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 int tlen = dev->needed_tailroom;
1503 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1506 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1509 /* FIXME: Save some space for broken drivers that write a hard
1510 * header at transmission time by themselves. PPP is the notable
1511 * one here. This should really be fixed at the driver level.
1513 skb_reserve(skb, reserved);
1514 skb_reset_network_header(skb);
1516 /* Try to align data part correctly */
1521 skb_reset_network_header(skb);
1523 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1529 if (len > (dev->mtu + dev->hard_header_len)) {
1530 /* Earlier code assumed this would be a VLAN pkt,
1531 * double-check this now that we have the actual
1534 struct ethhdr *ehdr;
1535 skb_reset_mac_header(skb);
1536 ehdr = eth_hdr(skb);
1537 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1543 skb->protocol = proto;
1545 skb->priority = sk->sk_priority;
1546 skb->mark = sk->sk_mark;
1547 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1551 dev_queue_xmit(skb);
1562 static unsigned int run_filter(const struct sk_buff *skb,
1563 const struct sock *sk,
1566 struct sk_filter *filter;
1569 filter = rcu_dereference(sk->sk_filter);
1571 res = SK_RUN_FILTER(filter, skb);
1578 * This function makes lazy skb cloning in hope that most of packets
1579 * are discarded by BPF.
1581 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1582 * and skb->cb are mangled. It works because (and until) packets
1583 * falling here are owned by current CPU. Output packets are cloned
1584 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1585 * sequencially, so that if we return skb to original state on exit,
1586 * we will not harm anyone.
1589 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1590 struct packet_type *pt, struct net_device *orig_dev)
1593 struct sockaddr_ll *sll;
1594 struct packet_sock *po;
1595 u8 *skb_head = skb->data;
1596 int skb_len = skb->len;
1597 unsigned int snaplen, res;
1599 if (skb->pkt_type == PACKET_LOOPBACK)
1602 sk = pt->af_packet_priv;
1605 if (!net_eq(dev_net(dev), sock_net(sk)))
1610 if (dev->header_ops) {
1611 /* The device has an explicit notion of ll header,
1612 * exported to higher levels.
1614 * Otherwise, the device hides details of its frame
1615 * structure, so that corresponding packet head is
1616 * never delivered to user.
1618 if (sk->sk_type != SOCK_DGRAM)
1619 skb_push(skb, skb->data - skb_mac_header(skb));
1620 else if (skb->pkt_type == PACKET_OUTGOING) {
1621 /* Special case: outgoing packets have ll header at head */
1622 skb_pull(skb, skb_network_offset(skb));
1628 res = run_filter(skb, sk, snaplen);
1630 goto drop_n_restore;
1634 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1635 (unsigned)sk->sk_rcvbuf)
1638 if (skb_shared(skb)) {
1639 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1643 if (skb_head != skb->data) {
1644 skb->data = skb_head;
1651 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1654 sll = &PACKET_SKB_CB(skb)->sa.ll;
1655 sll->sll_family = AF_PACKET;
1656 sll->sll_hatype = dev->type;
1657 sll->sll_protocol = skb->protocol;
1658 sll->sll_pkttype = skb->pkt_type;
1659 if (unlikely(po->origdev))
1660 sll->sll_ifindex = orig_dev->ifindex;
1662 sll->sll_ifindex = dev->ifindex;
1664 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1666 PACKET_SKB_CB(skb)->origlen = skb->len;
1668 if (pskb_trim(skb, snaplen))
1671 skb_set_owner_r(skb, sk);
1675 /* drop conntrack reference */
1678 spin_lock(&sk->sk_receive_queue.lock);
1679 po->stats.tp_packets++;
1680 skb->dropcount = atomic_read(&sk->sk_drops);
1681 __skb_queue_tail(&sk->sk_receive_queue, skb);
1682 spin_unlock(&sk->sk_receive_queue.lock);
1683 sk->sk_data_ready(sk, skb->len);
1687 spin_lock(&sk->sk_receive_queue.lock);
1688 po->stats.tp_drops++;
1689 atomic_inc(&sk->sk_drops);
1690 spin_unlock(&sk->sk_receive_queue.lock);
1693 if (skb_head != skb->data && skb_shared(skb)) {
1694 skb->data = skb_head;
1702 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1703 struct packet_type *pt, struct net_device *orig_dev)
1706 struct packet_sock *po;
1707 struct sockaddr_ll *sll;
1709 struct tpacket_hdr *h1;
1710 struct tpacket2_hdr *h2;
1711 struct tpacket3_hdr *h3;
1714 u8 *skb_head = skb->data;
1715 int skb_len = skb->len;
1716 unsigned int snaplen, res;
1717 unsigned long status = TP_STATUS_USER;
1718 unsigned short macoff, netoff, hdrlen;
1719 struct sk_buff *copy_skb = NULL;
1722 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1724 if (skb->pkt_type == PACKET_LOOPBACK)
1727 sk = pt->af_packet_priv;
1730 if (!net_eq(dev_net(dev), sock_net(sk)))
1733 if (dev->header_ops) {
1734 if (sk->sk_type != SOCK_DGRAM)
1735 skb_push(skb, skb->data - skb_mac_header(skb));
1736 else if (skb->pkt_type == PACKET_OUTGOING) {
1737 /* Special case: outgoing packets have ll header at head */
1738 skb_pull(skb, skb_network_offset(skb));
1742 if (skb->ip_summed == CHECKSUM_PARTIAL)
1743 status |= TP_STATUS_CSUMNOTREADY;
1747 res = run_filter(skb, sk, snaplen);
1749 goto drop_n_restore;
1753 if (sk->sk_type == SOCK_DGRAM) {
1754 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1757 unsigned maclen = skb_network_offset(skb);
1758 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1759 (maclen < 16 ? 16 : maclen)) +
1761 macoff = netoff - maclen;
1763 if (po->tp_version <= TPACKET_V2) {
1764 if (macoff + snaplen > po->rx_ring.frame_size) {
1765 if (po->copy_thresh &&
1766 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1767 < (unsigned)sk->sk_rcvbuf) {
1768 if (skb_shared(skb)) {
1769 copy_skb = skb_clone(skb, GFP_ATOMIC);
1771 copy_skb = skb_get(skb);
1772 skb_head = skb->data;
1775 skb_set_owner_r(copy_skb, sk);
1777 snaplen = po->rx_ring.frame_size - macoff;
1778 if ((int)snaplen < 0)
1782 spin_lock(&sk->sk_receive_queue.lock);
1783 h.raw = packet_current_rx_frame(po, skb,
1784 TP_STATUS_KERNEL, (macoff+snaplen));
1787 if (po->tp_version <= TPACKET_V2) {
1788 packet_increment_rx_head(po, &po->rx_ring);
1790 * LOSING will be reported till you read the stats,
1791 * because it's COR - Clear On Read.
1792 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1795 if (po->stats.tp_drops)
1796 status |= TP_STATUS_LOSING;
1798 po->stats.tp_packets++;
1800 status |= TP_STATUS_COPY;
1801 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1803 spin_unlock(&sk->sk_receive_queue.lock);
1805 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1807 switch (po->tp_version) {
1809 h.h1->tp_len = skb->len;
1810 h.h1->tp_snaplen = snaplen;
1811 h.h1->tp_mac = macoff;
1812 h.h1->tp_net = netoff;
1813 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1814 && shhwtstamps->syststamp.tv64)
1815 tv = ktime_to_timeval(shhwtstamps->syststamp);
1816 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1817 && shhwtstamps->hwtstamp.tv64)
1818 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1819 else if (skb->tstamp.tv64)
1820 tv = ktime_to_timeval(skb->tstamp);
1822 do_gettimeofday(&tv);
1823 h.h1->tp_sec = tv.tv_sec;
1824 h.h1->tp_usec = tv.tv_usec;
1825 hdrlen = sizeof(*h.h1);
1828 h.h2->tp_len = skb->len;
1829 h.h2->tp_snaplen = snaplen;
1830 h.h2->tp_mac = macoff;
1831 h.h2->tp_net = netoff;
1832 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1833 && shhwtstamps->syststamp.tv64)
1834 ts = ktime_to_timespec(shhwtstamps->syststamp);
1835 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1836 && shhwtstamps->hwtstamp.tv64)
1837 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1838 else if (skb->tstamp.tv64)
1839 ts = ktime_to_timespec(skb->tstamp);
1841 getnstimeofday(&ts);
1842 h.h2->tp_sec = ts.tv_sec;
1843 h.h2->tp_nsec = ts.tv_nsec;
1844 if (vlan_tx_tag_present(skb)) {
1845 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1846 status |= TP_STATUS_VLAN_VALID;
1848 h.h2->tp_vlan_tci = 0;
1850 h.h2->tp_padding = 0;
1851 hdrlen = sizeof(*h.h2);
1854 /* tp_nxt_offset,vlan are already populated above.
1855 * So DONT clear those fields here
1857 h.h3->tp_status |= status;
1858 h.h3->tp_len = skb->len;
1859 h.h3->tp_snaplen = snaplen;
1860 h.h3->tp_mac = macoff;
1861 h.h3->tp_net = netoff;
1862 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1863 && shhwtstamps->syststamp.tv64)
1864 ts = ktime_to_timespec(shhwtstamps->syststamp);
1865 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1866 && shhwtstamps->hwtstamp.tv64)
1867 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1868 else if (skb->tstamp.tv64)
1869 ts = ktime_to_timespec(skb->tstamp);
1871 getnstimeofday(&ts);
1872 h.h3->tp_sec = ts.tv_sec;
1873 h.h3->tp_nsec = ts.tv_nsec;
1874 hdrlen = sizeof(*h.h3);
1880 sll = h.raw + TPACKET_ALIGN(hdrlen);
1881 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1882 sll->sll_family = AF_PACKET;
1883 sll->sll_hatype = dev->type;
1884 sll->sll_protocol = skb->protocol;
1885 sll->sll_pkttype = skb->pkt_type;
1886 if (unlikely(po->origdev))
1887 sll->sll_ifindex = orig_dev->ifindex;
1889 sll->sll_ifindex = dev->ifindex;
1892 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1896 if (po->tp_version <= TPACKET_V2) {
1897 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1898 + macoff + snaplen);
1899 for (start = h.raw; start < end; start += PAGE_SIZE)
1900 flush_dcache_page(pgv_to_page(start));
1905 if (po->tp_version <= TPACKET_V2)
1906 __packet_set_status(po, h.raw, status);
1908 prb_clear_blk_fill_status(&po->rx_ring);
1910 sk->sk_data_ready(sk, 0);
1913 if (skb_head != skb->data && skb_shared(skb)) {
1914 skb->data = skb_head;
1922 po->stats.tp_drops++;
1923 spin_unlock(&sk->sk_receive_queue.lock);
1925 sk->sk_data_ready(sk, 0);
1926 kfree_skb(copy_skb);
1927 goto drop_n_restore;
1930 static void tpacket_destruct_skb(struct sk_buff *skb)
1932 struct packet_sock *po = pkt_sk(skb->sk);
1935 if (likely(po->tx_ring.pg_vec)) {
1936 ph = skb_shinfo(skb)->destructor_arg;
1937 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1938 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1939 atomic_dec(&po->tx_ring.pending);
1940 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1946 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1947 void *frame, struct net_device *dev, int size_max,
1948 __be16 proto, unsigned char *addr, int hlen)
1951 struct tpacket_hdr *h1;
1952 struct tpacket2_hdr *h2;
1955 int to_write, offset, len, tp_len, nr_frags, len_max;
1956 struct socket *sock = po->sk.sk_socket;
1963 skb->protocol = proto;
1965 skb->priority = po->sk.sk_priority;
1966 skb->mark = po->sk.sk_mark;
1967 skb_shinfo(skb)->destructor_arg = ph.raw;
1969 switch (po->tp_version) {
1971 tp_len = ph.h2->tp_len;
1974 tp_len = ph.h1->tp_len;
1977 if (unlikely(tp_len > size_max)) {
1978 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1982 skb_reserve(skb, hlen);
1983 skb_reset_network_header(skb);
1985 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1988 if (sock->type == SOCK_DGRAM) {
1989 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1991 if (unlikely(err < 0))
1993 } else if (dev->hard_header_len) {
1994 /* net device doesn't like empty head */
1995 if (unlikely(tp_len <= dev->hard_header_len)) {
1996 pr_err("packet size is too short (%d < %d)\n",
1997 tp_len, dev->hard_header_len);
2001 skb_push(skb, dev->hard_header_len);
2002 err = skb_store_bits(skb, 0, data,
2003 dev->hard_header_len);
2007 data += dev->hard_header_len;
2008 to_write -= dev->hard_header_len;
2012 offset = offset_in_page(data);
2013 len_max = PAGE_SIZE - offset;
2014 len = ((to_write > len_max) ? len_max : to_write);
2016 skb->data_len = to_write;
2017 skb->len += to_write;
2018 skb->truesize += to_write;
2019 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2021 while (likely(to_write)) {
2022 nr_frags = skb_shinfo(skb)->nr_frags;
2024 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2025 pr_err("Packet exceed the number of skb frags(%lu)\n",
2030 page = pgv_to_page(data);
2032 flush_dcache_page(page);
2034 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2037 len_max = PAGE_SIZE;
2038 len = ((to_write > len_max) ? len_max : to_write);
2044 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2046 struct sk_buff *skb;
2047 struct net_device *dev;
2049 bool need_rls_dev = false;
2050 int err, reserve = 0;
2052 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2053 int tp_len, size_max;
2054 unsigned char *addr;
2059 mutex_lock(&po->pg_vec_lock);
2062 if (saddr == NULL) {
2063 dev = po->prot_hook.dev;
2068 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2070 if (msg->msg_namelen < (saddr->sll_halen
2071 + offsetof(struct sockaddr_ll,
2074 proto = saddr->sll_protocol;
2075 addr = saddr->sll_addr;
2076 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2077 need_rls_dev = true;
2081 if (unlikely(dev == NULL))
2084 reserve = dev->hard_header_len;
2087 if (unlikely(!(dev->flags & IFF_UP)))
2090 size_max = po->tx_ring.frame_size
2091 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2093 if (size_max > dev->mtu + reserve)
2094 size_max = dev->mtu + reserve;
2097 ph = packet_current_frame(po, &po->tx_ring,
2098 TP_STATUS_SEND_REQUEST);
2100 if (unlikely(ph == NULL)) {
2105 status = TP_STATUS_SEND_REQUEST;
2106 hlen = LL_RESERVED_SPACE(dev);
2107 tlen = dev->needed_tailroom;
2108 skb = sock_alloc_send_skb(&po->sk,
2109 hlen + tlen + sizeof(struct sockaddr_ll),
2112 if (unlikely(skb == NULL))
2115 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2118 if (unlikely(tp_len < 0)) {
2120 __packet_set_status(po, ph,
2121 TP_STATUS_AVAILABLE);
2122 packet_increment_head(&po->tx_ring);
2126 status = TP_STATUS_WRONG_FORMAT;
2132 skb->destructor = tpacket_destruct_skb;
2133 __packet_set_status(po, ph, TP_STATUS_SENDING);
2134 atomic_inc(&po->tx_ring.pending);
2136 status = TP_STATUS_SEND_REQUEST;
2137 err = dev_queue_xmit(skb);
2138 if (unlikely(err > 0)) {
2139 err = net_xmit_errno(err);
2140 if (err && __packet_get_status(po, ph) ==
2141 TP_STATUS_AVAILABLE) {
2142 /* skb was destructed already */
2147 * skb was dropped but not destructed yet;
2148 * let's treat it like congestion or err < 0
2152 packet_increment_head(&po->tx_ring);
2154 } while (likely((ph != NULL) ||
2155 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2156 (atomic_read(&po->tx_ring.pending))))
2163 __packet_set_status(po, ph, status);
2169 mutex_unlock(&po->pg_vec_lock);
2173 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2174 size_t reserve, size_t len,
2175 size_t linear, int noblock,
2178 struct sk_buff *skb;
2180 /* Under a page? Don't bother with paged skb. */
2181 if (prepad + len < PAGE_SIZE || !linear)
2184 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2189 skb_reserve(skb, reserve);
2190 skb_put(skb, linear);
2191 skb->data_len = len - linear;
2192 skb->len += len - linear;
2197 static int packet_snd(struct socket *sock,
2198 struct msghdr *msg, size_t len)
2200 struct sock *sk = sock->sk;
2201 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2202 struct sk_buff *skb;
2203 struct net_device *dev;
2205 bool need_rls_dev = false;
2206 unsigned char *addr;
2207 int err, reserve = 0;
2208 struct virtio_net_hdr vnet_hdr = { 0 };
2211 struct packet_sock *po = pkt_sk(sk);
2212 unsigned short gso_type = 0;
2216 * Get and verify the address.
2219 if (saddr == NULL) {
2220 dev = po->prot_hook.dev;
2225 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2227 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2229 proto = saddr->sll_protocol;
2230 addr = saddr->sll_addr;
2231 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2232 need_rls_dev = true;
2238 if (sock->type == SOCK_RAW)
2239 reserve = dev->hard_header_len;
2242 if (!(dev->flags & IFF_UP))
2245 if (po->has_vnet_hdr) {
2246 vnet_hdr_len = sizeof(vnet_hdr);
2249 if (len < vnet_hdr_len)
2252 len -= vnet_hdr_len;
2254 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2259 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2260 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2262 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2263 vnet_hdr.csum_offset + 2;
2266 if (vnet_hdr.hdr_len > len)
2269 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2270 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2271 case VIRTIO_NET_HDR_GSO_TCPV4:
2272 gso_type = SKB_GSO_TCPV4;
2274 case VIRTIO_NET_HDR_GSO_TCPV6:
2275 gso_type = SKB_GSO_TCPV6;
2277 case VIRTIO_NET_HDR_GSO_UDP:
2278 gso_type = SKB_GSO_UDP;
2284 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2285 gso_type |= SKB_GSO_TCP_ECN;
2287 if (vnet_hdr.gso_size == 0)
2294 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2298 hlen = LL_RESERVED_SPACE(dev);
2299 tlen = dev->needed_tailroom;
2300 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2301 msg->msg_flags & MSG_DONTWAIT, &err);
2305 skb_set_network_header(skb, reserve);
2308 if (sock->type == SOCK_DGRAM &&
2309 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2312 /* Returns -EFAULT on error */
2313 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2316 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2320 if (!gso_type && (len > dev->mtu + reserve)) {
2321 /* Earlier code assumed this would be a VLAN pkt,
2322 * double-check this now that we have the actual
2325 struct ethhdr *ehdr;
2326 skb_reset_mac_header(skb);
2327 ehdr = eth_hdr(skb);
2328 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2334 skb->protocol = proto;
2336 skb->priority = sk->sk_priority;
2337 skb->mark = sk->sk_mark;
2339 if (po->has_vnet_hdr) {
2340 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2341 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2342 vnet_hdr.csum_offset)) {
2348 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2349 skb_shinfo(skb)->gso_type = gso_type;
2351 /* Header must be checked, and gso_segs computed. */
2352 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2353 skb_shinfo(skb)->gso_segs = 0;
2355 len += vnet_hdr_len;
2362 err = dev_queue_xmit(skb);
2363 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2374 if (dev && need_rls_dev)
2380 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2381 struct msghdr *msg, size_t len)
2383 struct sock *sk = sock->sk;
2384 struct packet_sock *po = pkt_sk(sk);
2385 if (po->tx_ring.pg_vec)
2386 return tpacket_snd(po, msg);
2388 return packet_snd(sock, msg, len);
2392 * Close a PACKET socket. This is fairly simple. We immediately go
2393 * to 'closed' state and remove our protocol entry in the device list.
2396 static int packet_release(struct socket *sock)
2398 struct sock *sk = sock->sk;
2399 struct packet_sock *po;
2401 union tpacket_req_u req_u;
2409 spin_lock_bh(&net->packet.sklist_lock);
2410 sk_del_node_init_rcu(sk);
2411 sock_prot_inuse_add(net, sk->sk_prot, -1);
2412 spin_unlock_bh(&net->packet.sklist_lock);
2414 spin_lock(&po->bind_lock);
2415 unregister_prot_hook(sk, false);
2416 if (po->prot_hook.dev) {
2417 dev_put(po->prot_hook.dev);
2418 po->prot_hook.dev = NULL;
2420 spin_unlock(&po->bind_lock);
2422 packet_flush_mclist(sk);
2424 memset(&req_u, 0, sizeof(req_u));
2426 if (po->rx_ring.pg_vec)
2427 packet_set_ring(sk, &req_u, 1, 0);
2429 if (po->tx_ring.pg_vec)
2430 packet_set_ring(sk, &req_u, 1, 1);
2436 * Now the socket is dead. No more input will appear.
2443 skb_queue_purge(&sk->sk_receive_queue);
2444 sk_refcnt_debug_release(sk);
2451 * Attach a packet hook.
2454 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2456 struct packet_sock *po = pkt_sk(sk);
2463 spin_lock(&po->bind_lock);
2464 unregister_prot_hook(sk, true);
2466 po->prot_hook.type = protocol;
2467 if (po->prot_hook.dev)
2468 dev_put(po->prot_hook.dev);
2469 po->prot_hook.dev = dev;
2471 po->ifindex = dev ? dev->ifindex : 0;
2476 if (!dev || (dev->flags & IFF_UP)) {
2477 register_prot_hook(sk);
2479 sk->sk_err = ENETDOWN;
2480 if (!sock_flag(sk, SOCK_DEAD))
2481 sk->sk_error_report(sk);
2485 spin_unlock(&po->bind_lock);
2491 * Bind a packet socket to a device
2494 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2497 struct sock *sk = sock->sk;
2499 struct net_device *dev;
2506 if (addr_len != sizeof(struct sockaddr))
2508 strlcpy(name, uaddr->sa_data, sizeof(name));
2510 dev = dev_get_by_name(sock_net(sk), name);
2512 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2516 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2518 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2519 struct sock *sk = sock->sk;
2520 struct net_device *dev = NULL;
2528 if (addr_len < sizeof(struct sockaddr_ll))
2530 if (sll->sll_family != AF_PACKET)
2533 if (sll->sll_ifindex) {
2535 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2539 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2545 static struct proto packet_proto = {
2547 .owner = THIS_MODULE,
2548 .obj_size = sizeof(struct packet_sock),
2552 * Create a packet of type SOCK_PACKET.
2555 static int packet_create(struct net *net, struct socket *sock, int protocol,
2559 struct packet_sock *po;
2560 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2563 if (!capable(CAP_NET_RAW))
2565 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2566 sock->type != SOCK_PACKET)
2567 return -ESOCKTNOSUPPORT;
2569 sock->state = SS_UNCONNECTED;
2572 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2576 sock->ops = &packet_ops;
2577 if (sock->type == SOCK_PACKET)
2578 sock->ops = &packet_ops_spkt;
2580 sock_init_data(sock, sk);
2583 sk->sk_family = PF_PACKET;
2586 sk->sk_destruct = packet_sock_destruct;
2587 sk_refcnt_debug_inc(sk);
2590 * Attach a protocol block
2593 spin_lock_init(&po->bind_lock);
2594 mutex_init(&po->pg_vec_lock);
2595 po->prot_hook.func = packet_rcv;
2597 if (sock->type == SOCK_PACKET)
2598 po->prot_hook.func = packet_rcv_spkt;
2600 po->prot_hook.af_packet_priv = sk;
2603 po->prot_hook.type = proto;
2604 register_prot_hook(sk);
2607 spin_lock_bh(&net->packet.sklist_lock);
2608 sk_add_node_rcu(sk, &net->packet.sklist);
2609 sock_prot_inuse_add(net, &packet_proto, 1);
2610 spin_unlock_bh(&net->packet.sklist_lock);
2617 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2619 struct sock_exterr_skb *serr;
2620 struct sk_buff *skb, *skb2;
2624 skb = skb_dequeue(&sk->sk_error_queue);
2630 msg->msg_flags |= MSG_TRUNC;
2633 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2637 sock_recv_timestamp(msg, sk, skb);
2639 serr = SKB_EXT_ERR(skb);
2640 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2641 sizeof(serr->ee), &serr->ee);
2643 msg->msg_flags |= MSG_ERRQUEUE;
2646 /* Reset and regenerate socket error */
2647 spin_lock_bh(&sk->sk_error_queue.lock);
2649 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2650 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2651 spin_unlock_bh(&sk->sk_error_queue.lock);
2652 sk->sk_error_report(sk);
2654 spin_unlock_bh(&sk->sk_error_queue.lock);
2663 * Pull a packet from our receive queue and hand it to the user.
2664 * If necessary we block.
2667 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2668 struct msghdr *msg, size_t len, int flags)
2670 struct sock *sk = sock->sk;
2671 struct sk_buff *skb;
2673 struct sockaddr_ll *sll;
2674 int vnet_hdr_len = 0;
2677 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2681 /* What error should we return now? EUNATTACH? */
2682 if (pkt_sk(sk)->ifindex < 0)
2686 if (flags & MSG_ERRQUEUE) {
2687 err = packet_recv_error(sk, msg, len);
2692 * Call the generic datagram receiver. This handles all sorts
2693 * of horrible races and re-entrancy so we can forget about it
2694 * in the protocol layers.
2696 * Now it will return ENETDOWN, if device have just gone down,
2697 * but then it will block.
2700 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2703 * An error occurred so return it. Because skb_recv_datagram()
2704 * handles the blocking we don't see and worry about blocking
2711 if (pkt_sk(sk)->has_vnet_hdr) {
2712 struct virtio_net_hdr vnet_hdr = { 0 };
2715 vnet_hdr_len = sizeof(vnet_hdr);
2716 if (len < vnet_hdr_len)
2719 len -= vnet_hdr_len;
2721 if (skb_is_gso(skb)) {
2722 struct skb_shared_info *sinfo = skb_shinfo(skb);
2724 /* This is a hint as to how much should be linear. */
2725 vnet_hdr.hdr_len = skb_headlen(skb);
2726 vnet_hdr.gso_size = sinfo->gso_size;
2727 if (sinfo->gso_type & SKB_GSO_TCPV4)
2728 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2729 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2730 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2731 else if (sinfo->gso_type & SKB_GSO_UDP)
2732 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2733 else if (sinfo->gso_type & SKB_GSO_FCOE)
2737 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2738 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2740 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2742 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2743 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2744 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2745 vnet_hdr.csum_offset = skb->csum_offset;
2746 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2747 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2748 } /* else everything is zero */
2750 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2757 * If the address length field is there to be filled in, we fill
2761 sll = &PACKET_SKB_CB(skb)->sa.ll;
2762 if (sock->type == SOCK_PACKET)
2763 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2765 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2768 * You lose any data beyond the buffer you gave. If it worries a
2769 * user program they can ask the device for its MTU anyway.
2775 msg->msg_flags |= MSG_TRUNC;
2778 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2782 sock_recv_ts_and_drops(msg, sk, skb);
2785 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2788 if (pkt_sk(sk)->auxdata) {
2789 struct tpacket_auxdata aux;
2791 aux.tp_status = TP_STATUS_USER;
2792 if (skb->ip_summed == CHECKSUM_PARTIAL)
2793 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2794 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2795 aux.tp_snaplen = skb->len;
2797 aux.tp_net = skb_network_offset(skb);
2798 if (vlan_tx_tag_present(skb)) {
2799 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2800 aux.tp_status |= TP_STATUS_VLAN_VALID;
2802 aux.tp_vlan_tci = 0;
2805 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2809 * Free or return the buffer as appropriate. Again this
2810 * hides all the races and re-entrancy issues from us.
2812 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2815 skb_free_datagram(sk, skb);
2820 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2821 int *uaddr_len, int peer)
2823 struct net_device *dev;
2824 struct sock *sk = sock->sk;
2829 uaddr->sa_family = AF_PACKET;
2831 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2833 strncpy(uaddr->sa_data, dev->name, 14);
2835 memset(uaddr->sa_data, 0, 14);
2837 *uaddr_len = sizeof(*uaddr);
2842 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2843 int *uaddr_len, int peer)
2845 struct net_device *dev;
2846 struct sock *sk = sock->sk;
2847 struct packet_sock *po = pkt_sk(sk);
2848 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2853 sll->sll_family = AF_PACKET;
2854 sll->sll_ifindex = po->ifindex;
2855 sll->sll_protocol = po->num;
2856 sll->sll_pkttype = 0;
2858 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2860 sll->sll_hatype = dev->type;
2861 sll->sll_halen = dev->addr_len;
2862 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2864 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2868 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2873 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2877 case PACKET_MR_MULTICAST:
2878 if (i->alen != dev->addr_len)
2881 return dev_mc_add(dev, i->addr);
2883 return dev_mc_del(dev, i->addr);
2885 case PACKET_MR_PROMISC:
2886 return dev_set_promiscuity(dev, what);
2888 case PACKET_MR_ALLMULTI:
2889 return dev_set_allmulti(dev, what);
2891 case PACKET_MR_UNICAST:
2892 if (i->alen != dev->addr_len)
2895 return dev_uc_add(dev, i->addr);
2897 return dev_uc_del(dev, i->addr);
2905 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2907 for ( ; i; i = i->next) {
2908 if (i->ifindex == dev->ifindex)
2909 packet_dev_mc(dev, i, what);
2913 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2915 struct packet_sock *po = pkt_sk(sk);
2916 struct packet_mclist *ml, *i;
2917 struct net_device *dev;
2923 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2928 if (mreq->mr_alen > dev->addr_len)
2932 i = kmalloc(sizeof(*i), GFP_KERNEL);
2937 for (ml = po->mclist; ml; ml = ml->next) {
2938 if (ml->ifindex == mreq->mr_ifindex &&
2939 ml->type == mreq->mr_type &&
2940 ml->alen == mreq->mr_alen &&
2941 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2943 /* Free the new element ... */
2949 i->type = mreq->mr_type;
2950 i->ifindex = mreq->mr_ifindex;
2951 i->alen = mreq->mr_alen;
2952 memcpy(i->addr, mreq->mr_address, i->alen);
2954 i->next = po->mclist;
2956 err = packet_dev_mc(dev, i, 1);
2958 po->mclist = i->next;
2967 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2969 struct packet_mclist *ml, **mlp;
2973 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2974 if (ml->ifindex == mreq->mr_ifindex &&
2975 ml->type == mreq->mr_type &&
2976 ml->alen == mreq->mr_alen &&
2977 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2978 if (--ml->count == 0) {
2979 struct net_device *dev;
2981 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2983 packet_dev_mc(dev, ml, -1);
2991 return -EADDRNOTAVAIL;
2994 static void packet_flush_mclist(struct sock *sk)
2996 struct packet_sock *po = pkt_sk(sk);
2997 struct packet_mclist *ml;
3003 while ((ml = po->mclist) != NULL) {
3004 struct net_device *dev;
3006 po->mclist = ml->next;
3007 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3009 packet_dev_mc(dev, ml, -1);
3016 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3018 struct sock *sk = sock->sk;
3019 struct packet_sock *po = pkt_sk(sk);
3022 if (level != SOL_PACKET)
3023 return -ENOPROTOOPT;
3026 case PACKET_ADD_MEMBERSHIP:
3027 case PACKET_DROP_MEMBERSHIP:
3029 struct packet_mreq_max mreq;
3031 memset(&mreq, 0, sizeof(mreq));
3032 if (len < sizeof(struct packet_mreq))
3034 if (len > sizeof(mreq))
3036 if (copy_from_user(&mreq, optval, len))
3038 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3040 if (optname == PACKET_ADD_MEMBERSHIP)
3041 ret = packet_mc_add(sk, &mreq);
3043 ret = packet_mc_drop(sk, &mreq);
3047 case PACKET_RX_RING:
3048 case PACKET_TX_RING:
3050 union tpacket_req_u req_u;
3053 switch (po->tp_version) {
3056 len = sizeof(req_u.req);
3060 len = sizeof(req_u.req3);
3065 if (pkt_sk(sk)->has_vnet_hdr)
3067 if (copy_from_user(&req_u.req, optval, len))
3069 return packet_set_ring(sk, &req_u, 0,
3070 optname == PACKET_TX_RING);
3072 case PACKET_COPY_THRESH:
3076 if (optlen != sizeof(val))
3078 if (copy_from_user(&val, optval, sizeof(val)))
3081 pkt_sk(sk)->copy_thresh = val;
3084 case PACKET_VERSION:
3088 if (optlen != sizeof(val))
3090 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3092 if (copy_from_user(&val, optval, sizeof(val)))
3098 po->tp_version = val;
3104 case PACKET_RESERVE:
3108 if (optlen != sizeof(val))
3110 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3112 if (copy_from_user(&val, optval, sizeof(val)))
3114 po->tp_reserve = val;
3121 if (optlen != sizeof(val))
3123 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3125 if (copy_from_user(&val, optval, sizeof(val)))
3127 po->tp_loss = !!val;
3130 case PACKET_AUXDATA:
3134 if (optlen < sizeof(val))
3136 if (copy_from_user(&val, optval, sizeof(val)))
3139 po->auxdata = !!val;
3142 case PACKET_ORIGDEV:
3146 if (optlen < sizeof(val))
3148 if (copy_from_user(&val, optval, sizeof(val)))
3151 po->origdev = !!val;
3154 case PACKET_VNET_HDR:
3158 if (sock->type != SOCK_RAW)
3160 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3162 if (optlen < sizeof(val))
3164 if (copy_from_user(&val, optval, sizeof(val)))
3167 po->has_vnet_hdr = !!val;
3170 case PACKET_TIMESTAMP:
3174 if (optlen != sizeof(val))
3176 if (copy_from_user(&val, optval, sizeof(val)))
3179 po->tp_tstamp = val;
3186 if (optlen != sizeof(val))
3188 if (copy_from_user(&val, optval, sizeof(val)))
3191 return fanout_add(sk, val & 0xffff, val >> 16);
3194 return -ENOPROTOOPT;
3198 static int packet_getsockopt(struct socket *sock, int level, int optname,
3199 char __user *optval, int __user *optlen)
3203 struct sock *sk = sock->sk;
3204 struct packet_sock *po = pkt_sk(sk);
3206 struct tpacket_stats st;
3207 union tpacket_stats_u st_u;
3209 if (level != SOL_PACKET)
3210 return -ENOPROTOOPT;
3212 if (get_user(len, optlen))
3219 case PACKET_STATISTICS:
3220 if (po->tp_version == TPACKET_V3) {
3221 len = sizeof(struct tpacket_stats_v3);
3223 if (len > sizeof(struct tpacket_stats))
3224 len = sizeof(struct tpacket_stats);
3226 spin_lock_bh(&sk->sk_receive_queue.lock);
3227 if (po->tp_version == TPACKET_V3) {
3228 memcpy(&st_u.stats3, &po->stats,
3229 sizeof(struct tpacket_stats));
3230 st_u.stats3.tp_freeze_q_cnt =
3231 po->stats_u.stats3.tp_freeze_q_cnt;
3232 st_u.stats3.tp_packets += po->stats.tp_drops;
3233 data = &st_u.stats3;
3236 st.tp_packets += st.tp_drops;
3239 memset(&po->stats, 0, sizeof(st));
3240 spin_unlock_bh(&sk->sk_receive_queue.lock);
3242 case PACKET_AUXDATA:
3243 if (len > sizeof(int))
3249 case PACKET_ORIGDEV:
3250 if (len > sizeof(int))
3256 case PACKET_VNET_HDR:
3257 if (len > sizeof(int))
3259 val = po->has_vnet_hdr;
3263 case PACKET_VERSION:
3264 if (len > sizeof(int))
3266 val = po->tp_version;
3270 if (len > sizeof(int))
3272 if (copy_from_user(&val, optval, len))
3276 val = sizeof(struct tpacket_hdr);
3279 val = sizeof(struct tpacket2_hdr);
3282 val = sizeof(struct tpacket3_hdr);
3289 case PACKET_RESERVE:
3290 if (len > sizeof(unsigned int))
3291 len = sizeof(unsigned int);
3292 val = po->tp_reserve;
3296 if (len > sizeof(unsigned int))
3297 len = sizeof(unsigned int);
3301 case PACKET_TIMESTAMP:
3302 if (len > sizeof(int))
3304 val = po->tp_tstamp;
3308 if (len > sizeof(int))
3311 ((u32)po->fanout->id |
3312 ((u32)po->fanout->type << 16)) :
3317 return -ENOPROTOOPT;
3320 if (put_user(len, optlen))
3322 if (copy_to_user(optval, data, len))
3328 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3331 struct hlist_node *node;
3332 struct net_device *dev = data;
3333 struct net *net = dev_net(dev);
3336 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3337 struct packet_sock *po = pkt_sk(sk);
3340 case NETDEV_UNREGISTER:
3342 packet_dev_mclist(dev, po->mclist, -1);
3346 if (dev->ifindex == po->ifindex) {
3347 spin_lock(&po->bind_lock);
3349 __unregister_prot_hook(sk, false);
3350 sk->sk_err = ENETDOWN;
3351 if (!sock_flag(sk, SOCK_DEAD))
3352 sk->sk_error_report(sk);
3354 if (msg == NETDEV_UNREGISTER) {
3356 if (po->prot_hook.dev)
3357 dev_put(po->prot_hook.dev);
3358 po->prot_hook.dev = NULL;
3360 spin_unlock(&po->bind_lock);
3364 if (dev->ifindex == po->ifindex) {
3365 spin_lock(&po->bind_lock);
3367 register_prot_hook(sk);
3368 spin_unlock(&po->bind_lock);
3378 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3381 struct sock *sk = sock->sk;
3386 int amount = sk_wmem_alloc_get(sk);
3388 return put_user(amount, (int __user *)arg);
3392 struct sk_buff *skb;
3395 spin_lock_bh(&sk->sk_receive_queue.lock);
3396 skb = skb_peek(&sk->sk_receive_queue);
3399 spin_unlock_bh(&sk->sk_receive_queue.lock);
3400 return put_user(amount, (int __user *)arg);
3403 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3405 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3415 case SIOCGIFBRDADDR:
3416 case SIOCSIFBRDADDR:
3417 case SIOCGIFNETMASK:
3418 case SIOCSIFNETMASK:
3419 case SIOCGIFDSTADDR:
3420 case SIOCSIFDSTADDR:
3422 return inet_dgram_ops.ioctl(sock, cmd, arg);
3426 return -ENOIOCTLCMD;
3431 static unsigned int packet_poll(struct file *file, struct socket *sock,
3434 struct sock *sk = sock->sk;
3435 struct packet_sock *po = pkt_sk(sk);
3436 unsigned int mask = datagram_poll(file, sock, wait);
3438 spin_lock_bh(&sk->sk_receive_queue.lock);
3439 if (po->rx_ring.pg_vec) {
3440 if (!packet_previous_rx_frame(po, &po->rx_ring,
3442 mask |= POLLIN | POLLRDNORM;
3444 spin_unlock_bh(&sk->sk_receive_queue.lock);
3445 spin_lock_bh(&sk->sk_write_queue.lock);
3446 if (po->tx_ring.pg_vec) {
3447 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3448 mask |= POLLOUT | POLLWRNORM;
3450 spin_unlock_bh(&sk->sk_write_queue.lock);
3455 /* Dirty? Well, I still did not learn better way to account
3459 static void packet_mm_open(struct vm_area_struct *vma)
3461 struct file *file = vma->vm_file;
3462 struct socket *sock = file->private_data;
3463 struct sock *sk = sock->sk;
3466 atomic_inc(&pkt_sk(sk)->mapped);
3469 static void packet_mm_close(struct vm_area_struct *vma)
3471 struct file *file = vma->vm_file;
3472 struct socket *sock = file->private_data;
3473 struct sock *sk = sock->sk;
3476 atomic_dec(&pkt_sk(sk)->mapped);
3479 static const struct vm_operations_struct packet_mmap_ops = {
3480 .open = packet_mm_open,
3481 .close = packet_mm_close,
3484 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3489 for (i = 0; i < len; i++) {
3490 if (likely(pg_vec[i].buffer)) {
3491 if (is_vmalloc_addr(pg_vec[i].buffer))
3492 vfree(pg_vec[i].buffer);
3494 free_pages((unsigned long)pg_vec[i].buffer,
3496 pg_vec[i].buffer = NULL;
3502 static char *alloc_one_pg_vec_page(unsigned long order)
3504 char *buffer = NULL;
3505 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3506 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3508 buffer = (char *) __get_free_pages(gfp_flags, order);
3514 * __get_free_pages failed, fall back to vmalloc
3516 buffer = vzalloc((1 << order) * PAGE_SIZE);
3522 * vmalloc failed, lets dig into swap here
3524 gfp_flags &= ~__GFP_NORETRY;
3525 buffer = (char *)__get_free_pages(gfp_flags, order);
3530 * complete and utter failure
3535 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3537 unsigned int block_nr = req->tp_block_nr;
3541 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3542 if (unlikely(!pg_vec))
3545 for (i = 0; i < block_nr; i++) {
3546 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3547 if (unlikely(!pg_vec[i].buffer))
3548 goto out_free_pgvec;
3555 free_pg_vec(pg_vec, order, block_nr);
3560 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3561 int closing, int tx_ring)
3563 struct pgv *pg_vec = NULL;
3564 struct packet_sock *po = pkt_sk(sk);
3565 int was_running, order = 0;
3566 struct packet_ring_buffer *rb;
3567 struct sk_buff_head *rb_queue;
3570 /* Added to avoid minimal code churn */
3571 struct tpacket_req *req = &req_u->req;
3573 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3574 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3575 WARN(1, "Tx-ring is not supported.\n");
3579 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3580 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3584 if (atomic_read(&po->mapped))
3586 if (atomic_read(&rb->pending))
3590 if (req->tp_block_nr) {
3591 /* Sanity tests and some calculations */
3593 if (unlikely(rb->pg_vec))
3596 switch (po->tp_version) {
3598 po->tp_hdrlen = TPACKET_HDRLEN;
3601 po->tp_hdrlen = TPACKET2_HDRLEN;
3604 po->tp_hdrlen = TPACKET3_HDRLEN;
3609 if (unlikely((int)req->tp_block_size <= 0))
3611 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3613 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3616 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3619 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3620 if (unlikely(rb->frames_per_block <= 0))
3622 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3627 order = get_order(req->tp_block_size);
3628 pg_vec = alloc_pg_vec(req, order);
3629 if (unlikely(!pg_vec))
3631 switch (po->tp_version) {
3633 /* Transmit path is not supported. We checked
3634 * it above but just being paranoid
3637 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3646 if (unlikely(req->tp_frame_nr))
3652 /* Detach socket from network */
3653 spin_lock(&po->bind_lock);
3654 was_running = po->running;
3658 __unregister_prot_hook(sk, false);
3660 spin_unlock(&po->bind_lock);
3665 mutex_lock(&po->pg_vec_lock);
3666 if (closing || atomic_read(&po->mapped) == 0) {
3668 spin_lock_bh(&rb_queue->lock);
3669 swap(rb->pg_vec, pg_vec);
3670 rb->frame_max = (req->tp_frame_nr - 1);
3672 rb->frame_size = req->tp_frame_size;
3673 spin_unlock_bh(&rb_queue->lock);
3675 swap(rb->pg_vec_order, order);
3676 swap(rb->pg_vec_len, req->tp_block_nr);
3678 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3679 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3680 tpacket_rcv : packet_rcv;
3681 skb_queue_purge(rb_queue);
3682 if (atomic_read(&po->mapped))
3683 pr_err("packet_mmap: vma is busy: %d\n",
3684 atomic_read(&po->mapped));
3686 mutex_unlock(&po->pg_vec_lock);
3688 spin_lock(&po->bind_lock);
3691 register_prot_hook(sk);
3693 spin_unlock(&po->bind_lock);
3694 if (closing && (po->tp_version > TPACKET_V2)) {
3695 /* Because we don't support block-based V3 on tx-ring */
3697 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3702 free_pg_vec(pg_vec, order, req->tp_block_nr);
3707 static int packet_mmap(struct file *file, struct socket *sock,
3708 struct vm_area_struct *vma)
3710 struct sock *sk = sock->sk;
3711 struct packet_sock *po = pkt_sk(sk);
3712 unsigned long size, expected_size;
3713 struct packet_ring_buffer *rb;
3714 unsigned long start;
3721 mutex_lock(&po->pg_vec_lock);
3724 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3726 expected_size += rb->pg_vec_len
3732 if (expected_size == 0)
3735 size = vma->vm_end - vma->vm_start;
3736 if (size != expected_size)
3739 start = vma->vm_start;
3740 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3741 if (rb->pg_vec == NULL)
3744 for (i = 0; i < rb->pg_vec_len; i++) {
3746 void *kaddr = rb->pg_vec[i].buffer;
3749 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3750 page = pgv_to_page(kaddr);
3751 err = vm_insert_page(vma, start, page);
3760 atomic_inc(&po->mapped);
3761 vma->vm_ops = &packet_mmap_ops;
3765 mutex_unlock(&po->pg_vec_lock);
3769 static const struct proto_ops packet_ops_spkt = {
3770 .family = PF_PACKET,
3771 .owner = THIS_MODULE,
3772 .release = packet_release,
3773 .bind = packet_bind_spkt,
3774 .connect = sock_no_connect,
3775 .socketpair = sock_no_socketpair,
3776 .accept = sock_no_accept,
3777 .getname = packet_getname_spkt,
3778 .poll = datagram_poll,
3779 .ioctl = packet_ioctl,
3780 .listen = sock_no_listen,
3781 .shutdown = sock_no_shutdown,
3782 .setsockopt = sock_no_setsockopt,
3783 .getsockopt = sock_no_getsockopt,
3784 .sendmsg = packet_sendmsg_spkt,
3785 .recvmsg = packet_recvmsg,
3786 .mmap = sock_no_mmap,
3787 .sendpage = sock_no_sendpage,
3790 static const struct proto_ops packet_ops = {
3791 .family = PF_PACKET,
3792 .owner = THIS_MODULE,
3793 .release = packet_release,
3794 .bind = packet_bind,
3795 .connect = sock_no_connect,
3796 .socketpair = sock_no_socketpair,
3797 .accept = sock_no_accept,
3798 .getname = packet_getname,
3799 .poll = packet_poll,
3800 .ioctl = packet_ioctl,
3801 .listen = sock_no_listen,
3802 .shutdown = sock_no_shutdown,
3803 .setsockopt = packet_setsockopt,
3804 .getsockopt = packet_getsockopt,
3805 .sendmsg = packet_sendmsg,
3806 .recvmsg = packet_recvmsg,
3807 .mmap = packet_mmap,
3808 .sendpage = sock_no_sendpage,
3811 static const struct net_proto_family packet_family_ops = {
3812 .family = PF_PACKET,
3813 .create = packet_create,
3814 .owner = THIS_MODULE,
3817 static struct notifier_block packet_netdev_notifier = {
3818 .notifier_call = packet_notifier,
3821 #ifdef CONFIG_PROC_FS
3823 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3826 struct net *net = seq_file_net(seq);
3829 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3832 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3834 struct net *net = seq_file_net(seq);
3835 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3838 static void packet_seq_stop(struct seq_file *seq, void *v)
3844 static int packet_seq_show(struct seq_file *seq, void *v)
3846 if (v == SEQ_START_TOKEN)
3847 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3849 struct sock *s = sk_entry(v);
3850 const struct packet_sock *po = pkt_sk(s);
3853 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3855 atomic_read(&s->sk_refcnt),
3860 atomic_read(&s->sk_rmem_alloc),
3868 static const struct seq_operations packet_seq_ops = {
3869 .start = packet_seq_start,
3870 .next = packet_seq_next,
3871 .stop = packet_seq_stop,
3872 .show = packet_seq_show,
3875 static int packet_seq_open(struct inode *inode, struct file *file)
3877 return seq_open_net(inode, file, &packet_seq_ops,
3878 sizeof(struct seq_net_private));
3881 static const struct file_operations packet_seq_fops = {
3882 .owner = THIS_MODULE,
3883 .open = packet_seq_open,
3885 .llseek = seq_lseek,
3886 .release = seq_release_net,
3891 static int __net_init packet_net_init(struct net *net)
3893 spin_lock_init(&net->packet.sklist_lock);
3894 INIT_HLIST_HEAD(&net->packet.sklist);
3896 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3902 static void __net_exit packet_net_exit(struct net *net)
3904 proc_net_remove(net, "packet");
3907 static struct pernet_operations packet_net_ops = {
3908 .init = packet_net_init,
3909 .exit = packet_net_exit,
3913 static void __exit packet_exit(void)
3915 unregister_netdevice_notifier(&packet_netdev_notifier);
3916 unregister_pernet_subsys(&packet_net_ops);
3917 sock_unregister(PF_PACKET);
3918 proto_unregister(&packet_proto);
3921 static int __init packet_init(void)
3923 int rc = proto_register(&packet_proto, 0);
3928 sock_register(&packet_family_ops);
3929 register_pernet_subsys(&packet_net_ops);
3930 register_netdevice_notifier(&packet_netdev_notifier);
3935 module_init(packet_init);
3936 module_exit(packet_exit);
3937 MODULE_LICENSE("GPL");
3938 MODULE_ALIAS_NETPROTO(PF_PACKET);