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 prb_init_ft_ops(p1, req_u);
569 prb_setup_retire_blk_timer(po, tx_ring);
570 prb_open_block(p1, pbd);
573 /* Do NOT update the last_blk_num first.
574 * Assumes sk_buff_head lock is held.
576 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
578 mod_timer(&pkc->retire_blk_timer,
579 jiffies + pkc->tov_in_jiffies);
580 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
585 * 1) We refresh the timer only when we open a block.
586 * By doing this we don't waste cycles refreshing the timer
587 * on packet-by-packet basis.
589 * With a 1MB block-size, on a 1Gbps line, it will take
590 * i) ~8 ms to fill a block + ii) memcpy etc.
591 * In this cut we are not accounting for the memcpy time.
593 * So, if the user sets the 'tmo' to 10ms then the timer
594 * will never fire while the block is still getting filled
595 * (which is what we want). However, the user could choose
596 * to close a block early and that's fine.
598 * But when the timer does fire, we check whether or not to refresh it.
599 * Since the tmo granularity is in msecs, it is not too expensive
600 * to refresh the timer, lets say every '8' msecs.
601 * Either the user can set the 'tmo' or we can derive it based on
602 * a) line-speed and b) block-size.
603 * prb_calc_retire_blk_tmo() calculates the tmo.
606 static void prb_retire_rx_blk_timer_expired(unsigned long data)
608 struct packet_sock *po = (struct packet_sock *)data;
609 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
611 struct tpacket_block_desc *pbd;
613 spin_lock(&po->sk.sk_receive_queue.lock);
615 frozen = prb_queue_frozen(pkc);
616 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
618 if (unlikely(pkc->delete_blk_timer))
621 /* We only need to plug the race when the block is partially filled.
623 * lock(); increment BLOCK_NUM_PKTS; unlock()
624 * copy_bits() is in progress ...
625 * timer fires on other cpu:
626 * we can't retire the current block because copy_bits
630 if (BLOCK_NUM_PKTS(pbd)) {
631 while (atomic_read(&pkc->blk_fill_in_prog)) {
632 /* Waiting for skb_copy_bits to finish... */
637 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
639 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
640 if (!prb_dispatch_next_block(pkc, po))
645 /* Case 1. Queue was frozen because user-space was
648 if (prb_curr_blk_in_use(pkc, pbd)) {
650 * Ok, user-space is still behind.
651 * So just refresh the timer.
655 /* Case 2. queue was frozen,user-space caught up,
656 * now the link went idle && the timer fired.
657 * We don't have a block to close.So we open this
658 * block and restart the timer.
659 * opening a block thaws the queue,restarts timer
660 * Thawing/timer-refresh is a side effect.
662 prb_open_block(pkc, pbd);
669 _prb_refresh_rx_retire_blk_timer(pkc);
672 spin_unlock(&po->sk.sk_receive_queue.lock);
675 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
676 struct tpacket_block_desc *pbd1, __u32 status)
678 /* Flush everything minus the block header */
680 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
685 /* Skip the block header(we know header WILL fit in 4K) */
688 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
689 for (; start < end; start += PAGE_SIZE)
690 flush_dcache_page(pgv_to_page(start));
695 /* Now update the block status. */
697 BLOCK_STATUS(pbd1) = status;
699 /* Flush the block header */
701 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
703 flush_dcache_page(pgv_to_page(start));
713 * 2) Increment active_blk_num
715 * Note:We DONT refresh the timer on purpose.
716 * Because almost always the next block will be opened.
718 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
719 struct tpacket_block_desc *pbd1,
720 struct packet_sock *po, unsigned int stat)
722 __u32 status = TP_STATUS_USER | stat;
724 struct tpacket3_hdr *last_pkt;
725 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
727 if (po->stats.stats3.tp_drops)
728 status |= TP_STATUS_LOSING;
730 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
731 last_pkt->tp_next_offset = 0;
733 /* Get the ts of the last pkt */
734 if (BLOCK_NUM_PKTS(pbd1)) {
735 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
736 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
738 /* Ok, we tmo'd - so get the current time */
741 h1->ts_last_pkt.ts_sec = ts.tv_sec;
742 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
747 /* Flush the block */
748 prb_flush_block(pkc1, pbd1, status);
750 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
753 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
755 pkc->reset_pending_on_curr_blk = 0;
759 * Side effect of opening a block:
761 * 1) prb_queue is thawed.
762 * 2) retire_blk_timer is refreshed.
765 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
766 struct tpacket_block_desc *pbd1)
769 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
773 /* We could have just memset this but we will lose the
774 * flexibility of making the priv area sticky
777 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
778 BLOCK_NUM_PKTS(pbd1) = 0;
779 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
783 h1->ts_first_pkt.ts_sec = ts.tv_sec;
784 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
786 pkc1->pkblk_start = (char *)pbd1;
787 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
789 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
790 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
792 pbd1->version = pkc1->version;
793 pkc1->prev = pkc1->nxt_offset;
794 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
796 prb_thaw_queue(pkc1);
797 _prb_refresh_rx_retire_blk_timer(pkc1);
803 * Queue freeze logic:
804 * 1) Assume tp_block_nr = 8 blocks.
805 * 2) At time 't0', user opens Rx ring.
806 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
807 * 4) user-space is either sleeping or processing block '0'.
808 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
809 * it will close block-7,loop around and try to fill block '0'.
811 * __packet_lookup_frame_in_block
812 * prb_retire_current_block()
813 * prb_dispatch_next_block()
814 * |->(BLOCK_STATUS == USER) evaluates to true
815 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
816 * 6) Now there are two cases:
817 * 6.1) Link goes idle right after the queue is frozen.
818 * But remember, the last open_block() refreshed the timer.
819 * When this timer expires,it will refresh itself so that we can
820 * re-open block-0 in near future.
821 * 6.2) Link is busy and keeps on receiving packets. This is a simple
822 * case and __packet_lookup_frame_in_block will check if block-0
823 * is free and can now be re-used.
825 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
826 struct packet_sock *po)
828 pkc->reset_pending_on_curr_blk = 1;
829 po->stats.stats3.tp_freeze_q_cnt++;
832 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
835 * If the next block is free then we will dispatch it
836 * and return a good offset.
837 * Else, we will freeze the queue.
838 * So, caller must check the return value.
840 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
841 struct packet_sock *po)
843 struct tpacket_block_desc *pbd;
847 /* 1. Get current block num */
848 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
850 /* 2. If this block is currently in_use then freeze the queue */
851 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
852 prb_freeze_queue(pkc, po);
858 * open this block and return the offset where the first packet
859 * needs to get stored.
861 prb_open_block(pkc, pbd);
862 return (void *)pkc->nxt_offset;
865 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
866 struct packet_sock *po, unsigned int status)
868 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
870 /* retire/close the current block */
871 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
873 * Plug the case where copy_bits() is in progress on
874 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
875 * have space to copy the pkt in the current block and
876 * called prb_retire_current_block()
878 * We don't need to worry about the TMO case because
879 * the timer-handler already handled this case.
881 if (!(status & TP_STATUS_BLK_TMO)) {
882 while (atomic_read(&pkc->blk_fill_in_prog)) {
883 /* Waiting for skb_copy_bits to finish... */
887 prb_close_block(pkc, pbd, po, status);
892 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
893 struct tpacket_block_desc *pbd)
895 return TP_STATUS_USER & BLOCK_STATUS(pbd);
898 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
900 return pkc->reset_pending_on_curr_blk;
903 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
905 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
906 atomic_dec(&pkc->blk_fill_in_prog);
909 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
910 struct tpacket3_hdr *ppd)
912 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
915 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
916 struct tpacket3_hdr *ppd)
918 ppd->hv1.tp_rxhash = 0;
921 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
922 struct tpacket3_hdr *ppd)
924 if (vlan_tx_tag_present(pkc->skb)) {
925 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
926 ppd->tp_status = TP_STATUS_VLAN_VALID;
928 ppd->hv1.tp_vlan_tci = 0;
929 ppd->tp_status = TP_STATUS_AVAILABLE;
933 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
934 struct tpacket3_hdr *ppd)
936 prb_fill_vlan_info(pkc, ppd);
938 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
939 prb_fill_rxhash(pkc, ppd);
941 prb_clear_rxhash(pkc, ppd);
944 static void prb_fill_curr_block(char *curr,
945 struct tpacket_kbdq_core *pkc,
946 struct tpacket_block_desc *pbd,
949 struct tpacket3_hdr *ppd;
951 ppd = (struct tpacket3_hdr *)curr;
952 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
954 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
955 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
956 BLOCK_NUM_PKTS(pbd) += 1;
957 atomic_inc(&pkc->blk_fill_in_prog);
958 prb_run_all_ft_ops(pkc, ppd);
961 /* Assumes caller has the sk->rx_queue.lock */
962 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
968 struct tpacket_kbdq_core *pkc;
969 struct tpacket_block_desc *pbd;
972 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
973 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
975 /* Queue is frozen when user space is lagging behind */
976 if (prb_queue_frozen(pkc)) {
978 * Check if that last block which caused the queue to freeze,
979 * is still in_use by user-space.
981 if (prb_curr_blk_in_use(pkc, pbd)) {
982 /* Can't record this packet */
986 * Ok, the block was released by user-space.
987 * Now let's open that block.
988 * opening a block also thaws the queue.
989 * Thawing is a side effect.
991 prb_open_block(pkc, pbd);
996 curr = pkc->nxt_offset;
998 end = (char *)pbd + pkc->kblk_size;
1000 /* first try the current block */
1001 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1002 prb_fill_curr_block(curr, pkc, pbd, len);
1003 return (void *)curr;
1006 /* Ok, close the current block */
1007 prb_retire_current_block(pkc, po, 0);
1009 /* Now, try to dispatch the next block */
1010 curr = (char *)prb_dispatch_next_block(pkc, po);
1012 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1013 prb_fill_curr_block(curr, pkc, pbd, len);
1014 return (void *)curr;
1018 * No free blocks are available.user_space hasn't caught up yet.
1019 * Queue was just frozen and now this packet will get dropped.
1024 static void *packet_current_rx_frame(struct packet_sock *po,
1025 struct sk_buff *skb,
1026 int status, unsigned int len)
1029 switch (po->tp_version) {
1032 curr = packet_lookup_frame(po, &po->rx_ring,
1033 po->rx_ring.head, status);
1036 return __packet_lookup_frame_in_block(po, skb, status, len);
1038 WARN(1, "TPACKET version not supported\n");
1044 static void *prb_lookup_block(struct packet_sock *po,
1045 struct packet_ring_buffer *rb,
1049 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1050 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1052 if (status != BLOCK_STATUS(pbd))
1057 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1060 if (rb->prb_bdqc.kactive_blk_num)
1061 prev = rb->prb_bdqc.kactive_blk_num-1;
1063 prev = rb->prb_bdqc.knum_blocks-1;
1067 /* Assumes caller has held the rx_queue.lock */
1068 static void *__prb_previous_block(struct packet_sock *po,
1069 struct packet_ring_buffer *rb,
1072 unsigned int previous = prb_previous_blk_num(rb);
1073 return prb_lookup_block(po, rb, previous, status);
1076 static void *packet_previous_rx_frame(struct packet_sock *po,
1077 struct packet_ring_buffer *rb,
1080 if (po->tp_version <= TPACKET_V2)
1081 return packet_previous_frame(po, rb, status);
1083 return __prb_previous_block(po, rb, status);
1086 static void packet_increment_rx_head(struct packet_sock *po,
1087 struct packet_ring_buffer *rb)
1089 switch (po->tp_version) {
1092 return packet_increment_head(rb);
1095 WARN(1, "TPACKET version not supported.\n");
1101 static void *packet_previous_frame(struct packet_sock *po,
1102 struct packet_ring_buffer *rb,
1105 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1106 return packet_lookup_frame(po, rb, previous, status);
1109 static void packet_increment_head(struct packet_ring_buffer *buff)
1111 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1114 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1116 struct sock *sk = &po->sk;
1119 if (po->prot_hook.func != tpacket_rcv)
1120 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1123 spin_lock(&sk->sk_receive_queue.lock);
1124 if (po->tp_version == TPACKET_V3)
1125 has_room = prb_lookup_block(po, &po->rx_ring,
1126 po->rx_ring.prb_bdqc.kactive_blk_num,
1129 has_room = packet_lookup_frame(po, &po->rx_ring,
1132 spin_unlock(&sk->sk_receive_queue.lock);
1137 static void packet_sock_destruct(struct sock *sk)
1139 skb_queue_purge(&sk->sk_error_queue);
1141 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1142 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1144 if (!sock_flag(sk, SOCK_DEAD)) {
1145 pr_err("Attempt to release alive packet socket: %p\n", sk);
1149 sk_refcnt_debug_dec(sk);
1152 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1154 int x = atomic_read(&f->rr_cur) + 1;
1162 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1163 struct sk_buff *skb,
1166 return (((u64)skb->rxhash) * num) >> 32;
1169 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1170 struct sk_buff *skb,
1175 cur = atomic_read(&f->rr_cur);
1176 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1177 fanout_rr_next(f, num))) != cur)
1182 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1183 struct sk_buff *skb,
1186 return smp_processor_id() % num;
1189 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1190 struct sk_buff *skb,
1191 unsigned int idx, unsigned int skip,
1196 i = j = min_t(int, f->next[idx], num - 1);
1198 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1210 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1212 return f->flags & (flag >> 8);
1215 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1216 struct packet_type *pt, struct net_device *orig_dev)
1218 struct packet_fanout *f = pt->af_packet_priv;
1219 unsigned int num = f->num_members;
1220 struct packet_sock *po;
1223 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1230 case PACKET_FANOUT_HASH:
1232 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1233 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1237 skb_get_rxhash(skb);
1238 idx = fanout_demux_hash(f, skb, num);
1240 case PACKET_FANOUT_LB:
1241 idx = fanout_demux_lb(f, skb, num);
1243 case PACKET_FANOUT_CPU:
1244 idx = fanout_demux_cpu(f, skb, num);
1246 case PACKET_FANOUT_ROLLOVER:
1247 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1251 po = pkt_sk(f->arr[idx]);
1252 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1253 unlikely(!packet_rcv_has_room(po, skb))) {
1254 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1255 po = pkt_sk(f->arr[idx]);
1258 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1261 DEFINE_MUTEX(fanout_mutex);
1262 EXPORT_SYMBOL_GPL(fanout_mutex);
1263 static LIST_HEAD(fanout_list);
1265 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1267 struct packet_fanout *f = po->fanout;
1269 spin_lock(&f->lock);
1270 f->arr[f->num_members] = sk;
1273 spin_unlock(&f->lock);
1276 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1278 struct packet_fanout *f = po->fanout;
1281 spin_lock(&f->lock);
1282 for (i = 0; i < f->num_members; i++) {
1283 if (f->arr[i] == sk)
1286 BUG_ON(i >= f->num_members);
1287 f->arr[i] = f->arr[f->num_members - 1];
1289 spin_unlock(&f->lock);
1292 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1294 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1300 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1302 struct packet_sock *po = pkt_sk(sk);
1303 struct packet_fanout *f, *match;
1304 u8 type = type_flags & 0xff;
1305 u8 flags = type_flags >> 8;
1309 case PACKET_FANOUT_ROLLOVER:
1310 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1312 case PACKET_FANOUT_HASH:
1313 case PACKET_FANOUT_LB:
1314 case PACKET_FANOUT_CPU:
1326 mutex_lock(&fanout_mutex);
1328 list_for_each_entry(f, &fanout_list, list) {
1330 read_pnet(&f->net) == sock_net(sk)) {
1336 if (match && match->flags != flags)
1340 match = kzalloc(sizeof(*match), GFP_KERNEL);
1343 write_pnet(&match->net, sock_net(sk));
1346 match->flags = flags;
1347 atomic_set(&match->rr_cur, 0);
1348 INIT_LIST_HEAD(&match->list);
1349 spin_lock_init(&match->lock);
1350 atomic_set(&match->sk_ref, 0);
1351 match->prot_hook.type = po->prot_hook.type;
1352 match->prot_hook.dev = po->prot_hook.dev;
1353 match->prot_hook.func = packet_rcv_fanout;
1354 match->prot_hook.af_packet_priv = match;
1355 match->prot_hook.id_match = match_fanout_group;
1356 dev_add_pack(&match->prot_hook);
1357 list_add(&match->list, &fanout_list);
1360 if (match->type == type &&
1361 match->prot_hook.type == po->prot_hook.type &&
1362 match->prot_hook.dev == po->prot_hook.dev) {
1364 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1365 __dev_remove_pack(&po->prot_hook);
1367 atomic_inc(&match->sk_ref);
1368 __fanout_link(sk, po);
1373 mutex_unlock(&fanout_mutex);
1377 static void fanout_release(struct sock *sk)
1379 struct packet_sock *po = pkt_sk(sk);
1380 struct packet_fanout *f;
1386 mutex_lock(&fanout_mutex);
1389 if (atomic_dec_and_test(&f->sk_ref)) {
1391 dev_remove_pack(&f->prot_hook);
1394 mutex_unlock(&fanout_mutex);
1397 static const struct proto_ops packet_ops;
1399 static const struct proto_ops packet_ops_spkt;
1401 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1402 struct packet_type *pt, struct net_device *orig_dev)
1405 struct sockaddr_pkt *spkt;
1408 * When we registered the protocol we saved the socket in the data
1409 * field for just this event.
1412 sk = pt->af_packet_priv;
1415 * Yank back the headers [hope the device set this
1416 * right or kerboom...]
1418 * Incoming packets have ll header pulled,
1421 * For outgoing ones skb->data == skb_mac_header(skb)
1422 * so that this procedure is noop.
1425 if (skb->pkt_type == PACKET_LOOPBACK)
1428 if (!net_eq(dev_net(dev), sock_net(sk)))
1431 skb = skb_share_check(skb, GFP_ATOMIC);
1435 /* drop any routing info */
1438 /* drop conntrack reference */
1441 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1443 skb_push(skb, skb->data - skb_mac_header(skb));
1446 * The SOCK_PACKET socket receives _all_ frames.
1449 spkt->spkt_family = dev->type;
1450 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1451 spkt->spkt_protocol = skb->protocol;
1454 * Charge the memory to the socket. This is done specifically
1455 * to prevent sockets using all the memory up.
1458 if (sock_queue_rcv_skb(sk, skb) == 0)
1469 * Output a raw packet to a device layer. This bypasses all the other
1470 * protocol layers and you must therefore supply it with a complete frame
1473 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1474 struct msghdr *msg, size_t len)
1476 struct sock *sk = sock->sk;
1477 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1478 struct sk_buff *skb = NULL;
1479 struct net_device *dev;
1485 * Get and verify the address.
1489 if (msg->msg_namelen < sizeof(struct sockaddr))
1491 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1492 proto = saddr->spkt_protocol;
1494 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1497 * Find the device first to size check it
1500 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1503 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1509 if (!(dev->flags & IFF_UP))
1513 * You may not queue a frame bigger than the mtu. This is the lowest level
1514 * raw protocol and you must do your own fragmentation at this level.
1517 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1518 if (!netif_supports_nofcs(dev)) {
1519 err = -EPROTONOSUPPORT;
1522 extra_len = 4; /* We're doing our own CRC */
1526 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1530 size_t reserved = LL_RESERVED_SPACE(dev);
1531 int tlen = dev->needed_tailroom;
1532 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1535 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1538 /* FIXME: Save some space for broken drivers that write a hard
1539 * header at transmission time by themselves. PPP is the notable
1540 * one here. This should really be fixed at the driver level.
1542 skb_reserve(skb, reserved);
1543 skb_reset_network_header(skb);
1545 /* Try to align data part correctly */
1550 skb_reset_network_header(skb);
1552 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1558 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1559 /* Earlier code assumed this would be a VLAN pkt,
1560 * double-check this now that we have the actual
1563 struct ethhdr *ehdr;
1564 skb_reset_mac_header(skb);
1565 ehdr = eth_hdr(skb);
1566 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1572 skb->protocol = proto;
1574 skb->priority = sk->sk_priority;
1575 skb->mark = sk->sk_mark;
1577 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1579 if (unlikely(extra_len == 4))
1582 skb_probe_transport_header(skb, 0);
1584 dev_queue_xmit(skb);
1595 static unsigned int run_filter(const struct sk_buff *skb,
1596 const struct sock *sk,
1599 struct sk_filter *filter;
1602 filter = rcu_dereference(sk->sk_filter);
1604 res = SK_RUN_FILTER(filter, skb);
1611 * This function makes lazy skb cloning in hope that most of packets
1612 * are discarded by BPF.
1614 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1615 * and skb->cb are mangled. It works because (and until) packets
1616 * falling here are owned by current CPU. Output packets are cloned
1617 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1618 * sequencially, so that if we return skb to original state on exit,
1619 * we will not harm anyone.
1622 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1623 struct packet_type *pt, struct net_device *orig_dev)
1626 struct sockaddr_ll *sll;
1627 struct packet_sock *po;
1628 u8 *skb_head = skb->data;
1629 int skb_len = skb->len;
1630 unsigned int snaplen, res;
1632 if (skb->pkt_type == PACKET_LOOPBACK)
1635 sk = pt->af_packet_priv;
1638 if (!net_eq(dev_net(dev), sock_net(sk)))
1643 if (dev->header_ops) {
1644 /* The device has an explicit notion of ll header,
1645 * exported to higher levels.
1647 * Otherwise, the device hides details of its frame
1648 * structure, so that corresponding packet head is
1649 * never delivered to user.
1651 if (sk->sk_type != SOCK_DGRAM)
1652 skb_push(skb, skb->data - skb_mac_header(skb));
1653 else if (skb->pkt_type == PACKET_OUTGOING) {
1654 /* Special case: outgoing packets have ll header at head */
1655 skb_pull(skb, skb_network_offset(skb));
1661 res = run_filter(skb, sk, snaplen);
1663 goto drop_n_restore;
1667 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1670 if (skb_shared(skb)) {
1671 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1675 if (skb_head != skb->data) {
1676 skb->data = skb_head;
1683 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1686 sll = &PACKET_SKB_CB(skb)->sa.ll;
1687 sll->sll_family = AF_PACKET;
1688 sll->sll_hatype = dev->type;
1689 sll->sll_protocol = skb->protocol;
1690 sll->sll_pkttype = skb->pkt_type;
1691 if (unlikely(po->origdev))
1692 sll->sll_ifindex = orig_dev->ifindex;
1694 sll->sll_ifindex = dev->ifindex;
1696 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1698 PACKET_SKB_CB(skb)->origlen = skb->len;
1700 if (pskb_trim(skb, snaplen))
1703 skb_set_owner_r(skb, sk);
1707 /* drop conntrack reference */
1710 spin_lock(&sk->sk_receive_queue.lock);
1711 po->stats.stats1.tp_packets++;
1712 skb->dropcount = atomic_read(&sk->sk_drops);
1713 __skb_queue_tail(&sk->sk_receive_queue, skb);
1714 spin_unlock(&sk->sk_receive_queue.lock);
1715 sk->sk_data_ready(sk, skb->len);
1719 spin_lock(&sk->sk_receive_queue.lock);
1720 po->stats.stats1.tp_drops++;
1721 atomic_inc(&sk->sk_drops);
1722 spin_unlock(&sk->sk_receive_queue.lock);
1725 if (skb_head != skb->data && skb_shared(skb)) {
1726 skb->data = skb_head;
1734 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1735 struct packet_type *pt, struct net_device *orig_dev)
1738 struct packet_sock *po;
1739 struct sockaddr_ll *sll;
1740 union tpacket_uhdr h;
1741 u8 *skb_head = skb->data;
1742 int skb_len = skb->len;
1743 unsigned int snaplen, res;
1744 unsigned long status = TP_STATUS_USER;
1745 unsigned short macoff, netoff, hdrlen;
1746 struct sk_buff *copy_skb = NULL;
1750 if (skb->pkt_type == PACKET_LOOPBACK)
1753 sk = pt->af_packet_priv;
1756 if (!net_eq(dev_net(dev), sock_net(sk)))
1759 if (dev->header_ops) {
1760 if (sk->sk_type != SOCK_DGRAM)
1761 skb_push(skb, skb->data - skb_mac_header(skb));
1762 else if (skb->pkt_type == PACKET_OUTGOING) {
1763 /* Special case: outgoing packets have ll header at head */
1764 skb_pull(skb, skb_network_offset(skb));
1768 if (skb->ip_summed == CHECKSUM_PARTIAL)
1769 status |= TP_STATUS_CSUMNOTREADY;
1773 res = run_filter(skb, sk, snaplen);
1775 goto drop_n_restore;
1779 if (sk->sk_type == SOCK_DGRAM) {
1780 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1783 unsigned int maclen = skb_network_offset(skb);
1784 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1785 (maclen < 16 ? 16 : maclen)) +
1787 macoff = netoff - maclen;
1789 if (po->tp_version <= TPACKET_V2) {
1790 if (macoff + snaplen > po->rx_ring.frame_size) {
1791 if (po->copy_thresh &&
1792 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1793 if (skb_shared(skb)) {
1794 copy_skb = skb_clone(skb, GFP_ATOMIC);
1796 copy_skb = skb_get(skb);
1797 skb_head = skb->data;
1800 skb_set_owner_r(copy_skb, sk);
1802 snaplen = po->rx_ring.frame_size - macoff;
1803 if ((int)snaplen < 0)
1807 spin_lock(&sk->sk_receive_queue.lock);
1808 h.raw = packet_current_rx_frame(po, skb,
1809 TP_STATUS_KERNEL, (macoff+snaplen));
1812 if (po->tp_version <= TPACKET_V2) {
1813 packet_increment_rx_head(po, &po->rx_ring);
1815 * LOSING will be reported till you read the stats,
1816 * because it's COR - Clear On Read.
1817 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1820 if (po->stats.stats1.tp_drops)
1821 status |= TP_STATUS_LOSING;
1823 po->stats.stats1.tp_packets++;
1825 status |= TP_STATUS_COPY;
1826 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1828 spin_unlock(&sk->sk_receive_queue.lock);
1830 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1832 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1833 getnstimeofday(&ts);
1835 status |= ts_status;
1837 switch (po->tp_version) {
1839 h.h1->tp_len = skb->len;
1840 h.h1->tp_snaplen = snaplen;
1841 h.h1->tp_mac = macoff;
1842 h.h1->tp_net = netoff;
1843 h.h1->tp_sec = ts.tv_sec;
1844 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1845 hdrlen = sizeof(*h.h1);
1848 h.h2->tp_len = skb->len;
1849 h.h2->tp_snaplen = snaplen;
1850 h.h2->tp_mac = macoff;
1851 h.h2->tp_net = netoff;
1852 h.h2->tp_sec = ts.tv_sec;
1853 h.h2->tp_nsec = ts.tv_nsec;
1854 if (vlan_tx_tag_present(skb)) {
1855 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1856 status |= TP_STATUS_VLAN_VALID;
1858 h.h2->tp_vlan_tci = 0;
1860 h.h2->tp_padding = 0;
1861 hdrlen = sizeof(*h.h2);
1864 /* tp_nxt_offset,vlan are already populated above.
1865 * So DONT clear those fields here
1867 h.h3->tp_status |= status;
1868 h.h3->tp_len = skb->len;
1869 h.h3->tp_snaplen = snaplen;
1870 h.h3->tp_mac = macoff;
1871 h.h3->tp_net = netoff;
1872 h.h3->tp_sec = ts.tv_sec;
1873 h.h3->tp_nsec = ts.tv_nsec;
1874 hdrlen = sizeof(*h.h3);
1880 sll = h.raw + TPACKET_ALIGN(hdrlen);
1881 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1882 sll->sll_family = AF_PACKET;
1883 sll->sll_hatype = dev->type;
1884 sll->sll_protocol = skb->protocol;
1885 sll->sll_pkttype = skb->pkt_type;
1886 if (unlikely(po->origdev))
1887 sll->sll_ifindex = orig_dev->ifindex;
1889 sll->sll_ifindex = dev->ifindex;
1892 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1896 if (po->tp_version <= TPACKET_V2) {
1897 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1898 + macoff + snaplen);
1899 for (start = h.raw; start < end; start += PAGE_SIZE)
1900 flush_dcache_page(pgv_to_page(start));
1905 if (po->tp_version <= TPACKET_V2)
1906 __packet_set_status(po, h.raw, status);
1908 prb_clear_blk_fill_status(&po->rx_ring);
1910 sk->sk_data_ready(sk, 0);
1913 if (skb_head != skb->data && skb_shared(skb)) {
1914 skb->data = skb_head;
1922 po->stats.stats1.tp_drops++;
1923 spin_unlock(&sk->sk_receive_queue.lock);
1925 sk->sk_data_ready(sk, 0);
1926 kfree_skb(copy_skb);
1927 goto drop_n_restore;
1930 static void tpacket_destruct_skb(struct sk_buff *skb)
1932 struct packet_sock *po = pkt_sk(skb->sk);
1935 if (likely(po->tx_ring.pg_vec)) {
1938 ph = skb_shinfo(skb)->destructor_arg;
1939 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1940 atomic_dec(&po->tx_ring.pending);
1942 ts = __packet_set_timestamp(po, ph, skb);
1943 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1949 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1950 void *frame, struct net_device *dev, int size_max,
1951 __be16 proto, unsigned char *addr, int hlen)
1953 union tpacket_uhdr ph;
1954 int to_write, offset, len, tp_len, nr_frags, len_max;
1955 struct socket *sock = po->sk.sk_socket;
1962 skb->protocol = proto;
1964 skb->priority = po->sk.sk_priority;
1965 skb->mark = po->sk.sk_mark;
1966 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1967 skb_shinfo(skb)->destructor_arg = ph.raw;
1969 switch (po->tp_version) {
1971 tp_len = ph.h2->tp_len;
1974 tp_len = ph.h1->tp_len;
1977 if (unlikely(tp_len > size_max)) {
1978 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1982 skb_reserve(skb, hlen);
1983 skb_reset_network_header(skb);
1984 skb_probe_transport_header(skb, 0);
1986 if (po->tp_tx_has_off) {
1987 int off_min, off_max, off;
1988 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1989 off_max = po->tx_ring.frame_size - tp_len;
1990 if (sock->type == SOCK_DGRAM) {
1991 switch (po->tp_version) {
1993 off = ph.h2->tp_net;
1996 off = ph.h1->tp_net;
2000 switch (po->tp_version) {
2002 off = ph.h2->tp_mac;
2005 off = ph.h1->tp_mac;
2009 if (unlikely((off < off_min) || (off_max < off)))
2011 data = ph.raw + off;
2013 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2017 if (sock->type == SOCK_DGRAM) {
2018 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2020 if (unlikely(err < 0))
2022 } else if (dev->hard_header_len) {
2023 /* net device doesn't like empty head */
2024 if (unlikely(tp_len <= dev->hard_header_len)) {
2025 pr_err("packet size is too short (%d < %d)\n",
2026 tp_len, dev->hard_header_len);
2030 skb_push(skb, dev->hard_header_len);
2031 err = skb_store_bits(skb, 0, data,
2032 dev->hard_header_len);
2036 data += dev->hard_header_len;
2037 to_write -= dev->hard_header_len;
2040 offset = offset_in_page(data);
2041 len_max = PAGE_SIZE - offset;
2042 len = ((to_write > len_max) ? len_max : to_write);
2044 skb->data_len = to_write;
2045 skb->len += to_write;
2046 skb->truesize += to_write;
2047 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2049 while (likely(to_write)) {
2050 nr_frags = skb_shinfo(skb)->nr_frags;
2052 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2053 pr_err("Packet exceed the number of skb frags(%lu)\n",
2058 page = pgv_to_page(data);
2060 flush_dcache_page(page);
2062 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2065 len_max = PAGE_SIZE;
2066 len = ((to_write > len_max) ? len_max : to_write);
2072 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2074 struct sk_buff *skb;
2075 struct net_device *dev;
2077 int err, reserve = 0;
2079 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2080 int tp_len, size_max;
2081 unsigned char *addr;
2083 int status = TP_STATUS_AVAILABLE;
2086 mutex_lock(&po->pg_vec_lock);
2088 if (likely(saddr == NULL)) {
2089 dev = packet_cached_dev_get(po);
2094 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2096 if (msg->msg_namelen < (saddr->sll_halen
2097 + offsetof(struct sockaddr_ll,
2100 proto = saddr->sll_protocol;
2101 addr = saddr->sll_addr;
2102 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2106 if (unlikely(dev == NULL))
2109 if (unlikely(!(dev->flags & IFF_UP)))
2112 reserve = dev->hard_header_len;
2114 size_max = po->tx_ring.frame_size
2115 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2117 if (size_max > dev->mtu + reserve)
2118 size_max = dev->mtu + reserve;
2121 ph = packet_current_frame(po, &po->tx_ring,
2122 TP_STATUS_SEND_REQUEST);
2124 if (unlikely(ph == NULL)) {
2129 status = TP_STATUS_SEND_REQUEST;
2130 hlen = LL_RESERVED_SPACE(dev);
2131 tlen = dev->needed_tailroom;
2132 skb = sock_alloc_send_skb(&po->sk,
2133 hlen + tlen + sizeof(struct sockaddr_ll),
2136 if (unlikely(skb == NULL))
2139 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2142 if (unlikely(tp_len < 0)) {
2144 __packet_set_status(po, ph,
2145 TP_STATUS_AVAILABLE);
2146 packet_increment_head(&po->tx_ring);
2150 status = TP_STATUS_WRONG_FORMAT;
2156 skb->destructor = tpacket_destruct_skb;
2157 __packet_set_status(po, ph, TP_STATUS_SENDING);
2158 atomic_inc(&po->tx_ring.pending);
2160 status = TP_STATUS_SEND_REQUEST;
2161 err = dev_queue_xmit(skb);
2162 if (unlikely(err > 0)) {
2163 err = net_xmit_errno(err);
2164 if (err && __packet_get_status(po, ph) ==
2165 TP_STATUS_AVAILABLE) {
2166 /* skb was destructed already */
2171 * skb was dropped but not destructed yet;
2172 * let's treat it like congestion or err < 0
2176 packet_increment_head(&po->tx_ring);
2178 } while (likely((ph != NULL) ||
2179 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2180 (atomic_read(&po->tx_ring.pending))))
2187 __packet_set_status(po, ph, status);
2192 mutex_unlock(&po->pg_vec_lock);
2196 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2197 size_t reserve, size_t len,
2198 size_t linear, int noblock,
2201 struct sk_buff *skb;
2203 /* Under a page? Don't bother with paged skb. */
2204 if (prepad + len < PAGE_SIZE || !linear)
2207 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2212 skb_reserve(skb, reserve);
2213 skb_put(skb, linear);
2214 skb->data_len = len - linear;
2215 skb->len += len - linear;
2220 static int packet_snd(struct socket *sock,
2221 struct msghdr *msg, size_t len)
2223 struct sock *sk = sock->sk;
2224 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2225 struct sk_buff *skb;
2226 struct net_device *dev;
2228 unsigned char *addr;
2229 int err, reserve = 0;
2230 struct virtio_net_hdr vnet_hdr = { 0 };
2233 struct packet_sock *po = pkt_sk(sk);
2234 unsigned short gso_type = 0;
2239 * Get and verify the address.
2242 if (likely(saddr == NULL)) {
2243 dev = packet_cached_dev_get(po);
2248 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2250 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2252 proto = saddr->sll_protocol;
2253 addr = saddr->sll_addr;
2254 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2258 if (unlikely(dev == NULL))
2261 if (unlikely(!(dev->flags & IFF_UP)))
2264 if (sock->type == SOCK_RAW)
2265 reserve = dev->hard_header_len;
2266 if (po->has_vnet_hdr) {
2267 vnet_hdr_len = sizeof(vnet_hdr);
2270 if (len < vnet_hdr_len)
2273 len -= vnet_hdr_len;
2275 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2280 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2281 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2283 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2284 vnet_hdr.csum_offset + 2;
2287 if (vnet_hdr.hdr_len > len)
2290 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2291 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2292 case VIRTIO_NET_HDR_GSO_TCPV4:
2293 gso_type = SKB_GSO_TCPV4;
2295 case VIRTIO_NET_HDR_GSO_TCPV6:
2296 gso_type = SKB_GSO_TCPV6;
2298 case VIRTIO_NET_HDR_GSO_UDP:
2299 gso_type = SKB_GSO_UDP;
2305 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2306 gso_type |= SKB_GSO_TCP_ECN;
2308 if (vnet_hdr.gso_size == 0)
2314 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2315 if (!netif_supports_nofcs(dev)) {
2316 err = -EPROTONOSUPPORT;
2319 extra_len = 4; /* We're doing our own CRC */
2323 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2327 hlen = LL_RESERVED_SPACE(dev);
2328 tlen = dev->needed_tailroom;
2329 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2330 msg->msg_flags & MSG_DONTWAIT, &err);
2334 skb_set_network_header(skb, reserve);
2337 if (sock->type == SOCK_DGRAM &&
2338 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2341 /* Returns -EFAULT on error */
2342 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2346 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2348 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2349 /* Earlier code assumed this would be a VLAN pkt,
2350 * double-check this now that we have the actual
2353 struct ethhdr *ehdr;
2354 skb_reset_mac_header(skb);
2355 ehdr = eth_hdr(skb);
2356 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2362 skb->protocol = proto;
2364 skb->priority = sk->sk_priority;
2365 skb->mark = sk->sk_mark;
2367 if (po->has_vnet_hdr) {
2368 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2369 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2370 vnet_hdr.csum_offset)) {
2376 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2377 skb_shinfo(skb)->gso_type = gso_type;
2379 /* Header must be checked, and gso_segs computed. */
2380 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2381 skb_shinfo(skb)->gso_segs = 0;
2383 len += vnet_hdr_len;
2386 skb_probe_transport_header(skb, reserve);
2388 if (unlikely(extra_len == 4))
2395 err = dev_queue_xmit(skb);
2396 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2412 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2413 struct msghdr *msg, size_t len)
2415 struct sock *sk = sock->sk;
2416 struct packet_sock *po = pkt_sk(sk);
2417 if (po->tx_ring.pg_vec)
2418 return tpacket_snd(po, msg);
2420 return packet_snd(sock, msg, len);
2424 * Close a PACKET socket. This is fairly simple. We immediately go
2425 * to 'closed' state and remove our protocol entry in the device list.
2428 static int packet_release(struct socket *sock)
2430 struct sock *sk = sock->sk;
2431 struct packet_sock *po;
2433 union tpacket_req_u req_u;
2441 mutex_lock(&net->packet.sklist_lock);
2442 sk_del_node_init_rcu(sk);
2443 mutex_unlock(&net->packet.sklist_lock);
2446 sock_prot_inuse_add(net, sk->sk_prot, -1);
2449 spin_lock(&po->bind_lock);
2450 unregister_prot_hook(sk, false);
2451 packet_cached_dev_reset(po);
2453 if (po->prot_hook.dev) {
2454 dev_put(po->prot_hook.dev);
2455 po->prot_hook.dev = NULL;
2457 spin_unlock(&po->bind_lock);
2459 packet_flush_mclist(sk);
2461 if (po->rx_ring.pg_vec) {
2462 memset(&req_u, 0, sizeof(req_u));
2463 packet_set_ring(sk, &req_u, 1, 0);
2466 if (po->tx_ring.pg_vec) {
2467 memset(&req_u, 0, sizeof(req_u));
2468 packet_set_ring(sk, &req_u, 1, 1);
2475 * Now the socket is dead. No more input will appear.
2482 skb_queue_purge(&sk->sk_receive_queue);
2483 sk_refcnt_debug_release(sk);
2490 * Attach a packet hook.
2493 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2495 struct packet_sock *po = pkt_sk(sk);
2506 spin_lock(&po->bind_lock);
2507 unregister_prot_hook(sk, true);
2510 po->prot_hook.type = protocol;
2511 if (po->prot_hook.dev)
2512 dev_put(po->prot_hook.dev);
2514 po->prot_hook.dev = dev;
2515 po->ifindex = dev ? dev->ifindex : 0;
2517 packet_cached_dev_assign(po, dev);
2522 if (!dev || (dev->flags & IFF_UP)) {
2523 register_prot_hook(sk);
2525 sk->sk_err = ENETDOWN;
2526 if (!sock_flag(sk, SOCK_DEAD))
2527 sk->sk_error_report(sk);
2531 spin_unlock(&po->bind_lock);
2537 * Bind a packet socket to a device
2540 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2543 struct sock *sk = sock->sk;
2545 struct net_device *dev;
2552 if (addr_len != sizeof(struct sockaddr))
2554 strlcpy(name, uaddr->sa_data, sizeof(name));
2556 dev = dev_get_by_name(sock_net(sk), name);
2558 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2562 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2564 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2565 struct sock *sk = sock->sk;
2566 struct net_device *dev = NULL;
2574 if (addr_len < sizeof(struct sockaddr_ll))
2576 if (sll->sll_family != AF_PACKET)
2579 if (sll->sll_ifindex) {
2581 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2585 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2591 static struct proto packet_proto = {
2593 .owner = THIS_MODULE,
2594 .obj_size = sizeof(struct packet_sock),
2598 * Create a packet of type SOCK_PACKET.
2601 static int packet_create(struct net *net, struct socket *sock, int protocol,
2605 struct packet_sock *po;
2606 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2609 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2611 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2612 sock->type != SOCK_PACKET)
2613 return -ESOCKTNOSUPPORT;
2615 sock->state = SS_UNCONNECTED;
2618 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2622 sock->ops = &packet_ops;
2623 if (sock->type == SOCK_PACKET)
2624 sock->ops = &packet_ops_spkt;
2626 sock_init_data(sock, sk);
2629 sk->sk_family = PF_PACKET;
2632 packet_cached_dev_reset(po);
2634 sk->sk_destruct = packet_sock_destruct;
2635 sk_refcnt_debug_inc(sk);
2638 * Attach a protocol block
2641 spin_lock_init(&po->bind_lock);
2642 mutex_init(&po->pg_vec_lock);
2643 po->prot_hook.func = packet_rcv;
2645 if (sock->type == SOCK_PACKET)
2646 po->prot_hook.func = packet_rcv_spkt;
2648 po->prot_hook.af_packet_priv = sk;
2651 po->prot_hook.type = proto;
2652 register_prot_hook(sk);
2655 mutex_lock(&net->packet.sklist_lock);
2656 sk_add_node_rcu(sk, &net->packet.sklist);
2657 mutex_unlock(&net->packet.sklist_lock);
2660 sock_prot_inuse_add(net, &packet_proto, 1);
2668 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2670 struct sock_exterr_skb *serr;
2671 struct sk_buff *skb, *skb2;
2675 skb = skb_dequeue(&sk->sk_error_queue);
2681 msg->msg_flags |= MSG_TRUNC;
2684 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2688 sock_recv_timestamp(msg, sk, skb);
2690 serr = SKB_EXT_ERR(skb);
2691 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2692 sizeof(serr->ee), &serr->ee);
2694 msg->msg_flags |= MSG_ERRQUEUE;
2697 /* Reset and regenerate socket error */
2698 spin_lock_bh(&sk->sk_error_queue.lock);
2700 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2701 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2702 spin_unlock_bh(&sk->sk_error_queue.lock);
2703 sk->sk_error_report(sk);
2705 spin_unlock_bh(&sk->sk_error_queue.lock);
2714 * Pull a packet from our receive queue and hand it to the user.
2715 * If necessary we block.
2718 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2719 struct msghdr *msg, size_t len, int flags)
2721 struct sock *sk = sock->sk;
2722 struct sk_buff *skb;
2724 int vnet_hdr_len = 0;
2727 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2731 /* What error should we return now? EUNATTACH? */
2732 if (pkt_sk(sk)->ifindex < 0)
2736 if (flags & MSG_ERRQUEUE) {
2737 err = packet_recv_error(sk, msg, len);
2742 * Call the generic datagram receiver. This handles all sorts
2743 * of horrible races and re-entrancy so we can forget about it
2744 * in the protocol layers.
2746 * Now it will return ENETDOWN, if device have just gone down,
2747 * but then it will block.
2750 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2753 * An error occurred so return it. Because skb_recv_datagram()
2754 * handles the blocking we don't see and worry about blocking
2761 if (pkt_sk(sk)->has_vnet_hdr) {
2762 struct virtio_net_hdr vnet_hdr = { 0 };
2765 vnet_hdr_len = sizeof(vnet_hdr);
2766 if (len < vnet_hdr_len)
2769 len -= vnet_hdr_len;
2771 if (skb_is_gso(skb)) {
2772 struct skb_shared_info *sinfo = skb_shinfo(skb);
2774 /* This is a hint as to how much should be linear. */
2775 vnet_hdr.hdr_len = skb_headlen(skb);
2776 vnet_hdr.gso_size = sinfo->gso_size;
2777 if (sinfo->gso_type & SKB_GSO_TCPV4)
2778 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2779 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2780 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2781 else if (sinfo->gso_type & SKB_GSO_UDP)
2782 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2783 else if (sinfo->gso_type & SKB_GSO_FCOE)
2787 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2788 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2790 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2792 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2793 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2794 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2795 vnet_hdr.csum_offset = skb->csum_offset;
2796 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2797 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2798 } /* else everything is zero */
2800 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2806 /* You lose any data beyond the buffer you gave. If it worries
2807 * a user program they can ask the device for its MTU
2813 msg->msg_flags |= MSG_TRUNC;
2816 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2820 sock_recv_ts_and_drops(msg, sk, skb);
2822 if (msg->msg_name) {
2823 /* If the address length field is there to be filled
2824 * in, we fill it in now.
2826 if (sock->type == SOCK_PACKET) {
2827 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2829 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2830 msg->msg_namelen = sll->sll_halen +
2831 offsetof(struct sockaddr_ll, sll_addr);
2833 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2837 if (pkt_sk(sk)->auxdata) {
2838 struct tpacket_auxdata aux;
2840 aux.tp_status = TP_STATUS_USER;
2841 if (skb->ip_summed == CHECKSUM_PARTIAL)
2842 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2843 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2844 aux.tp_snaplen = skb->len;
2846 aux.tp_net = skb_network_offset(skb);
2847 if (vlan_tx_tag_present(skb)) {
2848 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2849 aux.tp_status |= TP_STATUS_VLAN_VALID;
2851 aux.tp_vlan_tci = 0;
2854 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2858 * Free or return the buffer as appropriate. Again this
2859 * hides all the races and re-entrancy issues from us.
2861 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2864 skb_free_datagram(sk, skb);
2869 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2870 int *uaddr_len, int peer)
2872 struct net_device *dev;
2873 struct sock *sk = sock->sk;
2878 uaddr->sa_family = AF_PACKET;
2879 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2881 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2883 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2885 *uaddr_len = sizeof(*uaddr);
2890 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2891 int *uaddr_len, int peer)
2893 struct net_device *dev;
2894 struct sock *sk = sock->sk;
2895 struct packet_sock *po = pkt_sk(sk);
2896 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2901 sll->sll_family = AF_PACKET;
2902 sll->sll_ifindex = po->ifindex;
2903 sll->sll_protocol = po->num;
2904 sll->sll_pkttype = 0;
2906 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2908 sll->sll_hatype = dev->type;
2909 sll->sll_halen = dev->addr_len;
2910 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2912 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2916 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2921 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2925 case PACKET_MR_MULTICAST:
2926 if (i->alen != dev->addr_len)
2929 return dev_mc_add(dev, i->addr);
2931 return dev_mc_del(dev, i->addr);
2933 case PACKET_MR_PROMISC:
2934 return dev_set_promiscuity(dev, what);
2936 case PACKET_MR_ALLMULTI:
2937 return dev_set_allmulti(dev, what);
2939 case PACKET_MR_UNICAST:
2940 if (i->alen != dev->addr_len)
2943 return dev_uc_add(dev, i->addr);
2945 return dev_uc_del(dev, i->addr);
2953 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2955 for ( ; i; i = i->next) {
2956 if (i->ifindex == dev->ifindex)
2957 packet_dev_mc(dev, i, what);
2961 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2963 struct packet_sock *po = pkt_sk(sk);
2964 struct packet_mclist *ml, *i;
2965 struct net_device *dev;
2971 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2976 if (mreq->mr_alen > dev->addr_len)
2980 i = kmalloc(sizeof(*i), GFP_KERNEL);
2985 for (ml = po->mclist; ml; ml = ml->next) {
2986 if (ml->ifindex == mreq->mr_ifindex &&
2987 ml->type == mreq->mr_type &&
2988 ml->alen == mreq->mr_alen &&
2989 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2991 /* Free the new element ... */
2997 i->type = mreq->mr_type;
2998 i->ifindex = mreq->mr_ifindex;
2999 i->alen = mreq->mr_alen;
3000 memcpy(i->addr, mreq->mr_address, i->alen);
3002 i->next = po->mclist;
3004 err = packet_dev_mc(dev, i, 1);
3006 po->mclist = i->next;
3015 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3017 struct packet_mclist *ml, **mlp;
3021 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3022 if (ml->ifindex == mreq->mr_ifindex &&
3023 ml->type == mreq->mr_type &&
3024 ml->alen == mreq->mr_alen &&
3025 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3026 if (--ml->count == 0) {
3027 struct net_device *dev;
3029 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3031 packet_dev_mc(dev, ml, -1);
3039 return -EADDRNOTAVAIL;
3042 static void packet_flush_mclist(struct sock *sk)
3044 struct packet_sock *po = pkt_sk(sk);
3045 struct packet_mclist *ml;
3051 while ((ml = po->mclist) != NULL) {
3052 struct net_device *dev;
3054 po->mclist = ml->next;
3055 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3057 packet_dev_mc(dev, ml, -1);
3064 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3066 struct sock *sk = sock->sk;
3067 struct packet_sock *po = pkt_sk(sk);
3070 if (level != SOL_PACKET)
3071 return -ENOPROTOOPT;
3074 case PACKET_ADD_MEMBERSHIP:
3075 case PACKET_DROP_MEMBERSHIP:
3077 struct packet_mreq_max mreq;
3079 memset(&mreq, 0, sizeof(mreq));
3080 if (len < sizeof(struct packet_mreq))
3082 if (len > sizeof(mreq))
3084 if (copy_from_user(&mreq, optval, len))
3086 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3088 if (optname == PACKET_ADD_MEMBERSHIP)
3089 ret = packet_mc_add(sk, &mreq);
3091 ret = packet_mc_drop(sk, &mreq);
3095 case PACKET_RX_RING:
3096 case PACKET_TX_RING:
3098 union tpacket_req_u req_u;
3101 switch (po->tp_version) {
3104 len = sizeof(req_u.req);
3108 len = sizeof(req_u.req3);
3113 if (pkt_sk(sk)->has_vnet_hdr)
3115 if (copy_from_user(&req_u.req, optval, len))
3117 return packet_set_ring(sk, &req_u, 0,
3118 optname == PACKET_TX_RING);
3120 case PACKET_COPY_THRESH:
3124 if (optlen != sizeof(val))
3126 if (copy_from_user(&val, optval, sizeof(val)))
3129 pkt_sk(sk)->copy_thresh = val;
3132 case PACKET_VERSION:
3136 if (optlen != sizeof(val))
3138 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3140 if (copy_from_user(&val, optval, sizeof(val)))
3146 po->tp_version = val;
3152 case PACKET_RESERVE:
3156 if (optlen != sizeof(val))
3158 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3160 if (copy_from_user(&val, optval, sizeof(val)))
3162 po->tp_reserve = val;
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_loss = !!val;
3178 case PACKET_AUXDATA:
3182 if (optlen < sizeof(val))
3184 if (copy_from_user(&val, optval, sizeof(val)))
3187 po->auxdata = !!val;
3190 case PACKET_ORIGDEV:
3194 if (optlen < sizeof(val))
3196 if (copy_from_user(&val, optval, sizeof(val)))
3199 po->origdev = !!val;
3202 case PACKET_VNET_HDR:
3206 if (sock->type != SOCK_RAW)
3208 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3210 if (optlen < sizeof(val))
3212 if (copy_from_user(&val, optval, sizeof(val)))
3215 po->has_vnet_hdr = !!val;
3218 case PACKET_TIMESTAMP:
3222 if (optlen != sizeof(val))
3224 if (copy_from_user(&val, optval, sizeof(val)))
3227 po->tp_tstamp = val;
3234 if (optlen != sizeof(val))
3236 if (copy_from_user(&val, optval, sizeof(val)))
3239 return fanout_add(sk, val & 0xffff, val >> 16);
3241 case PACKET_TX_HAS_OFF:
3245 if (optlen != sizeof(val))
3247 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3249 if (copy_from_user(&val, optval, sizeof(val)))
3251 po->tp_tx_has_off = !!val;
3255 return -ENOPROTOOPT;
3259 static int packet_getsockopt(struct socket *sock, int level, int optname,
3260 char __user *optval, int __user *optlen)
3263 int val, lv = sizeof(val);
3264 struct sock *sk = sock->sk;
3265 struct packet_sock *po = pkt_sk(sk);
3267 union tpacket_stats_u st;
3269 if (level != SOL_PACKET)
3270 return -ENOPROTOOPT;
3272 if (get_user(len, optlen))
3279 case PACKET_STATISTICS:
3280 spin_lock_bh(&sk->sk_receive_queue.lock);
3281 memcpy(&st, &po->stats, sizeof(st));
3282 memset(&po->stats, 0, sizeof(po->stats));
3283 spin_unlock_bh(&sk->sk_receive_queue.lock);
3285 if (po->tp_version == TPACKET_V3) {
3286 lv = sizeof(struct tpacket_stats_v3);
3287 st.stats3.tp_packets += st.stats3.tp_drops;
3290 lv = sizeof(struct tpacket_stats);
3291 st.stats1.tp_packets += st.stats1.tp_drops;
3296 case PACKET_AUXDATA:
3299 case PACKET_ORIGDEV:
3302 case PACKET_VNET_HDR:
3303 val = po->has_vnet_hdr;
3305 case PACKET_VERSION:
3306 val = po->tp_version;
3309 if (len > sizeof(int))
3311 if (copy_from_user(&val, optval, len))
3315 val = sizeof(struct tpacket_hdr);
3318 val = sizeof(struct tpacket2_hdr);
3321 val = sizeof(struct tpacket3_hdr);
3327 case PACKET_RESERVE:
3328 val = po->tp_reserve;
3333 case PACKET_TIMESTAMP:
3334 val = po->tp_tstamp;
3338 ((u32)po->fanout->id |
3339 ((u32)po->fanout->type << 16) |
3340 ((u32)po->fanout->flags << 24)) :
3343 case PACKET_TX_HAS_OFF:
3344 val = po->tp_tx_has_off;
3347 return -ENOPROTOOPT;
3352 if (put_user(len, optlen))
3354 if (copy_to_user(optval, data, len))
3360 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3363 struct net_device *dev = data;
3364 struct net *net = dev_net(dev);
3367 sk_for_each_rcu(sk, &net->packet.sklist) {
3368 struct packet_sock *po = pkt_sk(sk);
3371 case NETDEV_UNREGISTER:
3373 packet_dev_mclist(dev, po->mclist, -1);
3377 if (dev->ifindex == po->ifindex) {
3378 spin_lock(&po->bind_lock);
3380 __unregister_prot_hook(sk, false);
3381 sk->sk_err = ENETDOWN;
3382 if (!sock_flag(sk, SOCK_DEAD))
3383 sk->sk_error_report(sk);
3385 if (msg == NETDEV_UNREGISTER) {
3386 packet_cached_dev_reset(po);
3388 if (po->prot_hook.dev)
3389 dev_put(po->prot_hook.dev);
3390 po->prot_hook.dev = NULL;
3392 spin_unlock(&po->bind_lock);
3396 if (dev->ifindex == po->ifindex) {
3397 spin_lock(&po->bind_lock);
3399 register_prot_hook(sk);
3400 spin_unlock(&po->bind_lock);
3410 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3413 struct sock *sk = sock->sk;
3418 int amount = sk_wmem_alloc_get(sk);
3420 return put_user(amount, (int __user *)arg);
3424 struct sk_buff *skb;
3427 spin_lock_bh(&sk->sk_receive_queue.lock);
3428 skb = skb_peek(&sk->sk_receive_queue);
3431 spin_unlock_bh(&sk->sk_receive_queue.lock);
3432 return put_user(amount, (int __user *)arg);
3435 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3437 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3447 case SIOCGIFBRDADDR:
3448 case SIOCSIFBRDADDR:
3449 case SIOCGIFNETMASK:
3450 case SIOCSIFNETMASK:
3451 case SIOCGIFDSTADDR:
3452 case SIOCSIFDSTADDR:
3454 return inet_dgram_ops.ioctl(sock, cmd, arg);
3458 return -ENOIOCTLCMD;
3463 static unsigned int packet_poll(struct file *file, struct socket *sock,
3466 struct sock *sk = sock->sk;
3467 struct packet_sock *po = pkt_sk(sk);
3468 unsigned int mask = datagram_poll(file, sock, wait);
3470 spin_lock_bh(&sk->sk_receive_queue.lock);
3471 if (po->rx_ring.pg_vec) {
3472 if (!packet_previous_rx_frame(po, &po->rx_ring,
3474 mask |= POLLIN | POLLRDNORM;
3476 spin_unlock_bh(&sk->sk_receive_queue.lock);
3477 spin_lock_bh(&sk->sk_write_queue.lock);
3478 if (po->tx_ring.pg_vec) {
3479 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3480 mask |= POLLOUT | POLLWRNORM;
3482 spin_unlock_bh(&sk->sk_write_queue.lock);
3487 /* Dirty? Well, I still did not learn better way to account
3491 static void packet_mm_open(struct vm_area_struct *vma)
3493 struct file *file = vma->vm_file;
3494 struct socket *sock = file->private_data;
3495 struct sock *sk = sock->sk;
3498 atomic_inc(&pkt_sk(sk)->mapped);
3501 static void packet_mm_close(struct vm_area_struct *vma)
3503 struct file *file = vma->vm_file;
3504 struct socket *sock = file->private_data;
3505 struct sock *sk = sock->sk;
3508 atomic_dec(&pkt_sk(sk)->mapped);
3511 static const struct vm_operations_struct packet_mmap_ops = {
3512 .open = packet_mm_open,
3513 .close = packet_mm_close,
3516 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3521 for (i = 0; i < len; i++) {
3522 if (likely(pg_vec[i].buffer)) {
3523 if (is_vmalloc_addr(pg_vec[i].buffer))
3524 vfree(pg_vec[i].buffer);
3526 free_pages((unsigned long)pg_vec[i].buffer,
3528 pg_vec[i].buffer = NULL;
3534 static char *alloc_one_pg_vec_page(unsigned long order)
3536 char *buffer = NULL;
3537 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3538 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3540 buffer = (char *) __get_free_pages(gfp_flags, order);
3546 * __get_free_pages failed, fall back to vmalloc
3548 buffer = vzalloc((1 << order) * PAGE_SIZE);
3554 * vmalloc failed, lets dig into swap here
3556 gfp_flags &= ~__GFP_NORETRY;
3557 buffer = (char *)__get_free_pages(gfp_flags, order);
3562 * complete and utter failure
3567 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3569 unsigned int block_nr = req->tp_block_nr;
3573 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3574 if (unlikely(!pg_vec))
3577 for (i = 0; i < block_nr; i++) {
3578 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3579 if (unlikely(!pg_vec[i].buffer))
3580 goto out_free_pgvec;
3587 free_pg_vec(pg_vec, order, block_nr);
3592 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3593 int closing, int tx_ring)
3595 struct pgv *pg_vec = NULL;
3596 struct packet_sock *po = pkt_sk(sk);
3597 int was_running, order = 0;
3598 struct packet_ring_buffer *rb;
3599 struct sk_buff_head *rb_queue;
3602 /* Added to avoid minimal code churn */
3603 struct tpacket_req *req = &req_u->req;
3605 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3606 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3607 WARN(1, "Tx-ring is not supported.\n");
3611 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3612 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3616 if (atomic_read(&po->mapped))
3618 if (atomic_read(&rb->pending))
3622 if (req->tp_block_nr) {
3623 /* Sanity tests and some calculations */
3625 if (unlikely(rb->pg_vec))
3628 switch (po->tp_version) {
3630 po->tp_hdrlen = TPACKET_HDRLEN;
3633 po->tp_hdrlen = TPACKET2_HDRLEN;
3636 po->tp_hdrlen = TPACKET3_HDRLEN;
3641 if (unlikely((int)req->tp_block_size <= 0))
3643 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3645 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3648 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3651 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3652 if (unlikely(rb->frames_per_block <= 0))
3654 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3659 order = get_order(req->tp_block_size);
3660 pg_vec = alloc_pg_vec(req, order);
3661 if (unlikely(!pg_vec))
3663 switch (po->tp_version) {
3665 /* Transmit path is not supported. We checked
3666 * it above but just being paranoid
3669 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3678 if (unlikely(req->tp_frame_nr))
3684 /* Detach socket from network */
3685 spin_lock(&po->bind_lock);
3686 was_running = po->running;
3690 __unregister_prot_hook(sk, false);
3692 spin_unlock(&po->bind_lock);
3697 mutex_lock(&po->pg_vec_lock);
3698 if (closing || atomic_read(&po->mapped) == 0) {
3700 spin_lock_bh(&rb_queue->lock);
3701 swap(rb->pg_vec, pg_vec);
3702 rb->frame_max = (req->tp_frame_nr - 1);
3704 rb->frame_size = req->tp_frame_size;
3705 spin_unlock_bh(&rb_queue->lock);
3707 swap(rb->pg_vec_order, order);
3708 swap(rb->pg_vec_len, req->tp_block_nr);
3710 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3711 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3712 tpacket_rcv : packet_rcv;
3713 skb_queue_purge(rb_queue);
3714 if (atomic_read(&po->mapped))
3715 pr_err("packet_mmap: vma is busy: %d\n",
3716 atomic_read(&po->mapped));
3718 mutex_unlock(&po->pg_vec_lock);
3720 spin_lock(&po->bind_lock);
3723 register_prot_hook(sk);
3725 spin_unlock(&po->bind_lock);
3726 if (closing && (po->tp_version > TPACKET_V2)) {
3727 /* Because we don't support block-based V3 on tx-ring */
3729 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3734 free_pg_vec(pg_vec, order, req->tp_block_nr);
3739 static int packet_mmap(struct file *file, struct socket *sock,
3740 struct vm_area_struct *vma)
3742 struct sock *sk = sock->sk;
3743 struct packet_sock *po = pkt_sk(sk);
3744 unsigned long size, expected_size;
3745 struct packet_ring_buffer *rb;
3746 unsigned long start;
3753 mutex_lock(&po->pg_vec_lock);
3756 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3758 expected_size += rb->pg_vec_len
3764 if (expected_size == 0)
3767 size = vma->vm_end - vma->vm_start;
3768 if (size != expected_size)
3771 start = vma->vm_start;
3772 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3773 if (rb->pg_vec == NULL)
3776 for (i = 0; i < rb->pg_vec_len; i++) {
3778 void *kaddr = rb->pg_vec[i].buffer;
3781 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3782 page = pgv_to_page(kaddr);
3783 err = vm_insert_page(vma, start, page);
3792 atomic_inc(&po->mapped);
3793 vma->vm_ops = &packet_mmap_ops;
3797 mutex_unlock(&po->pg_vec_lock);
3801 static const struct proto_ops packet_ops_spkt = {
3802 .family = PF_PACKET,
3803 .owner = THIS_MODULE,
3804 .release = packet_release,
3805 .bind = packet_bind_spkt,
3806 .connect = sock_no_connect,
3807 .socketpair = sock_no_socketpair,
3808 .accept = sock_no_accept,
3809 .getname = packet_getname_spkt,
3810 .poll = datagram_poll,
3811 .ioctl = packet_ioctl,
3812 .listen = sock_no_listen,
3813 .shutdown = sock_no_shutdown,
3814 .setsockopt = sock_no_setsockopt,
3815 .getsockopt = sock_no_getsockopt,
3816 .sendmsg = packet_sendmsg_spkt,
3817 .recvmsg = packet_recvmsg,
3818 .mmap = sock_no_mmap,
3819 .sendpage = sock_no_sendpage,
3822 static const struct proto_ops packet_ops = {
3823 .family = PF_PACKET,
3824 .owner = THIS_MODULE,
3825 .release = packet_release,
3826 .bind = packet_bind,
3827 .connect = sock_no_connect,
3828 .socketpair = sock_no_socketpair,
3829 .accept = sock_no_accept,
3830 .getname = packet_getname,
3831 .poll = packet_poll,
3832 .ioctl = packet_ioctl,
3833 .listen = sock_no_listen,
3834 .shutdown = sock_no_shutdown,
3835 .setsockopt = packet_setsockopt,
3836 .getsockopt = packet_getsockopt,
3837 .sendmsg = packet_sendmsg,
3838 .recvmsg = packet_recvmsg,
3839 .mmap = packet_mmap,
3840 .sendpage = sock_no_sendpage,
3843 static const struct net_proto_family packet_family_ops = {
3844 .family = PF_PACKET,
3845 .create = packet_create,
3846 .owner = THIS_MODULE,
3849 static struct notifier_block packet_netdev_notifier = {
3850 .notifier_call = packet_notifier,
3853 #ifdef CONFIG_PROC_FS
3855 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3858 struct net *net = seq_file_net(seq);
3861 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3864 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3866 struct net *net = seq_file_net(seq);
3867 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3870 static void packet_seq_stop(struct seq_file *seq, void *v)
3876 static int packet_seq_show(struct seq_file *seq, void *v)
3878 if (v == SEQ_START_TOKEN)
3879 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3881 struct sock *s = sk_entry(v);
3882 const struct packet_sock *po = pkt_sk(s);
3885 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3887 atomic_read(&s->sk_refcnt),
3892 atomic_read(&s->sk_rmem_alloc),
3893 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3900 static const struct seq_operations packet_seq_ops = {
3901 .start = packet_seq_start,
3902 .next = packet_seq_next,
3903 .stop = packet_seq_stop,
3904 .show = packet_seq_show,
3907 static int packet_seq_open(struct inode *inode, struct file *file)
3909 return seq_open_net(inode, file, &packet_seq_ops,
3910 sizeof(struct seq_net_private));
3913 static const struct file_operations packet_seq_fops = {
3914 .owner = THIS_MODULE,
3915 .open = packet_seq_open,
3917 .llseek = seq_lseek,
3918 .release = seq_release_net,
3923 static int __net_init packet_net_init(struct net *net)
3925 mutex_init(&net->packet.sklist_lock);
3926 INIT_HLIST_HEAD(&net->packet.sklist);
3928 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3934 static void __net_exit packet_net_exit(struct net *net)
3936 remove_proc_entry("packet", net->proc_net);
3939 static struct pernet_operations packet_net_ops = {
3940 .init = packet_net_init,
3941 .exit = packet_net_exit,
3945 static void __exit packet_exit(void)
3947 unregister_netdevice_notifier(&packet_netdev_notifier);
3948 unregister_pernet_subsys(&packet_net_ops);
3949 sock_unregister(PF_PACKET);
3950 proto_unregister(&packet_proto);
3953 static int __init packet_init(void)
3955 int rc = proto_register(&packet_proto, 0);
3960 sock_register(&packet_family_ops);
3961 register_pernet_subsys(&packet_net_ops);
3962 register_netdevice_notifier(&packet_netdev_notifier);
3967 module_init(packet_init);
3968 module_exit(packet_exit);
3969 MODULE_LICENSE("GPL");
3970 MODULE_ALIAS_NETPROTO(PF_PACKET);