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 packet_type prot_hook ____cacheline_aligned_in_smp;
301 #define PACKET_FANOUT_MAX 256
303 struct packet_fanout {
307 unsigned int num_members;
312 struct list_head list;
313 struct sock *arr[PACKET_FANOUT_MAX];
316 struct packet_type prot_hook ____cacheline_aligned_in_smp;
319 struct packet_skb_cb {
320 unsigned int origlen;
322 struct sockaddr_pkt pkt;
323 struct sockaddr_ll ll;
327 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
329 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
330 #define GET_PBLOCK_DESC(x, bid) \
331 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
333 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
334 #define GET_NEXT_PRB_BLK_NUM(x) \
335 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336 ((x)->kactive_blk_num+1) : 0)
338 static struct packet_sock *pkt_sk(struct sock *sk)
340 return (struct packet_sock *)sk;
343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
346 /* register_prot_hook must be invoked with the po->bind_lock held,
347 * or from a context in which asynchronous accesses to the packet
348 * socket is not possible (packet_create()).
350 static void register_prot_hook(struct sock *sk)
352 struct packet_sock *po = pkt_sk(sk);
355 __fanout_link(sk, po);
357 dev_add_pack(&po->prot_hook);
363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364 * held. If the sync parameter is true, we will temporarily drop
365 * the po->bind_lock and do a synchronize_net to make sure no
366 * asynchronous packet processing paths still refer to the elements
367 * of po->prot_hook. If the sync parameter is false, it is the
368 * callers responsibility to take care of this.
370 static void __unregister_prot_hook(struct sock *sk, bool sync)
372 struct packet_sock *po = pkt_sk(sk);
376 __fanout_unlink(sk, po);
378 __dev_remove_pack(&po->prot_hook);
382 spin_unlock(&po->bind_lock);
384 spin_lock(&po->bind_lock);
388 static void unregister_prot_hook(struct sock *sk, bool sync)
390 struct packet_sock *po = pkt_sk(sk);
393 __unregister_prot_hook(sk, sync);
396 static inline __pure struct page *pgv_to_page(void *addr)
398 if (is_vmalloc_addr(addr))
399 return vmalloc_to_page(addr);
400 return virt_to_page(addr);
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
406 struct tpacket_hdr *h1;
407 struct tpacket2_hdr *h2;
412 switch (po->tp_version) {
414 h.h1->tp_status = status;
415 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
418 h.h2->tp_status = status;
419 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
423 WARN(1, "TPACKET version not supported.\n");
430 static int __packet_get_status(struct packet_sock *po, void *frame)
433 struct tpacket_hdr *h1;
434 struct tpacket2_hdr *h2;
441 switch (po->tp_version) {
443 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
444 return h.h1->tp_status;
446 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
447 return h.h2->tp_status;
450 WARN(1, "TPACKET version not supported.\n");
456 static void *packet_lookup_frame(struct packet_sock *po,
457 struct packet_ring_buffer *rb,
458 unsigned int position,
461 unsigned int pg_vec_pos, frame_offset;
463 struct tpacket_hdr *h1;
464 struct tpacket2_hdr *h2;
468 pg_vec_pos = position / rb->frames_per_block;
469 frame_offset = position % rb->frames_per_block;
471 h.raw = rb->pg_vec[pg_vec_pos].buffer +
472 (frame_offset * rb->frame_size);
474 if (status != __packet_get_status(po, h.raw))
480 static void *packet_current_frame(struct packet_sock *po,
481 struct packet_ring_buffer *rb,
484 return packet_lookup_frame(po, rb, rb->head, status);
487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
489 del_timer_sync(&pkc->retire_blk_timer);
492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
494 struct sk_buff_head *rb_queue)
496 struct tpacket_kbdq_core *pkc;
498 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
500 spin_lock(&rb_queue->lock);
501 pkc->delete_blk_timer = 1;
502 spin_unlock(&rb_queue->lock);
504 prb_del_retire_blk_timer(pkc);
507 static void prb_init_blk_timer(struct packet_sock *po,
508 struct tpacket_kbdq_core *pkc,
509 void (*func) (unsigned long))
511 init_timer(&pkc->retire_blk_timer);
512 pkc->retire_blk_timer.data = (long)po;
513 pkc->retire_blk_timer.function = func;
514 pkc->retire_blk_timer.expires = jiffies;
517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
519 struct tpacket_kbdq_core *pkc;
524 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529 int blk_size_in_bytes)
531 struct net_device *dev;
532 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
533 struct ethtool_cmd ecmd;
537 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
538 if (unlikely(!dev)) {
540 return DEFAULT_PRB_RETIRE_TOV;
542 err = __ethtool_get_settings(dev, &ecmd);
545 switch (ecmd.speed) {
555 * If the link speed is so slow you don't really
556 * need to worry about perf anyways
561 return DEFAULT_PRB_RETIRE_TOV;
565 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
578 union tpacket_req_u *req_u)
580 p1->feature_req_word = req_u->req3.tp_feature_req_word;
583 static void init_prb_bdqc(struct packet_sock *po,
584 struct packet_ring_buffer *rb,
586 union tpacket_req_u *req_u, int tx_ring)
588 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
589 struct tpacket_block_desc *pbd;
591 memset(p1, 0x0, sizeof(*p1));
593 p1->knxt_seq_num = 1;
595 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
596 p1->pkblk_start = (char *)pg_vec[0].buffer;
597 p1->kblk_size = req_u->req3.tp_block_size;
598 p1->knum_blocks = req_u->req3.tp_block_nr;
599 p1->hdrlen = po->tp_hdrlen;
600 p1->version = po->tp_version;
601 p1->last_kactive_blk_num = 0;
602 po->stats_u.stats3.tp_freeze_q_cnt = 0;
603 if (req_u->req3.tp_retire_blk_tov)
604 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
606 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
607 req_u->req3.tp_block_size);
608 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
609 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
611 prb_init_ft_ops(p1, req_u);
612 prb_setup_retire_blk_timer(po, tx_ring);
613 prb_open_block(p1, pbd);
616 /* Do NOT update the last_blk_num first.
617 * Assumes sk_buff_head lock is held.
619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
621 mod_timer(&pkc->retire_blk_timer,
622 jiffies + pkc->tov_in_jiffies);
623 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
628 * 1) We refresh the timer only when we open a block.
629 * By doing this we don't waste cycles refreshing the timer
630 * on packet-by-packet basis.
632 * With a 1MB block-size, on a 1Gbps line, it will take
633 * i) ~8 ms to fill a block + ii) memcpy etc.
634 * In this cut we are not accounting for the memcpy time.
636 * So, if the user sets the 'tmo' to 10ms then the timer
637 * will never fire while the block is still getting filled
638 * (which is what we want). However, the user could choose
639 * to close a block early and that's fine.
641 * But when the timer does fire, we check whether or not to refresh it.
642 * Since the tmo granularity is in msecs, it is not too expensive
643 * to refresh the timer, lets say every '8' msecs.
644 * Either the user can set the 'tmo' or we can derive it based on
645 * a) line-speed and b) block-size.
646 * prb_calc_retire_blk_tmo() calculates the tmo.
649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
651 struct packet_sock *po = (struct packet_sock *)data;
652 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
654 struct tpacket_block_desc *pbd;
656 spin_lock(&po->sk.sk_receive_queue.lock);
658 frozen = prb_queue_frozen(pkc);
659 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
661 if (unlikely(pkc->delete_blk_timer))
664 /* We only need to plug the race when the block is partially filled.
666 * lock(); increment BLOCK_NUM_PKTS; unlock()
667 * copy_bits() is in progress ...
668 * timer fires on other cpu:
669 * we can't retire the current block because copy_bits
673 if (BLOCK_NUM_PKTS(pbd)) {
674 while (atomic_read(&pkc->blk_fill_in_prog)) {
675 /* Waiting for skb_copy_bits to finish... */
680 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
682 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
683 if (!prb_dispatch_next_block(pkc, po))
688 /* Case 1. Queue was frozen because user-space was
691 if (prb_curr_blk_in_use(pkc, pbd)) {
693 * Ok, user-space is still behind.
694 * So just refresh the timer.
698 /* Case 2. queue was frozen,user-space caught up,
699 * now the link went idle && the timer fired.
700 * We don't have a block to close.So we open this
701 * block and restart the timer.
702 * opening a block thaws the queue,restarts timer
703 * Thawing/timer-refresh is a side effect.
705 prb_open_block(pkc, pbd);
712 _prb_refresh_rx_retire_blk_timer(pkc);
715 spin_unlock(&po->sk.sk_receive_queue.lock);
718 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
719 struct tpacket_block_desc *pbd1, __u32 status)
721 /* Flush everything minus the block header */
723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
728 /* Skip the block header(we know header WILL fit in 4K) */
731 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
732 for (; start < end; start += PAGE_SIZE)
733 flush_dcache_page(pgv_to_page(start));
738 /* Now update the block status. */
740 BLOCK_STATUS(pbd1) = status;
742 /* Flush the block header */
744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
746 flush_dcache_page(pgv_to_page(start));
756 * 2) Increment active_blk_num
758 * Note:We DONT refresh the timer on purpose.
759 * Because almost always the next block will be opened.
761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
762 struct tpacket_block_desc *pbd1,
763 struct packet_sock *po, unsigned int stat)
765 __u32 status = TP_STATUS_USER | stat;
767 struct tpacket3_hdr *last_pkt;
768 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
770 if (po->stats.tp_drops)
771 status |= TP_STATUS_LOSING;
773 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
774 last_pkt->tp_next_offset = 0;
776 /* Get the ts of the last pkt */
777 if (BLOCK_NUM_PKTS(pbd1)) {
778 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
779 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
781 /* Ok, we tmo'd - so get the current time */
784 h1->ts_last_pkt.ts_sec = ts.tv_sec;
785 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
790 /* Flush the block */
791 prb_flush_block(pkc1, pbd1, status);
793 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
796 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
798 pkc->reset_pending_on_curr_blk = 0;
802 * Side effect of opening a block:
804 * 1) prb_queue is thawed.
805 * 2) retire_blk_timer is refreshed.
808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
809 struct tpacket_block_desc *pbd1)
812 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
816 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
818 /* We could have just memset this but we will lose the
819 * flexibility of making the priv area sticky
821 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
822 BLOCK_NUM_PKTS(pbd1) = 0;
823 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827 pkc1->pkblk_start = (char *)pbd1;
828 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
829 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
830 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835 prb_thaw_queue(pkc1);
836 _prb_refresh_rx_retire_blk_timer(pkc1);
843 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
844 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats_u.stats3.tp_freeze_q_cnt++;
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
890 struct tpacket_block_desc *pbd;
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
905 * open this block and return the offset where the first packet
906 * needs to get stored.
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 while (atomic_read(&pkc->blk_fill_in_prog)) {
930 /* Waiting for skb_copy_bits to finish... */
934 prb_close_block(pkc, pbd, po, status);
938 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
943 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
944 struct tpacket_block_desc *pbd)
946 return TP_STATUS_USER & BLOCK_STATUS(pbd);
949 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
951 return pkc->reset_pending_on_curr_blk;
954 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
956 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
957 atomic_dec(&pkc->blk_fill_in_prog);
960 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
963 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
966 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
967 struct tpacket3_hdr *ppd)
969 ppd->hv1.tp_rxhash = 0;
972 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
973 struct tpacket3_hdr *ppd)
975 if (vlan_tx_tag_present(pkc->skb)) {
976 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
977 ppd->tp_status = TP_STATUS_VLAN_VALID;
979 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
984 struct tpacket3_hdr *ppd)
986 prb_fill_vlan_info(pkc, ppd);
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
991 prb_clear_rxhash(pkc, ppd);
994 static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
999 struct tpacket3_hdr *ppd;
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 atomic_inc(&pkc->blk_fill_in_prog);
1008 prb_run_all_ft_ops(pkc, ppd);
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1022 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1031 if (prb_curr_blk_in_use(pkc, pbd)) {
1032 /* Can't record this packet */
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1041 prb_open_block(pkc, pbd);
1046 curr = pkc->nxt_offset;
1048 end = (char *) ((char *)pbd + pkc->kblk_size);
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1079 switch (po->tp_version) {
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1086 return __packet_lookup_frame_in_block(po, skb, status, len);
1088 WARN(1, "TPACKET version not supported\n");
1094 static void *prb_lookup_block(struct packet_sock *po,
1095 struct packet_ring_buffer *rb,
1096 unsigned int previous,
1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1102 if (status != BLOCK_STATUS(pbd))
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1113 prev = rb->prb_bdqc.knum_blocks-1;
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1133 return __prb_previous_block(po, rb, status);
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137 struct packet_ring_buffer *rb)
1139 switch (po->tp_version) {
1142 return packet_increment_head(rb);
1145 WARN(1, "TPACKET version not supported.\n");
1151 static void *packet_previous_frame(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1164 static void packet_sock_destruct(struct sock *sk)
1166 skb_queue_purge(&sk->sk_error_queue);
1168 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1171 if (!sock_flag(sk, SOCK_DEAD)) {
1172 pr_err("Attempt to release alive packet socket: %p\n", sk);
1176 sk_refcnt_debug_dec(sk);
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1181 int x = atomic_read(&f->rr_cur) + 1;
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1191 u32 idx, hash = skb->rxhash;
1193 idx = ((u64)hash * num) >> 32;
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1202 cur = atomic_read(&f->rr_cur);
1203 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204 fanout_rr_next(f, num))) != cur)
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1211 unsigned int cpu = smp_processor_id();
1213 return f->arr[cpu % num];
1216 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217 struct packet_type *pt, struct net_device *orig_dev)
1219 struct packet_fanout *f = pt->af_packet_priv;
1220 unsigned int num = f->num_members;
1221 struct packet_sock *po;
1224 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1231 case PACKET_FANOUT_HASH:
1234 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1238 skb_get_rxhash(skb);
1239 sk = fanout_demux_hash(f, skb, num);
1241 case PACKET_FANOUT_LB:
1242 sk = fanout_demux_lb(f, skb, num);
1244 case PACKET_FANOUT_CPU:
1245 sk = fanout_demux_cpu(f, skb, num);
1251 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1254 static DEFINE_MUTEX(fanout_mutex);
1255 static LIST_HEAD(fanout_list);
1257 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1259 struct packet_fanout *f = po->fanout;
1261 spin_lock(&f->lock);
1262 f->arr[f->num_members] = sk;
1265 spin_unlock(&f->lock);
1268 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1270 struct packet_fanout *f = po->fanout;
1273 spin_lock(&f->lock);
1274 for (i = 0; i < f->num_members; i++) {
1275 if (f->arr[i] == sk)
1278 BUG_ON(i >= f->num_members);
1279 f->arr[i] = f->arr[f->num_members - 1];
1281 spin_unlock(&f->lock);
1284 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1286 struct packet_sock *po = pkt_sk(sk);
1287 struct packet_fanout *f, *match;
1288 u8 type = type_flags & 0xff;
1289 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1293 case PACKET_FANOUT_HASH:
1294 case PACKET_FANOUT_LB:
1295 case PACKET_FANOUT_CPU:
1307 mutex_lock(&fanout_mutex);
1309 list_for_each_entry(f, &fanout_list, list) {
1311 read_pnet(&f->net) == sock_net(sk)) {
1317 if (match && match->defrag != defrag)
1321 match = kzalloc(sizeof(*match), GFP_KERNEL);
1324 write_pnet(&match->net, sock_net(sk));
1327 match->defrag = defrag;
1328 atomic_set(&match->rr_cur, 0);
1329 INIT_LIST_HEAD(&match->list);
1330 spin_lock_init(&match->lock);
1331 atomic_set(&match->sk_ref, 0);
1332 match->prot_hook.type = po->prot_hook.type;
1333 match->prot_hook.dev = po->prot_hook.dev;
1334 match->prot_hook.func = packet_rcv_fanout;
1335 match->prot_hook.af_packet_priv = match;
1336 dev_add_pack(&match->prot_hook);
1337 list_add(&match->list, &fanout_list);
1340 if (match->type == type &&
1341 match->prot_hook.type == po->prot_hook.type &&
1342 match->prot_hook.dev == po->prot_hook.dev) {
1344 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1345 __dev_remove_pack(&po->prot_hook);
1347 atomic_inc(&match->sk_ref);
1348 __fanout_link(sk, po);
1353 mutex_unlock(&fanout_mutex);
1357 static void fanout_release(struct sock *sk)
1359 struct packet_sock *po = pkt_sk(sk);
1360 struct packet_fanout *f;
1368 mutex_lock(&fanout_mutex);
1369 if (atomic_dec_and_test(&f->sk_ref)) {
1371 dev_remove_pack(&f->prot_hook);
1374 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 int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2042 struct sk_buff *skb;
2043 struct net_device *dev;
2045 bool need_rls_dev = false;
2046 int err, reserve = 0;
2048 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2049 int tp_len, size_max;
2050 unsigned char *addr;
2054 mutex_lock(&po->pg_vec_lock);
2057 if (saddr == NULL) {
2058 dev = po->prot_hook.dev;
2063 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2065 if (msg->msg_namelen < (saddr->sll_halen
2066 + offsetof(struct sockaddr_ll,
2069 proto = saddr->sll_protocol;
2070 addr = saddr->sll_addr;
2071 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2072 need_rls_dev = true;
2076 if (unlikely(dev == NULL))
2079 reserve = dev->hard_header_len;
2082 if (unlikely(!(dev->flags & IFF_UP)))
2085 size_max = po->tx_ring.frame_size
2086 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2088 if (size_max > dev->mtu + reserve)
2089 size_max = dev->mtu + reserve;
2092 ph = packet_current_frame(po, &po->tx_ring,
2093 TP_STATUS_SEND_REQUEST);
2095 if (unlikely(ph == NULL)) {
2100 status = TP_STATUS_SEND_REQUEST;
2101 skb = sock_alloc_send_skb(&po->sk,
2102 LL_ALLOCATED_SPACE(dev)
2103 + sizeof(struct sockaddr_ll),
2106 if (unlikely(skb == NULL))
2109 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2112 if (unlikely(tp_len < 0)) {
2114 __packet_set_status(po, ph,
2115 TP_STATUS_AVAILABLE);
2116 packet_increment_head(&po->tx_ring);
2120 status = TP_STATUS_WRONG_FORMAT;
2126 skb->destructor = tpacket_destruct_skb;
2127 __packet_set_status(po, ph, TP_STATUS_SENDING);
2128 atomic_inc(&po->tx_ring.pending);
2130 status = TP_STATUS_SEND_REQUEST;
2131 err = dev_queue_xmit(skb);
2132 if (unlikely(err > 0)) {
2133 err = net_xmit_errno(err);
2134 if (err && __packet_get_status(po, ph) ==
2135 TP_STATUS_AVAILABLE) {
2136 /* skb was destructed already */
2141 * skb was dropped but not destructed yet;
2142 * let's treat it like congestion or err < 0
2146 packet_increment_head(&po->tx_ring);
2148 } while (likely((ph != NULL) ||
2149 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2150 (atomic_read(&po->tx_ring.pending))))
2157 __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 bool need_rls_dev = false;
2200 unsigned char *addr;
2201 int err, reserve = 0;
2202 struct virtio_net_hdr vnet_hdr = { 0 };
2205 struct packet_sock *po = pkt_sk(sk);
2206 unsigned short gso_type = 0;
2209 * Get and verify the address.
2212 if (saddr == NULL) {
2213 dev = po->prot_hook.dev;
2218 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2220 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2222 proto = saddr->sll_protocol;
2223 addr = saddr->sll_addr;
2224 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2225 need_rls_dev = true;
2231 if (sock->type == SOCK_RAW)
2232 reserve = dev->hard_header_len;
2235 if (!(dev->flags & IFF_UP))
2238 if (po->has_vnet_hdr) {
2239 vnet_hdr_len = sizeof(vnet_hdr);
2242 if (len < vnet_hdr_len)
2245 len -= vnet_hdr_len;
2247 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2252 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2253 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2255 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2256 vnet_hdr.csum_offset + 2;
2259 if (vnet_hdr.hdr_len > len)
2262 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2263 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2264 case VIRTIO_NET_HDR_GSO_TCPV4:
2265 gso_type = SKB_GSO_TCPV4;
2267 case VIRTIO_NET_HDR_GSO_TCPV6:
2268 gso_type = SKB_GSO_TCPV6;
2270 case VIRTIO_NET_HDR_GSO_UDP:
2271 gso_type = SKB_GSO_UDP;
2277 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2278 gso_type |= SKB_GSO_TCP_ECN;
2280 if (vnet_hdr.gso_size == 0)
2287 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2291 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2292 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2293 msg->msg_flags & MSG_DONTWAIT, &err);
2297 skb_set_network_header(skb, reserve);
2300 if (sock->type == SOCK_DGRAM &&
2301 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2304 /* Returns -EFAULT on error */
2305 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2308 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2312 if (!gso_type && (len > dev->mtu + reserve)) {
2313 /* Earlier code assumed this would be a VLAN pkt,
2314 * double-check this now that we have the actual
2317 struct ethhdr *ehdr;
2318 skb_reset_mac_header(skb);
2319 ehdr = eth_hdr(skb);
2320 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2326 skb->protocol = proto;
2328 skb->priority = sk->sk_priority;
2329 skb->mark = sk->sk_mark;
2331 if (po->has_vnet_hdr) {
2332 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2333 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2334 vnet_hdr.csum_offset)) {
2340 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2341 skb_shinfo(skb)->gso_type = gso_type;
2343 /* Header must be checked, and gso_segs computed. */
2344 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2345 skb_shinfo(skb)->gso_segs = 0;
2347 len += vnet_hdr_len;
2354 err = dev_queue_xmit(skb);
2355 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2366 if (dev && need_rls_dev)
2372 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2373 struct msghdr *msg, size_t len)
2375 struct sock *sk = sock->sk;
2376 struct packet_sock *po = pkt_sk(sk);
2377 if (po->tx_ring.pg_vec)
2378 return tpacket_snd(po, msg);
2380 return packet_snd(sock, msg, len);
2384 * Close a PACKET socket. This is fairly simple. We immediately go
2385 * to 'closed' state and remove our protocol entry in the device list.
2388 static int packet_release(struct socket *sock)
2390 struct sock *sk = sock->sk;
2391 struct packet_sock *po;
2393 union tpacket_req_u req_u;
2401 spin_lock_bh(&net->packet.sklist_lock);
2402 sk_del_node_init_rcu(sk);
2403 sock_prot_inuse_add(net, sk->sk_prot, -1);
2404 spin_unlock_bh(&net->packet.sklist_lock);
2406 spin_lock(&po->bind_lock);
2407 unregister_prot_hook(sk, false);
2408 if (po->prot_hook.dev) {
2409 dev_put(po->prot_hook.dev);
2410 po->prot_hook.dev = NULL;
2412 spin_unlock(&po->bind_lock);
2414 packet_flush_mclist(sk);
2416 memset(&req_u, 0, sizeof(req_u));
2418 if (po->rx_ring.pg_vec)
2419 packet_set_ring(sk, &req_u, 1, 0);
2421 if (po->tx_ring.pg_vec)
2422 packet_set_ring(sk, &req_u, 1, 1);
2428 * Now the socket is dead. No more input will appear.
2435 skb_queue_purge(&sk->sk_receive_queue);
2436 sk_refcnt_debug_release(sk);
2443 * Attach a packet hook.
2446 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2448 struct packet_sock *po = pkt_sk(sk);
2459 spin_lock(&po->bind_lock);
2460 unregister_prot_hook(sk, true);
2462 po->prot_hook.type = protocol;
2463 if (po->prot_hook.dev)
2464 dev_put(po->prot_hook.dev);
2465 po->prot_hook.dev = dev;
2467 po->ifindex = dev ? dev->ifindex : 0;
2472 if (!dev || (dev->flags & IFF_UP)) {
2473 register_prot_hook(sk);
2475 sk->sk_err = ENETDOWN;
2476 if (!sock_flag(sk, SOCK_DEAD))
2477 sk->sk_error_report(sk);
2481 spin_unlock(&po->bind_lock);
2487 * Bind a packet socket to a device
2490 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2493 struct sock *sk = sock->sk;
2495 struct net_device *dev;
2502 if (addr_len != sizeof(struct sockaddr))
2504 strlcpy(name, uaddr->sa_data, sizeof(name));
2506 dev = dev_get_by_name(sock_net(sk), name);
2508 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2512 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2514 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2515 struct sock *sk = sock->sk;
2516 struct net_device *dev = NULL;
2524 if (addr_len < sizeof(struct sockaddr_ll))
2526 if (sll->sll_family != AF_PACKET)
2529 if (sll->sll_ifindex) {
2531 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2535 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2541 static struct proto packet_proto = {
2543 .owner = THIS_MODULE,
2544 .obj_size = sizeof(struct packet_sock),
2548 * Create a packet of type SOCK_PACKET.
2551 static int packet_create(struct net *net, struct socket *sock, int protocol,
2555 struct packet_sock *po;
2556 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2559 if (!capable(CAP_NET_RAW))
2561 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2562 sock->type != SOCK_PACKET)
2563 return -ESOCKTNOSUPPORT;
2565 sock->state = SS_UNCONNECTED;
2568 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2572 sock->ops = &packet_ops;
2573 if (sock->type == SOCK_PACKET)
2574 sock->ops = &packet_ops_spkt;
2576 sock_init_data(sock, sk);
2579 sk->sk_family = PF_PACKET;
2582 sk->sk_destruct = packet_sock_destruct;
2583 sk_refcnt_debug_inc(sk);
2586 * Attach a protocol block
2589 spin_lock_init(&po->bind_lock);
2590 mutex_init(&po->pg_vec_lock);
2591 po->prot_hook.func = packet_rcv;
2593 if (sock->type == SOCK_PACKET)
2594 po->prot_hook.func = packet_rcv_spkt;
2596 po->prot_hook.af_packet_priv = sk;
2599 po->prot_hook.type = proto;
2600 register_prot_hook(sk);
2603 spin_lock_bh(&net->packet.sklist_lock);
2604 sk_add_node_rcu(sk, &net->packet.sklist);
2605 sock_prot_inuse_add(net, &packet_proto, 1);
2606 spin_unlock_bh(&net->packet.sklist_lock);
2613 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2615 struct sock_exterr_skb *serr;
2616 struct sk_buff *skb, *skb2;
2620 skb = skb_dequeue(&sk->sk_error_queue);
2626 msg->msg_flags |= MSG_TRUNC;
2629 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2633 sock_recv_timestamp(msg, sk, skb);
2635 serr = SKB_EXT_ERR(skb);
2636 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2637 sizeof(serr->ee), &serr->ee);
2639 msg->msg_flags |= MSG_ERRQUEUE;
2642 /* Reset and regenerate socket error */
2643 spin_lock_bh(&sk->sk_error_queue.lock);
2645 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2646 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2647 spin_unlock_bh(&sk->sk_error_queue.lock);
2648 sk->sk_error_report(sk);
2650 spin_unlock_bh(&sk->sk_error_queue.lock);
2659 * Pull a packet from our receive queue and hand it to the user.
2660 * If necessary we block.
2663 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2664 struct msghdr *msg, size_t len, int flags)
2666 struct sock *sk = sock->sk;
2667 struct sk_buff *skb;
2669 struct sockaddr_ll *sll;
2670 int vnet_hdr_len = 0;
2673 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2677 /* What error should we return now? EUNATTACH? */
2678 if (pkt_sk(sk)->ifindex < 0)
2682 if (flags & MSG_ERRQUEUE) {
2683 err = packet_recv_error(sk, msg, len);
2688 * Call the generic datagram receiver. This handles all sorts
2689 * of horrible races and re-entrancy so we can forget about it
2690 * in the protocol layers.
2692 * Now it will return ENETDOWN, if device have just gone down,
2693 * but then it will block.
2696 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2699 * An error occurred so return it. Because skb_recv_datagram()
2700 * handles the blocking we don't see and worry about blocking
2707 if (pkt_sk(sk)->has_vnet_hdr) {
2708 struct virtio_net_hdr vnet_hdr = { 0 };
2711 vnet_hdr_len = sizeof(vnet_hdr);
2712 if (len < vnet_hdr_len)
2715 len -= vnet_hdr_len;
2717 if (skb_is_gso(skb)) {
2718 struct skb_shared_info *sinfo = skb_shinfo(skb);
2720 /* This is a hint as to how much should be linear. */
2721 vnet_hdr.hdr_len = skb_headlen(skb);
2722 vnet_hdr.gso_size = sinfo->gso_size;
2723 if (sinfo->gso_type & SKB_GSO_TCPV4)
2724 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2725 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2726 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2727 else if (sinfo->gso_type & SKB_GSO_UDP)
2728 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2729 else if (sinfo->gso_type & SKB_GSO_FCOE)
2733 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2734 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2736 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2738 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2739 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2740 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2741 vnet_hdr.csum_offset = skb->csum_offset;
2742 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2743 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2744 } /* else everything is zero */
2746 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2753 * If the address length field is there to be filled in, we fill
2757 sll = &PACKET_SKB_CB(skb)->sa.ll;
2758 if (sock->type == SOCK_PACKET)
2759 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2761 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2764 * You lose any data beyond the buffer you gave. If it worries a
2765 * user program they can ask the device for its MTU anyway.
2771 msg->msg_flags |= MSG_TRUNC;
2774 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2778 sock_recv_ts_and_drops(msg, sk, skb);
2781 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2784 if (pkt_sk(sk)->auxdata) {
2785 struct tpacket_auxdata aux;
2787 aux.tp_status = TP_STATUS_USER;
2788 if (skb->ip_summed == CHECKSUM_PARTIAL)
2789 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2790 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2791 aux.tp_snaplen = skb->len;
2793 aux.tp_net = skb_network_offset(skb);
2794 if (vlan_tx_tag_present(skb)) {
2795 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2796 aux.tp_status |= TP_STATUS_VLAN_VALID;
2798 aux.tp_vlan_tci = 0;
2801 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2805 * Free or return the buffer as appropriate. Again this
2806 * hides all the races and re-entrancy issues from us.
2808 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2811 skb_free_datagram(sk, skb);
2816 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2817 int *uaddr_len, int peer)
2819 struct net_device *dev;
2820 struct sock *sk = sock->sk;
2825 uaddr->sa_family = AF_PACKET;
2827 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2829 strncpy(uaddr->sa_data, dev->name, 14);
2831 memset(uaddr->sa_data, 0, 14);
2833 *uaddr_len = sizeof(*uaddr);
2838 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2839 int *uaddr_len, int peer)
2841 struct net_device *dev;
2842 struct sock *sk = sock->sk;
2843 struct packet_sock *po = pkt_sk(sk);
2844 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2849 sll->sll_family = AF_PACKET;
2850 sll->sll_ifindex = po->ifindex;
2851 sll->sll_protocol = po->num;
2852 sll->sll_pkttype = 0;
2854 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2856 sll->sll_hatype = dev->type;
2857 sll->sll_halen = dev->addr_len;
2858 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2860 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2864 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2869 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2873 case PACKET_MR_MULTICAST:
2874 if (i->alen != dev->addr_len)
2877 return dev_mc_add(dev, i->addr);
2879 return dev_mc_del(dev, i->addr);
2881 case PACKET_MR_PROMISC:
2882 return dev_set_promiscuity(dev, what);
2884 case PACKET_MR_ALLMULTI:
2885 return dev_set_allmulti(dev, what);
2887 case PACKET_MR_UNICAST:
2888 if (i->alen != dev->addr_len)
2891 return dev_uc_add(dev, i->addr);
2893 return dev_uc_del(dev, i->addr);
2901 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2903 for ( ; i; i = i->next) {
2904 if (i->ifindex == dev->ifindex)
2905 packet_dev_mc(dev, i, what);
2909 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2911 struct packet_sock *po = pkt_sk(sk);
2912 struct packet_mclist *ml, *i;
2913 struct net_device *dev;
2919 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2924 if (mreq->mr_alen > dev->addr_len)
2928 i = kmalloc(sizeof(*i), GFP_KERNEL);
2933 for (ml = po->mclist; ml; ml = ml->next) {
2934 if (ml->ifindex == mreq->mr_ifindex &&
2935 ml->type == mreq->mr_type &&
2936 ml->alen == mreq->mr_alen &&
2937 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2939 /* Free the new element ... */
2945 i->type = mreq->mr_type;
2946 i->ifindex = mreq->mr_ifindex;
2947 i->alen = mreq->mr_alen;
2948 memcpy(i->addr, mreq->mr_address, i->alen);
2950 i->next = po->mclist;
2952 err = packet_dev_mc(dev, i, 1);
2954 po->mclist = i->next;
2963 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2965 struct packet_mclist *ml, **mlp;
2969 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2970 if (ml->ifindex == mreq->mr_ifindex &&
2971 ml->type == mreq->mr_type &&
2972 ml->alen == mreq->mr_alen &&
2973 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2974 if (--ml->count == 0) {
2975 struct net_device *dev;
2977 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2979 packet_dev_mc(dev, ml, -1);
2987 return -EADDRNOTAVAIL;
2990 static void packet_flush_mclist(struct sock *sk)
2992 struct packet_sock *po = pkt_sk(sk);
2993 struct packet_mclist *ml;
2999 while ((ml = po->mclist) != NULL) {
3000 struct net_device *dev;
3002 po->mclist = ml->next;
3003 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3005 packet_dev_mc(dev, ml, -1);
3012 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3014 struct sock *sk = sock->sk;
3015 struct packet_sock *po = pkt_sk(sk);
3018 if (level != SOL_PACKET)
3019 return -ENOPROTOOPT;
3022 case PACKET_ADD_MEMBERSHIP:
3023 case PACKET_DROP_MEMBERSHIP:
3025 struct packet_mreq_max mreq;
3027 memset(&mreq, 0, sizeof(mreq));
3028 if (len < sizeof(struct packet_mreq))
3030 if (len > sizeof(mreq))
3032 if (copy_from_user(&mreq, optval, len))
3034 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3036 if (optname == PACKET_ADD_MEMBERSHIP)
3037 ret = packet_mc_add(sk, &mreq);
3039 ret = packet_mc_drop(sk, &mreq);
3043 case PACKET_RX_RING:
3044 case PACKET_TX_RING:
3046 union tpacket_req_u req_u;
3049 switch (po->tp_version) {
3052 len = sizeof(req_u.req);
3056 len = sizeof(req_u.req3);
3061 if (pkt_sk(sk)->has_vnet_hdr)
3063 if (copy_from_user(&req_u.req, optval, len))
3065 return packet_set_ring(sk, &req_u, 0,
3066 optname == PACKET_TX_RING);
3068 case PACKET_COPY_THRESH:
3072 if (optlen != sizeof(val))
3074 if (copy_from_user(&val, optval, sizeof(val)))
3077 pkt_sk(sk)->copy_thresh = val;
3080 case PACKET_VERSION:
3084 if (optlen != sizeof(val))
3086 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3088 if (copy_from_user(&val, optval, sizeof(val)))
3094 po->tp_version = val;
3100 case PACKET_RESERVE:
3104 if (optlen != sizeof(val))
3106 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3108 if (copy_from_user(&val, optval, sizeof(val)))
3110 po->tp_reserve = val;
3117 if (optlen != sizeof(val))
3119 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3121 if (copy_from_user(&val, optval, sizeof(val)))
3123 po->tp_loss = !!val;
3126 case PACKET_AUXDATA:
3130 if (optlen < sizeof(val))
3132 if (copy_from_user(&val, optval, sizeof(val)))
3135 po->auxdata = !!val;
3138 case PACKET_ORIGDEV:
3142 if (optlen < sizeof(val))
3144 if (copy_from_user(&val, optval, sizeof(val)))
3147 po->origdev = !!val;
3150 case PACKET_VNET_HDR:
3154 if (sock->type != SOCK_RAW)
3156 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3158 if (optlen < sizeof(val))
3160 if (copy_from_user(&val, optval, sizeof(val)))
3163 po->has_vnet_hdr = !!val;
3166 case PACKET_TIMESTAMP:
3170 if (optlen != sizeof(val))
3172 if (copy_from_user(&val, optval, sizeof(val)))
3175 po->tp_tstamp = val;
3182 if (optlen != sizeof(val))
3184 if (copy_from_user(&val, optval, sizeof(val)))
3187 return fanout_add(sk, val & 0xffff, val >> 16);
3190 return -ENOPROTOOPT;
3194 static int packet_getsockopt(struct socket *sock, int level, int optname,
3195 char __user *optval, int __user *optlen)
3199 struct sock *sk = sock->sk;
3200 struct packet_sock *po = pkt_sk(sk);
3202 struct tpacket_stats st;
3203 union tpacket_stats_u st_u;
3205 if (level != SOL_PACKET)
3206 return -ENOPROTOOPT;
3208 if (get_user(len, optlen))
3215 case PACKET_STATISTICS:
3216 if (po->tp_version == TPACKET_V3) {
3217 len = sizeof(struct tpacket_stats_v3);
3219 if (len > sizeof(struct tpacket_stats))
3220 len = sizeof(struct tpacket_stats);
3222 spin_lock_bh(&sk->sk_receive_queue.lock);
3223 if (po->tp_version == TPACKET_V3) {
3224 memcpy(&st_u.stats3, &po->stats,
3225 sizeof(struct tpacket_stats));
3226 st_u.stats3.tp_freeze_q_cnt =
3227 po->stats_u.stats3.tp_freeze_q_cnt;
3228 st_u.stats3.tp_packets += po->stats.tp_drops;
3229 data = &st_u.stats3;
3232 st.tp_packets += st.tp_drops;
3235 memset(&po->stats, 0, sizeof(st));
3236 spin_unlock_bh(&sk->sk_receive_queue.lock);
3238 case PACKET_AUXDATA:
3239 if (len > sizeof(int))
3245 case PACKET_ORIGDEV:
3246 if (len > sizeof(int))
3252 case PACKET_VNET_HDR:
3253 if (len > sizeof(int))
3255 val = po->has_vnet_hdr;
3259 case PACKET_VERSION:
3260 if (len > sizeof(int))
3262 val = po->tp_version;
3266 if (len > sizeof(int))
3268 if (copy_from_user(&val, optval, len))
3272 val = sizeof(struct tpacket_hdr);
3275 val = sizeof(struct tpacket2_hdr);
3278 val = sizeof(struct tpacket3_hdr);
3285 case PACKET_RESERVE:
3286 if (len > sizeof(unsigned int))
3287 len = sizeof(unsigned int);
3288 val = po->tp_reserve;
3292 if (len > sizeof(unsigned int))
3293 len = sizeof(unsigned int);
3297 case PACKET_TIMESTAMP:
3298 if (len > sizeof(int))
3300 val = po->tp_tstamp;
3304 if (len > sizeof(int))
3307 ((u32)po->fanout->id |
3308 ((u32)po->fanout->type << 16)) :
3313 return -ENOPROTOOPT;
3316 if (put_user(len, optlen))
3318 if (copy_to_user(optval, data, len))
3324 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3327 struct hlist_node *node;
3328 struct net_device *dev = data;
3329 struct net *net = dev_net(dev);
3332 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3333 struct packet_sock *po = pkt_sk(sk);
3336 case NETDEV_UNREGISTER:
3338 packet_dev_mclist(dev, po->mclist, -1);
3342 if (dev->ifindex == po->ifindex) {
3343 spin_lock(&po->bind_lock);
3345 __unregister_prot_hook(sk, false);
3346 sk->sk_err = ENETDOWN;
3347 if (!sock_flag(sk, SOCK_DEAD))
3348 sk->sk_error_report(sk);
3350 if (msg == NETDEV_UNREGISTER) {
3352 if (po->prot_hook.dev)
3353 dev_put(po->prot_hook.dev);
3354 po->prot_hook.dev = NULL;
3356 spin_unlock(&po->bind_lock);
3360 if (dev->ifindex == po->ifindex) {
3361 spin_lock(&po->bind_lock);
3363 register_prot_hook(sk);
3364 spin_unlock(&po->bind_lock);
3374 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3377 struct sock *sk = sock->sk;
3382 int amount = sk_wmem_alloc_get(sk);
3384 return put_user(amount, (int __user *)arg);
3388 struct sk_buff *skb;
3391 spin_lock_bh(&sk->sk_receive_queue.lock);
3392 skb = skb_peek(&sk->sk_receive_queue);
3395 spin_unlock_bh(&sk->sk_receive_queue.lock);
3396 return put_user(amount, (int __user *)arg);
3399 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3401 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3411 case SIOCGIFBRDADDR:
3412 case SIOCSIFBRDADDR:
3413 case SIOCGIFNETMASK:
3414 case SIOCSIFNETMASK:
3415 case SIOCGIFDSTADDR:
3416 case SIOCSIFDSTADDR:
3418 return inet_dgram_ops.ioctl(sock, cmd, arg);
3422 return -ENOIOCTLCMD;
3427 static unsigned int packet_poll(struct file *file, struct socket *sock,
3430 struct sock *sk = sock->sk;
3431 struct packet_sock *po = pkt_sk(sk);
3432 unsigned int mask = datagram_poll(file, sock, wait);
3434 spin_lock_bh(&sk->sk_receive_queue.lock);
3435 if (po->rx_ring.pg_vec) {
3436 if (!packet_previous_rx_frame(po, &po->rx_ring,
3438 mask |= POLLIN | POLLRDNORM;
3440 spin_unlock_bh(&sk->sk_receive_queue.lock);
3441 spin_lock_bh(&sk->sk_write_queue.lock);
3442 if (po->tx_ring.pg_vec) {
3443 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3444 mask |= POLLOUT | POLLWRNORM;
3446 spin_unlock_bh(&sk->sk_write_queue.lock);
3451 /* Dirty? Well, I still did not learn better way to account
3455 static void packet_mm_open(struct vm_area_struct *vma)
3457 struct file *file = vma->vm_file;
3458 struct socket *sock = file->private_data;
3459 struct sock *sk = sock->sk;
3462 atomic_inc(&pkt_sk(sk)->mapped);
3465 static void packet_mm_close(struct vm_area_struct *vma)
3467 struct file *file = vma->vm_file;
3468 struct socket *sock = file->private_data;
3469 struct sock *sk = sock->sk;
3472 atomic_dec(&pkt_sk(sk)->mapped);
3475 static const struct vm_operations_struct packet_mmap_ops = {
3476 .open = packet_mm_open,
3477 .close = packet_mm_close,
3480 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3485 for (i = 0; i < len; i++) {
3486 if (likely(pg_vec[i].buffer)) {
3487 if (is_vmalloc_addr(pg_vec[i].buffer))
3488 vfree(pg_vec[i].buffer);
3490 free_pages((unsigned long)pg_vec[i].buffer,
3492 pg_vec[i].buffer = NULL;
3498 static char *alloc_one_pg_vec_page(unsigned long order)
3500 char *buffer = NULL;
3501 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3502 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3504 buffer = (char *) __get_free_pages(gfp_flags, order);
3510 * __get_free_pages failed, fall back to vmalloc
3512 buffer = vzalloc((1 << order) * PAGE_SIZE);
3518 * vmalloc failed, lets dig into swap here
3520 gfp_flags &= ~__GFP_NORETRY;
3521 buffer = (char *)__get_free_pages(gfp_flags, order);
3526 * complete and utter failure
3531 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3533 unsigned int block_nr = req->tp_block_nr;
3537 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3538 if (unlikely(!pg_vec))
3541 for (i = 0; i < block_nr; i++) {
3542 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3543 if (unlikely(!pg_vec[i].buffer))
3544 goto out_free_pgvec;
3551 free_pg_vec(pg_vec, order, block_nr);
3556 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3557 int closing, int tx_ring)
3559 struct pgv *pg_vec = NULL;
3560 struct packet_sock *po = pkt_sk(sk);
3561 int was_running, order = 0;
3562 struct packet_ring_buffer *rb;
3563 struct sk_buff_head *rb_queue;
3566 /* Added to avoid minimal code churn */
3567 struct tpacket_req *req = &req_u->req;
3569 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3570 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3571 WARN(1, "Tx-ring is not supported.\n");
3575 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3576 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3580 if (atomic_read(&po->mapped))
3582 if (atomic_read(&rb->pending))
3586 if (req->tp_block_nr) {
3587 /* Sanity tests and some calculations */
3589 if (unlikely(rb->pg_vec))
3592 switch (po->tp_version) {
3594 po->tp_hdrlen = TPACKET_HDRLEN;
3597 po->tp_hdrlen = TPACKET2_HDRLEN;
3600 po->tp_hdrlen = TPACKET3_HDRLEN;
3605 if (unlikely((int)req->tp_block_size <= 0))
3607 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3609 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3612 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3615 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3616 if (unlikely(rb->frames_per_block <= 0))
3618 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3623 order = get_order(req->tp_block_size);
3624 pg_vec = alloc_pg_vec(req, order);
3625 if (unlikely(!pg_vec))
3627 switch (po->tp_version) {
3629 /* Transmit path is not supported. We checked
3630 * it above but just being paranoid
3633 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3642 if (unlikely(req->tp_frame_nr))
3648 /* Detach socket from network */
3649 spin_lock(&po->bind_lock);
3650 was_running = po->running;
3654 __unregister_prot_hook(sk, false);
3656 spin_unlock(&po->bind_lock);
3661 mutex_lock(&po->pg_vec_lock);
3662 if (closing || atomic_read(&po->mapped) == 0) {
3664 spin_lock_bh(&rb_queue->lock);
3665 swap(rb->pg_vec, pg_vec);
3666 rb->frame_max = (req->tp_frame_nr - 1);
3668 rb->frame_size = req->tp_frame_size;
3669 spin_unlock_bh(&rb_queue->lock);
3671 swap(rb->pg_vec_order, order);
3672 swap(rb->pg_vec_len, req->tp_block_nr);
3674 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3675 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3676 tpacket_rcv : packet_rcv;
3677 skb_queue_purge(rb_queue);
3678 if (atomic_read(&po->mapped))
3679 pr_err("packet_mmap: vma is busy: %d\n",
3680 atomic_read(&po->mapped));
3682 mutex_unlock(&po->pg_vec_lock);
3684 spin_lock(&po->bind_lock);
3687 register_prot_hook(sk);
3689 spin_unlock(&po->bind_lock);
3690 if (closing && (po->tp_version > TPACKET_V2)) {
3691 /* Because we don't support block-based V3 on tx-ring */
3693 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3698 free_pg_vec(pg_vec, order, req->tp_block_nr);
3703 static int packet_mmap(struct file *file, struct socket *sock,
3704 struct vm_area_struct *vma)
3706 struct sock *sk = sock->sk;
3707 struct packet_sock *po = pkt_sk(sk);
3708 unsigned long size, expected_size;
3709 struct packet_ring_buffer *rb;
3710 unsigned long start;
3717 mutex_lock(&po->pg_vec_lock);
3720 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3722 expected_size += rb->pg_vec_len
3728 if (expected_size == 0)
3731 size = vma->vm_end - vma->vm_start;
3732 if (size != expected_size)
3735 start = vma->vm_start;
3736 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3737 if (rb->pg_vec == NULL)
3740 for (i = 0; i < rb->pg_vec_len; i++) {
3742 void *kaddr = rb->pg_vec[i].buffer;
3745 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3746 page = pgv_to_page(kaddr);
3747 err = vm_insert_page(vma, start, page);
3756 atomic_inc(&po->mapped);
3757 vma->vm_ops = &packet_mmap_ops;
3761 mutex_unlock(&po->pg_vec_lock);
3765 static const struct proto_ops packet_ops_spkt = {
3766 .family = PF_PACKET,
3767 .owner = THIS_MODULE,
3768 .release = packet_release,
3769 .bind = packet_bind_spkt,
3770 .connect = sock_no_connect,
3771 .socketpair = sock_no_socketpair,
3772 .accept = sock_no_accept,
3773 .getname = packet_getname_spkt,
3774 .poll = datagram_poll,
3775 .ioctl = packet_ioctl,
3776 .listen = sock_no_listen,
3777 .shutdown = sock_no_shutdown,
3778 .setsockopt = sock_no_setsockopt,
3779 .getsockopt = sock_no_getsockopt,
3780 .sendmsg = packet_sendmsg_spkt,
3781 .recvmsg = packet_recvmsg,
3782 .mmap = sock_no_mmap,
3783 .sendpage = sock_no_sendpage,
3786 static const struct proto_ops packet_ops = {
3787 .family = PF_PACKET,
3788 .owner = THIS_MODULE,
3789 .release = packet_release,
3790 .bind = packet_bind,
3791 .connect = sock_no_connect,
3792 .socketpair = sock_no_socketpair,
3793 .accept = sock_no_accept,
3794 .getname = packet_getname,
3795 .poll = packet_poll,
3796 .ioctl = packet_ioctl,
3797 .listen = sock_no_listen,
3798 .shutdown = sock_no_shutdown,
3799 .setsockopt = packet_setsockopt,
3800 .getsockopt = packet_getsockopt,
3801 .sendmsg = packet_sendmsg,
3802 .recvmsg = packet_recvmsg,
3803 .mmap = packet_mmap,
3804 .sendpage = sock_no_sendpage,
3807 static const struct net_proto_family packet_family_ops = {
3808 .family = PF_PACKET,
3809 .create = packet_create,
3810 .owner = THIS_MODULE,
3813 static struct notifier_block packet_netdev_notifier = {
3814 .notifier_call = packet_notifier,
3817 #ifdef CONFIG_PROC_FS
3819 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3822 struct net *net = seq_file_net(seq);
3825 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3828 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3830 struct net *net = seq_file_net(seq);
3831 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3834 static void packet_seq_stop(struct seq_file *seq, void *v)
3840 static int packet_seq_show(struct seq_file *seq, void *v)
3842 if (v == SEQ_START_TOKEN)
3843 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3845 struct sock *s = sk_entry(v);
3846 const struct packet_sock *po = pkt_sk(s);
3849 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3851 atomic_read(&s->sk_refcnt),
3856 atomic_read(&s->sk_rmem_alloc),
3864 static const struct seq_operations packet_seq_ops = {
3865 .start = packet_seq_start,
3866 .next = packet_seq_next,
3867 .stop = packet_seq_stop,
3868 .show = packet_seq_show,
3871 static int packet_seq_open(struct inode *inode, struct file *file)
3873 return seq_open_net(inode, file, &packet_seq_ops,
3874 sizeof(struct seq_net_private));
3877 static const struct file_operations packet_seq_fops = {
3878 .owner = THIS_MODULE,
3879 .open = packet_seq_open,
3881 .llseek = seq_lseek,
3882 .release = seq_release_net,
3887 static int __net_init packet_net_init(struct net *net)
3889 spin_lock_init(&net->packet.sklist_lock);
3890 INIT_HLIST_HEAD(&net->packet.sklist);
3892 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3898 static void __net_exit packet_net_exit(struct net *net)
3900 proc_net_remove(net, "packet");
3903 static struct pernet_operations packet_net_ops = {
3904 .init = packet_net_init,
3905 .exit = packet_net_exit,
3909 static void __exit packet_exit(void)
3911 unregister_netdevice_notifier(&packet_netdev_notifier);
3912 unregister_pernet_subsys(&packet_net_ops);
3913 sock_unregister(PF_PACKET);
3914 proto_unregister(&packet_proto);
3917 static int __init packet_init(void)
3919 int rc = proto_register(&packet_proto, 0);
3924 sock_register(&packet_family_ops);
3925 register_pernet_subsys(&packet_net_ops);
3926 register_netdevice_notifier(&packet_netdev_notifier);
3931 module_init(packet_init);
3932 module_exit(packet_exit);
3933 MODULE_LICENSE("GPL");
3934 MODULE_ALIAS_NETPROTO(PF_PACKET);