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];
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 struct net_device *packet_cached_dev_get(struct packet_sock *po)
242 struct net_device *dev;
245 dev = rcu_dereference(po->cached_dev);
253 static void packet_cached_dev_assign(struct packet_sock *po,
254 struct net_device *dev)
256 rcu_assign_pointer(po->cached_dev, dev);
259 static void packet_cached_dev_reset(struct packet_sock *po)
261 RCU_INIT_POINTER(po->cached_dev, NULL);
264 /* register_prot_hook must be invoked with the po->bind_lock held,
265 * or from a context in which asynchronous accesses to the packet
266 * socket is not possible (packet_create()).
268 static void register_prot_hook(struct sock *sk)
270 struct packet_sock *po = pkt_sk(sk);
274 __fanout_link(sk, po);
276 dev_add_pack(&po->prot_hook);
283 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
284 * held. If the sync parameter is true, we will temporarily drop
285 * the po->bind_lock and do a synchronize_net to make sure no
286 * asynchronous packet processing paths still refer to the elements
287 * of po->prot_hook. If the sync parameter is false, it is the
288 * callers responsibility to take care of this.
290 static void __unregister_prot_hook(struct sock *sk, bool sync)
292 struct packet_sock *po = pkt_sk(sk);
297 __fanout_unlink(sk, po);
299 __dev_remove_pack(&po->prot_hook);
304 spin_unlock(&po->bind_lock);
306 spin_lock(&po->bind_lock);
310 static void unregister_prot_hook(struct sock *sk, bool sync)
312 struct packet_sock *po = pkt_sk(sk);
315 __unregister_prot_hook(sk, sync);
318 static inline __pure struct page *pgv_to_page(void *addr)
320 if (is_vmalloc_addr(addr))
321 return vmalloc_to_page(addr);
322 return virt_to_page(addr);
325 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
327 union tpacket_uhdr h;
330 switch (po->tp_version) {
332 h.h1->tp_status = status;
333 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
336 h.h2->tp_status = status;
337 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
341 WARN(1, "TPACKET version not supported.\n");
348 static int __packet_get_status(struct packet_sock *po, void *frame)
350 union tpacket_uhdr h;
355 switch (po->tp_version) {
357 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
358 return h.h1->tp_status;
360 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
361 return h.h2->tp_status;
364 WARN(1, "TPACKET version not supported.\n");
370 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
373 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
376 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
377 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
378 return TP_STATUS_TS_SYS_HARDWARE;
379 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
380 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
381 return TP_STATUS_TS_RAW_HARDWARE;
384 if (ktime_to_timespec_cond(skb->tstamp, ts))
385 return TP_STATUS_TS_SOFTWARE;
390 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
393 union tpacket_uhdr h;
397 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
401 switch (po->tp_version) {
403 h.h1->tp_sec = ts.tv_sec;
404 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
407 h.h2->tp_sec = ts.tv_sec;
408 h.h2->tp_nsec = ts.tv_nsec;
412 WARN(1, "TPACKET version not supported.\n");
416 /* one flush is safe, as both fields always lie on the same cacheline */
417 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
423 static void *packet_lookup_frame(struct packet_sock *po,
424 struct packet_ring_buffer *rb,
425 unsigned int position,
428 unsigned int pg_vec_pos, frame_offset;
429 union tpacket_uhdr h;
431 pg_vec_pos = position / rb->frames_per_block;
432 frame_offset = position % rb->frames_per_block;
434 h.raw = rb->pg_vec[pg_vec_pos].buffer +
435 (frame_offset * rb->frame_size);
437 if (status != __packet_get_status(po, h.raw))
443 static void *packet_current_frame(struct packet_sock *po,
444 struct packet_ring_buffer *rb,
447 return packet_lookup_frame(po, rb, rb->head, status);
450 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
452 del_timer_sync(&pkc->retire_blk_timer);
455 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
457 struct sk_buff_head *rb_queue)
459 struct tpacket_kbdq_core *pkc;
461 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
463 spin_lock_bh(&rb_queue->lock);
464 pkc->delete_blk_timer = 1;
465 spin_unlock_bh(&rb_queue->lock);
467 prb_del_retire_blk_timer(pkc);
470 static void prb_init_blk_timer(struct packet_sock *po,
471 struct tpacket_kbdq_core *pkc,
472 void (*func) (unsigned long))
474 init_timer(&pkc->retire_blk_timer);
475 pkc->retire_blk_timer.data = (long)po;
476 pkc->retire_blk_timer.function = func;
477 pkc->retire_blk_timer.expires = jiffies;
480 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
482 struct tpacket_kbdq_core *pkc;
487 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
488 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
491 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
492 int blk_size_in_bytes)
494 struct net_device *dev;
495 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
496 struct ethtool_cmd ecmd;
501 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
502 if (unlikely(!dev)) {
504 return DEFAULT_PRB_RETIRE_TOV;
506 err = __ethtool_get_settings(dev, &ecmd);
507 speed = ethtool_cmd_speed(&ecmd);
511 * If the link speed is so slow you don't really
512 * need to worry about perf anyways
514 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
515 return DEFAULT_PRB_RETIRE_TOV;
522 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
534 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
535 union tpacket_req_u *req_u)
537 p1->feature_req_word = req_u->req3.tp_feature_req_word;
540 static void init_prb_bdqc(struct packet_sock *po,
541 struct packet_ring_buffer *rb,
543 union tpacket_req_u *req_u, int tx_ring)
545 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
546 struct tpacket_block_desc *pbd;
548 memset(p1, 0x0, sizeof(*p1));
550 p1->knxt_seq_num = 1;
552 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
553 p1->pkblk_start = pg_vec[0].buffer;
554 p1->kblk_size = req_u->req3.tp_block_size;
555 p1->knum_blocks = req_u->req3.tp_block_nr;
556 p1->hdrlen = po->tp_hdrlen;
557 p1->version = po->tp_version;
558 p1->last_kactive_blk_num = 0;
559 po->stats.stats3.tp_freeze_q_cnt = 0;
560 if (req_u->req3.tp_retire_blk_tov)
561 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
563 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
564 req_u->req3.tp_block_size);
565 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
566 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
568 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
569 prb_init_ft_ops(p1, req_u);
570 prb_setup_retire_blk_timer(po, tx_ring);
571 prb_open_block(p1, pbd);
574 /* Do NOT update the last_blk_num first.
575 * Assumes sk_buff_head lock is held.
577 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
579 mod_timer(&pkc->retire_blk_timer,
580 jiffies + pkc->tov_in_jiffies);
581 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
586 * 1) We refresh the timer only when we open a block.
587 * By doing this we don't waste cycles refreshing the timer
588 * on packet-by-packet basis.
590 * With a 1MB block-size, on a 1Gbps line, it will take
591 * i) ~8 ms to fill a block + ii) memcpy etc.
592 * In this cut we are not accounting for the memcpy time.
594 * So, if the user sets the 'tmo' to 10ms then the timer
595 * will never fire while the block is still getting filled
596 * (which is what we want). However, the user could choose
597 * to close a block early and that's fine.
599 * But when the timer does fire, we check whether or not to refresh it.
600 * Since the tmo granularity is in msecs, it is not too expensive
601 * to refresh the timer, lets say every '8' msecs.
602 * Either the user can set the 'tmo' or we can derive it based on
603 * a) line-speed and b) block-size.
604 * prb_calc_retire_blk_tmo() calculates the tmo.
607 static void prb_retire_rx_blk_timer_expired(unsigned long data)
609 struct packet_sock *po = (struct packet_sock *)data;
610 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
612 struct tpacket_block_desc *pbd;
614 spin_lock(&po->sk.sk_receive_queue.lock);
616 frozen = prb_queue_frozen(pkc);
617 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
619 if (unlikely(pkc->delete_blk_timer))
622 /* We only need to plug the race when the block is partially filled.
624 * lock(); increment BLOCK_NUM_PKTS; unlock()
625 * copy_bits() is in progress ...
626 * timer fires on other cpu:
627 * we can't retire the current block because copy_bits
631 if (BLOCK_NUM_PKTS(pbd)) {
632 while (atomic_read(&pkc->blk_fill_in_prog)) {
633 /* Waiting for skb_copy_bits to finish... */
638 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
640 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
641 if (!prb_dispatch_next_block(pkc, po))
646 /* Case 1. Queue was frozen because user-space was
649 if (prb_curr_blk_in_use(pkc, pbd)) {
651 * Ok, user-space is still behind.
652 * So just refresh the timer.
656 /* Case 2. queue was frozen,user-space caught up,
657 * now the link went idle && the timer fired.
658 * We don't have a block to close.So we open this
659 * block and restart the timer.
660 * opening a block thaws the queue,restarts timer
661 * Thawing/timer-refresh is a side effect.
663 prb_open_block(pkc, pbd);
670 _prb_refresh_rx_retire_blk_timer(pkc);
673 spin_unlock(&po->sk.sk_receive_queue.lock);
676 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
677 struct tpacket_block_desc *pbd1, __u32 status)
679 /* Flush everything minus the block header */
681 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
686 /* Skip the block header(we know header WILL fit in 4K) */
689 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
690 for (; start < end; start += PAGE_SIZE)
691 flush_dcache_page(pgv_to_page(start));
696 /* Now update the block status. */
698 BLOCK_STATUS(pbd1) = status;
700 /* Flush the block header */
702 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
704 flush_dcache_page(pgv_to_page(start));
714 * 2) Increment active_blk_num
716 * Note:We DONT refresh the timer on purpose.
717 * Because almost always the next block will be opened.
719 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
720 struct tpacket_block_desc *pbd1,
721 struct packet_sock *po, unsigned int stat)
723 __u32 status = TP_STATUS_USER | stat;
725 struct tpacket3_hdr *last_pkt;
726 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
728 if (po->stats.stats3.tp_drops)
729 status |= TP_STATUS_LOSING;
731 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
732 last_pkt->tp_next_offset = 0;
734 /* Get the ts of the last pkt */
735 if (BLOCK_NUM_PKTS(pbd1)) {
736 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
737 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
739 /* Ok, we tmo'd - so get the current time */
742 h1->ts_last_pkt.ts_sec = ts.tv_sec;
743 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
748 /* Flush the block */
749 prb_flush_block(pkc1, pbd1, status);
751 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
754 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
756 pkc->reset_pending_on_curr_blk = 0;
760 * Side effect of opening a block:
762 * 1) prb_queue is thawed.
763 * 2) retire_blk_timer is refreshed.
766 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
767 struct tpacket_block_desc *pbd1)
770 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
774 /* We could have just memset this but we will lose the
775 * flexibility of making the priv area sticky
778 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
779 BLOCK_NUM_PKTS(pbd1) = 0;
780 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
784 h1->ts_first_pkt.ts_sec = ts.tv_sec;
785 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
787 pkc1->pkblk_start = (char *)pbd1;
788 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
790 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
791 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
793 pbd1->version = pkc1->version;
794 pkc1->prev = pkc1->nxt_offset;
795 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
797 prb_thaw_queue(pkc1);
798 _prb_refresh_rx_retire_blk_timer(pkc1);
804 * Queue freeze logic:
805 * 1) Assume tp_block_nr = 8 blocks.
806 * 2) At time 't0', user opens Rx ring.
807 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
808 * 4) user-space is either sleeping or processing block '0'.
809 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
810 * it will close block-7,loop around and try to fill block '0'.
812 * __packet_lookup_frame_in_block
813 * prb_retire_current_block()
814 * prb_dispatch_next_block()
815 * |->(BLOCK_STATUS == USER) evaluates to true
816 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
817 * 6) Now there are two cases:
818 * 6.1) Link goes idle right after the queue is frozen.
819 * But remember, the last open_block() refreshed the timer.
820 * When this timer expires,it will refresh itself so that we can
821 * re-open block-0 in near future.
822 * 6.2) Link is busy and keeps on receiving packets. This is a simple
823 * case and __packet_lookup_frame_in_block will check if block-0
824 * is free and can now be re-used.
826 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
827 struct packet_sock *po)
829 pkc->reset_pending_on_curr_blk = 1;
830 po->stats.stats3.tp_freeze_q_cnt++;
833 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
836 * If the next block is free then we will dispatch it
837 * and return a good offset.
838 * Else, we will freeze the queue.
839 * So, caller must check the return value.
841 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
842 struct packet_sock *po)
844 struct tpacket_block_desc *pbd;
848 /* 1. Get current block num */
849 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
851 /* 2. If this block is currently in_use then freeze the queue */
852 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
853 prb_freeze_queue(pkc, po);
859 * open this block and return the offset where the first packet
860 * needs to get stored.
862 prb_open_block(pkc, pbd);
863 return (void *)pkc->nxt_offset;
866 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
867 struct packet_sock *po, unsigned int status)
869 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
871 /* retire/close the current block */
872 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
874 * Plug the case where copy_bits() is in progress on
875 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
876 * have space to copy the pkt in the current block and
877 * called prb_retire_current_block()
879 * We don't need to worry about the TMO case because
880 * the timer-handler already handled this case.
882 if (!(status & TP_STATUS_BLK_TMO)) {
883 while (atomic_read(&pkc->blk_fill_in_prog)) {
884 /* Waiting for skb_copy_bits to finish... */
888 prb_close_block(pkc, pbd, po, status);
893 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
894 struct tpacket_block_desc *pbd)
896 return TP_STATUS_USER & BLOCK_STATUS(pbd);
899 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
901 return pkc->reset_pending_on_curr_blk;
904 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
906 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
907 atomic_dec(&pkc->blk_fill_in_prog);
910 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
911 struct tpacket3_hdr *ppd)
913 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
916 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
917 struct tpacket3_hdr *ppd)
919 ppd->hv1.tp_rxhash = 0;
922 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
923 struct tpacket3_hdr *ppd)
925 if (vlan_tx_tag_present(pkc->skb)) {
926 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
927 ppd->tp_status = TP_STATUS_VLAN_VALID;
929 ppd->hv1.tp_vlan_tci = 0;
930 ppd->tp_status = TP_STATUS_AVAILABLE;
934 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
935 struct tpacket3_hdr *ppd)
937 prb_fill_vlan_info(pkc, ppd);
939 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
940 prb_fill_rxhash(pkc, ppd);
942 prb_clear_rxhash(pkc, ppd);
945 static void prb_fill_curr_block(char *curr,
946 struct tpacket_kbdq_core *pkc,
947 struct tpacket_block_desc *pbd,
950 struct tpacket3_hdr *ppd;
952 ppd = (struct tpacket3_hdr *)curr;
953 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
955 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
956 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
957 BLOCK_NUM_PKTS(pbd) += 1;
958 atomic_inc(&pkc->blk_fill_in_prog);
959 prb_run_all_ft_ops(pkc, ppd);
962 /* Assumes caller has the sk->rx_queue.lock */
963 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
969 struct tpacket_kbdq_core *pkc;
970 struct tpacket_block_desc *pbd;
973 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
974 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
976 /* Queue is frozen when user space is lagging behind */
977 if (prb_queue_frozen(pkc)) {
979 * Check if that last block which caused the queue to freeze,
980 * is still in_use by user-space.
982 if (prb_curr_blk_in_use(pkc, pbd)) {
983 /* Can't record this packet */
987 * Ok, the block was released by user-space.
988 * Now let's open that block.
989 * opening a block also thaws the queue.
990 * Thawing is a side effect.
992 prb_open_block(pkc, pbd);
997 curr = pkc->nxt_offset;
999 end = (char *)pbd + pkc->kblk_size;
1001 /* first try the current block */
1002 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1003 prb_fill_curr_block(curr, pkc, pbd, len);
1004 return (void *)curr;
1007 /* Ok, close the current block */
1008 prb_retire_current_block(pkc, po, 0);
1010 /* Now, try to dispatch the next block */
1011 curr = (char *)prb_dispatch_next_block(pkc, po);
1013 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1014 prb_fill_curr_block(curr, pkc, pbd, len);
1015 return (void *)curr;
1019 * No free blocks are available.user_space hasn't caught up yet.
1020 * Queue was just frozen and now this packet will get dropped.
1025 static void *packet_current_rx_frame(struct packet_sock *po,
1026 struct sk_buff *skb,
1027 int status, unsigned int len)
1030 switch (po->tp_version) {
1033 curr = packet_lookup_frame(po, &po->rx_ring,
1034 po->rx_ring.head, status);
1037 return __packet_lookup_frame_in_block(po, skb, status, len);
1039 WARN(1, "TPACKET version not supported\n");
1045 static void *prb_lookup_block(struct packet_sock *po,
1046 struct packet_ring_buffer *rb,
1050 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1051 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1053 if (status != BLOCK_STATUS(pbd))
1058 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1061 if (rb->prb_bdqc.kactive_blk_num)
1062 prev = rb->prb_bdqc.kactive_blk_num-1;
1064 prev = rb->prb_bdqc.knum_blocks-1;
1068 /* Assumes caller has held the rx_queue.lock */
1069 static void *__prb_previous_block(struct packet_sock *po,
1070 struct packet_ring_buffer *rb,
1073 unsigned int previous = prb_previous_blk_num(rb);
1074 return prb_lookup_block(po, rb, previous, status);
1077 static void *packet_previous_rx_frame(struct packet_sock *po,
1078 struct packet_ring_buffer *rb,
1081 if (po->tp_version <= TPACKET_V2)
1082 return packet_previous_frame(po, rb, status);
1084 return __prb_previous_block(po, rb, status);
1087 static void packet_increment_rx_head(struct packet_sock *po,
1088 struct packet_ring_buffer *rb)
1090 switch (po->tp_version) {
1093 return packet_increment_head(rb);
1096 WARN(1, "TPACKET version not supported.\n");
1102 static void *packet_previous_frame(struct packet_sock *po,
1103 struct packet_ring_buffer *rb,
1106 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1107 return packet_lookup_frame(po, rb, previous, status);
1110 static void packet_increment_head(struct packet_ring_buffer *buff)
1112 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1115 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1117 struct sock *sk = &po->sk;
1120 if (po->prot_hook.func != tpacket_rcv)
1121 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1124 spin_lock(&sk->sk_receive_queue.lock);
1125 if (po->tp_version == TPACKET_V3)
1126 has_room = prb_lookup_block(po, &po->rx_ring,
1127 po->rx_ring.prb_bdqc.kactive_blk_num,
1130 has_room = packet_lookup_frame(po, &po->rx_ring,
1133 spin_unlock(&sk->sk_receive_queue.lock);
1138 static void packet_sock_destruct(struct sock *sk)
1140 skb_queue_purge(&sk->sk_error_queue);
1142 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1143 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1145 if (!sock_flag(sk, SOCK_DEAD)) {
1146 pr_err("Attempt to release alive packet socket: %p\n", sk);
1150 sk_refcnt_debug_dec(sk);
1153 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1155 int x = atomic_read(&f->rr_cur) + 1;
1163 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1164 struct sk_buff *skb,
1167 return (((u64)skb->rxhash) * num) >> 32;
1170 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1171 struct sk_buff *skb,
1176 cur = atomic_read(&f->rr_cur);
1177 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1178 fanout_rr_next(f, num))) != cur)
1183 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1184 struct sk_buff *skb,
1187 return smp_processor_id() % num;
1190 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1191 struct sk_buff *skb,
1192 unsigned int idx, unsigned int skip,
1197 i = j = min_t(int, f->next[idx], num - 1);
1199 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1211 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1213 return f->flags & (flag >> 8);
1216 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217 struct packet_type *pt, struct net_device *orig_dev)
1219 struct packet_fanout *f = pt->af_packet_priv;
1220 unsigned int num = f->num_members;
1221 struct packet_sock *po;
1224 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1231 case PACKET_FANOUT_HASH:
1233 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1234 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1238 skb_get_rxhash(skb);
1239 idx = fanout_demux_hash(f, skb, num);
1241 case PACKET_FANOUT_LB:
1242 idx = fanout_demux_lb(f, skb, num);
1244 case PACKET_FANOUT_CPU:
1245 idx = fanout_demux_cpu(f, skb, num);
1247 case PACKET_FANOUT_ROLLOVER:
1248 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1252 po = pkt_sk(f->arr[idx]);
1253 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1254 unlikely(!packet_rcv_has_room(po, skb))) {
1255 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1256 po = pkt_sk(f->arr[idx]);
1259 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1262 DEFINE_MUTEX(fanout_mutex);
1263 EXPORT_SYMBOL_GPL(fanout_mutex);
1264 static LIST_HEAD(fanout_list);
1266 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1268 struct packet_fanout *f = po->fanout;
1270 spin_lock(&f->lock);
1271 f->arr[f->num_members] = sk;
1274 spin_unlock(&f->lock);
1277 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1279 struct packet_fanout *f = po->fanout;
1282 spin_lock(&f->lock);
1283 for (i = 0; i < f->num_members; i++) {
1284 if (f->arr[i] == sk)
1287 BUG_ON(i >= f->num_members);
1288 f->arr[i] = f->arr[f->num_members - 1];
1290 spin_unlock(&f->lock);
1293 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1295 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1301 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1303 struct packet_sock *po = pkt_sk(sk);
1304 struct packet_fanout *f, *match;
1305 u8 type = type_flags & 0xff;
1306 u8 flags = type_flags >> 8;
1310 case PACKET_FANOUT_ROLLOVER:
1311 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1313 case PACKET_FANOUT_HASH:
1314 case PACKET_FANOUT_LB:
1315 case PACKET_FANOUT_CPU:
1327 mutex_lock(&fanout_mutex);
1329 list_for_each_entry(f, &fanout_list, list) {
1331 read_pnet(&f->net) == sock_net(sk)) {
1337 if (match && match->flags != flags)
1341 match = kzalloc(sizeof(*match), GFP_KERNEL);
1344 write_pnet(&match->net, sock_net(sk));
1347 match->flags = flags;
1348 atomic_set(&match->rr_cur, 0);
1349 INIT_LIST_HEAD(&match->list);
1350 spin_lock_init(&match->lock);
1351 atomic_set(&match->sk_ref, 0);
1352 match->prot_hook.type = po->prot_hook.type;
1353 match->prot_hook.dev = po->prot_hook.dev;
1354 match->prot_hook.func = packet_rcv_fanout;
1355 match->prot_hook.af_packet_priv = match;
1356 match->prot_hook.id_match = match_fanout_group;
1357 dev_add_pack(&match->prot_hook);
1358 list_add(&match->list, &fanout_list);
1361 if (match->type == type &&
1362 match->prot_hook.type == po->prot_hook.type &&
1363 match->prot_hook.dev == po->prot_hook.dev) {
1365 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1366 __dev_remove_pack(&po->prot_hook);
1368 atomic_inc(&match->sk_ref);
1369 __fanout_link(sk, po);
1374 mutex_unlock(&fanout_mutex);
1378 static void fanout_release(struct sock *sk)
1380 struct packet_sock *po = pkt_sk(sk);
1381 struct packet_fanout *f;
1387 mutex_lock(&fanout_mutex);
1390 if (atomic_dec_and_test(&f->sk_ref)) {
1392 dev_remove_pack(&f->prot_hook);
1395 mutex_unlock(&fanout_mutex);
1398 static const struct proto_ops packet_ops;
1400 static const struct proto_ops packet_ops_spkt;
1402 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1403 struct packet_type *pt, struct net_device *orig_dev)
1406 struct sockaddr_pkt *spkt;
1409 * When we registered the protocol we saved the socket in the data
1410 * field for just this event.
1413 sk = pt->af_packet_priv;
1416 * Yank back the headers [hope the device set this
1417 * right or kerboom...]
1419 * Incoming packets have ll header pulled,
1422 * For outgoing ones skb->data == skb_mac_header(skb)
1423 * so that this procedure is noop.
1426 if (skb->pkt_type == PACKET_LOOPBACK)
1429 if (!net_eq(dev_net(dev), sock_net(sk)))
1432 skb = skb_share_check(skb, GFP_ATOMIC);
1436 /* drop any routing info */
1439 /* drop conntrack reference */
1442 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1444 skb_push(skb, skb->data - skb_mac_header(skb));
1447 * The SOCK_PACKET socket receives _all_ frames.
1450 spkt->spkt_family = dev->type;
1451 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1452 spkt->spkt_protocol = skb->protocol;
1455 * Charge the memory to the socket. This is done specifically
1456 * to prevent sockets using all the memory up.
1459 if (sock_queue_rcv_skb(sk, skb) == 0)
1470 * Output a raw packet to a device layer. This bypasses all the other
1471 * protocol layers and you must therefore supply it with a complete frame
1474 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1475 struct msghdr *msg, size_t len)
1477 struct sock *sk = sock->sk;
1478 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1479 struct sk_buff *skb = NULL;
1480 struct net_device *dev;
1486 * Get and verify the address.
1490 if (msg->msg_namelen < sizeof(struct sockaddr))
1492 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1493 proto = saddr->spkt_protocol;
1495 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1498 * Find the device first to size check it
1501 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1504 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1510 if (!(dev->flags & IFF_UP))
1514 * You may not queue a frame bigger than the mtu. This is the lowest level
1515 * raw protocol and you must do your own fragmentation at this level.
1518 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1519 if (!netif_supports_nofcs(dev)) {
1520 err = -EPROTONOSUPPORT;
1523 extra_len = 4; /* We're doing our own CRC */
1527 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1531 size_t reserved = LL_RESERVED_SPACE(dev);
1532 int tlen = dev->needed_tailroom;
1533 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1536 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1539 /* FIXME: Save some space for broken drivers that write a hard
1540 * header at transmission time by themselves. PPP is the notable
1541 * one here. This should really be fixed at the driver level.
1543 skb_reserve(skb, reserved);
1544 skb_reset_network_header(skb);
1546 /* Try to align data part correctly */
1551 skb_reset_network_header(skb);
1553 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1559 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1560 /* Earlier code assumed this would be a VLAN pkt,
1561 * double-check this now that we have the actual
1564 struct ethhdr *ehdr;
1565 skb_reset_mac_header(skb);
1566 ehdr = eth_hdr(skb);
1567 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1573 skb->protocol = proto;
1575 skb->priority = sk->sk_priority;
1576 skb->mark = sk->sk_mark;
1578 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1580 if (unlikely(extra_len == 4))
1583 skb_probe_transport_header(skb, 0);
1585 dev_queue_xmit(skb);
1596 static unsigned int run_filter(const struct sk_buff *skb,
1597 const struct sock *sk,
1600 struct sk_filter *filter;
1603 filter = rcu_dereference(sk->sk_filter);
1605 res = SK_RUN_FILTER(filter, skb);
1612 * This function makes lazy skb cloning in hope that most of packets
1613 * are discarded by BPF.
1615 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1616 * and skb->cb are mangled. It works because (and until) packets
1617 * falling here are owned by current CPU. Output packets are cloned
1618 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1619 * sequencially, so that if we return skb to original state on exit,
1620 * we will not harm anyone.
1623 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1624 struct packet_type *pt, struct net_device *orig_dev)
1627 struct sockaddr_ll *sll;
1628 struct packet_sock *po;
1629 u8 *skb_head = skb->data;
1630 int skb_len = skb->len;
1631 unsigned int snaplen, res;
1633 if (skb->pkt_type == PACKET_LOOPBACK)
1636 sk = pt->af_packet_priv;
1639 if (!net_eq(dev_net(dev), sock_net(sk)))
1644 if (dev->header_ops) {
1645 /* The device has an explicit notion of ll header,
1646 * exported to higher levels.
1648 * Otherwise, the device hides details of its frame
1649 * structure, so that corresponding packet head is
1650 * never delivered to user.
1652 if (sk->sk_type != SOCK_DGRAM)
1653 skb_push(skb, skb->data - skb_mac_header(skb));
1654 else if (skb->pkt_type == PACKET_OUTGOING) {
1655 /* Special case: outgoing packets have ll header at head */
1656 skb_pull(skb, skb_network_offset(skb));
1662 res = run_filter(skb, sk, snaplen);
1664 goto drop_n_restore;
1668 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1671 if (skb_shared(skb)) {
1672 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1676 if (skb_head != skb->data) {
1677 skb->data = skb_head;
1684 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1687 sll = &PACKET_SKB_CB(skb)->sa.ll;
1688 sll->sll_family = AF_PACKET;
1689 sll->sll_hatype = dev->type;
1690 sll->sll_protocol = skb->protocol;
1691 sll->sll_pkttype = skb->pkt_type;
1692 if (unlikely(po->origdev))
1693 sll->sll_ifindex = orig_dev->ifindex;
1695 sll->sll_ifindex = dev->ifindex;
1697 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1699 PACKET_SKB_CB(skb)->origlen = skb->len;
1701 if (pskb_trim(skb, snaplen))
1704 skb_set_owner_r(skb, sk);
1708 /* drop conntrack reference */
1711 spin_lock(&sk->sk_receive_queue.lock);
1712 po->stats.stats1.tp_packets++;
1713 skb->dropcount = atomic_read(&sk->sk_drops);
1714 __skb_queue_tail(&sk->sk_receive_queue, skb);
1715 spin_unlock(&sk->sk_receive_queue.lock);
1716 sk->sk_data_ready(sk, skb->len);
1720 spin_lock(&sk->sk_receive_queue.lock);
1721 po->stats.stats1.tp_drops++;
1722 atomic_inc(&sk->sk_drops);
1723 spin_unlock(&sk->sk_receive_queue.lock);
1726 if (skb_head != skb->data && skb_shared(skb)) {
1727 skb->data = skb_head;
1735 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1736 struct packet_type *pt, struct net_device *orig_dev)
1739 struct packet_sock *po;
1740 struct sockaddr_ll *sll;
1741 union tpacket_uhdr h;
1742 u8 *skb_head = skb->data;
1743 int skb_len = skb->len;
1744 unsigned int snaplen, res;
1745 unsigned long status = TP_STATUS_USER;
1746 unsigned short macoff, netoff, hdrlen;
1747 struct sk_buff *copy_skb = NULL;
1751 if (skb->pkt_type == PACKET_LOOPBACK)
1754 sk = pt->af_packet_priv;
1757 if (!net_eq(dev_net(dev), sock_net(sk)))
1760 if (dev->header_ops) {
1761 if (sk->sk_type != SOCK_DGRAM)
1762 skb_push(skb, skb->data - skb_mac_header(skb));
1763 else if (skb->pkt_type == PACKET_OUTGOING) {
1764 /* Special case: outgoing packets have ll header at head */
1765 skb_pull(skb, skb_network_offset(skb));
1769 if (skb->ip_summed == CHECKSUM_PARTIAL)
1770 status |= TP_STATUS_CSUMNOTREADY;
1774 res = run_filter(skb, sk, snaplen);
1776 goto drop_n_restore;
1780 if (sk->sk_type == SOCK_DGRAM) {
1781 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1784 unsigned int maclen = skb_network_offset(skb);
1785 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1786 (maclen < 16 ? 16 : maclen)) +
1788 macoff = netoff - maclen;
1790 if (po->tp_version <= TPACKET_V2) {
1791 if (macoff + snaplen > po->rx_ring.frame_size) {
1792 if (po->copy_thresh &&
1793 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1794 if (skb_shared(skb)) {
1795 copy_skb = skb_clone(skb, GFP_ATOMIC);
1797 copy_skb = skb_get(skb);
1798 skb_head = skb->data;
1801 skb_set_owner_r(copy_skb, sk);
1803 snaplen = po->rx_ring.frame_size - macoff;
1804 if ((int)snaplen < 0)
1807 } else if (unlikely(macoff + snaplen >
1808 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1811 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1812 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1813 snaplen, nval, macoff);
1815 if (unlikely((int)snaplen < 0)) {
1817 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1820 spin_lock(&sk->sk_receive_queue.lock);
1821 h.raw = packet_current_rx_frame(po, skb,
1822 TP_STATUS_KERNEL, (macoff+snaplen));
1825 if (po->tp_version <= TPACKET_V2) {
1826 packet_increment_rx_head(po, &po->rx_ring);
1828 * LOSING will be reported till you read the stats,
1829 * because it's COR - Clear On Read.
1830 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1833 if (po->stats.stats1.tp_drops)
1834 status |= TP_STATUS_LOSING;
1836 po->stats.stats1.tp_packets++;
1838 status |= TP_STATUS_COPY;
1839 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1841 spin_unlock(&sk->sk_receive_queue.lock);
1843 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1845 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1846 getnstimeofday(&ts);
1848 status |= ts_status;
1850 switch (po->tp_version) {
1852 h.h1->tp_len = skb->len;
1853 h.h1->tp_snaplen = snaplen;
1854 h.h1->tp_mac = macoff;
1855 h.h1->tp_net = netoff;
1856 h.h1->tp_sec = ts.tv_sec;
1857 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1858 hdrlen = sizeof(*h.h1);
1861 h.h2->tp_len = skb->len;
1862 h.h2->tp_snaplen = snaplen;
1863 h.h2->tp_mac = macoff;
1864 h.h2->tp_net = netoff;
1865 h.h2->tp_sec = ts.tv_sec;
1866 h.h2->tp_nsec = ts.tv_nsec;
1867 if (vlan_tx_tag_present(skb)) {
1868 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1869 status |= TP_STATUS_VLAN_VALID;
1871 h.h2->tp_vlan_tci = 0;
1873 h.h2->tp_padding = 0;
1874 hdrlen = sizeof(*h.h2);
1877 /* tp_nxt_offset,vlan are already populated above.
1878 * So DONT clear those fields here
1880 h.h3->tp_status |= status;
1881 h.h3->tp_len = skb->len;
1882 h.h3->tp_snaplen = snaplen;
1883 h.h3->tp_mac = macoff;
1884 h.h3->tp_net = netoff;
1885 h.h3->tp_sec = ts.tv_sec;
1886 h.h3->tp_nsec = ts.tv_nsec;
1887 hdrlen = sizeof(*h.h3);
1893 sll = h.raw + TPACKET_ALIGN(hdrlen);
1894 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1895 sll->sll_family = AF_PACKET;
1896 sll->sll_hatype = dev->type;
1897 sll->sll_protocol = skb->protocol;
1898 sll->sll_pkttype = skb->pkt_type;
1899 if (unlikely(po->origdev))
1900 sll->sll_ifindex = orig_dev->ifindex;
1902 sll->sll_ifindex = dev->ifindex;
1905 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1909 if (po->tp_version <= TPACKET_V2) {
1910 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1911 + macoff + snaplen);
1912 for (start = h.raw; start < end; start += PAGE_SIZE)
1913 flush_dcache_page(pgv_to_page(start));
1918 if (po->tp_version <= TPACKET_V2)
1919 __packet_set_status(po, h.raw, status);
1921 prb_clear_blk_fill_status(&po->rx_ring);
1923 sk->sk_data_ready(sk, 0);
1926 if (skb_head != skb->data && skb_shared(skb)) {
1927 skb->data = skb_head;
1935 po->stats.stats1.tp_drops++;
1936 spin_unlock(&sk->sk_receive_queue.lock);
1938 sk->sk_data_ready(sk, 0);
1939 kfree_skb(copy_skb);
1940 goto drop_n_restore;
1943 static void tpacket_destruct_skb(struct sk_buff *skb)
1945 struct packet_sock *po = pkt_sk(skb->sk);
1948 if (likely(po->tx_ring.pg_vec)) {
1951 ph = skb_shinfo(skb)->destructor_arg;
1952 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1953 atomic_dec(&po->tx_ring.pending);
1955 ts = __packet_set_timestamp(po, ph, skb);
1956 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1962 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1963 void *frame, struct net_device *dev, int size_max,
1964 __be16 proto, unsigned char *addr, int hlen)
1966 union tpacket_uhdr ph;
1967 int to_write, offset, len, tp_len, nr_frags, len_max;
1968 struct socket *sock = po->sk.sk_socket;
1975 skb->protocol = proto;
1977 skb->priority = po->sk.sk_priority;
1978 skb->mark = po->sk.sk_mark;
1979 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1980 skb_shinfo(skb)->destructor_arg = ph.raw;
1982 switch (po->tp_version) {
1984 tp_len = ph.h2->tp_len;
1987 tp_len = ph.h1->tp_len;
1990 if (unlikely(tp_len > size_max)) {
1991 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1995 skb_reserve(skb, hlen);
1996 skb_reset_network_header(skb);
1997 skb_probe_transport_header(skb, 0);
1999 if (po->tp_tx_has_off) {
2000 int off_min, off_max, off;
2001 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2002 off_max = po->tx_ring.frame_size - tp_len;
2003 if (sock->type == SOCK_DGRAM) {
2004 switch (po->tp_version) {
2006 off = ph.h2->tp_net;
2009 off = ph.h1->tp_net;
2013 switch (po->tp_version) {
2015 off = ph.h2->tp_mac;
2018 off = ph.h1->tp_mac;
2022 if (unlikely((off < off_min) || (off_max < off)))
2024 data = ph.raw + off;
2026 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2030 if (sock->type == SOCK_DGRAM) {
2031 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2033 if (unlikely(err < 0))
2035 } else if (dev->hard_header_len) {
2036 /* net device doesn't like empty head */
2037 if (unlikely(tp_len <= dev->hard_header_len)) {
2038 pr_err("packet size is too short (%d < %d)\n",
2039 tp_len, dev->hard_header_len);
2043 skb_push(skb, dev->hard_header_len);
2044 err = skb_store_bits(skb, 0, data,
2045 dev->hard_header_len);
2049 data += dev->hard_header_len;
2050 to_write -= dev->hard_header_len;
2053 offset = offset_in_page(data);
2054 len_max = PAGE_SIZE - offset;
2055 len = ((to_write > len_max) ? len_max : to_write);
2057 skb->data_len = to_write;
2058 skb->len += to_write;
2059 skb->truesize += to_write;
2060 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2062 while (likely(to_write)) {
2063 nr_frags = skb_shinfo(skb)->nr_frags;
2065 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2066 pr_err("Packet exceed the number of skb frags(%lu)\n",
2071 page = pgv_to_page(data);
2073 flush_dcache_page(page);
2075 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2078 len_max = PAGE_SIZE;
2079 len = ((to_write > len_max) ? len_max : to_write);
2085 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2087 struct sk_buff *skb;
2088 struct net_device *dev;
2090 int err, reserve = 0;
2092 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2093 int tp_len, size_max;
2094 unsigned char *addr;
2096 int status = TP_STATUS_AVAILABLE;
2099 mutex_lock(&po->pg_vec_lock);
2101 if (likely(saddr == NULL)) {
2102 dev = packet_cached_dev_get(po);
2107 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2109 if (msg->msg_namelen < (saddr->sll_halen
2110 + offsetof(struct sockaddr_ll,
2113 proto = saddr->sll_protocol;
2114 addr = saddr->sll_addr;
2115 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2119 if (unlikely(dev == NULL))
2122 if (unlikely(!(dev->flags & IFF_UP)))
2125 reserve = dev->hard_header_len;
2127 size_max = po->tx_ring.frame_size
2128 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2130 if (size_max > dev->mtu + reserve)
2131 size_max = dev->mtu + reserve;
2134 ph = packet_current_frame(po, &po->tx_ring,
2135 TP_STATUS_SEND_REQUEST);
2137 if (unlikely(ph == NULL)) {
2142 status = TP_STATUS_SEND_REQUEST;
2143 hlen = LL_RESERVED_SPACE(dev);
2144 tlen = dev->needed_tailroom;
2145 skb = sock_alloc_send_skb(&po->sk,
2146 hlen + tlen + sizeof(struct sockaddr_ll),
2149 if (unlikely(skb == NULL))
2152 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2155 if (unlikely(tp_len < 0)) {
2157 __packet_set_status(po, ph,
2158 TP_STATUS_AVAILABLE);
2159 packet_increment_head(&po->tx_ring);
2163 status = TP_STATUS_WRONG_FORMAT;
2169 skb->destructor = tpacket_destruct_skb;
2170 __packet_set_status(po, ph, TP_STATUS_SENDING);
2171 atomic_inc(&po->tx_ring.pending);
2173 status = TP_STATUS_SEND_REQUEST;
2174 err = dev_queue_xmit(skb);
2175 if (unlikely(err > 0)) {
2176 err = net_xmit_errno(err);
2177 if (err && __packet_get_status(po, ph) ==
2178 TP_STATUS_AVAILABLE) {
2179 /* skb was destructed already */
2184 * skb was dropped but not destructed yet;
2185 * let's treat it like congestion or err < 0
2189 packet_increment_head(&po->tx_ring);
2191 } while (likely((ph != NULL) ||
2192 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2193 (atomic_read(&po->tx_ring.pending))))
2200 __packet_set_status(po, ph, status);
2205 mutex_unlock(&po->pg_vec_lock);
2209 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2210 size_t reserve, size_t len,
2211 size_t linear, int noblock,
2214 struct sk_buff *skb;
2216 /* Under a page? Don't bother with paged skb. */
2217 if (prepad + len < PAGE_SIZE || !linear)
2220 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2225 skb_reserve(skb, reserve);
2226 skb_put(skb, linear);
2227 skb->data_len = len - linear;
2228 skb->len += len - linear;
2233 static int packet_snd(struct socket *sock,
2234 struct msghdr *msg, size_t len)
2236 struct sock *sk = sock->sk;
2237 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2238 struct sk_buff *skb;
2239 struct net_device *dev;
2241 unsigned char *addr;
2242 int err, reserve = 0;
2243 struct virtio_net_hdr vnet_hdr = { 0 };
2246 struct packet_sock *po = pkt_sk(sk);
2247 unsigned short gso_type = 0;
2252 * Get and verify the address.
2255 if (likely(saddr == NULL)) {
2256 dev = packet_cached_dev_get(po);
2261 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2263 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2265 proto = saddr->sll_protocol;
2266 addr = saddr->sll_addr;
2267 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2271 if (unlikely(dev == NULL))
2274 if (unlikely(!(dev->flags & IFF_UP)))
2277 if (sock->type == SOCK_RAW)
2278 reserve = dev->hard_header_len;
2279 if (po->has_vnet_hdr) {
2280 vnet_hdr_len = sizeof(vnet_hdr);
2283 if (len < vnet_hdr_len)
2286 len -= vnet_hdr_len;
2288 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2293 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2294 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2296 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2297 vnet_hdr.csum_offset + 2;
2300 if (vnet_hdr.hdr_len > len)
2303 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2304 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2305 case VIRTIO_NET_HDR_GSO_TCPV4:
2306 gso_type = SKB_GSO_TCPV4;
2308 case VIRTIO_NET_HDR_GSO_TCPV6:
2309 gso_type = SKB_GSO_TCPV6;
2311 case VIRTIO_NET_HDR_GSO_UDP:
2312 gso_type = SKB_GSO_UDP;
2318 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2319 gso_type |= SKB_GSO_TCP_ECN;
2321 if (vnet_hdr.gso_size == 0)
2327 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2328 if (!netif_supports_nofcs(dev)) {
2329 err = -EPROTONOSUPPORT;
2332 extra_len = 4; /* We're doing our own CRC */
2336 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2340 hlen = LL_RESERVED_SPACE(dev);
2341 tlen = dev->needed_tailroom;
2342 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2343 msg->msg_flags & MSG_DONTWAIT, &err);
2347 skb_set_network_header(skb, reserve);
2350 if (sock->type == SOCK_DGRAM &&
2351 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2354 /* Returns -EFAULT on error */
2355 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2359 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2361 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2362 /* Earlier code assumed this would be a VLAN pkt,
2363 * double-check this now that we have the actual
2366 struct ethhdr *ehdr;
2367 skb_reset_mac_header(skb);
2368 ehdr = eth_hdr(skb);
2369 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2375 skb->protocol = proto;
2377 skb->priority = sk->sk_priority;
2378 skb->mark = sk->sk_mark;
2380 if (po->has_vnet_hdr) {
2381 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2382 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2383 vnet_hdr.csum_offset)) {
2389 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2390 skb_shinfo(skb)->gso_type = gso_type;
2392 /* Header must be checked, and gso_segs computed. */
2393 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2394 skb_shinfo(skb)->gso_segs = 0;
2396 len += vnet_hdr_len;
2399 skb_probe_transport_header(skb, reserve);
2401 if (unlikely(extra_len == 4))
2408 err = dev_queue_xmit(skb);
2409 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2425 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2426 struct msghdr *msg, size_t len)
2428 struct sock *sk = sock->sk;
2429 struct packet_sock *po = pkt_sk(sk);
2430 if (po->tx_ring.pg_vec)
2431 return tpacket_snd(po, msg);
2433 return packet_snd(sock, msg, len);
2437 * Close a PACKET socket. This is fairly simple. We immediately go
2438 * to 'closed' state and remove our protocol entry in the device list.
2441 static int packet_release(struct socket *sock)
2443 struct sock *sk = sock->sk;
2444 struct packet_sock *po;
2446 union tpacket_req_u req_u;
2454 mutex_lock(&net->packet.sklist_lock);
2455 sk_del_node_init_rcu(sk);
2456 mutex_unlock(&net->packet.sklist_lock);
2459 sock_prot_inuse_add(net, sk->sk_prot, -1);
2462 spin_lock(&po->bind_lock);
2463 unregister_prot_hook(sk, false);
2464 packet_cached_dev_reset(po);
2466 if (po->prot_hook.dev) {
2467 dev_put(po->prot_hook.dev);
2468 po->prot_hook.dev = NULL;
2470 spin_unlock(&po->bind_lock);
2472 packet_flush_mclist(sk);
2474 if (po->rx_ring.pg_vec) {
2475 memset(&req_u, 0, sizeof(req_u));
2476 packet_set_ring(sk, &req_u, 1, 0);
2479 if (po->tx_ring.pg_vec) {
2480 memset(&req_u, 0, sizeof(req_u));
2481 packet_set_ring(sk, &req_u, 1, 1);
2488 * Now the socket is dead. No more input will appear.
2495 skb_queue_purge(&sk->sk_receive_queue);
2496 sk_refcnt_debug_release(sk);
2503 * Attach a packet hook.
2506 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2508 struct packet_sock *po = pkt_sk(sk);
2519 spin_lock(&po->bind_lock);
2520 unregister_prot_hook(sk, true);
2523 po->prot_hook.type = protocol;
2524 if (po->prot_hook.dev)
2525 dev_put(po->prot_hook.dev);
2527 po->prot_hook.dev = dev;
2528 po->ifindex = dev ? dev->ifindex : 0;
2530 packet_cached_dev_assign(po, dev);
2535 if (!dev || (dev->flags & IFF_UP)) {
2536 register_prot_hook(sk);
2538 sk->sk_err = ENETDOWN;
2539 if (!sock_flag(sk, SOCK_DEAD))
2540 sk->sk_error_report(sk);
2544 spin_unlock(&po->bind_lock);
2550 * Bind a packet socket to a device
2553 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2556 struct sock *sk = sock->sk;
2558 struct net_device *dev;
2565 if (addr_len != sizeof(struct sockaddr))
2567 strlcpy(name, uaddr->sa_data, sizeof(name));
2569 dev = dev_get_by_name(sock_net(sk), name);
2571 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2575 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2577 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2578 struct sock *sk = sock->sk;
2579 struct net_device *dev = NULL;
2587 if (addr_len < sizeof(struct sockaddr_ll))
2589 if (sll->sll_family != AF_PACKET)
2592 if (sll->sll_ifindex) {
2594 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2598 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2604 static struct proto packet_proto = {
2606 .owner = THIS_MODULE,
2607 .obj_size = sizeof(struct packet_sock),
2611 * Create a packet of type SOCK_PACKET.
2614 static int packet_create(struct net *net, struct socket *sock, int protocol,
2618 struct packet_sock *po;
2619 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2622 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2624 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2625 sock->type != SOCK_PACKET)
2626 return -ESOCKTNOSUPPORT;
2628 sock->state = SS_UNCONNECTED;
2631 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2635 sock->ops = &packet_ops;
2636 if (sock->type == SOCK_PACKET)
2637 sock->ops = &packet_ops_spkt;
2639 sock_init_data(sock, sk);
2642 sk->sk_family = PF_PACKET;
2645 packet_cached_dev_reset(po);
2647 sk->sk_destruct = packet_sock_destruct;
2648 sk_refcnt_debug_inc(sk);
2651 * Attach a protocol block
2654 spin_lock_init(&po->bind_lock);
2655 mutex_init(&po->pg_vec_lock);
2656 po->prot_hook.func = packet_rcv;
2658 if (sock->type == SOCK_PACKET)
2659 po->prot_hook.func = packet_rcv_spkt;
2661 po->prot_hook.af_packet_priv = sk;
2664 po->prot_hook.type = proto;
2665 register_prot_hook(sk);
2668 mutex_lock(&net->packet.sklist_lock);
2669 sk_add_node_rcu(sk, &net->packet.sklist);
2670 mutex_unlock(&net->packet.sklist_lock);
2673 sock_prot_inuse_add(net, &packet_proto, 1);
2681 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2683 struct sock_exterr_skb *serr;
2684 struct sk_buff *skb, *skb2;
2688 skb = skb_dequeue(&sk->sk_error_queue);
2694 msg->msg_flags |= MSG_TRUNC;
2697 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2701 sock_recv_timestamp(msg, sk, skb);
2703 serr = SKB_EXT_ERR(skb);
2704 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2705 sizeof(serr->ee), &serr->ee);
2707 msg->msg_flags |= MSG_ERRQUEUE;
2710 /* Reset and regenerate socket error */
2711 spin_lock_bh(&sk->sk_error_queue.lock);
2713 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2714 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2715 spin_unlock_bh(&sk->sk_error_queue.lock);
2716 sk->sk_error_report(sk);
2718 spin_unlock_bh(&sk->sk_error_queue.lock);
2727 * Pull a packet from our receive queue and hand it to the user.
2728 * If necessary we block.
2731 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2732 struct msghdr *msg, size_t len, int flags)
2734 struct sock *sk = sock->sk;
2735 struct sk_buff *skb;
2737 int vnet_hdr_len = 0;
2740 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2744 /* What error should we return now? EUNATTACH? */
2745 if (pkt_sk(sk)->ifindex < 0)
2749 if (flags & MSG_ERRQUEUE) {
2750 err = packet_recv_error(sk, msg, len);
2755 * Call the generic datagram receiver. This handles all sorts
2756 * of horrible races and re-entrancy so we can forget about it
2757 * in the protocol layers.
2759 * Now it will return ENETDOWN, if device have just gone down,
2760 * but then it will block.
2763 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2766 * An error occurred so return it. Because skb_recv_datagram()
2767 * handles the blocking we don't see and worry about blocking
2774 if (pkt_sk(sk)->has_vnet_hdr) {
2775 struct virtio_net_hdr vnet_hdr = { 0 };
2778 vnet_hdr_len = sizeof(vnet_hdr);
2779 if (len < vnet_hdr_len)
2782 len -= vnet_hdr_len;
2784 if (skb_is_gso(skb)) {
2785 struct skb_shared_info *sinfo = skb_shinfo(skb);
2787 /* This is a hint as to how much should be linear. */
2788 vnet_hdr.hdr_len = skb_headlen(skb);
2789 vnet_hdr.gso_size = sinfo->gso_size;
2790 if (sinfo->gso_type & SKB_GSO_TCPV4)
2791 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2792 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2793 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2794 else if (sinfo->gso_type & SKB_GSO_UDP)
2795 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2796 else if (sinfo->gso_type & SKB_GSO_FCOE)
2800 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2801 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2803 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2805 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2806 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2807 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2808 vnet_hdr.csum_offset = skb->csum_offset;
2809 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2810 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2811 } /* else everything is zero */
2813 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2819 /* You lose any data beyond the buffer you gave. If it worries
2820 * a user program they can ask the device for its MTU
2826 msg->msg_flags |= MSG_TRUNC;
2829 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2833 sock_recv_ts_and_drops(msg, sk, skb);
2835 if (msg->msg_name) {
2836 /* If the address length field is there to be filled
2837 * in, we fill it in now.
2839 if (sock->type == SOCK_PACKET) {
2840 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2842 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2843 msg->msg_namelen = sll->sll_halen +
2844 offsetof(struct sockaddr_ll, sll_addr);
2846 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2850 if (pkt_sk(sk)->auxdata) {
2851 struct tpacket_auxdata aux;
2853 aux.tp_status = TP_STATUS_USER;
2854 if (skb->ip_summed == CHECKSUM_PARTIAL)
2855 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2856 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2857 aux.tp_snaplen = skb->len;
2859 aux.tp_net = skb_network_offset(skb);
2860 if (vlan_tx_tag_present(skb)) {
2861 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2862 aux.tp_status |= TP_STATUS_VLAN_VALID;
2864 aux.tp_vlan_tci = 0;
2867 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2871 * Free or return the buffer as appropriate. Again this
2872 * hides all the races and re-entrancy issues from us.
2874 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2877 skb_free_datagram(sk, skb);
2882 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2883 int *uaddr_len, int peer)
2885 struct net_device *dev;
2886 struct sock *sk = sock->sk;
2891 uaddr->sa_family = AF_PACKET;
2892 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2894 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2896 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2898 *uaddr_len = sizeof(*uaddr);
2903 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2904 int *uaddr_len, int peer)
2906 struct net_device *dev;
2907 struct sock *sk = sock->sk;
2908 struct packet_sock *po = pkt_sk(sk);
2909 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2914 sll->sll_family = AF_PACKET;
2915 sll->sll_ifindex = po->ifindex;
2916 sll->sll_protocol = po->num;
2917 sll->sll_pkttype = 0;
2919 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2921 sll->sll_hatype = dev->type;
2922 sll->sll_halen = dev->addr_len;
2923 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2925 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2929 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2934 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2938 case PACKET_MR_MULTICAST:
2939 if (i->alen != dev->addr_len)
2942 return dev_mc_add(dev, i->addr);
2944 return dev_mc_del(dev, i->addr);
2946 case PACKET_MR_PROMISC:
2947 return dev_set_promiscuity(dev, what);
2949 case PACKET_MR_ALLMULTI:
2950 return dev_set_allmulti(dev, what);
2952 case PACKET_MR_UNICAST:
2953 if (i->alen != dev->addr_len)
2956 return dev_uc_add(dev, i->addr);
2958 return dev_uc_del(dev, i->addr);
2966 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2968 for ( ; i; i = i->next) {
2969 if (i->ifindex == dev->ifindex)
2970 packet_dev_mc(dev, i, what);
2974 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2976 struct packet_sock *po = pkt_sk(sk);
2977 struct packet_mclist *ml, *i;
2978 struct net_device *dev;
2984 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2989 if (mreq->mr_alen > dev->addr_len)
2993 i = kmalloc(sizeof(*i), GFP_KERNEL);
2998 for (ml = po->mclist; ml; ml = ml->next) {
2999 if (ml->ifindex == mreq->mr_ifindex &&
3000 ml->type == mreq->mr_type &&
3001 ml->alen == mreq->mr_alen &&
3002 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3004 /* Free the new element ... */
3010 i->type = mreq->mr_type;
3011 i->ifindex = mreq->mr_ifindex;
3012 i->alen = mreq->mr_alen;
3013 memcpy(i->addr, mreq->mr_address, i->alen);
3015 i->next = po->mclist;
3017 err = packet_dev_mc(dev, i, 1);
3019 po->mclist = i->next;
3028 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3030 struct packet_mclist *ml, **mlp;
3034 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3035 if (ml->ifindex == mreq->mr_ifindex &&
3036 ml->type == mreq->mr_type &&
3037 ml->alen == mreq->mr_alen &&
3038 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3039 if (--ml->count == 0) {
3040 struct net_device *dev;
3042 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3044 packet_dev_mc(dev, ml, -1);
3052 return -EADDRNOTAVAIL;
3055 static void packet_flush_mclist(struct sock *sk)
3057 struct packet_sock *po = pkt_sk(sk);
3058 struct packet_mclist *ml;
3064 while ((ml = po->mclist) != NULL) {
3065 struct net_device *dev;
3067 po->mclist = ml->next;
3068 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3070 packet_dev_mc(dev, ml, -1);
3077 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3079 struct sock *sk = sock->sk;
3080 struct packet_sock *po = pkt_sk(sk);
3083 if (level != SOL_PACKET)
3084 return -ENOPROTOOPT;
3087 case PACKET_ADD_MEMBERSHIP:
3088 case PACKET_DROP_MEMBERSHIP:
3090 struct packet_mreq_max mreq;
3092 memset(&mreq, 0, sizeof(mreq));
3093 if (len < sizeof(struct packet_mreq))
3095 if (len > sizeof(mreq))
3097 if (copy_from_user(&mreq, optval, len))
3099 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3101 if (optname == PACKET_ADD_MEMBERSHIP)
3102 ret = packet_mc_add(sk, &mreq);
3104 ret = packet_mc_drop(sk, &mreq);
3108 case PACKET_RX_RING:
3109 case PACKET_TX_RING:
3111 union tpacket_req_u req_u;
3114 switch (po->tp_version) {
3117 len = sizeof(req_u.req);
3121 len = sizeof(req_u.req3);
3126 if (pkt_sk(sk)->has_vnet_hdr)
3128 if (copy_from_user(&req_u.req, optval, len))
3130 return packet_set_ring(sk, &req_u, 0,
3131 optname == PACKET_TX_RING);
3133 case PACKET_COPY_THRESH:
3137 if (optlen != sizeof(val))
3139 if (copy_from_user(&val, optval, sizeof(val)))
3142 pkt_sk(sk)->copy_thresh = val;
3145 case PACKET_VERSION:
3149 if (optlen != sizeof(val))
3151 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3153 if (copy_from_user(&val, optval, sizeof(val)))
3159 po->tp_version = val;
3165 case PACKET_RESERVE:
3169 if (optlen != sizeof(val))
3171 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3173 if (copy_from_user(&val, optval, sizeof(val)))
3175 po->tp_reserve = val;
3182 if (optlen != sizeof(val))
3184 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3186 if (copy_from_user(&val, optval, sizeof(val)))
3188 po->tp_loss = !!val;
3191 case PACKET_AUXDATA:
3195 if (optlen < sizeof(val))
3197 if (copy_from_user(&val, optval, sizeof(val)))
3200 po->auxdata = !!val;
3203 case PACKET_ORIGDEV:
3207 if (optlen < sizeof(val))
3209 if (copy_from_user(&val, optval, sizeof(val)))
3212 po->origdev = !!val;
3215 case PACKET_VNET_HDR:
3219 if (sock->type != SOCK_RAW)
3221 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3223 if (optlen < sizeof(val))
3225 if (copy_from_user(&val, optval, sizeof(val)))
3228 po->has_vnet_hdr = !!val;
3231 case PACKET_TIMESTAMP:
3235 if (optlen != sizeof(val))
3237 if (copy_from_user(&val, optval, sizeof(val)))
3240 po->tp_tstamp = val;
3247 if (optlen != sizeof(val))
3249 if (copy_from_user(&val, optval, sizeof(val)))
3252 return fanout_add(sk, val & 0xffff, val >> 16);
3254 case PACKET_TX_HAS_OFF:
3258 if (optlen != sizeof(val))
3260 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3262 if (copy_from_user(&val, optval, sizeof(val)))
3264 po->tp_tx_has_off = !!val;
3268 return -ENOPROTOOPT;
3272 static int packet_getsockopt(struct socket *sock, int level, int optname,
3273 char __user *optval, int __user *optlen)
3276 int val, lv = sizeof(val);
3277 struct sock *sk = sock->sk;
3278 struct packet_sock *po = pkt_sk(sk);
3280 union tpacket_stats_u st;
3282 if (level != SOL_PACKET)
3283 return -ENOPROTOOPT;
3285 if (get_user(len, optlen))
3292 case PACKET_STATISTICS:
3293 spin_lock_bh(&sk->sk_receive_queue.lock);
3294 memcpy(&st, &po->stats, sizeof(st));
3295 memset(&po->stats, 0, sizeof(po->stats));
3296 spin_unlock_bh(&sk->sk_receive_queue.lock);
3298 if (po->tp_version == TPACKET_V3) {
3299 lv = sizeof(struct tpacket_stats_v3);
3300 st.stats3.tp_packets += st.stats3.tp_drops;
3303 lv = sizeof(struct tpacket_stats);
3304 st.stats1.tp_packets += st.stats1.tp_drops;
3309 case PACKET_AUXDATA:
3312 case PACKET_ORIGDEV:
3315 case PACKET_VNET_HDR:
3316 val = po->has_vnet_hdr;
3318 case PACKET_VERSION:
3319 val = po->tp_version;
3322 if (len > sizeof(int))
3324 if (copy_from_user(&val, optval, len))
3328 val = sizeof(struct tpacket_hdr);
3331 val = sizeof(struct tpacket2_hdr);
3334 val = sizeof(struct tpacket3_hdr);
3340 case PACKET_RESERVE:
3341 val = po->tp_reserve;
3346 case PACKET_TIMESTAMP:
3347 val = po->tp_tstamp;
3351 ((u32)po->fanout->id |
3352 ((u32)po->fanout->type << 16) |
3353 ((u32)po->fanout->flags << 24)) :
3356 case PACKET_TX_HAS_OFF:
3357 val = po->tp_tx_has_off;
3360 return -ENOPROTOOPT;
3365 if (put_user(len, optlen))
3367 if (copy_to_user(optval, data, len))
3373 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3376 struct net_device *dev = data;
3377 struct net *net = dev_net(dev);
3380 sk_for_each_rcu(sk, &net->packet.sklist) {
3381 struct packet_sock *po = pkt_sk(sk);
3384 case NETDEV_UNREGISTER:
3386 packet_dev_mclist(dev, po->mclist, -1);
3390 if (dev->ifindex == po->ifindex) {
3391 spin_lock(&po->bind_lock);
3393 __unregister_prot_hook(sk, false);
3394 sk->sk_err = ENETDOWN;
3395 if (!sock_flag(sk, SOCK_DEAD))
3396 sk->sk_error_report(sk);
3398 if (msg == NETDEV_UNREGISTER) {
3399 packet_cached_dev_reset(po);
3401 if (po->prot_hook.dev)
3402 dev_put(po->prot_hook.dev);
3403 po->prot_hook.dev = NULL;
3405 spin_unlock(&po->bind_lock);
3409 if (dev->ifindex == po->ifindex) {
3410 spin_lock(&po->bind_lock);
3412 register_prot_hook(sk);
3413 spin_unlock(&po->bind_lock);
3423 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3426 struct sock *sk = sock->sk;
3431 int amount = sk_wmem_alloc_get(sk);
3433 return put_user(amount, (int __user *)arg);
3437 struct sk_buff *skb;
3440 spin_lock_bh(&sk->sk_receive_queue.lock);
3441 skb = skb_peek(&sk->sk_receive_queue);
3444 spin_unlock_bh(&sk->sk_receive_queue.lock);
3445 return put_user(amount, (int __user *)arg);
3448 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3450 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3460 case SIOCGIFBRDADDR:
3461 case SIOCSIFBRDADDR:
3462 case SIOCGIFNETMASK:
3463 case SIOCSIFNETMASK:
3464 case SIOCGIFDSTADDR:
3465 case SIOCSIFDSTADDR:
3467 return inet_dgram_ops.ioctl(sock, cmd, arg);
3471 return -ENOIOCTLCMD;
3476 static unsigned int packet_poll(struct file *file, struct socket *sock,
3479 struct sock *sk = sock->sk;
3480 struct packet_sock *po = pkt_sk(sk);
3481 unsigned int mask = datagram_poll(file, sock, wait);
3483 spin_lock_bh(&sk->sk_receive_queue.lock);
3484 if (po->rx_ring.pg_vec) {
3485 if (!packet_previous_rx_frame(po, &po->rx_ring,
3487 mask |= POLLIN | POLLRDNORM;
3489 spin_unlock_bh(&sk->sk_receive_queue.lock);
3490 spin_lock_bh(&sk->sk_write_queue.lock);
3491 if (po->tx_ring.pg_vec) {
3492 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3493 mask |= POLLOUT | POLLWRNORM;
3495 spin_unlock_bh(&sk->sk_write_queue.lock);
3500 /* Dirty? Well, I still did not learn better way to account
3504 static void packet_mm_open(struct vm_area_struct *vma)
3506 struct file *file = vma->vm_file;
3507 struct socket *sock = file->private_data;
3508 struct sock *sk = sock->sk;
3511 atomic_inc(&pkt_sk(sk)->mapped);
3514 static void packet_mm_close(struct vm_area_struct *vma)
3516 struct file *file = vma->vm_file;
3517 struct socket *sock = file->private_data;
3518 struct sock *sk = sock->sk;
3521 atomic_dec(&pkt_sk(sk)->mapped);
3524 static const struct vm_operations_struct packet_mmap_ops = {
3525 .open = packet_mm_open,
3526 .close = packet_mm_close,
3529 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3534 for (i = 0; i < len; i++) {
3535 if (likely(pg_vec[i].buffer)) {
3536 if (is_vmalloc_addr(pg_vec[i].buffer))
3537 vfree(pg_vec[i].buffer);
3539 free_pages((unsigned long)pg_vec[i].buffer,
3541 pg_vec[i].buffer = NULL;
3547 static char *alloc_one_pg_vec_page(unsigned long order)
3549 char *buffer = NULL;
3550 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3551 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3553 buffer = (char *) __get_free_pages(gfp_flags, order);
3559 * __get_free_pages failed, fall back to vmalloc
3561 buffer = vzalloc((1 << order) * PAGE_SIZE);
3567 * vmalloc failed, lets dig into swap here
3569 gfp_flags &= ~__GFP_NORETRY;
3570 buffer = (char *)__get_free_pages(gfp_flags, order);
3575 * complete and utter failure
3580 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3582 unsigned int block_nr = req->tp_block_nr;
3586 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3587 if (unlikely(!pg_vec))
3590 for (i = 0; i < block_nr; i++) {
3591 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3592 if (unlikely(!pg_vec[i].buffer))
3593 goto out_free_pgvec;
3600 free_pg_vec(pg_vec, order, block_nr);
3605 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3606 int closing, int tx_ring)
3608 struct pgv *pg_vec = NULL;
3609 struct packet_sock *po = pkt_sk(sk);
3610 int was_running, order = 0;
3611 struct packet_ring_buffer *rb;
3612 struct sk_buff_head *rb_queue;
3615 /* Added to avoid minimal code churn */
3616 struct tpacket_req *req = &req_u->req;
3618 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3619 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3620 WARN(1, "Tx-ring is not supported.\n");
3624 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3625 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3629 if (atomic_read(&po->mapped))
3631 if (atomic_read(&rb->pending))
3635 if (req->tp_block_nr) {
3636 /* Sanity tests and some calculations */
3638 if (unlikely(rb->pg_vec))
3641 switch (po->tp_version) {
3643 po->tp_hdrlen = TPACKET_HDRLEN;
3646 po->tp_hdrlen = TPACKET2_HDRLEN;
3649 po->tp_hdrlen = TPACKET3_HDRLEN;
3654 if (unlikely((int)req->tp_block_size <= 0))
3656 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3658 if (po->tp_version >= TPACKET_V3 &&
3659 (int)(req->tp_block_size -
3660 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3662 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3665 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3668 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3669 if (unlikely(rb->frames_per_block <= 0))
3671 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3676 order = get_order(req->tp_block_size);
3677 pg_vec = alloc_pg_vec(req, order);
3678 if (unlikely(!pg_vec))
3680 switch (po->tp_version) {
3682 /* Transmit path is not supported. We checked
3683 * it above but just being paranoid
3686 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3695 if (unlikely(req->tp_frame_nr))
3701 /* Detach socket from network */
3702 spin_lock(&po->bind_lock);
3703 was_running = po->running;
3707 __unregister_prot_hook(sk, false);
3709 spin_unlock(&po->bind_lock);
3714 mutex_lock(&po->pg_vec_lock);
3715 if (closing || atomic_read(&po->mapped) == 0) {
3717 spin_lock_bh(&rb_queue->lock);
3718 swap(rb->pg_vec, pg_vec);
3719 rb->frame_max = (req->tp_frame_nr - 1);
3721 rb->frame_size = req->tp_frame_size;
3722 spin_unlock_bh(&rb_queue->lock);
3724 swap(rb->pg_vec_order, order);
3725 swap(rb->pg_vec_len, req->tp_block_nr);
3727 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3728 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3729 tpacket_rcv : packet_rcv;
3730 skb_queue_purge(rb_queue);
3731 if (atomic_read(&po->mapped))
3732 pr_err("packet_mmap: vma is busy: %d\n",
3733 atomic_read(&po->mapped));
3735 mutex_unlock(&po->pg_vec_lock);
3737 spin_lock(&po->bind_lock);
3740 register_prot_hook(sk);
3742 spin_unlock(&po->bind_lock);
3743 if (closing && (po->tp_version > TPACKET_V2)) {
3744 /* Because we don't support block-based V3 on tx-ring */
3746 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3751 free_pg_vec(pg_vec, order, req->tp_block_nr);
3756 static int packet_mmap(struct file *file, struct socket *sock,
3757 struct vm_area_struct *vma)
3759 struct sock *sk = sock->sk;
3760 struct packet_sock *po = pkt_sk(sk);
3761 unsigned long size, expected_size;
3762 struct packet_ring_buffer *rb;
3763 unsigned long start;
3770 mutex_lock(&po->pg_vec_lock);
3773 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3775 expected_size += rb->pg_vec_len
3781 if (expected_size == 0)
3784 size = vma->vm_end - vma->vm_start;
3785 if (size != expected_size)
3788 start = vma->vm_start;
3789 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3790 if (rb->pg_vec == NULL)
3793 for (i = 0; i < rb->pg_vec_len; i++) {
3795 void *kaddr = rb->pg_vec[i].buffer;
3798 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3799 page = pgv_to_page(kaddr);
3800 err = vm_insert_page(vma, start, page);
3809 atomic_inc(&po->mapped);
3810 vma->vm_ops = &packet_mmap_ops;
3814 mutex_unlock(&po->pg_vec_lock);
3818 static const struct proto_ops packet_ops_spkt = {
3819 .family = PF_PACKET,
3820 .owner = THIS_MODULE,
3821 .release = packet_release,
3822 .bind = packet_bind_spkt,
3823 .connect = sock_no_connect,
3824 .socketpair = sock_no_socketpair,
3825 .accept = sock_no_accept,
3826 .getname = packet_getname_spkt,
3827 .poll = datagram_poll,
3828 .ioctl = packet_ioctl,
3829 .listen = sock_no_listen,
3830 .shutdown = sock_no_shutdown,
3831 .setsockopt = sock_no_setsockopt,
3832 .getsockopt = sock_no_getsockopt,
3833 .sendmsg = packet_sendmsg_spkt,
3834 .recvmsg = packet_recvmsg,
3835 .mmap = sock_no_mmap,
3836 .sendpage = sock_no_sendpage,
3839 static const struct proto_ops packet_ops = {
3840 .family = PF_PACKET,
3841 .owner = THIS_MODULE,
3842 .release = packet_release,
3843 .bind = packet_bind,
3844 .connect = sock_no_connect,
3845 .socketpair = sock_no_socketpair,
3846 .accept = sock_no_accept,
3847 .getname = packet_getname,
3848 .poll = packet_poll,
3849 .ioctl = packet_ioctl,
3850 .listen = sock_no_listen,
3851 .shutdown = sock_no_shutdown,
3852 .setsockopt = packet_setsockopt,
3853 .getsockopt = packet_getsockopt,
3854 .sendmsg = packet_sendmsg,
3855 .recvmsg = packet_recvmsg,
3856 .mmap = packet_mmap,
3857 .sendpage = sock_no_sendpage,
3860 static const struct net_proto_family packet_family_ops = {
3861 .family = PF_PACKET,
3862 .create = packet_create,
3863 .owner = THIS_MODULE,
3866 static struct notifier_block packet_netdev_notifier = {
3867 .notifier_call = packet_notifier,
3870 #ifdef CONFIG_PROC_FS
3872 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3875 struct net *net = seq_file_net(seq);
3878 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3881 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3883 struct net *net = seq_file_net(seq);
3884 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3887 static void packet_seq_stop(struct seq_file *seq, void *v)
3893 static int packet_seq_show(struct seq_file *seq, void *v)
3895 if (v == SEQ_START_TOKEN)
3896 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3898 struct sock *s = sk_entry(v);
3899 const struct packet_sock *po = pkt_sk(s);
3902 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3904 atomic_read(&s->sk_refcnt),
3909 atomic_read(&s->sk_rmem_alloc),
3910 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3917 static const struct seq_operations packet_seq_ops = {
3918 .start = packet_seq_start,
3919 .next = packet_seq_next,
3920 .stop = packet_seq_stop,
3921 .show = packet_seq_show,
3924 static int packet_seq_open(struct inode *inode, struct file *file)
3926 return seq_open_net(inode, file, &packet_seq_ops,
3927 sizeof(struct seq_net_private));
3930 static const struct file_operations packet_seq_fops = {
3931 .owner = THIS_MODULE,
3932 .open = packet_seq_open,
3934 .llseek = seq_lseek,
3935 .release = seq_release_net,
3940 static int __net_init packet_net_init(struct net *net)
3942 mutex_init(&net->packet.sklist_lock);
3943 INIT_HLIST_HEAD(&net->packet.sklist);
3945 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3951 static void __net_exit packet_net_exit(struct net *net)
3953 remove_proc_entry("packet", net->proc_net);
3956 static struct pernet_operations packet_net_ops = {
3957 .init = packet_net_init,
3958 .exit = packet_net_exit,
3962 static void __exit packet_exit(void)
3964 unregister_netdevice_notifier(&packet_netdev_notifier);
3965 unregister_pernet_subsys(&packet_net_ops);
3966 sock_unregister(PF_PACKET);
3967 proto_unregister(&packet_proto);
3970 static int __init packet_init(void)
3972 int rc = proto_register(&packet_proto, 0);
3977 sock_register(&packet_family_ops);
3978 register_pernet_subsys(&packet_net_ops);
3979 register_netdevice_notifier(&packet_netdev_notifier);
3984 module_init(packet_init);
3985 module_exit(packet_exit);
3986 MODULE_LICENSE("GPL");
3987 MODULE_ALIAS_NETPROTO(PF_PACKET);