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 if (f->num_members == 1)
1246 dev_add_pack(&f->prot_hook);
1247 spin_unlock(&f->lock);
1250 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1252 struct packet_fanout *f = po->fanout;
1255 spin_lock(&f->lock);
1256 for (i = 0; i < f->num_members; i++) {
1257 if (f->arr[i] == sk)
1260 BUG_ON(i >= f->num_members);
1261 f->arr[i] = f->arr[f->num_members - 1];
1263 if (f->num_members == 0)
1264 __dev_remove_pack(&f->prot_hook);
1265 spin_unlock(&f->lock);
1268 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1270 if (sk->sk_family != PF_PACKET)
1273 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1276 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1278 struct packet_sock *po = pkt_sk(sk);
1279 struct packet_fanout *f, *match;
1280 u8 type = type_flags & 0xff;
1281 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1285 case PACKET_FANOUT_HASH:
1286 case PACKET_FANOUT_LB:
1287 case PACKET_FANOUT_CPU:
1293 mutex_lock(&fanout_mutex);
1304 list_for_each_entry(f, &fanout_list, list) {
1306 read_pnet(&f->net) == sock_net(sk)) {
1312 if (match && match->defrag != defrag)
1316 match = kzalloc(sizeof(*match), GFP_KERNEL);
1319 write_pnet(&match->net, sock_net(sk));
1322 match->defrag = defrag;
1323 atomic_set(&match->rr_cur, 0);
1324 INIT_LIST_HEAD(&match->list);
1325 spin_lock_init(&match->lock);
1326 atomic_set(&match->sk_ref, 0);
1327 match->prot_hook.type = po->prot_hook.type;
1328 match->prot_hook.dev = po->prot_hook.dev;
1329 match->prot_hook.func = packet_rcv_fanout;
1330 match->prot_hook.af_packet_priv = match;
1331 match->prot_hook.id_match = match_fanout_group;
1332 list_add(&match->list, &fanout_list);
1335 if (match->type == type &&
1336 match->prot_hook.type == po->prot_hook.type &&
1337 match->prot_hook.dev == po->prot_hook.dev) {
1339 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1340 __dev_remove_pack(&po->prot_hook);
1342 atomic_inc(&match->sk_ref);
1343 __fanout_link(sk, po);
1348 mutex_unlock(&fanout_mutex);
1352 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1353 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1354 * It is the responsibility of the caller to call fanout_release_data() and
1355 * free the returned packet_fanout (after synchronize_net())
1357 static struct packet_fanout *fanout_release(struct sock *sk)
1359 struct packet_sock *po = pkt_sk(sk);
1360 struct packet_fanout *f;
1362 mutex_lock(&fanout_mutex);
1367 if (atomic_dec_and_test(&f->sk_ref))
1372 mutex_unlock(&fanout_mutex);
1377 static const struct proto_ops packet_ops;
1379 static const struct proto_ops packet_ops_spkt;
1381 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382 struct packet_type *pt, struct net_device *orig_dev)
1385 struct sockaddr_pkt *spkt;
1388 * When we registered the protocol we saved the socket in the data
1389 * field for just this event.
1392 sk = pt->af_packet_priv;
1395 * Yank back the headers [hope the device set this
1396 * right or kerboom...]
1398 * Incoming packets have ll header pulled,
1401 * For outgoing ones skb->data == skb_mac_header(skb)
1402 * so that this procedure is noop.
1405 if (skb->pkt_type == PACKET_LOOPBACK)
1408 if (!net_eq(dev_net(dev), sock_net(sk)))
1411 skb = skb_share_check(skb, GFP_ATOMIC);
1415 /* drop any routing info */
1418 /* drop conntrack reference */
1421 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1423 skb_push(skb, skb->data - skb_mac_header(skb));
1426 * The SOCK_PACKET socket receives _all_ frames.
1429 spkt->spkt_family = dev->type;
1430 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431 spkt->spkt_protocol = skb->protocol;
1434 * Charge the memory to the socket. This is done specifically
1435 * to prevent sockets using all the memory up.
1438 if (sock_queue_rcv_skb(sk, skb) == 0)
1449 * Output a raw packet to a device layer. This bypasses all the other
1450 * protocol layers and you must therefore supply it with a complete frame
1453 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454 struct msghdr *msg, size_t len)
1456 struct sock *sk = sock->sk;
1457 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1458 struct sk_buff *skb = NULL;
1459 struct net_device *dev;
1464 * Get and verify the address.
1468 if (msg->msg_namelen < sizeof(struct sockaddr))
1470 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471 proto = saddr->spkt_protocol;
1473 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1476 * Find the device first to size check it
1479 saddr->spkt_device[13] = 0;
1482 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1488 if (!(dev->flags & IFF_UP))
1492 * You may not queue a frame bigger than the mtu. This is the lowest level
1493 * raw protocol and you must do your own fragmentation at this level.
1497 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1501 size_t reserved = LL_RESERVED_SPACE(dev);
1502 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1505 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1508 /* FIXME: Save some space for broken drivers that write a hard
1509 * header at transmission time by themselves. PPP is the notable
1510 * one here. This should really be fixed at the driver level.
1512 skb_reserve(skb, reserved);
1513 skb_reset_network_header(skb);
1515 /* Try to align data part correctly */
1520 skb_reset_network_header(skb);
1522 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1528 if (len > (dev->mtu + dev->hard_header_len)) {
1529 /* Earlier code assumed this would be a VLAN pkt,
1530 * double-check this now that we have the actual
1533 struct ethhdr *ehdr;
1534 skb_reset_mac_header(skb);
1535 ehdr = eth_hdr(skb);
1536 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1542 skb->protocol = proto;
1544 skb->priority = sk->sk_priority;
1545 skb->mark = sk->sk_mark;
1546 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1550 dev_queue_xmit(skb);
1561 static unsigned int run_filter(const struct sk_buff *skb,
1562 const struct sock *sk,
1565 struct sk_filter *filter;
1568 filter = rcu_dereference(sk->sk_filter);
1570 res = SK_RUN_FILTER(filter, skb);
1577 * This function makes lazy skb cloning in hope that most of packets
1578 * are discarded by BPF.
1580 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1581 * and skb->cb are mangled. It works because (and until) packets
1582 * falling here are owned by current CPU. Output packets are cloned
1583 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1584 * sequencially, so that if we return skb to original state on exit,
1585 * we will not harm anyone.
1588 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1589 struct packet_type *pt, struct net_device *orig_dev)
1592 struct sockaddr_ll *sll;
1593 struct packet_sock *po;
1594 u8 *skb_head = skb->data;
1595 int skb_len = skb->len;
1596 unsigned int snaplen, res;
1598 if (skb->pkt_type == PACKET_LOOPBACK)
1601 sk = pt->af_packet_priv;
1604 if (!net_eq(dev_net(dev), sock_net(sk)))
1609 if (dev->header_ops) {
1610 /* The device has an explicit notion of ll header,
1611 * exported to higher levels.
1613 * Otherwise, the device hides details of its frame
1614 * structure, so that corresponding packet head is
1615 * never delivered to user.
1617 if (sk->sk_type != SOCK_DGRAM)
1618 skb_push(skb, skb->data - skb_mac_header(skb));
1619 else if (skb->pkt_type == PACKET_OUTGOING) {
1620 /* Special case: outgoing packets have ll header at head */
1621 skb_pull(skb, skb_network_offset(skb));
1627 res = run_filter(skb, sk, snaplen);
1629 goto drop_n_restore;
1633 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1636 if (skb_shared(skb)) {
1637 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1641 if (skb_head != skb->data) {
1642 skb->data = skb_head;
1649 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1652 sll = &PACKET_SKB_CB(skb)->sa.ll;
1653 sll->sll_family = AF_PACKET;
1654 sll->sll_hatype = dev->type;
1655 sll->sll_protocol = skb->protocol;
1656 sll->sll_pkttype = skb->pkt_type;
1657 if (unlikely(po->origdev))
1658 sll->sll_ifindex = orig_dev->ifindex;
1660 sll->sll_ifindex = dev->ifindex;
1662 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1664 PACKET_SKB_CB(skb)->origlen = skb->len;
1666 if (pskb_trim(skb, snaplen))
1669 skb_set_owner_r(skb, sk);
1673 /* drop conntrack reference */
1676 spin_lock(&sk->sk_receive_queue.lock);
1677 po->stats.tp_packets++;
1678 skb->dropcount = atomic_read(&sk->sk_drops);
1679 __skb_queue_tail(&sk->sk_receive_queue, skb);
1680 spin_unlock(&sk->sk_receive_queue.lock);
1681 sk->sk_data_ready(sk, skb->len);
1685 spin_lock(&sk->sk_receive_queue.lock);
1686 po->stats.tp_drops++;
1687 atomic_inc(&sk->sk_drops);
1688 spin_unlock(&sk->sk_receive_queue.lock);
1691 if (skb_head != skb->data && skb_shared(skb)) {
1692 skb->data = skb_head;
1700 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1701 struct packet_type *pt, struct net_device *orig_dev)
1704 struct packet_sock *po;
1705 struct sockaddr_ll *sll;
1707 struct tpacket_hdr *h1;
1708 struct tpacket2_hdr *h2;
1709 struct tpacket3_hdr *h3;
1712 u8 *skb_head = skb->data;
1713 int skb_len = skb->len;
1714 unsigned int snaplen, res;
1715 unsigned long status = TP_STATUS_USER;
1716 unsigned short macoff, netoff, hdrlen;
1717 struct sk_buff *copy_skb = NULL;
1720 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1722 if (skb->pkt_type == PACKET_LOOPBACK)
1725 sk = pt->af_packet_priv;
1728 if (!net_eq(dev_net(dev), sock_net(sk)))
1731 if (dev->header_ops) {
1732 if (sk->sk_type != SOCK_DGRAM)
1733 skb_push(skb, skb->data - skb_mac_header(skb));
1734 else if (skb->pkt_type == PACKET_OUTGOING) {
1735 /* Special case: outgoing packets have ll header at head */
1736 skb_pull(skb, skb_network_offset(skb));
1740 if (skb->ip_summed == CHECKSUM_PARTIAL)
1741 status |= TP_STATUS_CSUMNOTREADY;
1745 res = run_filter(skb, sk, snaplen);
1747 goto drop_n_restore;
1751 if (sk->sk_type == SOCK_DGRAM) {
1752 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1755 unsigned maclen = skb_network_offset(skb);
1756 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1757 (maclen < 16 ? 16 : maclen)) +
1759 macoff = netoff - maclen;
1761 if (po->tp_version <= TPACKET_V2) {
1762 if (macoff + snaplen > po->rx_ring.frame_size) {
1763 if (po->copy_thresh &&
1764 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1765 if (skb_shared(skb)) {
1766 copy_skb = skb_clone(skb, GFP_ATOMIC);
1768 copy_skb = skb_get(skb);
1769 skb_head = skb->data;
1772 skb_set_owner_r(copy_skb, sk);
1774 snaplen = po->rx_ring.frame_size - macoff;
1775 if ((int)snaplen < 0)
1779 spin_lock(&sk->sk_receive_queue.lock);
1780 h.raw = packet_current_rx_frame(po, skb,
1781 TP_STATUS_KERNEL, (macoff+snaplen));
1784 if (po->tp_version <= TPACKET_V2) {
1785 packet_increment_rx_head(po, &po->rx_ring);
1787 * LOSING will be reported till you read the stats,
1788 * because it's COR - Clear On Read.
1789 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1792 if (po->stats.tp_drops)
1793 status |= TP_STATUS_LOSING;
1795 po->stats.tp_packets++;
1797 status |= TP_STATUS_COPY;
1798 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1800 spin_unlock(&sk->sk_receive_queue.lock);
1802 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1804 switch (po->tp_version) {
1806 h.h1->tp_len = skb->len;
1807 h.h1->tp_snaplen = snaplen;
1808 h.h1->tp_mac = macoff;
1809 h.h1->tp_net = netoff;
1810 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1811 && shhwtstamps->syststamp.tv64)
1812 tv = ktime_to_timeval(shhwtstamps->syststamp);
1813 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1814 && shhwtstamps->hwtstamp.tv64)
1815 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1816 else if (skb->tstamp.tv64)
1817 tv = ktime_to_timeval(skb->tstamp);
1819 do_gettimeofday(&tv);
1820 h.h1->tp_sec = tv.tv_sec;
1821 h.h1->tp_usec = tv.tv_usec;
1822 hdrlen = sizeof(*h.h1);
1825 h.h2->tp_len = skb->len;
1826 h.h2->tp_snaplen = snaplen;
1827 h.h2->tp_mac = macoff;
1828 h.h2->tp_net = netoff;
1829 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1830 && shhwtstamps->syststamp.tv64)
1831 ts = ktime_to_timespec(shhwtstamps->syststamp);
1832 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1833 && shhwtstamps->hwtstamp.tv64)
1834 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1835 else if (skb->tstamp.tv64)
1836 ts = ktime_to_timespec(skb->tstamp);
1838 getnstimeofday(&ts);
1839 h.h2->tp_sec = ts.tv_sec;
1840 h.h2->tp_nsec = ts.tv_nsec;
1841 if (vlan_tx_tag_present(skb)) {
1842 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1843 status |= TP_STATUS_VLAN_VALID;
1845 h.h2->tp_vlan_tci = 0;
1847 h.h2->tp_padding = 0;
1848 hdrlen = sizeof(*h.h2);
1851 /* tp_nxt_offset,vlan are already populated above.
1852 * So DONT clear those fields here
1854 h.h3->tp_status |= status;
1855 h.h3->tp_len = skb->len;
1856 h.h3->tp_snaplen = snaplen;
1857 h.h3->tp_mac = macoff;
1858 h.h3->tp_net = netoff;
1859 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1860 && shhwtstamps->syststamp.tv64)
1861 ts = ktime_to_timespec(shhwtstamps->syststamp);
1862 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1863 && shhwtstamps->hwtstamp.tv64)
1864 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1865 else if (skb->tstamp.tv64)
1866 ts = ktime_to_timespec(skb->tstamp);
1868 getnstimeofday(&ts);
1869 h.h3->tp_sec = ts.tv_sec;
1870 h.h3->tp_nsec = ts.tv_nsec;
1871 hdrlen = sizeof(*h.h3);
1877 sll = h.raw + TPACKET_ALIGN(hdrlen);
1878 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1879 sll->sll_family = AF_PACKET;
1880 sll->sll_hatype = dev->type;
1881 sll->sll_protocol = skb->protocol;
1882 sll->sll_pkttype = skb->pkt_type;
1883 if (unlikely(po->origdev))
1884 sll->sll_ifindex = orig_dev->ifindex;
1886 sll->sll_ifindex = dev->ifindex;
1889 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1893 if (po->tp_version <= TPACKET_V2) {
1894 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1895 + macoff + snaplen);
1896 for (start = h.raw; start < end; start += PAGE_SIZE)
1897 flush_dcache_page(pgv_to_page(start));
1902 if (po->tp_version <= TPACKET_V2)
1903 __packet_set_status(po, h.raw, status);
1905 prb_clear_blk_fill_status(&po->rx_ring);
1907 sk->sk_data_ready(sk, 0);
1910 if (skb_head != skb->data && skb_shared(skb)) {
1911 skb->data = skb_head;
1919 po->stats.tp_drops++;
1920 spin_unlock(&sk->sk_receive_queue.lock);
1922 sk->sk_data_ready(sk, 0);
1923 kfree_skb(copy_skb);
1924 goto drop_n_restore;
1927 static void tpacket_destruct_skb(struct sk_buff *skb)
1929 struct packet_sock *po = pkt_sk(skb->sk);
1932 if (likely(po->tx_ring.pg_vec)) {
1933 ph = skb_shinfo(skb)->destructor_arg;
1934 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1935 atomic_dec(&po->tx_ring.pending);
1936 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1942 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1943 void *frame, struct net_device *dev, int size_max,
1944 __be16 proto, unsigned char *addr)
1947 struct tpacket_hdr *h1;
1948 struct tpacket2_hdr *h2;
1951 int to_write, offset, len, tp_len, nr_frags, len_max;
1952 struct socket *sock = po->sk.sk_socket;
1959 skb->protocol = proto;
1961 skb->priority = po->sk.sk_priority;
1962 skb->mark = po->sk.sk_mark;
1963 skb_shinfo(skb)->destructor_arg = ph.raw;
1965 switch (po->tp_version) {
1967 tp_len = ph.h2->tp_len;
1970 tp_len = ph.h1->tp_len;
1973 if (unlikely(tp_len > size_max)) {
1974 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1978 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1979 skb_reset_network_header(skb);
1981 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1984 if (sock->type == SOCK_DGRAM) {
1985 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1987 if (unlikely(err < 0))
1989 } else if (dev->hard_header_len) {
1990 /* net device doesn't like empty head */
1991 if (unlikely(tp_len <= dev->hard_header_len)) {
1992 pr_err("packet size is too short (%d < %d)\n",
1993 tp_len, dev->hard_header_len);
1997 skb_push(skb, dev->hard_header_len);
1998 err = skb_store_bits(skb, 0, data,
1999 dev->hard_header_len);
2003 data += dev->hard_header_len;
2004 to_write -= dev->hard_header_len;
2008 offset = offset_in_page(data);
2009 len_max = PAGE_SIZE - offset;
2010 len = ((to_write > len_max) ? len_max : to_write);
2012 skb->data_len = to_write;
2013 skb->len += to_write;
2014 skb->truesize += to_write;
2015 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2017 while (likely(to_write)) {
2018 nr_frags = skb_shinfo(skb)->nr_frags;
2020 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2021 pr_err("Packet exceed the number of skb frags(%lu)\n",
2026 page = pgv_to_page(data);
2028 flush_dcache_page(page);
2030 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2033 len_max = PAGE_SIZE;
2034 len = ((to_write > len_max) ? len_max : to_write);
2040 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2042 struct net_device *dev;
2045 dev = rcu_dereference(po->cached_dev);
2053 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2055 struct sk_buff *skb;
2056 struct net_device *dev;
2058 int err, reserve = 0;
2060 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2061 int tp_len, size_max;
2062 unsigned char *addr;
2066 mutex_lock(&po->pg_vec_lock);
2069 if (saddr == NULL) {
2070 dev = packet_cached_dev_get(po);
2075 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2077 if (msg->msg_namelen < (saddr->sll_halen
2078 + offsetof(struct sockaddr_ll,
2081 proto = saddr->sll_protocol;
2082 addr = saddr->sll_addr;
2083 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2087 if (unlikely(dev == NULL))
2090 if (unlikely(!(dev->flags & IFF_UP)))
2093 reserve = dev->hard_header_len;
2095 size_max = po->tx_ring.frame_size
2096 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2098 if (size_max > dev->mtu + reserve)
2099 size_max = dev->mtu + reserve;
2102 ph = packet_current_frame(po, &po->tx_ring,
2103 TP_STATUS_SEND_REQUEST);
2105 if (unlikely(ph == NULL)) {
2110 status = TP_STATUS_SEND_REQUEST;
2111 skb = sock_alloc_send_skb(&po->sk,
2112 LL_ALLOCATED_SPACE(dev)
2113 + sizeof(struct sockaddr_ll),
2116 if (unlikely(skb == NULL))
2119 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2122 if (unlikely(tp_len < 0)) {
2124 __packet_set_status(po, ph,
2125 TP_STATUS_AVAILABLE);
2126 packet_increment_head(&po->tx_ring);
2130 status = TP_STATUS_WRONG_FORMAT;
2136 skb->destructor = tpacket_destruct_skb;
2137 __packet_set_status(po, ph, TP_STATUS_SENDING);
2138 atomic_inc(&po->tx_ring.pending);
2140 status = TP_STATUS_SEND_REQUEST;
2141 err = dev_queue_xmit(skb);
2142 if (unlikely(err > 0)) {
2143 err = net_xmit_errno(err);
2144 if (err && __packet_get_status(po, ph) ==
2145 TP_STATUS_AVAILABLE) {
2146 /* skb was destructed already */
2151 * skb was dropped but not destructed yet;
2152 * let's treat it like congestion or err < 0
2156 packet_increment_head(&po->tx_ring);
2158 } while (likely((ph != NULL) ||
2159 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2160 (atomic_read(&po->tx_ring.pending))))
2167 __packet_set_status(po, ph, status);
2172 mutex_unlock(&po->pg_vec_lock);
2176 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2177 size_t reserve, size_t len,
2178 size_t linear, int noblock,
2181 struct sk_buff *skb;
2183 /* Under a page? Don't bother with paged skb. */
2184 if (prepad + len < PAGE_SIZE || !linear)
2187 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2192 skb_reserve(skb, reserve);
2193 skb_put(skb, linear);
2194 skb->data_len = len - linear;
2195 skb->len += len - linear;
2200 static int packet_snd(struct socket *sock,
2201 struct msghdr *msg, size_t len)
2203 struct sock *sk = sock->sk;
2204 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2205 struct sk_buff *skb;
2206 struct net_device *dev;
2208 unsigned char *addr;
2209 int err, reserve = 0;
2210 struct virtio_net_hdr vnet_hdr = { 0 };
2213 struct packet_sock *po = pkt_sk(sk);
2214 unsigned short gso_type = 0;
2217 * Get and verify the address.
2220 if (saddr == NULL) {
2221 dev = packet_cached_dev_get(po);
2226 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2228 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2230 proto = saddr->sll_protocol;
2231 addr = saddr->sll_addr;
2232 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2236 if (unlikely(dev == NULL))
2239 if (unlikely(!(dev->flags & IFF_UP)))
2242 if (sock->type == SOCK_RAW)
2243 reserve = dev->hard_header_len;
2244 if (po->has_vnet_hdr) {
2245 vnet_hdr_len = sizeof(vnet_hdr);
2248 if (len < vnet_hdr_len)
2251 len -= vnet_hdr_len;
2253 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2258 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2259 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2261 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2262 vnet_hdr.csum_offset + 2;
2265 if (vnet_hdr.hdr_len > len)
2268 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2269 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2270 case VIRTIO_NET_HDR_GSO_TCPV4:
2271 gso_type = SKB_GSO_TCPV4;
2273 case VIRTIO_NET_HDR_GSO_TCPV6:
2274 gso_type = SKB_GSO_TCPV6;
2276 case VIRTIO_NET_HDR_GSO_UDP:
2277 gso_type = SKB_GSO_UDP;
2283 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2284 gso_type |= SKB_GSO_TCP_ECN;
2286 if (vnet_hdr.gso_size == 0)
2293 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2297 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2298 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2299 msg->msg_flags & MSG_DONTWAIT, &err);
2303 skb_set_network_header(skb, reserve);
2306 if (sock->type == SOCK_DGRAM &&
2307 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2310 /* Returns -EFAULT on error */
2311 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2314 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2318 if (!gso_type && (len > dev->mtu + reserve)) {
2319 /* Earlier code assumed this would be a VLAN pkt,
2320 * double-check this now that we have the actual
2323 struct ethhdr *ehdr;
2324 skb_reset_mac_header(skb);
2325 ehdr = eth_hdr(skb);
2326 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2332 skb->protocol = proto;
2334 skb->priority = sk->sk_priority;
2335 skb->mark = sk->sk_mark;
2337 if (po->has_vnet_hdr) {
2338 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2339 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2340 vnet_hdr.csum_offset)) {
2346 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2347 skb_shinfo(skb)->gso_type = gso_type;
2349 /* Header must be checked, and gso_segs computed. */
2350 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2351 skb_shinfo(skb)->gso_segs = 0;
2353 len += vnet_hdr_len;
2360 err = dev_queue_xmit(skb);
2361 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2377 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2378 struct msghdr *msg, size_t len)
2380 struct sock *sk = sock->sk;
2381 struct packet_sock *po = pkt_sk(sk);
2382 if (po->tx_ring.pg_vec)
2383 return tpacket_snd(po, msg);
2385 return packet_snd(sock, msg, len);
2389 * Close a PACKET socket. This is fairly simple. We immediately go
2390 * to 'closed' state and remove our protocol entry in the device list.
2393 static int packet_release(struct socket *sock)
2395 struct sock *sk = sock->sk;
2396 struct packet_sock *po;
2397 struct packet_fanout *f;
2399 union tpacket_req_u req_u;
2407 spin_lock_bh(&net->packet.sklist_lock);
2408 sk_del_node_init_rcu(sk);
2409 sock_prot_inuse_add(net, sk->sk_prot, -1);
2410 spin_unlock_bh(&net->packet.sklist_lock);
2412 spin_lock(&po->bind_lock);
2413 unregister_prot_hook(sk, false);
2414 if (po->prot_hook.dev) {
2415 dev_put(po->prot_hook.dev);
2416 po->prot_hook.dev = NULL;
2418 spin_unlock(&po->bind_lock);
2420 packet_flush_mclist(sk);
2422 if (po->rx_ring.pg_vec) {
2423 memset(&req_u, 0, sizeof(req_u));
2424 packet_set_ring(sk, &req_u, 1, 0);
2427 if (po->tx_ring.pg_vec) {
2428 memset(&req_u, 0, sizeof(req_u));
2429 packet_set_ring(sk, &req_u, 1, 1);
2432 f = fanout_release(sk);
2439 * Now the socket is dead. No more input will appear.
2446 skb_queue_purge(&sk->sk_receive_queue);
2447 sk_refcnt_debug_release(sk);
2454 * Attach a packet hook.
2457 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2459 struct packet_sock *po = pkt_sk(sk);
2470 spin_lock(&po->bind_lock);
2471 unregister_prot_hook(sk, true);
2473 po->prot_hook.type = protocol;
2474 if (po->prot_hook.dev)
2475 dev_put(po->prot_hook.dev);
2476 po->prot_hook.dev = dev;
2478 po->ifindex = dev ? dev->ifindex : 0;
2483 if (!dev || (dev->flags & IFF_UP)) {
2484 register_prot_hook(sk);
2486 sk->sk_err = ENETDOWN;
2487 if (!sock_flag(sk, SOCK_DEAD))
2488 sk->sk_error_report(sk);
2492 spin_unlock(&po->bind_lock);
2498 * Bind a packet socket to a device
2501 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2504 struct sock *sk = sock->sk;
2506 struct net_device *dev;
2513 if (addr_len != sizeof(struct sockaddr))
2515 strlcpy(name, uaddr->sa_data, sizeof(name));
2517 dev = dev_get_by_name(sock_net(sk), name);
2519 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2523 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2525 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2526 struct sock *sk = sock->sk;
2527 struct net_device *dev = NULL;
2535 if (addr_len < sizeof(struct sockaddr_ll))
2537 if (sll->sll_family != AF_PACKET)
2540 if (sll->sll_ifindex) {
2542 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2546 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2552 static struct proto packet_proto = {
2554 .owner = THIS_MODULE,
2555 .obj_size = sizeof(struct packet_sock),
2559 * Create a packet of type SOCK_PACKET.
2562 static int packet_create(struct net *net, struct socket *sock, int protocol,
2566 struct packet_sock *po;
2567 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2570 if (!capable(CAP_NET_RAW))
2572 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2573 sock->type != SOCK_PACKET)
2574 return -ESOCKTNOSUPPORT;
2576 sock->state = SS_UNCONNECTED;
2579 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2583 sock->ops = &packet_ops;
2584 if (sock->type == SOCK_PACKET)
2585 sock->ops = &packet_ops_spkt;
2587 sock_init_data(sock, sk);
2590 sk->sk_family = PF_PACKET;
2592 RCU_INIT_POINTER(po->cached_dev, NULL);
2594 sk->sk_destruct = packet_sock_destruct;
2595 sk_refcnt_debug_inc(sk);
2598 * Attach a protocol block
2601 spin_lock_init(&po->bind_lock);
2602 mutex_init(&po->pg_vec_lock);
2603 po->prot_hook.func = packet_rcv;
2605 if (sock->type == SOCK_PACKET)
2606 po->prot_hook.func = packet_rcv_spkt;
2608 po->prot_hook.af_packet_priv = sk;
2611 po->prot_hook.type = proto;
2612 register_prot_hook(sk);
2615 spin_lock_bh(&net->packet.sklist_lock);
2616 sk_add_node_rcu(sk, &net->packet.sklist);
2617 sock_prot_inuse_add(net, &packet_proto, 1);
2618 spin_unlock_bh(&net->packet.sklist_lock);
2625 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2627 struct sock_exterr_skb *serr;
2628 struct sk_buff *skb, *skb2;
2632 skb = skb_dequeue(&sk->sk_error_queue);
2638 msg->msg_flags |= MSG_TRUNC;
2641 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2645 sock_recv_timestamp(msg, sk, skb);
2647 serr = SKB_EXT_ERR(skb);
2648 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2649 sizeof(serr->ee), &serr->ee);
2651 msg->msg_flags |= MSG_ERRQUEUE;
2654 /* Reset and regenerate socket error */
2655 spin_lock_bh(&sk->sk_error_queue.lock);
2657 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2658 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2659 spin_unlock_bh(&sk->sk_error_queue.lock);
2660 sk->sk_error_report(sk);
2662 spin_unlock_bh(&sk->sk_error_queue.lock);
2671 * Pull a packet from our receive queue and hand it to the user.
2672 * If necessary we block.
2675 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2676 struct msghdr *msg, size_t len, int flags)
2678 struct sock *sk = sock->sk;
2679 struct sk_buff *skb;
2681 int vnet_hdr_len = 0;
2684 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2688 /* What error should we return now? EUNATTACH? */
2689 if (pkt_sk(sk)->ifindex < 0)
2693 if (flags & MSG_ERRQUEUE) {
2694 err = packet_recv_error(sk, msg, len);
2699 * Call the generic datagram receiver. This handles all sorts
2700 * of horrible races and re-entrancy so we can forget about it
2701 * in the protocol layers.
2703 * Now it will return ENETDOWN, if device have just gone down,
2704 * but then it will block.
2707 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2710 * An error occurred so return it. Because skb_recv_datagram()
2711 * handles the blocking we don't see and worry about blocking
2718 if (pkt_sk(sk)->has_vnet_hdr) {
2719 struct virtio_net_hdr vnet_hdr = { 0 };
2722 vnet_hdr_len = sizeof(vnet_hdr);
2723 if (len < vnet_hdr_len)
2726 len -= vnet_hdr_len;
2728 if (skb_is_gso(skb)) {
2729 struct skb_shared_info *sinfo = skb_shinfo(skb);
2731 /* This is a hint as to how much should be linear. */
2732 vnet_hdr.hdr_len = skb_headlen(skb);
2733 vnet_hdr.gso_size = sinfo->gso_size;
2734 if (sinfo->gso_type & SKB_GSO_TCPV4)
2735 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2736 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2737 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2738 else if (sinfo->gso_type & SKB_GSO_UDP)
2739 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2740 else if (sinfo->gso_type & SKB_GSO_FCOE)
2744 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2745 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2747 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2749 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2750 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2751 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2752 vnet_hdr.csum_offset = skb->csum_offset;
2753 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2754 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2755 } /* else everything is zero */
2757 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2763 /* You lose any data beyond the buffer you gave. If it worries
2764 * a user program they can ask the device for its MTU
2770 msg->msg_flags |= MSG_TRUNC;
2773 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2777 sock_recv_ts_and_drops(msg, sk, skb);
2779 if (msg->msg_name) {
2780 /* If the address length field is there to be filled
2781 * in, we fill it in now.
2783 if (sock->type == SOCK_PACKET) {
2784 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2786 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2787 msg->msg_namelen = sll->sll_halen +
2788 offsetof(struct sockaddr_ll, sll_addr);
2790 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2794 if (pkt_sk(sk)->auxdata) {
2795 struct tpacket_auxdata aux;
2797 aux.tp_status = TP_STATUS_USER;
2798 if (skb->ip_summed == CHECKSUM_PARTIAL)
2799 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2800 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2801 aux.tp_snaplen = skb->len;
2803 aux.tp_net = skb_network_offset(skb);
2804 if (vlan_tx_tag_present(skb)) {
2805 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2806 aux.tp_status |= TP_STATUS_VLAN_VALID;
2808 aux.tp_vlan_tci = 0;
2811 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2815 * Free or return the buffer as appropriate. Again this
2816 * hides all the races and re-entrancy issues from us.
2818 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2821 skb_free_datagram(sk, skb);
2826 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2827 int *uaddr_len, int peer)
2829 struct net_device *dev;
2830 struct sock *sk = sock->sk;
2835 uaddr->sa_family = AF_PACKET;
2836 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2838 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2840 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2842 *uaddr_len = sizeof(*uaddr);
2847 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2848 int *uaddr_len, int peer)
2850 struct net_device *dev;
2851 struct sock *sk = sock->sk;
2852 struct packet_sock *po = pkt_sk(sk);
2853 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2858 sll->sll_family = AF_PACKET;
2859 sll->sll_ifindex = po->ifindex;
2860 sll->sll_protocol = po->num;
2861 sll->sll_pkttype = 0;
2863 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2865 sll->sll_hatype = dev->type;
2866 sll->sll_halen = dev->addr_len;
2867 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2869 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2873 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2878 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2882 case PACKET_MR_MULTICAST:
2883 if (i->alen != dev->addr_len)
2886 return dev_mc_add(dev, i->addr);
2888 return dev_mc_del(dev, i->addr);
2890 case PACKET_MR_PROMISC:
2891 return dev_set_promiscuity(dev, what);
2893 case PACKET_MR_ALLMULTI:
2894 return dev_set_allmulti(dev, what);
2896 case PACKET_MR_UNICAST:
2897 if (i->alen != dev->addr_len)
2900 return dev_uc_add(dev, i->addr);
2902 return dev_uc_del(dev, i->addr);
2910 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2912 for ( ; i; i = i->next) {
2913 if (i->ifindex == dev->ifindex)
2914 packet_dev_mc(dev, i, what);
2918 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2920 struct packet_sock *po = pkt_sk(sk);
2921 struct packet_mclist *ml, *i;
2922 struct net_device *dev;
2928 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2933 if (mreq->mr_alen > dev->addr_len)
2937 i = kmalloc(sizeof(*i), GFP_KERNEL);
2942 for (ml = po->mclist; ml; ml = ml->next) {
2943 if (ml->ifindex == mreq->mr_ifindex &&
2944 ml->type == mreq->mr_type &&
2945 ml->alen == mreq->mr_alen &&
2946 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2948 /* Free the new element ... */
2954 i->type = mreq->mr_type;
2955 i->ifindex = mreq->mr_ifindex;
2956 i->alen = mreq->mr_alen;
2957 memcpy(i->addr, mreq->mr_address, i->alen);
2959 i->next = po->mclist;
2961 err = packet_dev_mc(dev, i, 1);
2963 po->mclist = i->next;
2972 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2974 struct packet_mclist *ml, **mlp;
2978 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2979 if (ml->ifindex == mreq->mr_ifindex &&
2980 ml->type == mreq->mr_type &&
2981 ml->alen == mreq->mr_alen &&
2982 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2983 if (--ml->count == 0) {
2984 struct net_device *dev;
2986 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2988 packet_dev_mc(dev, ml, -1);
2996 return -EADDRNOTAVAIL;
2999 static void packet_flush_mclist(struct sock *sk)
3001 struct packet_sock *po = pkt_sk(sk);
3002 struct packet_mclist *ml;
3008 while ((ml = po->mclist) != NULL) {
3009 struct net_device *dev;
3011 po->mclist = ml->next;
3012 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3014 packet_dev_mc(dev, ml, -1);
3021 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3023 struct sock *sk = sock->sk;
3024 struct packet_sock *po = pkt_sk(sk);
3027 if (level != SOL_PACKET)
3028 return -ENOPROTOOPT;
3031 case PACKET_ADD_MEMBERSHIP:
3032 case PACKET_DROP_MEMBERSHIP:
3034 struct packet_mreq_max mreq;
3036 memset(&mreq, 0, sizeof(mreq));
3037 if (len < sizeof(struct packet_mreq))
3039 if (len > sizeof(mreq))
3041 if (copy_from_user(&mreq, optval, len))
3043 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3045 if (optname == PACKET_ADD_MEMBERSHIP)
3046 ret = packet_mc_add(sk, &mreq);
3048 ret = packet_mc_drop(sk, &mreq);
3052 case PACKET_RX_RING:
3053 case PACKET_TX_RING:
3055 union tpacket_req_u req_u;
3058 switch (po->tp_version) {
3061 len = sizeof(req_u.req);
3065 len = sizeof(req_u.req3);
3070 if (pkt_sk(sk)->has_vnet_hdr)
3072 if (copy_from_user(&req_u.req, optval, len))
3074 return packet_set_ring(sk, &req_u, 0,
3075 optname == PACKET_TX_RING);
3077 case PACKET_COPY_THRESH:
3081 if (optlen != sizeof(val))
3083 if (copy_from_user(&val, optval, sizeof(val)))
3086 pkt_sk(sk)->copy_thresh = val;
3089 case PACKET_VERSION:
3093 if (optlen != sizeof(val))
3095 if (copy_from_user(&val, optval, sizeof(val)))
3106 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3109 po->tp_version = val;
3115 case PACKET_RESERVE:
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_reserve = val;
3132 if (optlen != sizeof(val))
3134 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3136 if (copy_from_user(&val, optval, sizeof(val)))
3138 po->tp_loss = !!val;
3141 case PACKET_AUXDATA:
3145 if (optlen < sizeof(val))
3147 if (copy_from_user(&val, optval, sizeof(val)))
3150 po->auxdata = !!val;
3153 case PACKET_ORIGDEV:
3157 if (optlen < sizeof(val))
3159 if (copy_from_user(&val, optval, sizeof(val)))
3162 po->origdev = !!val;
3165 case PACKET_VNET_HDR:
3169 if (sock->type != SOCK_RAW)
3171 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3173 if (optlen < sizeof(val))
3175 if (copy_from_user(&val, optval, sizeof(val)))
3178 po->has_vnet_hdr = !!val;
3181 case PACKET_TIMESTAMP:
3185 if (optlen != sizeof(val))
3187 if (copy_from_user(&val, optval, sizeof(val)))
3190 po->tp_tstamp = val;
3197 if (optlen != sizeof(val))
3199 if (copy_from_user(&val, optval, sizeof(val)))
3202 return fanout_add(sk, val & 0xffff, val >> 16);
3205 return -ENOPROTOOPT;
3209 static int packet_getsockopt(struct socket *sock, int level, int optname,
3210 char __user *optval, int __user *optlen)
3214 struct sock *sk = sock->sk;
3215 struct packet_sock *po = pkt_sk(sk);
3217 struct tpacket_stats st;
3218 union tpacket_stats_u st_u;
3220 if (level != SOL_PACKET)
3221 return -ENOPROTOOPT;
3223 if (get_user(len, optlen))
3230 case PACKET_STATISTICS:
3231 if (po->tp_version == TPACKET_V3) {
3232 len = sizeof(struct tpacket_stats_v3);
3234 if (len > sizeof(struct tpacket_stats))
3235 len = sizeof(struct tpacket_stats);
3237 spin_lock_bh(&sk->sk_receive_queue.lock);
3238 if (po->tp_version == TPACKET_V3) {
3239 memcpy(&st_u.stats3, &po->stats,
3240 sizeof(struct tpacket_stats));
3241 st_u.stats3.tp_freeze_q_cnt =
3242 po->stats_u.stats3.tp_freeze_q_cnt;
3243 st_u.stats3.tp_packets += po->stats.tp_drops;
3244 data = &st_u.stats3;
3247 st.tp_packets += st.tp_drops;
3250 memset(&po->stats, 0, sizeof(st));
3251 spin_unlock_bh(&sk->sk_receive_queue.lock);
3253 case PACKET_AUXDATA:
3254 if (len > sizeof(int))
3260 case PACKET_ORIGDEV:
3261 if (len > sizeof(int))
3267 case PACKET_VNET_HDR:
3268 if (len > sizeof(int))
3270 val = po->has_vnet_hdr;
3274 case PACKET_VERSION:
3275 if (len > sizeof(int))
3277 val = po->tp_version;
3281 if (len > sizeof(int))
3283 if (copy_from_user(&val, optval, len))
3287 val = sizeof(struct tpacket_hdr);
3290 val = sizeof(struct tpacket2_hdr);
3293 val = sizeof(struct tpacket3_hdr);
3300 case PACKET_RESERVE:
3301 if (len > sizeof(unsigned int))
3302 len = sizeof(unsigned int);
3303 val = po->tp_reserve;
3307 if (len > sizeof(unsigned int))
3308 len = sizeof(unsigned int);
3312 case PACKET_TIMESTAMP:
3313 if (len > sizeof(int))
3315 val = po->tp_tstamp;
3319 if (len > sizeof(int))
3322 ((u32)po->fanout->id |
3323 ((u32)po->fanout->type << 16)) :
3328 return -ENOPROTOOPT;
3331 if (put_user(len, optlen))
3333 if (copy_to_user(optval, data, len))
3339 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3342 struct hlist_node *node;
3343 struct net_device *dev = data;
3344 struct net *net = dev_net(dev);
3347 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3348 struct packet_sock *po = pkt_sk(sk);
3351 case NETDEV_UNREGISTER:
3353 packet_dev_mclist(dev, po->mclist, -1);
3357 if (dev->ifindex == po->ifindex) {
3358 spin_lock(&po->bind_lock);
3360 __unregister_prot_hook(sk, false);
3361 sk->sk_err = ENETDOWN;
3362 if (!sock_flag(sk, SOCK_DEAD))
3363 sk->sk_error_report(sk);
3365 if (msg == NETDEV_UNREGISTER) {
3367 if (po->prot_hook.dev)
3368 dev_put(po->prot_hook.dev);
3369 po->prot_hook.dev = NULL;
3371 spin_unlock(&po->bind_lock);
3375 if (dev->ifindex == po->ifindex) {
3376 spin_lock(&po->bind_lock);
3378 register_prot_hook(sk);
3379 spin_unlock(&po->bind_lock);
3389 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3392 struct sock *sk = sock->sk;
3397 int amount = sk_wmem_alloc_get(sk);
3399 return put_user(amount, (int __user *)arg);
3403 struct sk_buff *skb;
3406 spin_lock_bh(&sk->sk_receive_queue.lock);
3407 skb = skb_peek(&sk->sk_receive_queue);
3410 spin_unlock_bh(&sk->sk_receive_queue.lock);
3411 return put_user(amount, (int __user *)arg);
3414 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3416 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3426 case SIOCGIFBRDADDR:
3427 case SIOCSIFBRDADDR:
3428 case SIOCGIFNETMASK:
3429 case SIOCSIFNETMASK:
3430 case SIOCGIFDSTADDR:
3431 case SIOCSIFDSTADDR:
3433 return inet_dgram_ops.ioctl(sock, cmd, arg);
3437 return -ENOIOCTLCMD;
3442 static unsigned int packet_poll(struct file *file, struct socket *sock,
3445 struct sock *sk = sock->sk;
3446 struct packet_sock *po = pkt_sk(sk);
3447 unsigned int mask = datagram_poll(file, sock, wait);
3449 spin_lock_bh(&sk->sk_receive_queue.lock);
3450 if (po->rx_ring.pg_vec) {
3451 if (!packet_previous_rx_frame(po, &po->rx_ring,
3453 mask |= POLLIN | POLLRDNORM;
3455 spin_unlock_bh(&sk->sk_receive_queue.lock);
3456 spin_lock_bh(&sk->sk_write_queue.lock);
3457 if (po->tx_ring.pg_vec) {
3458 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3459 mask |= POLLOUT | POLLWRNORM;
3461 spin_unlock_bh(&sk->sk_write_queue.lock);
3466 /* Dirty? Well, I still did not learn better way to account
3470 static void packet_mm_open(struct vm_area_struct *vma)
3472 struct file *file = vma->vm_file;
3473 struct socket *sock = file->private_data;
3474 struct sock *sk = sock->sk;
3477 atomic_inc(&pkt_sk(sk)->mapped);
3480 static void packet_mm_close(struct vm_area_struct *vma)
3482 struct file *file = vma->vm_file;
3483 struct socket *sock = file->private_data;
3484 struct sock *sk = sock->sk;
3487 atomic_dec(&pkt_sk(sk)->mapped);
3490 static const struct vm_operations_struct packet_mmap_ops = {
3491 .open = packet_mm_open,
3492 .close = packet_mm_close,
3495 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3500 for (i = 0; i < len; i++) {
3501 if (likely(pg_vec[i].buffer)) {
3502 if (is_vmalloc_addr(pg_vec[i].buffer))
3503 vfree(pg_vec[i].buffer);
3505 free_pages((unsigned long)pg_vec[i].buffer,
3507 pg_vec[i].buffer = NULL;
3513 static char *alloc_one_pg_vec_page(unsigned long order)
3515 char *buffer = NULL;
3516 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3517 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3519 buffer = (char *) __get_free_pages(gfp_flags, order);
3525 * __get_free_pages failed, fall back to vmalloc
3527 buffer = vzalloc((1 << order) * PAGE_SIZE);
3533 * vmalloc failed, lets dig into swap here
3535 gfp_flags &= ~__GFP_NORETRY;
3536 buffer = (char *)__get_free_pages(gfp_flags, order);
3541 * complete and utter failure
3546 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3548 unsigned int block_nr = req->tp_block_nr;
3552 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3553 if (unlikely(!pg_vec))
3556 for (i = 0; i < block_nr; i++) {
3557 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3558 if (unlikely(!pg_vec[i].buffer))
3559 goto out_free_pgvec;
3566 free_pg_vec(pg_vec, order, block_nr);
3571 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3572 int closing, int tx_ring)
3574 struct pgv *pg_vec = NULL;
3575 struct packet_sock *po = pkt_sk(sk);
3576 int was_running, order = 0;
3577 struct packet_ring_buffer *rb;
3578 struct sk_buff_head *rb_queue;
3581 /* Added to avoid minimal code churn */
3582 struct tpacket_req *req = &req_u->req;
3585 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3586 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3587 WARN(1, "Tx-ring is not supported.\n");
3591 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3592 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3596 if (atomic_read(&po->mapped))
3598 if (atomic_read(&rb->pending))
3602 if (req->tp_block_nr) {
3603 /* Sanity tests and some calculations */
3605 if (unlikely(rb->pg_vec))
3608 switch (po->tp_version) {
3610 po->tp_hdrlen = TPACKET_HDRLEN;
3613 po->tp_hdrlen = TPACKET2_HDRLEN;
3616 po->tp_hdrlen = TPACKET3_HDRLEN;
3621 if (unlikely((int)req->tp_block_size <= 0))
3623 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3625 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3628 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3631 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3632 if (unlikely(rb->frames_per_block <= 0))
3634 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3639 order = get_order(req->tp_block_size);
3640 pg_vec = alloc_pg_vec(req, order);
3641 if (unlikely(!pg_vec))
3643 switch (po->tp_version) {
3645 /* Transmit path is not supported. We checked
3646 * it above but just being paranoid
3649 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3658 if (unlikely(req->tp_frame_nr))
3663 /* Detach socket from network */
3664 spin_lock(&po->bind_lock);
3665 was_running = po->running;
3669 __unregister_prot_hook(sk, false);
3671 spin_unlock(&po->bind_lock);
3676 mutex_lock(&po->pg_vec_lock);
3677 if (closing || atomic_read(&po->mapped) == 0) {
3679 spin_lock_bh(&rb_queue->lock);
3680 swap(rb->pg_vec, pg_vec);
3681 rb->frame_max = (req->tp_frame_nr - 1);
3683 rb->frame_size = req->tp_frame_size;
3684 spin_unlock_bh(&rb_queue->lock);
3686 swap(rb->pg_vec_order, order);
3687 swap(rb->pg_vec_len, req->tp_block_nr);
3689 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3690 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3691 tpacket_rcv : packet_rcv;
3692 skb_queue_purge(rb_queue);
3693 if (atomic_read(&po->mapped))
3694 pr_err("packet_mmap: vma is busy: %d\n",
3695 atomic_read(&po->mapped));
3697 mutex_unlock(&po->pg_vec_lock);
3699 spin_lock(&po->bind_lock);
3702 register_prot_hook(sk);
3704 spin_unlock(&po->bind_lock);
3705 if (closing && (po->tp_version > TPACKET_V2)) {
3706 /* Because we don't support block-based V3 on tx-ring */
3708 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3712 free_pg_vec(pg_vec, order, req->tp_block_nr);
3718 static int packet_mmap(struct file *file, struct socket *sock,
3719 struct vm_area_struct *vma)
3721 struct sock *sk = sock->sk;
3722 struct packet_sock *po = pkt_sk(sk);
3723 unsigned long size, expected_size;
3724 struct packet_ring_buffer *rb;
3725 unsigned long start;
3732 mutex_lock(&po->pg_vec_lock);
3735 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3737 expected_size += rb->pg_vec_len
3743 if (expected_size == 0)
3746 size = vma->vm_end - vma->vm_start;
3747 if (size != expected_size)
3750 start = vma->vm_start;
3751 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3752 if (rb->pg_vec == NULL)
3755 for (i = 0; i < rb->pg_vec_len; i++) {
3757 void *kaddr = rb->pg_vec[i].buffer;
3760 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3761 page = pgv_to_page(kaddr);
3762 err = vm_insert_page(vma, start, page);
3771 atomic_inc(&po->mapped);
3772 vma->vm_ops = &packet_mmap_ops;
3776 mutex_unlock(&po->pg_vec_lock);
3780 static const struct proto_ops packet_ops_spkt = {
3781 .family = PF_PACKET,
3782 .owner = THIS_MODULE,
3783 .release = packet_release,
3784 .bind = packet_bind_spkt,
3785 .connect = sock_no_connect,
3786 .socketpair = sock_no_socketpair,
3787 .accept = sock_no_accept,
3788 .getname = packet_getname_spkt,
3789 .poll = datagram_poll,
3790 .ioctl = packet_ioctl,
3791 .listen = sock_no_listen,
3792 .shutdown = sock_no_shutdown,
3793 .setsockopt = sock_no_setsockopt,
3794 .getsockopt = sock_no_getsockopt,
3795 .sendmsg = packet_sendmsg_spkt,
3796 .recvmsg = packet_recvmsg,
3797 .mmap = sock_no_mmap,
3798 .sendpage = sock_no_sendpage,
3801 static const struct proto_ops packet_ops = {
3802 .family = PF_PACKET,
3803 .owner = THIS_MODULE,
3804 .release = packet_release,
3805 .bind = packet_bind,
3806 .connect = sock_no_connect,
3807 .socketpair = sock_no_socketpair,
3808 .accept = sock_no_accept,
3809 .getname = packet_getname,
3810 .poll = packet_poll,
3811 .ioctl = packet_ioctl,
3812 .listen = sock_no_listen,
3813 .shutdown = sock_no_shutdown,
3814 .setsockopt = packet_setsockopt,
3815 .getsockopt = packet_getsockopt,
3816 .sendmsg = packet_sendmsg,
3817 .recvmsg = packet_recvmsg,
3818 .mmap = packet_mmap,
3819 .sendpage = sock_no_sendpage,
3822 static const struct net_proto_family packet_family_ops = {
3823 .family = PF_PACKET,
3824 .create = packet_create,
3825 .owner = THIS_MODULE,
3828 static struct notifier_block packet_netdev_notifier = {
3829 .notifier_call = packet_notifier,
3832 #ifdef CONFIG_PROC_FS
3834 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3837 struct net *net = seq_file_net(seq);
3840 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3843 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3845 struct net *net = seq_file_net(seq);
3846 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3849 static void packet_seq_stop(struct seq_file *seq, void *v)
3855 static int packet_seq_show(struct seq_file *seq, void *v)
3857 if (v == SEQ_START_TOKEN)
3858 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3860 struct sock *s = sk_entry(v);
3861 const struct packet_sock *po = pkt_sk(s);
3864 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3866 atomic_read(&s->sk_refcnt),
3871 atomic_read(&s->sk_rmem_alloc),
3879 static const struct seq_operations packet_seq_ops = {
3880 .start = packet_seq_start,
3881 .next = packet_seq_next,
3882 .stop = packet_seq_stop,
3883 .show = packet_seq_show,
3886 static int packet_seq_open(struct inode *inode, struct file *file)
3888 return seq_open_net(inode, file, &packet_seq_ops,
3889 sizeof(struct seq_net_private));
3892 static const struct file_operations packet_seq_fops = {
3893 .owner = THIS_MODULE,
3894 .open = packet_seq_open,
3896 .llseek = seq_lseek,
3897 .release = seq_release_net,
3902 static int __net_init packet_net_init(struct net *net)
3904 spin_lock_init(&net->packet.sklist_lock);
3905 INIT_HLIST_HEAD(&net->packet.sklist);
3907 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3913 static void __net_exit packet_net_exit(struct net *net)
3915 proc_net_remove(net, "packet");
3918 static struct pernet_operations packet_net_ops = {
3919 .init = packet_net_init,
3920 .exit = packet_net_exit,
3924 static void __exit packet_exit(void)
3926 unregister_netdevice_notifier(&packet_netdev_notifier);
3927 unregister_pernet_subsys(&packet_net_ops);
3928 sock_unregister(PF_PACKET);
3929 proto_unregister(&packet_proto);
3932 static int __init packet_init(void)
3934 int rc = proto_register(&packet_proto, 0);
3939 sock_register(&packet_family_ops);
3940 register_pernet_subsys(&packet_net_ops);
3941 register_netdevice_notifier(&packet_netdev_notifier);
3946 module_init(packet_init);
3947 module_exit(packet_exit);
3948 MODULE_LICENSE("GPL");
3949 MODULE_ALIAS_NETPROTO(PF_PACKET);