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/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
80 #include <asm/cacheflush.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
94 #include <net/inet_common.h>
99 - if device has no dev->hard_header routine, it adds and removes ll header
100 inside itself. In this case ll header is invisible outside of device,
101 but higher levels still should reserve dev->hard_header_len.
102 Some devices are enough clever to reallocate skb, when header
103 will not fit to reserved space (tunnel), another ones are silly
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
111 Incoming, dev->hard_header!=NULL
112 mac_header -> ll header
115 Outgoing, dev->hard_header!=NULL
116 mac_header -> ll header
119 Incoming, dev->hard_header==NULL
120 mac_header -> UNKNOWN position. It is very likely, that it points to ll
121 header. PPP makes it, that is wrong, because introduce
122 assymetry between rx and tx paths.
125 Outgoing, dev->hard_header==NULL
126 mac_header -> data. ll header is still not built!
130 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
136 dev->hard_header != NULL
137 mac_header -> ll header
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144 We should set nh.raw on output to correct posistion,
145 packet classifier depends on it.
148 /* Private packet socket structures. */
150 struct packet_mclist {
151 struct packet_mclist *next;
156 unsigned char addr[MAX_ADDR_LEN];
158 /* identical to struct packet_mreq except it has
159 * a longer address field.
161 struct packet_mreq_max {
163 unsigned short mr_type;
164 unsigned short mr_alen;
165 unsigned char mr_address[MAX_ADDR_LEN];
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
182 unsigned int feature_req_word;
184 unsigned char reset_pending_on_curr_blk;
185 unsigned char delete_blk_timer;
186 unsigned short kactive_blk_num;
187 unsigned short blk_sizeof_priv;
189 /* last_kactive_blk_num:
190 * trick to see if user-space has caught up
191 * in order to avoid refreshing timer when every single pkt arrives.
193 unsigned short last_kactive_blk_num;
198 unsigned int knum_blocks;
199 uint64_t knxt_seq_num;
204 atomic_t blk_fill_in_prog;
206 /* Default is set to 8ms */
207 #define DEFAULT_PRB_RETIRE_TOV (8)
209 unsigned short retire_blk_tov;
210 unsigned short version;
211 unsigned long tov_in_jiffies;
213 /* timer to retire an outstanding block */
214 struct timer_list retire_blk_timer;
217 #define PGV_FROM_VMALLOC 1
222 struct packet_ring_buffer {
225 unsigned int frames_per_block;
226 unsigned int frame_size;
227 unsigned int frame_max;
229 unsigned int pg_vec_order;
230 unsigned int pg_vec_pages;
231 unsigned int pg_vec_len;
233 struct tpacket_kbdq_core prb_bdqc;
237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
239 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
240 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
241 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
248 static void *packet_previous_frame(struct packet_sock *po,
249 struct packet_ring_buffer *rb,
251 static void packet_increment_head(struct packet_ring_buffer *buff);
252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253 struct tpacket_block_desc *);
254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
255 struct packet_sock *);
256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
257 struct packet_sock *, unsigned int status);
258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
259 static void prb_open_block(struct tpacket_kbdq_core *,
260 struct tpacket_block_desc *);
261 static void prb_retire_rx_blk_timer_expired(unsigned long);
262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263 static void prb_init_blk_timer(struct packet_sock *,
264 struct tpacket_kbdq_core *,
265 void (*func) (unsigned long));
266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268 struct tpacket3_hdr *);
269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270 struct tpacket3_hdr *);
271 static void packet_flush_mclist(struct sock *sk);
273 struct packet_fanout;
275 /* struct sock has to be the first member of packet_sock */
277 struct packet_fanout *fanout;
278 struct tpacket_stats stats;
279 union tpacket_stats_u stats_u;
280 struct packet_ring_buffer rx_ring;
281 struct packet_ring_buffer tx_ring;
283 spinlock_t bind_lock;
284 struct mutex pg_vec_lock;
285 unsigned int running:1, /* prot_hook is attached*/
289 int ifindex; /* bound device */
291 struct packet_mclist *mclist;
293 enum tpacket_versions tp_version;
294 unsigned int tp_hdrlen;
295 unsigned int tp_reserve;
296 unsigned int tp_loss:1;
297 unsigned int tp_tstamp;
298 struct net_device __rcu *cached_dev;
299 struct packet_type prot_hook ____cacheline_aligned_in_smp;
302 #define PACKET_FANOUT_MAX 256
304 struct packet_fanout {
308 unsigned int num_members;
313 struct list_head list;
314 struct sock *arr[PACKET_FANOUT_MAX];
317 struct packet_type prot_hook ____cacheline_aligned_in_smp;
320 struct packet_skb_cb {
321 unsigned int origlen;
323 struct sockaddr_pkt pkt;
324 struct sockaddr_ll ll;
328 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
330 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
331 #define GET_PBLOCK_DESC(x, bid) \
332 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
333 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
334 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
335 #define GET_NEXT_PRB_BLK_NUM(x) \
336 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
337 ((x)->kactive_blk_num+1) : 0)
339 static struct packet_sock *pkt_sk(struct sock *sk)
341 return (struct packet_sock *)sk;
344 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
345 static void __fanout_link(struct sock *sk, struct packet_sock *po);
347 /* register_prot_hook must be invoked with the po->bind_lock held,
348 * or from a context in which asynchronous accesses to the packet
349 * socket is not possible (packet_create()).
351 static void register_prot_hook(struct sock *sk)
353 struct packet_sock *po = pkt_sk(sk);
357 __fanout_link(sk, po);
359 dev_add_pack(&po->prot_hook);
360 rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
368 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
369 * held. If the sync parameter is true, we will temporarily drop
370 * the po->bind_lock and do a synchronize_net to make sure no
371 * asynchronous packet processing paths still refer to the elements
372 * of po->prot_hook. If the sync parameter is false, it is the
373 * callers responsibility to take care of this.
375 static void __unregister_prot_hook(struct sock *sk, bool sync)
377 struct packet_sock *po = pkt_sk(sk);
381 __fanout_unlink(sk, po);
383 __dev_remove_pack(&po->prot_hook);
384 RCU_INIT_POINTER(po->cached_dev, NULL);
390 spin_unlock(&po->bind_lock);
392 spin_lock(&po->bind_lock);
396 static void unregister_prot_hook(struct sock *sk, bool sync)
398 struct packet_sock *po = pkt_sk(sk);
401 __unregister_prot_hook(sk, sync);
404 static inline __pure struct page *pgv_to_page(void *addr)
406 if (is_vmalloc_addr(addr))
407 return vmalloc_to_page(addr);
408 return virt_to_page(addr);
411 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
414 struct tpacket_hdr *h1;
415 struct tpacket2_hdr *h2;
420 switch (po->tp_version) {
422 h.h1->tp_status = status;
423 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
426 h.h2->tp_status = status;
427 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
431 WARN(1, "TPACKET version not supported.\n");
438 static int __packet_get_status(struct packet_sock *po, void *frame)
441 struct tpacket_hdr *h1;
442 struct tpacket2_hdr *h2;
449 switch (po->tp_version) {
451 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
452 return h.h1->tp_status;
454 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
455 return h.h2->tp_status;
458 WARN(1, "TPACKET version not supported.\n");
464 static void *packet_lookup_frame(struct packet_sock *po,
465 struct packet_ring_buffer *rb,
466 unsigned int position,
469 unsigned int pg_vec_pos, frame_offset;
471 struct tpacket_hdr *h1;
472 struct tpacket2_hdr *h2;
476 pg_vec_pos = position / rb->frames_per_block;
477 frame_offset = position % rb->frames_per_block;
479 h.raw = rb->pg_vec[pg_vec_pos].buffer +
480 (frame_offset * rb->frame_size);
482 if (status != __packet_get_status(po, h.raw))
488 static void *packet_current_frame(struct packet_sock *po,
489 struct packet_ring_buffer *rb,
492 return packet_lookup_frame(po, rb, rb->head, status);
495 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
497 del_timer_sync(&pkc->retire_blk_timer);
500 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
502 struct sk_buff_head *rb_queue)
504 struct tpacket_kbdq_core *pkc;
506 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
508 spin_lock_bh(&rb_queue->lock);
509 pkc->delete_blk_timer = 1;
510 spin_unlock_bh(&rb_queue->lock);
512 prb_del_retire_blk_timer(pkc);
515 static void prb_init_blk_timer(struct packet_sock *po,
516 struct tpacket_kbdq_core *pkc,
517 void (*func) (unsigned long))
519 init_timer(&pkc->retire_blk_timer);
520 pkc->retire_blk_timer.data = (long)po;
521 pkc->retire_blk_timer.function = func;
522 pkc->retire_blk_timer.expires = jiffies;
525 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
527 struct tpacket_kbdq_core *pkc;
532 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
533 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
536 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
537 int blk_size_in_bytes)
539 struct net_device *dev;
540 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
541 struct ethtool_cmd ecmd;
545 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
546 if (unlikely(!dev)) {
548 return DEFAULT_PRB_RETIRE_TOV;
550 err = __ethtool_get_settings(dev, &ecmd);
553 switch (ecmd.speed) {
563 * If the link speed is so slow you don't really
564 * need to worry about perf anyways
569 return DEFAULT_PRB_RETIRE_TOV;
573 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
585 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
586 union tpacket_req_u *req_u)
588 p1->feature_req_word = req_u->req3.tp_feature_req_word;
591 static void init_prb_bdqc(struct packet_sock *po,
592 struct packet_ring_buffer *rb,
594 union tpacket_req_u *req_u, int tx_ring)
596 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
597 struct tpacket_block_desc *pbd;
599 memset(p1, 0x0, sizeof(*p1));
601 p1->knxt_seq_num = 1;
603 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
604 p1->pkblk_start = (char *)pg_vec[0].buffer;
605 p1->kblk_size = req_u->req3.tp_block_size;
606 p1->knum_blocks = req_u->req3.tp_block_nr;
607 p1->hdrlen = po->tp_hdrlen;
608 p1->version = po->tp_version;
609 p1->last_kactive_blk_num = 0;
610 po->stats_u.stats3.tp_freeze_q_cnt = 0;
611 if (req_u->req3.tp_retire_blk_tov)
612 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
614 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
615 req_u->req3.tp_block_size);
616 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
617 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
619 prb_init_ft_ops(p1, req_u);
620 prb_setup_retire_blk_timer(po, tx_ring);
621 prb_open_block(p1, pbd);
624 /* Do NOT update the last_blk_num first.
625 * Assumes sk_buff_head lock is held.
627 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
629 mod_timer(&pkc->retire_blk_timer,
630 jiffies + pkc->tov_in_jiffies);
631 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
636 * 1) We refresh the timer only when we open a block.
637 * By doing this we don't waste cycles refreshing the timer
638 * on packet-by-packet basis.
640 * With a 1MB block-size, on a 1Gbps line, it will take
641 * i) ~8 ms to fill a block + ii) memcpy etc.
642 * In this cut we are not accounting for the memcpy time.
644 * So, if the user sets the 'tmo' to 10ms then the timer
645 * will never fire while the block is still getting filled
646 * (which is what we want). However, the user could choose
647 * to close a block early and that's fine.
649 * But when the timer does fire, we check whether or not to refresh it.
650 * Since the tmo granularity is in msecs, it is not too expensive
651 * to refresh the timer, lets say every '8' msecs.
652 * Either the user can set the 'tmo' or we can derive it based on
653 * a) line-speed and b) block-size.
654 * prb_calc_retire_blk_tmo() calculates the tmo.
657 static void prb_retire_rx_blk_timer_expired(unsigned long data)
659 struct packet_sock *po = (struct packet_sock *)data;
660 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
662 struct tpacket_block_desc *pbd;
664 spin_lock(&po->sk.sk_receive_queue.lock);
666 frozen = prb_queue_frozen(pkc);
667 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
669 if (unlikely(pkc->delete_blk_timer))
672 /* We only need to plug the race when the block is partially filled.
674 * lock(); increment BLOCK_NUM_PKTS; unlock()
675 * copy_bits() is in progress ...
676 * timer fires on other cpu:
677 * we can't retire the current block because copy_bits
681 if (BLOCK_NUM_PKTS(pbd)) {
682 while (atomic_read(&pkc->blk_fill_in_prog)) {
683 /* Waiting for skb_copy_bits to finish... */
688 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
690 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
691 if (!prb_dispatch_next_block(pkc, po))
696 /* Case 1. Queue was frozen because user-space was
699 if (prb_curr_blk_in_use(pkc, pbd)) {
701 * Ok, user-space is still behind.
702 * So just refresh the timer.
706 /* Case 2. queue was frozen,user-space caught up,
707 * now the link went idle && the timer fired.
708 * We don't have a block to close.So we open this
709 * block and restart the timer.
710 * opening a block thaws the queue,restarts timer
711 * Thawing/timer-refresh is a side effect.
713 prb_open_block(pkc, pbd);
720 _prb_refresh_rx_retire_blk_timer(pkc);
723 spin_unlock(&po->sk.sk_receive_queue.lock);
726 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
727 struct tpacket_block_desc *pbd1, __u32 status)
729 /* Flush everything minus the block header */
731 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
736 /* Skip the block header(we know header WILL fit in 4K) */
739 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
740 for (; start < end; start += PAGE_SIZE)
741 flush_dcache_page(pgv_to_page(start));
746 /* Now update the block status. */
748 BLOCK_STATUS(pbd1) = status;
750 /* Flush the block header */
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
754 flush_dcache_page(pgv_to_page(start));
764 * 2) Increment active_blk_num
766 * Note:We DONT refresh the timer on purpose.
767 * Because almost always the next block will be opened.
769 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
770 struct tpacket_block_desc *pbd1,
771 struct packet_sock *po, unsigned int stat)
773 __u32 status = TP_STATUS_USER | stat;
775 struct tpacket3_hdr *last_pkt;
776 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
778 if (po->stats.tp_drops)
779 status |= TP_STATUS_LOSING;
781 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
782 last_pkt->tp_next_offset = 0;
784 /* Get the ts of the last pkt */
785 if (BLOCK_NUM_PKTS(pbd1)) {
786 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
787 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
789 /* Ok, we tmo'd - so get the current time */
792 h1->ts_last_pkt.ts_sec = ts.tv_sec;
793 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
798 /* Flush the block */
799 prb_flush_block(pkc1, pbd1, status);
801 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
804 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
806 pkc->reset_pending_on_curr_blk = 0;
810 * Side effect of opening a block:
812 * 1) prb_queue is thawed.
813 * 2) retire_blk_timer is refreshed.
816 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
817 struct tpacket_block_desc *pbd1)
820 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
824 /* We could have just memset this but we will lose the
825 * flexibility of making the priv area sticky
827 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
828 BLOCK_NUM_PKTS(pbd1) = 0;
829 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 h1->ts_first_pkt.ts_sec = ts.tv_sec;
832 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
833 pkc1->pkblk_start = (char *)pbd1;
834 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
835 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
836 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
838 pbd1->version = pkc1->version;
839 pkc1->prev = pkc1->nxt_offset;
840 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
841 prb_thaw_queue(pkc1);
842 _prb_refresh_rx_retire_blk_timer(pkc1);
848 * Queue freeze logic:
849 * 1) Assume tp_block_nr = 8 blocks.
850 * 2) At time 't0', user opens Rx ring.
851 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
852 * 4) user-space is either sleeping or processing block '0'.
853 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
854 * it will close block-7,loop around and try to fill block '0'.
856 * __packet_lookup_frame_in_block
857 * prb_retire_current_block()
858 * prb_dispatch_next_block()
859 * |->(BLOCK_STATUS == USER) evaluates to true
860 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
861 * 6) Now there are two cases:
862 * 6.1) Link goes idle right after the queue is frozen.
863 * But remember, the last open_block() refreshed the timer.
864 * When this timer expires,it will refresh itself so that we can
865 * re-open block-0 in near future.
866 * 6.2) Link is busy and keeps on receiving packets. This is a simple
867 * case and __packet_lookup_frame_in_block will check if block-0
868 * is free and can now be re-used.
870 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
871 struct packet_sock *po)
873 pkc->reset_pending_on_curr_blk = 1;
874 po->stats_u.stats3.tp_freeze_q_cnt++;
877 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880 * If the next block is free then we will dispatch it
881 * and return a good offset.
882 * Else, we will freeze the queue.
883 * So, caller must check the return value.
885 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
886 struct packet_sock *po)
888 struct tpacket_block_desc *pbd;
892 /* 1. Get current block num */
893 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
895 /* 2. If this block is currently in_use then freeze the queue */
896 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
897 prb_freeze_queue(pkc, po);
903 * open this block and return the offset where the first packet
904 * needs to get stored.
906 prb_open_block(pkc, pbd);
907 return (void *)pkc->nxt_offset;
910 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
911 struct packet_sock *po, unsigned int status)
913 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
915 /* retire/close the current block */
916 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
918 * Plug the case where copy_bits() is in progress on
919 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
920 * have space to copy the pkt in the current block and
921 * called prb_retire_current_block()
923 * We don't need to worry about the TMO case because
924 * the timer-handler already handled this case.
926 if (!(status & TP_STATUS_BLK_TMO)) {
927 while (atomic_read(&pkc->blk_fill_in_prog)) {
928 /* Waiting for skb_copy_bits to finish... */
932 prb_close_block(pkc, pbd, po, status);
937 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
938 struct tpacket_block_desc *pbd)
940 return TP_STATUS_USER & BLOCK_STATUS(pbd);
943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
945 return pkc->reset_pending_on_curr_blk;
948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
950 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
951 atomic_dec(&pkc->blk_fill_in_prog);
954 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
955 struct tpacket3_hdr *ppd)
957 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
960 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
963 ppd->hv1.tp_rxhash = 0;
966 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
967 struct tpacket3_hdr *ppd)
969 if (vlan_tx_tag_present(pkc->skb)) {
970 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
971 ppd->tp_status = TP_STATUS_VLAN_VALID;
973 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
977 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
978 struct tpacket3_hdr *ppd)
980 prb_fill_vlan_info(pkc, ppd);
982 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
983 prb_fill_rxhash(pkc, ppd);
985 prb_clear_rxhash(pkc, ppd);
988 static void prb_fill_curr_block(char *curr,
989 struct tpacket_kbdq_core *pkc,
990 struct tpacket_block_desc *pbd,
993 struct tpacket3_hdr *ppd;
995 ppd = (struct tpacket3_hdr *)curr;
996 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
998 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
999 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1000 BLOCK_NUM_PKTS(pbd) += 1;
1001 atomic_inc(&pkc->blk_fill_in_prog);
1002 prb_run_all_ft_ops(pkc, ppd);
1005 /* Assumes caller has the sk->rx_queue.lock */
1006 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1007 struct sk_buff *skb,
1012 struct tpacket_kbdq_core *pkc;
1013 struct tpacket_block_desc *pbd;
1016 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1017 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1019 /* Queue is frozen when user space is lagging behind */
1020 if (prb_queue_frozen(pkc)) {
1022 * Check if that last block which caused the queue to freeze,
1023 * is still in_use by user-space.
1025 if (prb_curr_blk_in_use(pkc, pbd)) {
1026 /* Can't record this packet */
1030 * Ok, the block was released by user-space.
1031 * Now let's open that block.
1032 * opening a block also thaws the queue.
1033 * Thawing is a side effect.
1035 prb_open_block(pkc, pbd);
1040 curr = pkc->nxt_offset;
1042 end = (char *) ((char *)pbd + pkc->kblk_size);
1044 /* first try the current block */
1045 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1046 prb_fill_curr_block(curr, pkc, pbd, len);
1047 return (void *)curr;
1050 /* Ok, close the current block */
1051 prb_retire_current_block(pkc, po, 0);
1053 /* Now, try to dispatch the next block */
1054 curr = (char *)prb_dispatch_next_block(pkc, po);
1056 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057 prb_fill_curr_block(curr, pkc, pbd, len);
1058 return (void *)curr;
1062 * No free blocks are available.user_space hasn't caught up yet.
1063 * Queue was just frozen and now this packet will get dropped.
1068 static void *packet_current_rx_frame(struct packet_sock *po,
1069 struct sk_buff *skb,
1070 int status, unsigned int len)
1073 switch (po->tp_version) {
1076 curr = packet_lookup_frame(po, &po->rx_ring,
1077 po->rx_ring.head, status);
1080 return __packet_lookup_frame_in_block(po, skb, status, len);
1082 WARN(1, "TPACKET version not supported\n");
1088 static void *prb_lookup_block(struct packet_sock *po,
1089 struct packet_ring_buffer *rb,
1090 unsigned int previous,
1093 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1094 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1096 if (status != BLOCK_STATUS(pbd))
1101 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1104 if (rb->prb_bdqc.kactive_blk_num)
1105 prev = rb->prb_bdqc.kactive_blk_num-1;
1107 prev = rb->prb_bdqc.knum_blocks-1;
1111 /* Assumes caller has held the rx_queue.lock */
1112 static void *__prb_previous_block(struct packet_sock *po,
1113 struct packet_ring_buffer *rb,
1116 unsigned int previous = prb_previous_blk_num(rb);
1117 return prb_lookup_block(po, rb, previous, status);
1120 static void *packet_previous_rx_frame(struct packet_sock *po,
1121 struct packet_ring_buffer *rb,
1124 if (po->tp_version <= TPACKET_V2)
1125 return packet_previous_frame(po, rb, status);
1127 return __prb_previous_block(po, rb, status);
1130 static void packet_increment_rx_head(struct packet_sock *po,
1131 struct packet_ring_buffer *rb)
1133 switch (po->tp_version) {
1136 return packet_increment_head(rb);
1139 WARN(1, "TPACKET version not supported.\n");
1145 static void *packet_previous_frame(struct packet_sock *po,
1146 struct packet_ring_buffer *rb,
1149 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1150 return packet_lookup_frame(po, rb, previous, status);
1153 static void packet_increment_head(struct packet_ring_buffer *buff)
1155 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1158 static void packet_sock_destruct(struct sock *sk)
1160 skb_queue_purge(&sk->sk_error_queue);
1162 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1163 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1165 if (!sock_flag(sk, SOCK_DEAD)) {
1166 pr_err("Attempt to release alive packet socket: %p\n", sk);
1170 sk_refcnt_debug_dec(sk);
1173 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1175 u32 idx, hash = skb->rxhash;
1177 idx = ((u64)hash * num) >> 32;
1182 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1184 unsigned int val = atomic_inc_return(&f->rr_cur);
1186 return f->arr[val % num];
1189 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1191 unsigned int cpu = smp_processor_id();
1193 return f->arr[cpu % num];
1196 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1197 struct packet_type *pt, struct net_device *orig_dev)
1199 struct packet_fanout *f = pt->af_packet_priv;
1200 unsigned int num = ACCESS_ONCE(f->num_members);
1201 struct packet_sock *po;
1204 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1211 case PACKET_FANOUT_HASH:
1214 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1218 skb_get_rxhash(skb);
1219 sk = fanout_demux_hash(f, skb, num);
1221 case PACKET_FANOUT_LB:
1222 sk = fanout_demux_lb(f, skb, num);
1224 case PACKET_FANOUT_CPU:
1225 sk = fanout_demux_cpu(f, skb, num);
1231 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1234 static DEFINE_MUTEX(fanout_mutex);
1235 static LIST_HEAD(fanout_list);
1237 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1239 struct packet_fanout *f = po->fanout;
1241 spin_lock(&f->lock);
1242 f->arr[f->num_members] = sk;
1245 spin_unlock(&f->lock);
1248 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1250 struct packet_fanout *f = po->fanout;
1253 spin_lock(&f->lock);
1254 for (i = 0; i < f->num_members; i++) {
1255 if (f->arr[i] == sk)
1258 BUG_ON(i >= f->num_members);
1259 f->arr[i] = f->arr[f->num_members - 1];
1261 spin_unlock(&f->lock);
1264 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1266 if (sk->sk_family != PF_PACKET)
1269 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1272 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1274 struct packet_sock *po = pkt_sk(sk);
1275 struct packet_fanout *f, *match;
1276 u8 type = type_flags & 0xff;
1277 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1281 case PACKET_FANOUT_HASH:
1282 case PACKET_FANOUT_LB:
1283 case PACKET_FANOUT_CPU:
1289 mutex_lock(&fanout_mutex);
1300 list_for_each_entry(f, &fanout_list, list) {
1302 read_pnet(&f->net) == sock_net(sk)) {
1308 if (match && match->defrag != defrag)
1312 match = kzalloc(sizeof(*match), GFP_KERNEL);
1315 write_pnet(&match->net, sock_net(sk));
1318 match->defrag = defrag;
1319 atomic_set(&match->rr_cur, 0);
1320 INIT_LIST_HEAD(&match->list);
1321 spin_lock_init(&match->lock);
1322 atomic_set(&match->sk_ref, 0);
1323 match->prot_hook.type = po->prot_hook.type;
1324 match->prot_hook.dev = po->prot_hook.dev;
1325 match->prot_hook.func = packet_rcv_fanout;
1326 match->prot_hook.af_packet_priv = match;
1327 match->prot_hook.id_match = match_fanout_group;
1328 dev_add_pack(&match->prot_hook);
1329 list_add(&match->list, &fanout_list);
1332 if (match->type == type &&
1333 match->prot_hook.type == po->prot_hook.type &&
1334 match->prot_hook.dev == po->prot_hook.dev) {
1336 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1337 __dev_remove_pack(&po->prot_hook);
1339 atomic_inc(&match->sk_ref);
1340 __fanout_link(sk, po);
1345 mutex_unlock(&fanout_mutex);
1349 static void fanout_release(struct sock *sk)
1351 struct packet_sock *po = pkt_sk(sk);
1352 struct packet_fanout *f;
1354 mutex_lock(&fanout_mutex);
1359 if (atomic_dec_and_test(&f->sk_ref)) {
1361 dev_remove_pack(&f->prot_hook);
1365 mutex_unlock(&fanout_mutex);
1368 static const struct proto_ops packet_ops;
1370 static const struct proto_ops packet_ops_spkt;
1372 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1373 struct packet_type *pt, struct net_device *orig_dev)
1376 struct sockaddr_pkt *spkt;
1379 * When we registered the protocol we saved the socket in the data
1380 * field for just this event.
1383 sk = pt->af_packet_priv;
1386 * Yank back the headers [hope the device set this
1387 * right or kerboom...]
1389 * Incoming packets have ll header pulled,
1392 * For outgoing ones skb->data == skb_mac_header(skb)
1393 * so that this procedure is noop.
1396 if (skb->pkt_type == PACKET_LOOPBACK)
1399 if (!net_eq(dev_net(dev), sock_net(sk)))
1402 skb = skb_share_check(skb, GFP_ATOMIC);
1406 /* drop any routing info */
1409 /* drop conntrack reference */
1412 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1414 skb_push(skb, skb->data - skb_mac_header(skb));
1417 * The SOCK_PACKET socket receives _all_ frames.
1420 spkt->spkt_family = dev->type;
1421 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1422 spkt->spkt_protocol = skb->protocol;
1425 * Charge the memory to the socket. This is done specifically
1426 * to prevent sockets using all the memory up.
1429 if (sock_queue_rcv_skb(sk, skb) == 0)
1440 * Output a raw packet to a device layer. This bypasses all the other
1441 * protocol layers and you must therefore supply it with a complete frame
1444 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1445 struct msghdr *msg, size_t len)
1447 struct sock *sk = sock->sk;
1448 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1449 struct sk_buff *skb = NULL;
1450 struct net_device *dev;
1455 * Get and verify the address.
1459 if (msg->msg_namelen < sizeof(struct sockaddr))
1461 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1462 proto = saddr->spkt_protocol;
1464 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1467 * Find the device first to size check it
1470 saddr->spkt_device[13] = 0;
1473 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1479 if (!(dev->flags & IFF_UP))
1483 * You may not queue a frame bigger than the mtu. This is the lowest level
1484 * raw protocol and you must do your own fragmentation at this level.
1488 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1492 size_t reserved = LL_RESERVED_SPACE(dev);
1493 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1496 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1499 /* FIXME: Save some space for broken drivers that write a hard
1500 * header at transmission time by themselves. PPP is the notable
1501 * one here. This should really be fixed at the driver level.
1503 skb_reserve(skb, reserved);
1504 skb_reset_network_header(skb);
1506 /* Try to align data part correctly */
1511 skb_reset_network_header(skb);
1513 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1519 if (len > (dev->mtu + dev->hard_header_len)) {
1520 /* Earlier code assumed this would be a VLAN pkt,
1521 * double-check this now that we have the actual
1524 struct ethhdr *ehdr;
1525 skb_reset_mac_header(skb);
1526 ehdr = eth_hdr(skb);
1527 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1533 skb->protocol = proto;
1535 skb->priority = sk->sk_priority;
1536 skb->mark = sk->sk_mark;
1537 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1541 dev_queue_xmit(skb);
1552 static unsigned int run_filter(const struct sk_buff *skb,
1553 const struct sock *sk,
1556 struct sk_filter *filter;
1559 filter = rcu_dereference(sk->sk_filter);
1561 res = SK_RUN_FILTER(filter, skb);
1568 * This function makes lazy skb cloning in hope that most of packets
1569 * are discarded by BPF.
1571 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1572 * and skb->cb are mangled. It works because (and until) packets
1573 * falling here are owned by current CPU. Output packets are cloned
1574 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1575 * sequencially, so that if we return skb to original state on exit,
1576 * we will not harm anyone.
1579 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1580 struct packet_type *pt, struct net_device *orig_dev)
1583 struct sockaddr_ll *sll;
1584 struct packet_sock *po;
1585 u8 *skb_head = skb->data;
1586 int skb_len = skb->len;
1587 unsigned int snaplen, res;
1589 if (skb->pkt_type == PACKET_LOOPBACK)
1592 sk = pt->af_packet_priv;
1595 if (!net_eq(dev_net(dev), sock_net(sk)))
1600 if (dev->header_ops) {
1601 /* The device has an explicit notion of ll header,
1602 * exported to higher levels.
1604 * Otherwise, the device hides details of its frame
1605 * structure, so that corresponding packet head is
1606 * never delivered to user.
1608 if (sk->sk_type != SOCK_DGRAM)
1609 skb_push(skb, skb->data - skb_mac_header(skb));
1610 else if (skb->pkt_type == PACKET_OUTGOING) {
1611 /* Special case: outgoing packets have ll header at head */
1612 skb_pull(skb, skb_network_offset(skb));
1618 res = run_filter(skb, sk, snaplen);
1620 goto drop_n_restore;
1624 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1627 if (skb_shared(skb)) {
1628 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1632 if (skb_head != skb->data) {
1633 skb->data = skb_head;
1640 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1643 sll = &PACKET_SKB_CB(skb)->sa.ll;
1644 sll->sll_family = AF_PACKET;
1645 sll->sll_hatype = dev->type;
1646 sll->sll_protocol = skb->protocol;
1647 sll->sll_pkttype = skb->pkt_type;
1648 if (unlikely(po->origdev))
1649 sll->sll_ifindex = orig_dev->ifindex;
1651 sll->sll_ifindex = dev->ifindex;
1653 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1655 PACKET_SKB_CB(skb)->origlen = skb->len;
1657 if (pskb_trim(skb, snaplen))
1660 skb_set_owner_r(skb, sk);
1664 /* drop conntrack reference */
1667 spin_lock(&sk->sk_receive_queue.lock);
1668 po->stats.tp_packets++;
1669 skb->dropcount = atomic_read(&sk->sk_drops);
1670 __skb_queue_tail(&sk->sk_receive_queue, skb);
1671 spin_unlock(&sk->sk_receive_queue.lock);
1672 sk->sk_data_ready(sk, skb->len);
1676 spin_lock(&sk->sk_receive_queue.lock);
1677 po->stats.tp_drops++;
1678 atomic_inc(&sk->sk_drops);
1679 spin_unlock(&sk->sk_receive_queue.lock);
1682 if (skb_head != skb->data && skb_shared(skb)) {
1683 skb->data = skb_head;
1691 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1692 struct packet_type *pt, struct net_device *orig_dev)
1695 struct packet_sock *po;
1696 struct sockaddr_ll *sll;
1698 struct tpacket_hdr *h1;
1699 struct tpacket2_hdr *h2;
1700 struct tpacket3_hdr *h3;
1703 u8 *skb_head = skb->data;
1704 int skb_len = skb->len;
1705 unsigned int snaplen, res;
1706 unsigned long status = TP_STATUS_USER;
1707 unsigned short macoff, netoff, hdrlen;
1708 struct sk_buff *copy_skb = NULL;
1711 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1713 if (skb->pkt_type == PACKET_LOOPBACK)
1716 sk = pt->af_packet_priv;
1719 if (!net_eq(dev_net(dev), sock_net(sk)))
1722 if (dev->header_ops) {
1723 if (sk->sk_type != SOCK_DGRAM)
1724 skb_push(skb, skb->data - skb_mac_header(skb));
1725 else if (skb->pkt_type == PACKET_OUTGOING) {
1726 /* Special case: outgoing packets have ll header at head */
1727 skb_pull(skb, skb_network_offset(skb));
1731 if (skb->ip_summed == CHECKSUM_PARTIAL)
1732 status |= TP_STATUS_CSUMNOTREADY;
1736 res = run_filter(skb, sk, snaplen);
1738 goto drop_n_restore;
1742 if (sk->sk_type == SOCK_DGRAM) {
1743 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1746 unsigned maclen = skb_network_offset(skb);
1747 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1748 (maclen < 16 ? 16 : maclen)) +
1750 macoff = netoff - maclen;
1752 if (po->tp_version <= TPACKET_V2) {
1753 if (macoff + snaplen > po->rx_ring.frame_size) {
1754 if (po->copy_thresh &&
1755 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1756 if (skb_shared(skb)) {
1757 copy_skb = skb_clone(skb, GFP_ATOMIC);
1759 copy_skb = skb_get(skb);
1760 skb_head = skb->data;
1763 skb_set_owner_r(copy_skb, sk);
1765 snaplen = po->rx_ring.frame_size - macoff;
1766 if ((int)snaplen < 0)
1770 spin_lock(&sk->sk_receive_queue.lock);
1771 h.raw = packet_current_rx_frame(po, skb,
1772 TP_STATUS_KERNEL, (macoff+snaplen));
1775 if (po->tp_version <= TPACKET_V2) {
1776 packet_increment_rx_head(po, &po->rx_ring);
1778 * LOSING will be reported till you read the stats,
1779 * because it's COR - Clear On Read.
1780 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1783 if (po->stats.tp_drops)
1784 status |= TP_STATUS_LOSING;
1786 po->stats.tp_packets++;
1788 status |= TP_STATUS_COPY;
1789 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1791 spin_unlock(&sk->sk_receive_queue.lock);
1793 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1795 switch (po->tp_version) {
1797 h.h1->tp_len = skb->len;
1798 h.h1->tp_snaplen = snaplen;
1799 h.h1->tp_mac = macoff;
1800 h.h1->tp_net = netoff;
1801 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1802 && shhwtstamps->syststamp.tv64)
1803 tv = ktime_to_timeval(shhwtstamps->syststamp);
1804 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1805 && shhwtstamps->hwtstamp.tv64)
1806 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1807 else if (skb->tstamp.tv64)
1808 tv = ktime_to_timeval(skb->tstamp);
1810 do_gettimeofday(&tv);
1811 h.h1->tp_sec = tv.tv_sec;
1812 h.h1->tp_usec = tv.tv_usec;
1813 hdrlen = sizeof(*h.h1);
1816 h.h2->tp_len = skb->len;
1817 h.h2->tp_snaplen = snaplen;
1818 h.h2->tp_mac = macoff;
1819 h.h2->tp_net = netoff;
1820 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1821 && shhwtstamps->syststamp.tv64)
1822 ts = ktime_to_timespec(shhwtstamps->syststamp);
1823 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1824 && shhwtstamps->hwtstamp.tv64)
1825 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1826 else if (skb->tstamp.tv64)
1827 ts = ktime_to_timespec(skb->tstamp);
1829 getnstimeofday(&ts);
1830 h.h2->tp_sec = ts.tv_sec;
1831 h.h2->tp_nsec = ts.tv_nsec;
1832 if (vlan_tx_tag_present(skb)) {
1833 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1834 status |= TP_STATUS_VLAN_VALID;
1836 h.h2->tp_vlan_tci = 0;
1838 h.h2->tp_padding = 0;
1839 hdrlen = sizeof(*h.h2);
1842 /* tp_nxt_offset,vlan are already populated above.
1843 * So DONT clear those fields here
1845 h.h3->tp_status |= status;
1846 h.h3->tp_len = skb->len;
1847 h.h3->tp_snaplen = snaplen;
1848 h.h3->tp_mac = macoff;
1849 h.h3->tp_net = netoff;
1850 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1851 && shhwtstamps->syststamp.tv64)
1852 ts = ktime_to_timespec(shhwtstamps->syststamp);
1853 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1854 && shhwtstamps->hwtstamp.tv64)
1855 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1856 else if (skb->tstamp.tv64)
1857 ts = ktime_to_timespec(skb->tstamp);
1859 getnstimeofday(&ts);
1860 h.h3->tp_sec = ts.tv_sec;
1861 h.h3->tp_nsec = ts.tv_nsec;
1862 hdrlen = sizeof(*h.h3);
1868 sll = h.raw + TPACKET_ALIGN(hdrlen);
1869 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1870 sll->sll_family = AF_PACKET;
1871 sll->sll_hatype = dev->type;
1872 sll->sll_protocol = skb->protocol;
1873 sll->sll_pkttype = skb->pkt_type;
1874 if (unlikely(po->origdev))
1875 sll->sll_ifindex = orig_dev->ifindex;
1877 sll->sll_ifindex = dev->ifindex;
1880 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1884 if (po->tp_version <= TPACKET_V2) {
1885 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1886 + macoff + snaplen);
1887 for (start = h.raw; start < end; start += PAGE_SIZE)
1888 flush_dcache_page(pgv_to_page(start));
1893 if (po->tp_version <= TPACKET_V2)
1894 __packet_set_status(po, h.raw, status);
1896 prb_clear_blk_fill_status(&po->rx_ring);
1898 sk->sk_data_ready(sk, 0);
1901 if (skb_head != skb->data && skb_shared(skb)) {
1902 skb->data = skb_head;
1910 po->stats.tp_drops++;
1911 spin_unlock(&sk->sk_receive_queue.lock);
1913 sk->sk_data_ready(sk, 0);
1914 kfree_skb(copy_skb);
1915 goto drop_n_restore;
1918 static void tpacket_destruct_skb(struct sk_buff *skb)
1920 struct packet_sock *po = pkt_sk(skb->sk);
1923 if (likely(po->tx_ring.pg_vec)) {
1924 ph = skb_shinfo(skb)->destructor_arg;
1925 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1926 atomic_dec(&po->tx_ring.pending);
1927 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1933 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1934 void *frame, struct net_device *dev, int size_max,
1935 __be16 proto, unsigned char *addr)
1938 struct tpacket_hdr *h1;
1939 struct tpacket2_hdr *h2;
1942 int to_write, offset, len, tp_len, nr_frags, len_max;
1943 struct socket *sock = po->sk.sk_socket;
1950 skb->protocol = proto;
1952 skb->priority = po->sk.sk_priority;
1953 skb->mark = po->sk.sk_mark;
1954 skb_shinfo(skb)->destructor_arg = ph.raw;
1956 switch (po->tp_version) {
1958 tp_len = ph.h2->tp_len;
1961 tp_len = ph.h1->tp_len;
1964 if (unlikely(tp_len > size_max)) {
1965 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1969 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1970 skb_reset_network_header(skb);
1972 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1975 if (sock->type == SOCK_DGRAM) {
1976 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1978 if (unlikely(err < 0))
1980 } else if (dev->hard_header_len) {
1981 /* net device doesn't like empty head */
1982 if (unlikely(tp_len <= dev->hard_header_len)) {
1983 pr_err("packet size is too short (%d < %d)\n",
1984 tp_len, dev->hard_header_len);
1988 skb_push(skb, dev->hard_header_len);
1989 err = skb_store_bits(skb, 0, data,
1990 dev->hard_header_len);
1994 data += dev->hard_header_len;
1995 to_write -= dev->hard_header_len;
1999 offset = offset_in_page(data);
2000 len_max = PAGE_SIZE - offset;
2001 len = ((to_write > len_max) ? len_max : to_write);
2003 skb->data_len = to_write;
2004 skb->len += to_write;
2005 skb->truesize += to_write;
2006 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2008 while (likely(to_write)) {
2009 nr_frags = skb_shinfo(skb)->nr_frags;
2011 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2012 pr_err("Packet exceed the number of skb frags(%lu)\n",
2017 page = pgv_to_page(data);
2019 flush_dcache_page(page);
2021 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2024 len_max = PAGE_SIZE;
2025 len = ((to_write > len_max) ? len_max : to_write);
2031 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2033 struct net_device *dev;
2036 dev = rcu_dereference(po->cached_dev);
2044 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2046 struct sk_buff *skb;
2047 struct net_device *dev;
2049 int err, reserve = 0;
2051 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2052 int tp_len, size_max;
2053 unsigned char *addr;
2057 mutex_lock(&po->pg_vec_lock);
2060 if (saddr == NULL) {
2061 dev = packet_cached_dev_get(po);
2066 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2068 if (msg->msg_namelen < (saddr->sll_halen
2069 + offsetof(struct sockaddr_ll,
2072 proto = saddr->sll_protocol;
2073 addr = saddr->sll_addr;
2074 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2078 if (unlikely(dev == NULL))
2081 if (unlikely(!(dev->flags & IFF_UP)))
2084 reserve = dev->hard_header_len;
2086 size_max = po->tx_ring.frame_size
2087 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2089 if (size_max > dev->mtu + reserve)
2090 size_max = dev->mtu + reserve;
2093 ph = packet_current_frame(po, &po->tx_ring,
2094 TP_STATUS_SEND_REQUEST);
2096 if (unlikely(ph == NULL)) {
2101 status = TP_STATUS_SEND_REQUEST;
2102 skb = sock_alloc_send_skb(&po->sk,
2103 LL_ALLOCATED_SPACE(dev)
2104 + sizeof(struct sockaddr_ll),
2107 if (unlikely(skb == NULL))
2110 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2113 if (unlikely(tp_len < 0)) {
2115 __packet_set_status(po, ph,
2116 TP_STATUS_AVAILABLE);
2117 packet_increment_head(&po->tx_ring);
2121 status = TP_STATUS_WRONG_FORMAT;
2127 skb->destructor = tpacket_destruct_skb;
2128 __packet_set_status(po, ph, TP_STATUS_SENDING);
2129 atomic_inc(&po->tx_ring.pending);
2131 status = TP_STATUS_SEND_REQUEST;
2132 err = dev_queue_xmit(skb);
2133 if (unlikely(err > 0)) {
2134 err = net_xmit_errno(err);
2135 if (err && __packet_get_status(po, ph) ==
2136 TP_STATUS_AVAILABLE) {
2137 /* skb was destructed already */
2142 * skb was dropped but not destructed yet;
2143 * let's treat it like congestion or err < 0
2147 packet_increment_head(&po->tx_ring);
2149 } while (likely((ph != NULL) ||
2150 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2151 (atomic_read(&po->tx_ring.pending))))
2158 __packet_set_status(po, ph, status);
2163 mutex_unlock(&po->pg_vec_lock);
2167 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2168 size_t reserve, size_t len,
2169 size_t linear, int noblock,
2172 struct sk_buff *skb;
2174 /* Under a page? Don't bother with paged skb. */
2175 if (prepad + len < PAGE_SIZE || !linear)
2178 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2183 skb_reserve(skb, reserve);
2184 skb_put(skb, linear);
2185 skb->data_len = len - linear;
2186 skb->len += len - linear;
2191 static int packet_snd(struct socket *sock,
2192 struct msghdr *msg, size_t len)
2194 struct sock *sk = sock->sk;
2195 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2196 struct sk_buff *skb;
2197 struct net_device *dev;
2199 unsigned char *addr;
2200 int err, reserve = 0;
2201 struct virtio_net_hdr vnet_hdr = { 0 };
2204 struct packet_sock *po = pkt_sk(sk);
2205 unsigned short gso_type = 0;
2208 * Get and verify the address.
2211 if (saddr == NULL) {
2212 dev = packet_cached_dev_get(po);
2217 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2219 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2221 proto = saddr->sll_protocol;
2222 addr = saddr->sll_addr;
2223 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2227 if (unlikely(dev == NULL))
2230 if (unlikely(!(dev->flags & IFF_UP)))
2233 if (sock->type == SOCK_RAW)
2234 reserve = dev->hard_header_len;
2235 if (po->has_vnet_hdr) {
2236 vnet_hdr_len = sizeof(vnet_hdr);
2239 if (len < vnet_hdr_len)
2242 len -= vnet_hdr_len;
2244 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2249 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2250 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2252 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2253 vnet_hdr.csum_offset + 2;
2256 if (vnet_hdr.hdr_len > len)
2259 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2260 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2261 case VIRTIO_NET_HDR_GSO_TCPV4:
2262 gso_type = SKB_GSO_TCPV4;
2264 case VIRTIO_NET_HDR_GSO_TCPV6:
2265 gso_type = SKB_GSO_TCPV6;
2267 case VIRTIO_NET_HDR_GSO_UDP:
2268 gso_type = SKB_GSO_UDP;
2274 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2275 gso_type |= SKB_GSO_TCP_ECN;
2277 if (vnet_hdr.gso_size == 0)
2284 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2288 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2289 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2290 msg->msg_flags & MSG_DONTWAIT, &err);
2294 skb_set_network_header(skb, reserve);
2297 if (sock->type == SOCK_DGRAM &&
2298 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2301 /* Returns -EFAULT on error */
2302 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2305 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2309 if (!gso_type && (len > dev->mtu + reserve)) {
2310 /* Earlier code assumed this would be a VLAN pkt,
2311 * double-check this now that we have the actual
2314 struct ethhdr *ehdr;
2315 skb_reset_mac_header(skb);
2316 ehdr = eth_hdr(skb);
2317 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2323 skb->protocol = proto;
2325 skb->priority = sk->sk_priority;
2326 skb->mark = sk->sk_mark;
2328 if (po->has_vnet_hdr) {
2329 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2330 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2331 vnet_hdr.csum_offset)) {
2337 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2338 skb_shinfo(skb)->gso_type = gso_type;
2340 /* Header must be checked, and gso_segs computed. */
2341 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2342 skb_shinfo(skb)->gso_segs = 0;
2344 len += vnet_hdr_len;
2351 err = dev_queue_xmit(skb);
2352 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2368 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2369 struct msghdr *msg, size_t len)
2371 struct sock *sk = sock->sk;
2372 struct packet_sock *po = pkt_sk(sk);
2373 if (po->tx_ring.pg_vec)
2374 return tpacket_snd(po, msg);
2376 return packet_snd(sock, msg, len);
2380 * Close a PACKET socket. This is fairly simple. We immediately go
2381 * to 'closed' state and remove our protocol entry in the device list.
2384 static int packet_release(struct socket *sock)
2386 struct sock *sk = sock->sk;
2387 struct packet_sock *po;
2389 union tpacket_req_u req_u;
2397 spin_lock_bh(&net->packet.sklist_lock);
2398 sk_del_node_init_rcu(sk);
2399 sock_prot_inuse_add(net, sk->sk_prot, -1);
2400 spin_unlock_bh(&net->packet.sklist_lock);
2402 spin_lock(&po->bind_lock);
2403 unregister_prot_hook(sk, false);
2404 if (po->prot_hook.dev) {
2405 dev_put(po->prot_hook.dev);
2406 po->prot_hook.dev = NULL;
2408 spin_unlock(&po->bind_lock);
2410 packet_flush_mclist(sk);
2412 if (po->rx_ring.pg_vec) {
2413 memset(&req_u, 0, sizeof(req_u));
2414 packet_set_ring(sk, &req_u, 1, 0);
2417 if (po->tx_ring.pg_vec) {
2418 memset(&req_u, 0, sizeof(req_u));
2419 packet_set_ring(sk, &req_u, 1, 1);
2426 * Now the socket is dead. No more input will appear.
2433 skb_queue_purge(&sk->sk_receive_queue);
2434 sk_refcnt_debug_release(sk);
2441 * Attach a packet hook.
2444 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2446 struct packet_sock *po = pkt_sk(sk);
2457 spin_lock(&po->bind_lock);
2458 unregister_prot_hook(sk, true);
2460 po->prot_hook.type = protocol;
2461 if (po->prot_hook.dev)
2462 dev_put(po->prot_hook.dev);
2463 po->prot_hook.dev = dev;
2465 po->ifindex = dev ? dev->ifindex : 0;
2470 if (!dev || (dev->flags & IFF_UP)) {
2471 register_prot_hook(sk);
2473 sk->sk_err = ENETDOWN;
2474 if (!sock_flag(sk, SOCK_DEAD))
2475 sk->sk_error_report(sk);
2479 spin_unlock(&po->bind_lock);
2485 * Bind a packet socket to a device
2488 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2491 struct sock *sk = sock->sk;
2493 struct net_device *dev;
2500 if (addr_len != sizeof(struct sockaddr))
2502 strlcpy(name, uaddr->sa_data, sizeof(name));
2504 dev = dev_get_by_name(sock_net(sk), name);
2506 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2510 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2512 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2513 struct sock *sk = sock->sk;
2514 struct net_device *dev = NULL;
2522 if (addr_len < sizeof(struct sockaddr_ll))
2524 if (sll->sll_family != AF_PACKET)
2527 if (sll->sll_ifindex) {
2529 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2533 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2539 static struct proto packet_proto = {
2541 .owner = THIS_MODULE,
2542 .obj_size = sizeof(struct packet_sock),
2546 * Create a packet of type SOCK_PACKET.
2549 static int packet_create(struct net *net, struct socket *sock, int protocol,
2553 struct packet_sock *po;
2554 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2557 if (!capable(CAP_NET_RAW))
2559 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2560 sock->type != SOCK_PACKET)
2561 return -ESOCKTNOSUPPORT;
2563 sock->state = SS_UNCONNECTED;
2566 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2570 sock->ops = &packet_ops;
2571 if (sock->type == SOCK_PACKET)
2572 sock->ops = &packet_ops_spkt;
2574 sock_init_data(sock, sk);
2577 sk->sk_family = PF_PACKET;
2579 RCU_INIT_POINTER(po->cached_dev, NULL);
2581 sk->sk_destruct = packet_sock_destruct;
2582 sk_refcnt_debug_inc(sk);
2585 * Attach a protocol block
2588 spin_lock_init(&po->bind_lock);
2589 mutex_init(&po->pg_vec_lock);
2590 po->prot_hook.func = packet_rcv;
2592 if (sock->type == SOCK_PACKET)
2593 po->prot_hook.func = packet_rcv_spkt;
2595 po->prot_hook.af_packet_priv = sk;
2598 po->prot_hook.type = proto;
2599 register_prot_hook(sk);
2602 spin_lock_bh(&net->packet.sklist_lock);
2603 sk_add_node_rcu(sk, &net->packet.sklist);
2604 sock_prot_inuse_add(net, &packet_proto, 1);
2605 spin_unlock_bh(&net->packet.sklist_lock);
2612 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2614 struct sock_exterr_skb *serr;
2615 struct sk_buff *skb, *skb2;
2619 skb = skb_dequeue(&sk->sk_error_queue);
2625 msg->msg_flags |= MSG_TRUNC;
2628 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2632 sock_recv_timestamp(msg, sk, skb);
2634 serr = SKB_EXT_ERR(skb);
2635 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2636 sizeof(serr->ee), &serr->ee);
2638 msg->msg_flags |= MSG_ERRQUEUE;
2641 /* Reset and regenerate socket error */
2642 spin_lock_bh(&sk->sk_error_queue.lock);
2644 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2645 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2646 spin_unlock_bh(&sk->sk_error_queue.lock);
2647 sk->sk_error_report(sk);
2649 spin_unlock_bh(&sk->sk_error_queue.lock);
2658 * Pull a packet from our receive queue and hand it to the user.
2659 * If necessary we block.
2662 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2663 struct msghdr *msg, size_t len, int flags)
2665 struct sock *sk = sock->sk;
2666 struct sk_buff *skb;
2668 int vnet_hdr_len = 0;
2671 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2675 /* What error should we return now? EUNATTACH? */
2676 if (pkt_sk(sk)->ifindex < 0)
2680 if (flags & MSG_ERRQUEUE) {
2681 err = packet_recv_error(sk, msg, len);
2686 * Call the generic datagram receiver. This handles all sorts
2687 * of horrible races and re-entrancy so we can forget about it
2688 * in the protocol layers.
2690 * Now it will return ENETDOWN, if device have just gone down,
2691 * but then it will block.
2694 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2697 * An error occurred so return it. Because skb_recv_datagram()
2698 * handles the blocking we don't see and worry about blocking
2705 if (pkt_sk(sk)->has_vnet_hdr) {
2706 struct virtio_net_hdr vnet_hdr = { 0 };
2709 vnet_hdr_len = sizeof(vnet_hdr);
2710 if (len < vnet_hdr_len)
2713 len -= vnet_hdr_len;
2715 if (skb_is_gso(skb)) {
2716 struct skb_shared_info *sinfo = skb_shinfo(skb);
2718 /* This is a hint as to how much should be linear. */
2719 vnet_hdr.hdr_len = skb_headlen(skb);
2720 vnet_hdr.gso_size = sinfo->gso_size;
2721 if (sinfo->gso_type & SKB_GSO_TCPV4)
2722 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2723 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2724 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2725 else if (sinfo->gso_type & SKB_GSO_UDP)
2726 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2727 else if (sinfo->gso_type & SKB_GSO_FCOE)
2731 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2732 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2734 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2736 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2737 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2738 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2739 vnet_hdr.csum_offset = skb->csum_offset;
2740 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2741 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2742 } /* else everything is zero */
2744 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2750 /* You lose any data beyond the buffer you gave. If it worries
2751 * a user program they can ask the device for its MTU
2757 msg->msg_flags |= MSG_TRUNC;
2760 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2764 sock_recv_ts_and_drops(msg, sk, skb);
2766 if (msg->msg_name) {
2767 /* If the address length field is there to be filled
2768 * in, we fill it in now.
2770 if (sock->type == SOCK_PACKET) {
2771 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2773 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2774 msg->msg_namelen = sll->sll_halen +
2775 offsetof(struct sockaddr_ll, sll_addr);
2777 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2781 if (pkt_sk(sk)->auxdata) {
2782 struct tpacket_auxdata aux;
2784 aux.tp_status = TP_STATUS_USER;
2785 if (skb->ip_summed == CHECKSUM_PARTIAL)
2786 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2787 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2788 aux.tp_snaplen = skb->len;
2790 aux.tp_net = skb_network_offset(skb);
2791 if (vlan_tx_tag_present(skb)) {
2792 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2793 aux.tp_status |= TP_STATUS_VLAN_VALID;
2795 aux.tp_vlan_tci = 0;
2798 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2802 * Free or return the buffer as appropriate. Again this
2803 * hides all the races and re-entrancy issues from us.
2805 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2808 skb_free_datagram(sk, skb);
2813 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2814 int *uaddr_len, int peer)
2816 struct net_device *dev;
2817 struct sock *sk = sock->sk;
2822 uaddr->sa_family = AF_PACKET;
2823 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2825 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2827 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2829 *uaddr_len = sizeof(*uaddr);
2834 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2835 int *uaddr_len, int peer)
2837 struct net_device *dev;
2838 struct sock *sk = sock->sk;
2839 struct packet_sock *po = pkt_sk(sk);
2840 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2845 sll->sll_family = AF_PACKET;
2846 sll->sll_ifindex = po->ifindex;
2847 sll->sll_protocol = po->num;
2848 sll->sll_pkttype = 0;
2850 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2852 sll->sll_hatype = dev->type;
2853 sll->sll_halen = dev->addr_len;
2854 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2856 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2860 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2865 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2869 case PACKET_MR_MULTICAST:
2870 if (i->alen != dev->addr_len)
2873 return dev_mc_add(dev, i->addr);
2875 return dev_mc_del(dev, i->addr);
2877 case PACKET_MR_PROMISC:
2878 return dev_set_promiscuity(dev, what);
2880 case PACKET_MR_ALLMULTI:
2881 return dev_set_allmulti(dev, what);
2883 case PACKET_MR_UNICAST:
2884 if (i->alen != dev->addr_len)
2887 return dev_uc_add(dev, i->addr);
2889 return dev_uc_del(dev, i->addr);
2897 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2899 for ( ; i; i = i->next) {
2900 if (i->ifindex == dev->ifindex)
2901 packet_dev_mc(dev, i, what);
2905 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2907 struct packet_sock *po = pkt_sk(sk);
2908 struct packet_mclist *ml, *i;
2909 struct net_device *dev;
2915 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2920 if (mreq->mr_alen > dev->addr_len)
2924 i = kmalloc(sizeof(*i), GFP_KERNEL);
2929 for (ml = po->mclist; ml; ml = ml->next) {
2930 if (ml->ifindex == mreq->mr_ifindex &&
2931 ml->type == mreq->mr_type &&
2932 ml->alen == mreq->mr_alen &&
2933 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2935 /* Free the new element ... */
2941 i->type = mreq->mr_type;
2942 i->ifindex = mreq->mr_ifindex;
2943 i->alen = mreq->mr_alen;
2944 memcpy(i->addr, mreq->mr_address, i->alen);
2946 i->next = po->mclist;
2948 err = packet_dev_mc(dev, i, 1);
2950 po->mclist = i->next;
2959 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2961 struct packet_mclist *ml, **mlp;
2965 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2966 if (ml->ifindex == mreq->mr_ifindex &&
2967 ml->type == mreq->mr_type &&
2968 ml->alen == mreq->mr_alen &&
2969 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2970 if (--ml->count == 0) {
2971 struct net_device *dev;
2973 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2975 packet_dev_mc(dev, ml, -1);
2983 return -EADDRNOTAVAIL;
2986 static void packet_flush_mclist(struct sock *sk)
2988 struct packet_sock *po = pkt_sk(sk);
2989 struct packet_mclist *ml;
2995 while ((ml = po->mclist) != NULL) {
2996 struct net_device *dev;
2998 po->mclist = ml->next;
2999 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3001 packet_dev_mc(dev, ml, -1);
3008 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3010 struct sock *sk = sock->sk;
3011 struct packet_sock *po = pkt_sk(sk);
3014 if (level != SOL_PACKET)
3015 return -ENOPROTOOPT;
3018 case PACKET_ADD_MEMBERSHIP:
3019 case PACKET_DROP_MEMBERSHIP:
3021 struct packet_mreq_max mreq;
3023 memset(&mreq, 0, sizeof(mreq));
3024 if (len < sizeof(struct packet_mreq))
3026 if (len > sizeof(mreq))
3028 if (copy_from_user(&mreq, optval, len))
3030 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3032 if (optname == PACKET_ADD_MEMBERSHIP)
3033 ret = packet_mc_add(sk, &mreq);
3035 ret = packet_mc_drop(sk, &mreq);
3039 case PACKET_RX_RING:
3040 case PACKET_TX_RING:
3042 union tpacket_req_u req_u;
3045 switch (po->tp_version) {
3048 len = sizeof(req_u.req);
3052 len = sizeof(req_u.req3);
3057 if (pkt_sk(sk)->has_vnet_hdr)
3059 if (copy_from_user(&req_u.req, optval, len))
3061 return packet_set_ring(sk, &req_u, 0,
3062 optname == PACKET_TX_RING);
3064 case PACKET_COPY_THRESH:
3068 if (optlen != sizeof(val))
3070 if (copy_from_user(&val, optval, sizeof(val)))
3073 pkt_sk(sk)->copy_thresh = val;
3076 case PACKET_VERSION:
3080 if (optlen != sizeof(val))
3082 if (copy_from_user(&val, optval, sizeof(val)))
3093 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3096 po->tp_version = val;
3102 case PACKET_RESERVE:
3106 if (optlen != sizeof(val))
3108 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3110 if (copy_from_user(&val, optval, sizeof(val)))
3112 po->tp_reserve = val;
3119 if (optlen != sizeof(val))
3121 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3123 if (copy_from_user(&val, optval, sizeof(val)))
3125 po->tp_loss = !!val;
3128 case PACKET_AUXDATA:
3132 if (optlen < sizeof(val))
3134 if (copy_from_user(&val, optval, sizeof(val)))
3137 po->auxdata = !!val;
3140 case PACKET_ORIGDEV:
3144 if (optlen < sizeof(val))
3146 if (copy_from_user(&val, optval, sizeof(val)))
3149 po->origdev = !!val;
3152 case PACKET_VNET_HDR:
3156 if (sock->type != SOCK_RAW)
3158 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3160 if (optlen < sizeof(val))
3162 if (copy_from_user(&val, optval, sizeof(val)))
3165 po->has_vnet_hdr = !!val;
3168 case PACKET_TIMESTAMP:
3172 if (optlen != sizeof(val))
3174 if (copy_from_user(&val, optval, sizeof(val)))
3177 po->tp_tstamp = val;
3184 if (optlen != sizeof(val))
3186 if (copy_from_user(&val, optval, sizeof(val)))
3189 return fanout_add(sk, val & 0xffff, val >> 16);
3192 return -ENOPROTOOPT;
3196 static int packet_getsockopt(struct socket *sock, int level, int optname,
3197 char __user *optval, int __user *optlen)
3201 struct sock *sk = sock->sk;
3202 struct packet_sock *po = pkt_sk(sk);
3204 struct tpacket_stats st;
3205 union tpacket_stats_u st_u;
3207 if (level != SOL_PACKET)
3208 return -ENOPROTOOPT;
3210 if (get_user(len, optlen))
3217 case PACKET_STATISTICS:
3218 if (po->tp_version == TPACKET_V3) {
3219 len = sizeof(struct tpacket_stats_v3);
3221 if (len > sizeof(struct tpacket_stats))
3222 len = sizeof(struct tpacket_stats);
3224 spin_lock_bh(&sk->sk_receive_queue.lock);
3225 if (po->tp_version == TPACKET_V3) {
3226 memcpy(&st_u.stats3, &po->stats,
3227 sizeof(struct tpacket_stats));
3228 st_u.stats3.tp_freeze_q_cnt =
3229 po->stats_u.stats3.tp_freeze_q_cnt;
3230 st_u.stats3.tp_packets += po->stats.tp_drops;
3231 data = &st_u.stats3;
3234 st.tp_packets += st.tp_drops;
3237 memset(&po->stats, 0, sizeof(st));
3238 spin_unlock_bh(&sk->sk_receive_queue.lock);
3240 case PACKET_AUXDATA:
3241 if (len > sizeof(int))
3247 case PACKET_ORIGDEV:
3248 if (len > sizeof(int))
3254 case PACKET_VNET_HDR:
3255 if (len > sizeof(int))
3257 val = po->has_vnet_hdr;
3261 case PACKET_VERSION:
3262 if (len > sizeof(int))
3264 val = po->tp_version;
3268 if (len > sizeof(int))
3270 if (copy_from_user(&val, optval, len))
3274 val = sizeof(struct tpacket_hdr);
3277 val = sizeof(struct tpacket2_hdr);
3280 val = sizeof(struct tpacket3_hdr);
3287 case PACKET_RESERVE:
3288 if (len > sizeof(unsigned int))
3289 len = sizeof(unsigned int);
3290 val = po->tp_reserve;
3294 if (len > sizeof(unsigned int))
3295 len = sizeof(unsigned int);
3299 case PACKET_TIMESTAMP:
3300 if (len > sizeof(int))
3302 val = po->tp_tstamp;
3306 if (len > sizeof(int))
3309 ((u32)po->fanout->id |
3310 ((u32)po->fanout->type << 16)) :
3315 return -ENOPROTOOPT;
3318 if (put_user(len, optlen))
3320 if (copy_to_user(optval, data, len))
3326 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3329 struct hlist_node *node;
3330 struct net_device *dev = data;
3331 struct net *net = dev_net(dev);
3334 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3335 struct packet_sock *po = pkt_sk(sk);
3338 case NETDEV_UNREGISTER:
3340 packet_dev_mclist(dev, po->mclist, -1);
3344 if (dev->ifindex == po->ifindex) {
3345 spin_lock(&po->bind_lock);
3347 __unregister_prot_hook(sk, false);
3348 sk->sk_err = ENETDOWN;
3349 if (!sock_flag(sk, SOCK_DEAD))
3350 sk->sk_error_report(sk);
3352 if (msg == NETDEV_UNREGISTER) {
3355 if (po->prot_hook.dev)
3356 dev_put(po->prot_hook.dev);
3357 po->prot_hook.dev = NULL;
3359 spin_unlock(&po->bind_lock);
3363 if (dev->ifindex == po->ifindex) {
3364 spin_lock(&po->bind_lock);
3366 register_prot_hook(sk);
3367 spin_unlock(&po->bind_lock);
3377 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3380 struct sock *sk = sock->sk;
3385 int amount = sk_wmem_alloc_get(sk);
3387 return put_user(amount, (int __user *)arg);
3391 struct sk_buff *skb;
3394 spin_lock_bh(&sk->sk_receive_queue.lock);
3395 skb = skb_peek(&sk->sk_receive_queue);
3398 spin_unlock_bh(&sk->sk_receive_queue.lock);
3399 return put_user(amount, (int __user *)arg);
3402 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3404 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3414 case SIOCGIFBRDADDR:
3415 case SIOCSIFBRDADDR:
3416 case SIOCGIFNETMASK:
3417 case SIOCSIFNETMASK:
3418 case SIOCGIFDSTADDR:
3419 case SIOCSIFDSTADDR:
3421 return inet_dgram_ops.ioctl(sock, cmd, arg);
3425 return -ENOIOCTLCMD;
3430 static unsigned int packet_poll(struct file *file, struct socket *sock,
3433 struct sock *sk = sock->sk;
3434 struct packet_sock *po = pkt_sk(sk);
3435 unsigned int mask = datagram_poll(file, sock, wait);
3437 spin_lock_bh(&sk->sk_receive_queue.lock);
3438 if (po->rx_ring.pg_vec) {
3439 if (!packet_previous_rx_frame(po, &po->rx_ring,
3441 mask |= POLLIN | POLLRDNORM;
3443 spin_unlock_bh(&sk->sk_receive_queue.lock);
3444 spin_lock_bh(&sk->sk_write_queue.lock);
3445 if (po->tx_ring.pg_vec) {
3446 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3447 mask |= POLLOUT | POLLWRNORM;
3449 spin_unlock_bh(&sk->sk_write_queue.lock);
3454 /* Dirty? Well, I still did not learn better way to account
3458 static void packet_mm_open(struct vm_area_struct *vma)
3460 struct file *file = vma->vm_file;
3461 struct socket *sock = file->private_data;
3462 struct sock *sk = sock->sk;
3465 atomic_inc(&pkt_sk(sk)->mapped);
3468 static void packet_mm_close(struct vm_area_struct *vma)
3470 struct file *file = vma->vm_file;
3471 struct socket *sock = file->private_data;
3472 struct sock *sk = sock->sk;
3475 atomic_dec(&pkt_sk(sk)->mapped);
3478 static const struct vm_operations_struct packet_mmap_ops = {
3479 .open = packet_mm_open,
3480 .close = packet_mm_close,
3483 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3488 for (i = 0; i < len; i++) {
3489 if (likely(pg_vec[i].buffer)) {
3490 if (is_vmalloc_addr(pg_vec[i].buffer))
3491 vfree(pg_vec[i].buffer);
3493 free_pages((unsigned long)pg_vec[i].buffer,
3495 pg_vec[i].buffer = NULL;
3501 static char *alloc_one_pg_vec_page(unsigned long order)
3503 char *buffer = NULL;
3504 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3505 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3507 buffer = (char *) __get_free_pages(gfp_flags, order);
3513 * __get_free_pages failed, fall back to vmalloc
3515 buffer = vzalloc((1 << order) * PAGE_SIZE);
3521 * vmalloc failed, lets dig into swap here
3523 gfp_flags &= ~__GFP_NORETRY;
3524 buffer = (char *)__get_free_pages(gfp_flags, order);
3529 * complete and utter failure
3534 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3536 unsigned int block_nr = req->tp_block_nr;
3540 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3541 if (unlikely(!pg_vec))
3544 for (i = 0; i < block_nr; i++) {
3545 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3546 if (unlikely(!pg_vec[i].buffer))
3547 goto out_free_pgvec;
3554 free_pg_vec(pg_vec, order, block_nr);
3559 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3560 int closing, int tx_ring)
3562 struct pgv *pg_vec = NULL;
3563 struct packet_sock *po = pkt_sk(sk);
3564 int was_running, order = 0;
3565 struct packet_ring_buffer *rb;
3566 struct sk_buff_head *rb_queue;
3569 /* Added to avoid minimal code churn */
3570 struct tpacket_req *req = &req_u->req;
3573 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3574 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3575 WARN(1, "Tx-ring is not supported.\n");
3579 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3580 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3584 if (atomic_read(&po->mapped))
3586 if (atomic_read(&rb->pending))
3590 if (req->tp_block_nr) {
3591 /* Sanity tests and some calculations */
3593 if (unlikely(rb->pg_vec))
3596 switch (po->tp_version) {
3598 po->tp_hdrlen = TPACKET_HDRLEN;
3601 po->tp_hdrlen = TPACKET2_HDRLEN;
3604 po->tp_hdrlen = TPACKET3_HDRLEN;
3609 if (unlikely((int)req->tp_block_size <= 0))
3611 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3613 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3616 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3619 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3620 if (unlikely(rb->frames_per_block <= 0))
3622 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3627 order = get_order(req->tp_block_size);
3628 pg_vec = alloc_pg_vec(req, order);
3629 if (unlikely(!pg_vec))
3631 switch (po->tp_version) {
3633 /* Transmit path is not supported. We checked
3634 * it above but just being paranoid
3637 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3646 if (unlikely(req->tp_frame_nr))
3651 /* Detach socket from network */
3652 spin_lock(&po->bind_lock);
3653 was_running = po->running;
3657 __unregister_prot_hook(sk, false);
3659 spin_unlock(&po->bind_lock);
3664 mutex_lock(&po->pg_vec_lock);
3665 if (closing || atomic_read(&po->mapped) == 0) {
3667 spin_lock_bh(&rb_queue->lock);
3668 swap(rb->pg_vec, pg_vec);
3669 rb->frame_max = (req->tp_frame_nr - 1);
3671 rb->frame_size = req->tp_frame_size;
3672 spin_unlock_bh(&rb_queue->lock);
3674 swap(rb->pg_vec_order, order);
3675 swap(rb->pg_vec_len, req->tp_block_nr);
3677 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3678 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3679 tpacket_rcv : packet_rcv;
3680 skb_queue_purge(rb_queue);
3681 if (atomic_read(&po->mapped))
3682 pr_err("packet_mmap: vma is busy: %d\n",
3683 atomic_read(&po->mapped));
3685 mutex_unlock(&po->pg_vec_lock);
3687 spin_lock(&po->bind_lock);
3690 register_prot_hook(sk);
3692 spin_unlock(&po->bind_lock);
3693 if (closing && (po->tp_version > TPACKET_V2)) {
3694 /* Because we don't support block-based V3 on tx-ring */
3696 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3700 free_pg_vec(pg_vec, order, req->tp_block_nr);
3706 static int packet_mmap(struct file *file, struct socket *sock,
3707 struct vm_area_struct *vma)
3709 struct sock *sk = sock->sk;
3710 struct packet_sock *po = pkt_sk(sk);
3711 unsigned long size, expected_size;
3712 struct packet_ring_buffer *rb;
3713 unsigned long start;
3720 mutex_lock(&po->pg_vec_lock);
3723 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3725 expected_size += rb->pg_vec_len
3731 if (expected_size == 0)
3734 size = vma->vm_end - vma->vm_start;
3735 if (size != expected_size)
3738 start = vma->vm_start;
3739 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3740 if (rb->pg_vec == NULL)
3743 for (i = 0; i < rb->pg_vec_len; i++) {
3745 void *kaddr = rb->pg_vec[i].buffer;
3748 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3749 page = pgv_to_page(kaddr);
3750 err = vm_insert_page(vma, start, page);
3759 atomic_inc(&po->mapped);
3760 vma->vm_ops = &packet_mmap_ops;
3764 mutex_unlock(&po->pg_vec_lock);
3768 static const struct proto_ops packet_ops_spkt = {
3769 .family = PF_PACKET,
3770 .owner = THIS_MODULE,
3771 .release = packet_release,
3772 .bind = packet_bind_spkt,
3773 .connect = sock_no_connect,
3774 .socketpair = sock_no_socketpair,
3775 .accept = sock_no_accept,
3776 .getname = packet_getname_spkt,
3777 .poll = datagram_poll,
3778 .ioctl = packet_ioctl,
3779 .listen = sock_no_listen,
3780 .shutdown = sock_no_shutdown,
3781 .setsockopt = sock_no_setsockopt,
3782 .getsockopt = sock_no_getsockopt,
3783 .sendmsg = packet_sendmsg_spkt,
3784 .recvmsg = packet_recvmsg,
3785 .mmap = sock_no_mmap,
3786 .sendpage = sock_no_sendpage,
3789 static const struct proto_ops packet_ops = {
3790 .family = PF_PACKET,
3791 .owner = THIS_MODULE,
3792 .release = packet_release,
3793 .bind = packet_bind,
3794 .connect = sock_no_connect,
3795 .socketpair = sock_no_socketpair,
3796 .accept = sock_no_accept,
3797 .getname = packet_getname,
3798 .poll = packet_poll,
3799 .ioctl = packet_ioctl,
3800 .listen = sock_no_listen,
3801 .shutdown = sock_no_shutdown,
3802 .setsockopt = packet_setsockopt,
3803 .getsockopt = packet_getsockopt,
3804 .sendmsg = packet_sendmsg,
3805 .recvmsg = packet_recvmsg,
3806 .mmap = packet_mmap,
3807 .sendpage = sock_no_sendpage,
3810 static const struct net_proto_family packet_family_ops = {
3811 .family = PF_PACKET,
3812 .create = packet_create,
3813 .owner = THIS_MODULE,
3816 static struct notifier_block packet_netdev_notifier = {
3817 .notifier_call = packet_notifier,
3820 #ifdef CONFIG_PROC_FS
3822 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3825 struct net *net = seq_file_net(seq);
3828 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3831 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3833 struct net *net = seq_file_net(seq);
3834 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3837 static void packet_seq_stop(struct seq_file *seq, void *v)
3843 static int packet_seq_show(struct seq_file *seq, void *v)
3845 if (v == SEQ_START_TOKEN)
3846 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3848 struct sock *s = sk_entry(v);
3849 const struct packet_sock *po = pkt_sk(s);
3852 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3854 atomic_read(&s->sk_refcnt),
3859 atomic_read(&s->sk_rmem_alloc),
3867 static const struct seq_operations packet_seq_ops = {
3868 .start = packet_seq_start,
3869 .next = packet_seq_next,
3870 .stop = packet_seq_stop,
3871 .show = packet_seq_show,
3874 static int packet_seq_open(struct inode *inode, struct file *file)
3876 return seq_open_net(inode, file, &packet_seq_ops,
3877 sizeof(struct seq_net_private));
3880 static const struct file_operations packet_seq_fops = {
3881 .owner = THIS_MODULE,
3882 .open = packet_seq_open,
3884 .llseek = seq_lseek,
3885 .release = seq_release_net,
3890 static int __net_init packet_net_init(struct net *net)
3892 spin_lock_init(&net->packet.sklist_lock);
3893 INIT_HLIST_HEAD(&net->packet.sklist);
3895 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3901 static void __net_exit packet_net_exit(struct net *net)
3903 proc_net_remove(net, "packet");
3906 static struct pernet_operations packet_net_ops = {
3907 .init = packet_net_init,
3908 .exit = packet_net_exit,
3912 static void __exit packet_exit(void)
3914 unregister_netdevice_notifier(&packet_netdev_notifier);
3915 unregister_pernet_subsys(&packet_net_ops);
3916 sock_unregister(PF_PACKET);
3917 proto_unregister(&packet_proto);
3920 static int __init packet_init(void)
3922 int rc = proto_register(&packet_proto, 0);
3927 sock_register(&packet_family_ops);
3928 register_pernet_subsys(&packet_net_ops);
3929 register_netdevice_notifier(&packet_netdev_notifier);
3934 module_init(packet_init);
3935 module_exit(packet_exit);
3936 MODULE_LICENSE("GPL");
3937 MODULE_ALIAS_NETPROTO(PF_PACKET);