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3561 int closing, int tx_ring)
3563 struct pgv *pg_vec = NULL;
3564 struct packet_sock *po = pkt_sk(sk);
3565 int was_running, order = 0;
3566 struct packet_ring_buffer *rb;
3567 struct sk_buff_head *rb_queue;
3570 /* Added to avoid minimal code churn */
3571 struct tpacket_req *req = &req_u->req;
3573 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3574 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3575 WARN(1, "Tx-ring is not supported.\n");
3579 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3580 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3584 if (atomic_read(&po->mapped))
3586 if (atomic_read(&rb->pending))
3590 if (req->tp_block_nr) {
3591 /* Sanity tests and some calculations */
3593 if (unlikely(rb->pg_vec))
3596 switch (po->tp_version) {
3598 po->tp_hdrlen = TPACKET_HDRLEN;
3601 po->tp_hdrlen = TPACKET2_HDRLEN;
3604 po->tp_hdrlen = TPACKET3_HDRLEN;
3609 if (unlikely((int)req->tp_block_size <= 0))
3611 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3613 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3616 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3619 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3620 if (unlikely(rb->frames_per_block <= 0))
3622 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3627 order = get_order(req->tp_block_size);
3628 pg_vec = alloc_pg_vec(req, order);
3629 if (unlikely(!pg_vec))
3631 switch (po->tp_version) {
3633 /* Transmit path is not supported. We checked
3634 * it above but just being paranoid
3637 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3646 if (unlikely(req->tp_frame_nr))
3652 /* Detach socket from network */
3653 spin_lock(&po->bind_lock);
3654 was_running = po->running;
3658 __unregister_prot_hook(sk, false);
3660 spin_unlock(&po->bind_lock);
3665 mutex_lock(&po->pg_vec_lock);
3666 if (closing || atomic_read(&po->mapped) == 0) {
3668 spin_lock_bh(&rb_queue->lock);
3669 swap(rb->pg_vec, pg_vec);
3670 rb->frame_max = (req->tp_frame_nr - 1);
3672 rb->frame_size = req->tp_frame_size;
3673 spin_unlock_bh(&rb_queue->lock);
3675 swap(rb->pg_vec_order, order);
3676 swap(rb->pg_vec_len, req->tp_block_nr);
3678 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3679 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3680 tpacket_rcv : packet_rcv;
3681 skb_queue_purge(rb_queue);
3682 if (atomic_read(&po->mapped))
3683 pr_err("packet_mmap: vma is busy: %d\n",
3684 atomic_read(&po->mapped));
3686 mutex_unlock(&po->pg_vec_lock);
3688 spin_lock(&po->bind_lock);
3691 register_prot_hook(sk);
3693 spin_unlock(&po->bind_lock);
3694 if (closing && (po->tp_version > TPACKET_V2)) {
3695 /* Because we don't support block-based V3 on tx-ring */
3697 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3702 free_pg_vec(pg_vec, order, req->tp_block_nr);
3707 static int packet_mmap(struct file *file, struct socket *sock,
3708 struct vm_area_struct *vma)
3710 struct sock *sk = sock->sk;
3711 struct packet_sock *po = pkt_sk(sk);
3712 unsigned long size, expected_size;
3713 struct packet_ring_buffer *rb;
3714 unsigned long start;
3721 mutex_lock(&po->pg_vec_lock);
3724 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3726 expected_size += rb->pg_vec_len
3732 if (expected_size == 0)
3735 size = vma->vm_end - vma->vm_start;
3736 if (size != expected_size)
3739 start = vma->vm_start;
3740 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3741 if (rb->pg_vec == NULL)
3744 for (i = 0; i < rb->pg_vec_len; i++) {
3746 void *kaddr = rb->pg_vec[i].buffer;
3749 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3750 page = pgv_to_page(kaddr);
3751 err = vm_insert_page(vma, start, page);
3760 atomic_inc(&po->mapped);
3761 vma->vm_ops = &packet_mmap_ops;
3765 mutex_unlock(&po->pg_vec_lock);
3769 static const struct proto_ops packet_ops_spkt = {
3770 .family = PF_PACKET,
3771 .owner = THIS_MODULE,
3772 .release = packet_release,
3773 .bind = packet_bind_spkt,
3774 .connect = sock_no_connect,
3775 .socketpair = sock_no_socketpair,
3776 .accept = sock_no_accept,
3777 .getname = packet_getname_spkt,
3778 .poll = datagram_poll,
3779 .ioctl = packet_ioctl,
3780 .listen = sock_no_listen,
3781 .shutdown = sock_no_shutdown,
3782 .setsockopt = sock_no_setsockopt,
3783 .getsockopt = sock_no_getsockopt,
3784 .sendmsg = packet_sendmsg_spkt,
3785 .recvmsg = packet_recvmsg,
3786 .mmap = sock_no_mmap,
3787 .sendpage = sock_no_sendpage,
3790 static const struct proto_ops packet_ops = {
3791 .family = PF_PACKET,
3792 .owner = THIS_MODULE,
3793 .release = packet_release,
3794 .bind = packet_bind,
3795 .connect = sock_no_connect,
3796 .socketpair = sock_no_socketpair,
3797 .accept = sock_no_accept,
3798 .getname = packet_getname,
3799 .poll = packet_poll,
3800 .ioctl = packet_ioctl,
3801 .listen = sock_no_listen,
3802 .shutdown = sock_no_shutdown,
3803 .setsockopt = packet_setsockopt,
3804 .getsockopt = packet_getsockopt,
3805 .sendmsg = packet_sendmsg,
3806 .recvmsg = packet_recvmsg,
3807 .mmap = packet_mmap,
3808 .sendpage = sock_no_sendpage,
3811 static const struct net_proto_family packet_family_ops = {
3812 .family = PF_PACKET,
3813 .create = packet_create,
3814 .owner = THIS_MODULE,
3817 static struct notifier_block packet_netdev_notifier = {
3818 .notifier_call = packet_notifier,
3821 #ifdef CONFIG_PROC_FS
3823 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3826 struct net *net = seq_file_net(seq);
3829 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3832 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3834 struct net *net = seq_file_net(seq);
3835 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3838 static void packet_seq_stop(struct seq_file *seq, void *v)
3844 static int packet_seq_show(struct seq_file *seq, void *v)
3846 if (v == SEQ_START_TOKEN)
3847 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3849 struct sock *s = sk_entry(v);
3850 const struct packet_sock *po = pkt_sk(s);
3853 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3855 atomic_read(&s->sk_refcnt),
3860 atomic_read(&s->sk_rmem_alloc),
3861 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3868 static const struct seq_operations packet_seq_ops = {
3869 .start = packet_seq_start,
3870 .next = packet_seq_next,
3871 .stop = packet_seq_stop,
3872 .show = packet_seq_show,
3875 static int packet_seq_open(struct inode *inode, struct file *file)
3877 return seq_open_net(inode, file, &packet_seq_ops,
3878 sizeof(struct seq_net_private));
3881 static const struct file_operations packet_seq_fops = {
3882 .owner = THIS_MODULE,
3883 .open = packet_seq_open,
3885 .llseek = seq_lseek,
3886 .release = seq_release_net,
3891 static int __net_init packet_net_init(struct net *net)
3893 mutex_init(&net->packet.sklist_lock);
3894 INIT_HLIST_HEAD(&net->packet.sklist);
3896 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3902 static void __net_exit packet_net_exit(struct net *net)
3904 remove_proc_entry("packet", net->proc_net);
3907 static struct pernet_operations packet_net_ops = {
3908 .init = packet_net_init,
3909 .exit = packet_net_exit,
3913 static void __exit packet_exit(void)
3915 unregister_netdevice_notifier(&packet_netdev_notifier);
3916 unregister_pernet_subsys(&packet_net_ops);
3917 sock_unregister(PF_PACKET);
3918 proto_unregister(&packet_proto);
3921 static int __init packet_init(void)
3923 int rc = proto_register(&packet_proto, 0);
3928 sock_register(&packet_family_ops);
3929 register_pernet_subsys(&packet_net_ops);
3930 register_netdevice_notifier(&packet_netdev_notifier);
3935 module_init(packet_init);
3936 module_exit(packet_exit);
3937 MODULE_LICENSE("GPL");
3938 MODULE_ALIAS_NETPROTO(PF_PACKET);