2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
93 #include <net/inet_common.h>
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
218 struct packet_skb_cb {
219 unsigned int origlen;
221 struct sockaddr_pkt pkt;
222 struct sockaddr_ll ll;
226 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
228 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
229 #define GET_PBLOCK_DESC(x, bid) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
233 #define GET_NEXT_PRB_BLK_NUM(x) \
234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 ((x)->kactive_blk_num+1) : 0)
237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
240 /* register_prot_hook must be invoked with the po->bind_lock held,
241 * or from a context in which asynchronous accesses to the packet
242 * socket is not possible (packet_create()).
244 static void register_prot_hook(struct sock *sk)
246 struct packet_sock *po = pkt_sk(sk);
249 __fanout_link(sk, po);
251 dev_add_pack(&po->prot_hook);
257 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
258 * held. If the sync parameter is true, we will temporarily drop
259 * the po->bind_lock and do a synchronize_net to make sure no
260 * asynchronous packet processing paths still refer to the elements
261 * of po->prot_hook. If the sync parameter is false, it is the
262 * callers responsibility to take care of this.
264 static void __unregister_prot_hook(struct sock *sk, bool sync)
266 struct packet_sock *po = pkt_sk(sk);
270 __fanout_unlink(sk, po);
272 __dev_remove_pack(&po->prot_hook);
276 spin_unlock(&po->bind_lock);
278 spin_lock(&po->bind_lock);
282 static void unregister_prot_hook(struct sock *sk, bool sync)
284 struct packet_sock *po = pkt_sk(sk);
287 __unregister_prot_hook(sk, sync);
290 static inline __pure struct page *pgv_to_page(void *addr)
292 if (is_vmalloc_addr(addr))
293 return vmalloc_to_page(addr);
294 return virt_to_page(addr);
297 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
299 union tpacket_uhdr h;
302 switch (po->tp_version) {
304 h.h1->tp_status = status;
305 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
308 h.h2->tp_status = status;
309 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
313 WARN(1, "TPACKET version not supported.\n");
320 static int __packet_get_status(struct packet_sock *po, void *frame)
322 union tpacket_uhdr h;
327 switch (po->tp_version) {
329 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
330 return h.h1->tp_status;
332 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
333 return h.h2->tp_status;
336 WARN(1, "TPACKET version not supported.\n");
342 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
345 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
348 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
349 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
350 return TP_STATUS_TS_SYS_HARDWARE;
351 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
352 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
353 return TP_STATUS_TS_RAW_HARDWARE;
356 if (ktime_to_timespec_cond(skb->tstamp, ts))
357 return TP_STATUS_TS_SOFTWARE;
362 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
365 union tpacket_uhdr h;
369 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
373 switch (po->tp_version) {
375 h.h1->tp_sec = ts.tv_sec;
376 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
379 h.h2->tp_sec = ts.tv_sec;
380 h.h2->tp_nsec = ts.tv_nsec;
384 WARN(1, "TPACKET version not supported.\n");
388 /* one flush is safe, as both fields always lie on the same cacheline */
389 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
395 static void *packet_lookup_frame(struct packet_sock *po,
396 struct packet_ring_buffer *rb,
397 unsigned int position,
400 unsigned int pg_vec_pos, frame_offset;
401 union tpacket_uhdr h;
403 pg_vec_pos = position / rb->frames_per_block;
404 frame_offset = position % rb->frames_per_block;
406 h.raw = rb->pg_vec[pg_vec_pos].buffer +
407 (frame_offset * rb->frame_size);
409 if (status != __packet_get_status(po, h.raw))
415 static void *packet_current_frame(struct packet_sock *po,
416 struct packet_ring_buffer *rb,
419 return packet_lookup_frame(po, rb, rb->head, status);
422 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
424 del_timer_sync(&pkc->retire_blk_timer);
427 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
429 struct sk_buff_head *rb_queue)
431 struct tpacket_kbdq_core *pkc;
433 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
435 spin_lock(&rb_queue->lock);
436 pkc->delete_blk_timer = 1;
437 spin_unlock(&rb_queue->lock);
439 prb_del_retire_blk_timer(pkc);
442 static void prb_init_blk_timer(struct packet_sock *po,
443 struct tpacket_kbdq_core *pkc,
444 void (*func) (unsigned long))
446 init_timer(&pkc->retire_blk_timer);
447 pkc->retire_blk_timer.data = (long)po;
448 pkc->retire_blk_timer.function = func;
449 pkc->retire_blk_timer.expires = jiffies;
452 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
454 struct tpacket_kbdq_core *pkc;
459 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
460 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
463 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
464 int blk_size_in_bytes)
466 struct net_device *dev;
467 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
468 struct ethtool_cmd ecmd;
473 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
474 if (unlikely(!dev)) {
476 return DEFAULT_PRB_RETIRE_TOV;
478 err = __ethtool_get_settings(dev, &ecmd);
479 speed = ethtool_cmd_speed(&ecmd);
483 * If the link speed is so slow you don't really
484 * need to worry about perf anyways
486 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
487 return DEFAULT_PRB_RETIRE_TOV;
494 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
506 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
507 union tpacket_req_u *req_u)
509 p1->feature_req_word = req_u->req3.tp_feature_req_word;
512 static void init_prb_bdqc(struct packet_sock *po,
513 struct packet_ring_buffer *rb,
515 union tpacket_req_u *req_u, int tx_ring)
517 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
518 struct tpacket_block_desc *pbd;
520 memset(p1, 0x0, sizeof(*p1));
522 p1->knxt_seq_num = 1;
524 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
525 p1->pkblk_start = pg_vec[0].buffer;
526 p1->kblk_size = req_u->req3.tp_block_size;
527 p1->knum_blocks = req_u->req3.tp_block_nr;
528 p1->hdrlen = po->tp_hdrlen;
529 p1->version = po->tp_version;
530 p1->last_kactive_blk_num = 0;
531 po->stats.stats3.tp_freeze_q_cnt = 0;
532 if (req_u->req3.tp_retire_blk_tov)
533 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
535 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
536 req_u->req3.tp_block_size);
537 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
538 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
540 prb_init_ft_ops(p1, req_u);
541 prb_setup_retire_blk_timer(po, tx_ring);
542 prb_open_block(p1, pbd);
545 /* Do NOT update the last_blk_num first.
546 * Assumes sk_buff_head lock is held.
548 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
550 mod_timer(&pkc->retire_blk_timer,
551 jiffies + pkc->tov_in_jiffies);
552 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
557 * 1) We refresh the timer only when we open a block.
558 * By doing this we don't waste cycles refreshing the timer
559 * on packet-by-packet basis.
561 * With a 1MB block-size, on a 1Gbps line, it will take
562 * i) ~8 ms to fill a block + ii) memcpy etc.
563 * In this cut we are not accounting for the memcpy time.
565 * So, if the user sets the 'tmo' to 10ms then the timer
566 * will never fire while the block is still getting filled
567 * (which is what we want). However, the user could choose
568 * to close a block early and that's fine.
570 * But when the timer does fire, we check whether or not to refresh it.
571 * Since the tmo granularity is in msecs, it is not too expensive
572 * to refresh the timer, lets say every '8' msecs.
573 * Either the user can set the 'tmo' or we can derive it based on
574 * a) line-speed and b) block-size.
575 * prb_calc_retire_blk_tmo() calculates the tmo.
578 static void prb_retire_rx_blk_timer_expired(unsigned long data)
580 struct packet_sock *po = (struct packet_sock *)data;
581 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
583 struct tpacket_block_desc *pbd;
585 spin_lock(&po->sk.sk_receive_queue.lock);
587 frozen = prb_queue_frozen(pkc);
588 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
590 if (unlikely(pkc->delete_blk_timer))
593 /* We only need to plug the race when the block is partially filled.
595 * lock(); increment BLOCK_NUM_PKTS; unlock()
596 * copy_bits() is in progress ...
597 * timer fires on other cpu:
598 * we can't retire the current block because copy_bits
602 if (BLOCK_NUM_PKTS(pbd)) {
603 while (atomic_read(&pkc->blk_fill_in_prog)) {
604 /* Waiting for skb_copy_bits to finish... */
609 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
611 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
612 if (!prb_dispatch_next_block(pkc, po))
617 /* Case 1. Queue was frozen because user-space was
620 if (prb_curr_blk_in_use(pkc, pbd)) {
622 * Ok, user-space is still behind.
623 * So just refresh the timer.
627 /* Case 2. queue was frozen,user-space caught up,
628 * now the link went idle && the timer fired.
629 * We don't have a block to close.So we open this
630 * block and restart the timer.
631 * opening a block thaws the queue,restarts timer
632 * Thawing/timer-refresh is a side effect.
634 prb_open_block(pkc, pbd);
641 _prb_refresh_rx_retire_blk_timer(pkc);
644 spin_unlock(&po->sk.sk_receive_queue.lock);
647 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
648 struct tpacket_block_desc *pbd1, __u32 status)
650 /* Flush everything minus the block header */
652 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
657 /* Skip the block header(we know header WILL fit in 4K) */
660 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
661 for (; start < end; start += PAGE_SIZE)
662 flush_dcache_page(pgv_to_page(start));
667 /* Now update the block status. */
669 BLOCK_STATUS(pbd1) = status;
671 /* Flush the block header */
673 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
675 flush_dcache_page(pgv_to_page(start));
685 * 2) Increment active_blk_num
687 * Note:We DONT refresh the timer on purpose.
688 * Because almost always the next block will be opened.
690 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
691 struct tpacket_block_desc *pbd1,
692 struct packet_sock *po, unsigned int stat)
694 __u32 status = TP_STATUS_USER | stat;
696 struct tpacket3_hdr *last_pkt;
697 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
699 if (po->stats.stats3.tp_drops)
700 status |= TP_STATUS_LOSING;
702 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
703 last_pkt->tp_next_offset = 0;
705 /* Get the ts of the last pkt */
706 if (BLOCK_NUM_PKTS(pbd1)) {
707 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
708 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
710 /* Ok, we tmo'd - so get the current time */
713 h1->ts_last_pkt.ts_sec = ts.tv_sec;
714 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
719 /* Flush the block */
720 prb_flush_block(pkc1, pbd1, status);
722 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
725 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
727 pkc->reset_pending_on_curr_blk = 0;
731 * Side effect of opening a block:
733 * 1) prb_queue is thawed.
734 * 2) retire_blk_timer is refreshed.
737 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
738 struct tpacket_block_desc *pbd1)
741 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
745 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
747 /* We could have just memset this but we will lose the
748 * flexibility of making the priv area sticky
750 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
751 BLOCK_NUM_PKTS(pbd1) = 0;
752 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
754 h1->ts_first_pkt.ts_sec = ts.tv_sec;
755 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
756 pkc1->pkblk_start = (char *)pbd1;
757 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
758 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
759 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
760 pbd1->version = pkc1->version;
761 pkc1->prev = pkc1->nxt_offset;
762 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
763 prb_thaw_queue(pkc1);
764 _prb_refresh_rx_retire_blk_timer(pkc1);
771 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
772 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
778 * Queue freeze logic:
779 * 1) Assume tp_block_nr = 8 blocks.
780 * 2) At time 't0', user opens Rx ring.
781 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
782 * 4) user-space is either sleeping or processing block '0'.
783 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
784 * it will close block-7,loop around and try to fill block '0'.
786 * __packet_lookup_frame_in_block
787 * prb_retire_current_block()
788 * prb_dispatch_next_block()
789 * |->(BLOCK_STATUS == USER) evaluates to true
790 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
791 * 6) Now there are two cases:
792 * 6.1) Link goes idle right after the queue is frozen.
793 * But remember, the last open_block() refreshed the timer.
794 * When this timer expires,it will refresh itself so that we can
795 * re-open block-0 in near future.
796 * 6.2) Link is busy and keeps on receiving packets. This is a simple
797 * case and __packet_lookup_frame_in_block will check if block-0
798 * is free and can now be re-used.
800 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
801 struct packet_sock *po)
803 pkc->reset_pending_on_curr_blk = 1;
804 po->stats.stats3.tp_freeze_q_cnt++;
807 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
810 * If the next block is free then we will dispatch it
811 * and return a good offset.
812 * Else, we will freeze the queue.
813 * So, caller must check the return value.
815 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
816 struct packet_sock *po)
818 struct tpacket_block_desc *pbd;
822 /* 1. Get current block num */
823 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
825 /* 2. If this block is currently in_use then freeze the queue */
826 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
827 prb_freeze_queue(pkc, po);
833 * open this block and return the offset where the first packet
834 * needs to get stored.
836 prb_open_block(pkc, pbd);
837 return (void *)pkc->nxt_offset;
840 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
841 struct packet_sock *po, unsigned int status)
843 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
845 /* retire/close the current block */
846 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
848 * Plug the case where copy_bits() is in progress on
849 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
850 * have space to copy the pkt in the current block and
851 * called prb_retire_current_block()
853 * We don't need to worry about the TMO case because
854 * the timer-handler already handled this case.
856 if (!(status & TP_STATUS_BLK_TMO)) {
857 while (atomic_read(&pkc->blk_fill_in_prog)) {
858 /* Waiting for skb_copy_bits to finish... */
862 prb_close_block(pkc, pbd, po, status);
866 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
871 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
872 struct tpacket_block_desc *pbd)
874 return TP_STATUS_USER & BLOCK_STATUS(pbd);
877 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
879 return pkc->reset_pending_on_curr_blk;
882 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
884 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
885 atomic_dec(&pkc->blk_fill_in_prog);
888 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
889 struct tpacket3_hdr *ppd)
891 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
894 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
895 struct tpacket3_hdr *ppd)
897 ppd->hv1.tp_rxhash = 0;
900 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
901 struct tpacket3_hdr *ppd)
903 if (vlan_tx_tag_present(pkc->skb)) {
904 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
905 ppd->tp_status = TP_STATUS_VLAN_VALID;
907 ppd->hv1.tp_vlan_tci = 0;
908 ppd->tp_status = TP_STATUS_AVAILABLE;
912 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
913 struct tpacket3_hdr *ppd)
915 prb_fill_vlan_info(pkc, ppd);
917 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
918 prb_fill_rxhash(pkc, ppd);
920 prb_clear_rxhash(pkc, ppd);
923 static void prb_fill_curr_block(char *curr,
924 struct tpacket_kbdq_core *pkc,
925 struct tpacket_block_desc *pbd,
928 struct tpacket3_hdr *ppd;
930 ppd = (struct tpacket3_hdr *)curr;
931 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
933 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
934 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
935 BLOCK_NUM_PKTS(pbd) += 1;
936 atomic_inc(&pkc->blk_fill_in_prog);
937 prb_run_all_ft_ops(pkc, ppd);
940 /* Assumes caller has the sk->rx_queue.lock */
941 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
947 struct tpacket_kbdq_core *pkc;
948 struct tpacket_block_desc *pbd;
951 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
952 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
954 /* Queue is frozen when user space is lagging behind */
955 if (prb_queue_frozen(pkc)) {
957 * Check if that last block which caused the queue to freeze,
958 * is still in_use by user-space.
960 if (prb_curr_blk_in_use(pkc, pbd)) {
961 /* Can't record this packet */
965 * Ok, the block was released by user-space.
966 * Now let's open that block.
967 * opening a block also thaws the queue.
968 * Thawing is a side effect.
970 prb_open_block(pkc, pbd);
975 curr = pkc->nxt_offset;
977 end = (char *)pbd + pkc->kblk_size;
979 /* first try the current block */
980 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
981 prb_fill_curr_block(curr, pkc, pbd, len);
985 /* Ok, close the current block */
986 prb_retire_current_block(pkc, po, 0);
988 /* Now, try to dispatch the next block */
989 curr = (char *)prb_dispatch_next_block(pkc, po);
991 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
992 prb_fill_curr_block(curr, pkc, pbd, len);
997 * No free blocks are available.user_space hasn't caught up yet.
998 * Queue was just frozen and now this packet will get dropped.
1003 static void *packet_current_rx_frame(struct packet_sock *po,
1004 struct sk_buff *skb,
1005 int status, unsigned int len)
1008 switch (po->tp_version) {
1011 curr = packet_lookup_frame(po, &po->rx_ring,
1012 po->rx_ring.head, status);
1015 return __packet_lookup_frame_in_block(po, skb, status, len);
1017 WARN(1, "TPACKET version not supported\n");
1023 static void *prb_lookup_block(struct packet_sock *po,
1024 struct packet_ring_buffer *rb,
1028 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1029 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1031 if (status != BLOCK_STATUS(pbd))
1036 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1039 if (rb->prb_bdqc.kactive_blk_num)
1040 prev = rb->prb_bdqc.kactive_blk_num-1;
1042 prev = rb->prb_bdqc.knum_blocks-1;
1046 /* Assumes caller has held the rx_queue.lock */
1047 static void *__prb_previous_block(struct packet_sock *po,
1048 struct packet_ring_buffer *rb,
1051 unsigned int previous = prb_previous_blk_num(rb);
1052 return prb_lookup_block(po, rb, previous, status);
1055 static void *packet_previous_rx_frame(struct packet_sock *po,
1056 struct packet_ring_buffer *rb,
1059 if (po->tp_version <= TPACKET_V2)
1060 return packet_previous_frame(po, rb, status);
1062 return __prb_previous_block(po, rb, status);
1065 static void packet_increment_rx_head(struct packet_sock *po,
1066 struct packet_ring_buffer *rb)
1068 switch (po->tp_version) {
1071 return packet_increment_head(rb);
1074 WARN(1, "TPACKET version not supported.\n");
1080 static void *packet_previous_frame(struct packet_sock *po,
1081 struct packet_ring_buffer *rb,
1084 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1085 return packet_lookup_frame(po, rb, previous, status);
1088 static void packet_increment_head(struct packet_ring_buffer *buff)
1090 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1093 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1095 struct sock *sk = &po->sk;
1098 if (po->prot_hook.func != tpacket_rcv)
1099 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1102 spin_lock(&sk->sk_receive_queue.lock);
1103 if (po->tp_version == TPACKET_V3)
1104 has_room = prb_lookup_block(po, &po->rx_ring,
1105 po->rx_ring.prb_bdqc.kactive_blk_num,
1108 has_room = packet_lookup_frame(po, &po->rx_ring,
1111 spin_unlock(&sk->sk_receive_queue.lock);
1116 static void packet_sock_destruct(struct sock *sk)
1118 skb_queue_purge(&sk->sk_error_queue);
1120 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1121 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1123 if (!sock_flag(sk, SOCK_DEAD)) {
1124 pr_err("Attempt to release alive packet socket: %p\n", sk);
1128 sk_refcnt_debug_dec(sk);
1131 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1133 int x = atomic_read(&f->rr_cur) + 1;
1141 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1142 struct sk_buff *skb,
1145 return (((u64)skb->rxhash) * num) >> 32;
1148 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1149 struct sk_buff *skb,
1154 cur = atomic_read(&f->rr_cur);
1155 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1156 fanout_rr_next(f, num))) != cur)
1161 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1162 struct sk_buff *skb,
1165 return smp_processor_id() % num;
1168 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1169 struct sk_buff *skb,
1170 unsigned int idx, unsigned int skip,
1175 i = j = min_t(int, f->next[idx], num - 1);
1177 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1189 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1191 return f->flags & (flag >> 8);
1194 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1195 struct packet_type *pt, struct net_device *orig_dev)
1197 struct packet_fanout *f = pt->af_packet_priv;
1198 unsigned int num = f->num_members;
1199 struct packet_sock *po;
1202 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1209 case PACKET_FANOUT_HASH:
1211 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1212 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1216 skb_get_rxhash(skb);
1217 idx = fanout_demux_hash(f, skb, num);
1219 case PACKET_FANOUT_LB:
1220 idx = fanout_demux_lb(f, skb, num);
1222 case PACKET_FANOUT_CPU:
1223 idx = fanout_demux_cpu(f, skb, num);
1225 case PACKET_FANOUT_ROLLOVER:
1226 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1230 po = pkt_sk(f->arr[idx]);
1231 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1232 unlikely(!packet_rcv_has_room(po, skb))) {
1233 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1234 po = pkt_sk(f->arr[idx]);
1237 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1240 DEFINE_MUTEX(fanout_mutex);
1241 EXPORT_SYMBOL_GPL(fanout_mutex);
1242 static LIST_HEAD(fanout_list);
1244 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1246 struct packet_fanout *f = po->fanout;
1248 spin_lock(&f->lock);
1249 f->arr[f->num_members] = sk;
1252 spin_unlock(&f->lock);
1255 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1257 struct packet_fanout *f = po->fanout;
1260 spin_lock(&f->lock);
1261 for (i = 0; i < f->num_members; i++) {
1262 if (f->arr[i] == sk)
1265 BUG_ON(i >= f->num_members);
1266 f->arr[i] = f->arr[f->num_members - 1];
1268 spin_unlock(&f->lock);
1271 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1273 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1279 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1281 struct packet_sock *po = pkt_sk(sk);
1282 struct packet_fanout *f, *match;
1283 u8 type = type_flags & 0xff;
1284 u8 flags = type_flags >> 8;
1288 case PACKET_FANOUT_ROLLOVER:
1289 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1291 case PACKET_FANOUT_HASH:
1292 case PACKET_FANOUT_LB:
1293 case PACKET_FANOUT_CPU:
1305 mutex_lock(&fanout_mutex);
1307 list_for_each_entry(f, &fanout_list, list) {
1309 read_pnet(&f->net) == sock_net(sk)) {
1315 if (match && match->flags != flags)
1319 match = kzalloc(sizeof(*match), GFP_KERNEL);
1322 write_pnet(&match->net, sock_net(sk));
1325 match->flags = flags;
1326 atomic_set(&match->rr_cur, 0);
1327 INIT_LIST_HEAD(&match->list);
1328 spin_lock_init(&match->lock);
1329 atomic_set(&match->sk_ref, 0);
1330 match->prot_hook.type = po->prot_hook.type;
1331 match->prot_hook.dev = po->prot_hook.dev;
1332 match->prot_hook.func = packet_rcv_fanout;
1333 match->prot_hook.af_packet_priv = match;
1334 match->prot_hook.id_match = match_fanout_group;
1335 dev_add_pack(&match->prot_hook);
1336 list_add(&match->list, &fanout_list);
1339 if (match->type == type &&
1340 match->prot_hook.type == po->prot_hook.type &&
1341 match->prot_hook.dev == po->prot_hook.dev) {
1343 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1344 __dev_remove_pack(&po->prot_hook);
1346 atomic_inc(&match->sk_ref);
1347 __fanout_link(sk, po);
1352 mutex_unlock(&fanout_mutex);
1356 static void fanout_release(struct sock *sk)
1358 struct packet_sock *po = pkt_sk(sk);
1359 struct packet_fanout *f;
1365 mutex_lock(&fanout_mutex);
1368 if (atomic_dec_and_test(&f->sk_ref)) {
1370 dev_remove_pack(&f->prot_hook);
1373 mutex_unlock(&fanout_mutex);
1376 static const struct proto_ops packet_ops;
1378 static const struct proto_ops packet_ops_spkt;
1380 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1381 struct packet_type *pt, struct net_device *orig_dev)
1384 struct sockaddr_pkt *spkt;
1387 * When we registered the protocol we saved the socket in the data
1388 * field for just this event.
1391 sk = pt->af_packet_priv;
1394 * Yank back the headers [hope the device set this
1395 * right or kerboom...]
1397 * Incoming packets have ll header pulled,
1400 * For outgoing ones skb->data == skb_mac_header(skb)
1401 * so that this procedure is noop.
1404 if (skb->pkt_type == PACKET_LOOPBACK)
1407 if (!net_eq(dev_net(dev), sock_net(sk)))
1410 skb = skb_share_check(skb, GFP_ATOMIC);
1414 /* drop any routing info */
1417 /* drop conntrack reference */
1420 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1422 skb_push(skb, skb->data - skb_mac_header(skb));
1425 * The SOCK_PACKET socket receives _all_ frames.
1428 spkt->spkt_family = dev->type;
1429 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1430 spkt->spkt_protocol = skb->protocol;
1433 * Charge the memory to the socket. This is done specifically
1434 * to prevent sockets using all the memory up.
1437 if (sock_queue_rcv_skb(sk, skb) == 0)
1448 * Output a raw packet to a device layer. This bypasses all the other
1449 * protocol layers and you must therefore supply it with a complete frame
1452 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1453 struct msghdr *msg, size_t len)
1455 struct sock *sk = sock->sk;
1456 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1457 struct sk_buff *skb = NULL;
1458 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[sizeof(saddr->spkt_device) - 1] = 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.
1496 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1497 if (!netif_supports_nofcs(dev)) {
1498 err = -EPROTONOSUPPORT;
1501 extra_len = 4; /* We're doing our own CRC */
1505 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1509 size_t reserved = LL_RESERVED_SPACE(dev);
1510 int tlen = dev->needed_tailroom;
1511 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1514 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1517 /* FIXME: Save some space for broken drivers that write a hard
1518 * header at transmission time by themselves. PPP is the notable
1519 * one here. This should really be fixed at the driver level.
1521 skb_reserve(skb, reserved);
1522 skb_reset_network_header(skb);
1524 /* Try to align data part correctly */
1529 skb_reset_network_header(skb);
1531 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1537 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1538 /* Earlier code assumed this would be a VLAN pkt,
1539 * double-check this now that we have the actual
1542 struct ethhdr *ehdr;
1543 skb_reset_mac_header(skb);
1544 ehdr = eth_hdr(skb);
1545 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1551 skb->protocol = proto;
1553 skb->priority = sk->sk_priority;
1554 skb->mark = sk->sk_mark;
1556 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1558 if (unlikely(extra_len == 4))
1561 skb_probe_transport_header(skb, 0);
1563 dev_queue_xmit(skb);
1574 static unsigned int run_filter(const struct sk_buff *skb,
1575 const struct sock *sk,
1578 struct sk_filter *filter;
1581 filter = rcu_dereference(sk->sk_filter);
1583 res = SK_RUN_FILTER(filter, skb);
1590 * This function makes lazy skb cloning in hope that most of packets
1591 * are discarded by BPF.
1593 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1594 * and skb->cb are mangled. It works because (and until) packets
1595 * falling here are owned by current CPU. Output packets are cloned
1596 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1597 * sequencially, so that if we return skb to original state on exit,
1598 * we will not harm anyone.
1601 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1602 struct packet_type *pt, struct net_device *orig_dev)
1605 struct sockaddr_ll *sll;
1606 struct packet_sock *po;
1607 u8 *skb_head = skb->data;
1608 int skb_len = skb->len;
1609 unsigned int snaplen, res;
1611 if (skb->pkt_type == PACKET_LOOPBACK)
1614 sk = pt->af_packet_priv;
1617 if (!net_eq(dev_net(dev), sock_net(sk)))
1622 if (dev->header_ops) {
1623 /* The device has an explicit notion of ll header,
1624 * exported to higher levels.
1626 * Otherwise, the device hides details of its frame
1627 * structure, so that corresponding packet head is
1628 * never delivered to user.
1630 if (sk->sk_type != SOCK_DGRAM)
1631 skb_push(skb, skb->data - skb_mac_header(skb));
1632 else if (skb->pkt_type == PACKET_OUTGOING) {
1633 /* Special case: outgoing packets have ll header at head */
1634 skb_pull(skb, skb_network_offset(skb));
1640 res = run_filter(skb, sk, snaplen);
1642 goto drop_n_restore;
1646 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1649 if (skb_shared(skb)) {
1650 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1654 if (skb_head != skb->data) {
1655 skb->data = skb_head;
1662 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1665 sll = &PACKET_SKB_CB(skb)->sa.ll;
1666 sll->sll_family = AF_PACKET;
1667 sll->sll_hatype = dev->type;
1668 sll->sll_protocol = skb->protocol;
1669 sll->sll_pkttype = skb->pkt_type;
1670 if (unlikely(po->origdev))
1671 sll->sll_ifindex = orig_dev->ifindex;
1673 sll->sll_ifindex = dev->ifindex;
1675 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1677 PACKET_SKB_CB(skb)->origlen = skb->len;
1679 if (pskb_trim(skb, snaplen))
1682 skb_set_owner_r(skb, sk);
1686 /* drop conntrack reference */
1689 spin_lock(&sk->sk_receive_queue.lock);
1690 po->stats.stats1.tp_packets++;
1691 skb->dropcount = atomic_read(&sk->sk_drops);
1692 __skb_queue_tail(&sk->sk_receive_queue, skb);
1693 spin_unlock(&sk->sk_receive_queue.lock);
1694 sk->sk_data_ready(sk, skb->len);
1698 spin_lock(&sk->sk_receive_queue.lock);
1699 po->stats.stats1.tp_drops++;
1700 atomic_inc(&sk->sk_drops);
1701 spin_unlock(&sk->sk_receive_queue.lock);
1704 if (skb_head != skb->data && skb_shared(skb)) {
1705 skb->data = skb_head;
1713 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1714 struct packet_type *pt, struct net_device *orig_dev)
1717 struct packet_sock *po;
1718 struct sockaddr_ll *sll;
1719 union tpacket_uhdr h;
1720 u8 *skb_head = skb->data;
1721 int skb_len = skb->len;
1722 unsigned int snaplen, res;
1723 unsigned long status = TP_STATUS_USER;
1724 unsigned short macoff, netoff, hdrlen;
1725 struct sk_buff *copy_skb = NULL;
1729 if (skb->pkt_type == PACKET_LOOPBACK)
1732 sk = pt->af_packet_priv;
1735 if (!net_eq(dev_net(dev), sock_net(sk)))
1738 if (dev->header_ops) {
1739 if (sk->sk_type != SOCK_DGRAM)
1740 skb_push(skb, skb->data - skb_mac_header(skb));
1741 else if (skb->pkt_type == PACKET_OUTGOING) {
1742 /* Special case: outgoing packets have ll header at head */
1743 skb_pull(skb, skb_network_offset(skb));
1747 if (skb->ip_summed == CHECKSUM_PARTIAL)
1748 status |= TP_STATUS_CSUMNOTREADY;
1752 res = run_filter(skb, sk, snaplen);
1754 goto drop_n_restore;
1758 if (sk->sk_type == SOCK_DGRAM) {
1759 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1762 unsigned int maclen = skb_network_offset(skb);
1763 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1764 (maclen < 16 ? 16 : maclen)) +
1766 macoff = netoff - maclen;
1768 if (po->tp_version <= TPACKET_V2) {
1769 if (macoff + snaplen > po->rx_ring.frame_size) {
1770 if (po->copy_thresh &&
1771 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1772 if (skb_shared(skb)) {
1773 copy_skb = skb_clone(skb, GFP_ATOMIC);
1775 copy_skb = skb_get(skb);
1776 skb_head = skb->data;
1779 skb_set_owner_r(copy_skb, sk);
1781 snaplen = po->rx_ring.frame_size - macoff;
1782 if ((int)snaplen < 0)
1786 spin_lock(&sk->sk_receive_queue.lock);
1787 h.raw = packet_current_rx_frame(po, skb,
1788 TP_STATUS_KERNEL, (macoff+snaplen));
1791 if (po->tp_version <= TPACKET_V2) {
1792 packet_increment_rx_head(po, &po->rx_ring);
1794 * LOSING will be reported till you read the stats,
1795 * because it's COR - Clear On Read.
1796 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1799 if (po->stats.stats1.tp_drops)
1800 status |= TP_STATUS_LOSING;
1802 po->stats.stats1.tp_packets++;
1804 status |= TP_STATUS_COPY;
1805 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1807 spin_unlock(&sk->sk_receive_queue.lock);
1809 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1811 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1812 getnstimeofday(&ts);
1814 status |= ts_status;
1816 switch (po->tp_version) {
1818 h.h1->tp_len = skb->len;
1819 h.h1->tp_snaplen = snaplen;
1820 h.h1->tp_mac = macoff;
1821 h.h1->tp_net = netoff;
1822 h.h1->tp_sec = ts.tv_sec;
1823 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1824 hdrlen = sizeof(*h.h1);
1827 h.h2->tp_len = skb->len;
1828 h.h2->tp_snaplen = snaplen;
1829 h.h2->tp_mac = macoff;
1830 h.h2->tp_net = netoff;
1831 h.h2->tp_sec = ts.tv_sec;
1832 h.h2->tp_nsec = ts.tv_nsec;
1833 if (vlan_tx_tag_present(skb)) {
1834 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1835 status |= TP_STATUS_VLAN_VALID;
1837 h.h2->tp_vlan_tci = 0;
1839 h.h2->tp_padding = 0;
1840 hdrlen = sizeof(*h.h2);
1843 /* tp_nxt_offset,vlan are already populated above.
1844 * So DONT clear those fields here
1846 h.h3->tp_status |= status;
1847 h.h3->tp_len = skb->len;
1848 h.h3->tp_snaplen = snaplen;
1849 h.h3->tp_mac = macoff;
1850 h.h3->tp_net = netoff;
1851 h.h3->tp_sec = ts.tv_sec;
1852 h.h3->tp_nsec = ts.tv_nsec;
1853 hdrlen = sizeof(*h.h3);
1859 sll = h.raw + TPACKET_ALIGN(hdrlen);
1860 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1861 sll->sll_family = AF_PACKET;
1862 sll->sll_hatype = dev->type;
1863 sll->sll_protocol = skb->protocol;
1864 sll->sll_pkttype = skb->pkt_type;
1865 if (unlikely(po->origdev))
1866 sll->sll_ifindex = orig_dev->ifindex;
1868 sll->sll_ifindex = dev->ifindex;
1871 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1875 if (po->tp_version <= TPACKET_V2) {
1876 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1877 + macoff + snaplen);
1878 for (start = h.raw; start < end; start += PAGE_SIZE)
1879 flush_dcache_page(pgv_to_page(start));
1884 if (po->tp_version <= TPACKET_V2)
1885 __packet_set_status(po, h.raw, status);
1887 prb_clear_blk_fill_status(&po->rx_ring);
1889 sk->sk_data_ready(sk, 0);
1892 if (skb_head != skb->data && skb_shared(skb)) {
1893 skb->data = skb_head;
1901 po->stats.stats1.tp_drops++;
1902 spin_unlock(&sk->sk_receive_queue.lock);
1904 sk->sk_data_ready(sk, 0);
1905 kfree_skb(copy_skb);
1906 goto drop_n_restore;
1909 static void tpacket_destruct_skb(struct sk_buff *skb)
1911 struct packet_sock *po = pkt_sk(skb->sk);
1914 if (likely(po->tx_ring.pg_vec)) {
1917 ph = skb_shinfo(skb)->destructor_arg;
1918 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1919 atomic_dec(&po->tx_ring.pending);
1921 ts = __packet_set_timestamp(po, ph, skb);
1922 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1928 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1929 void *frame, struct net_device *dev, int size_max,
1930 __be16 proto, unsigned char *addr, int hlen)
1932 union tpacket_uhdr ph;
1933 int to_write, offset, len, tp_len, nr_frags, len_max;
1934 struct socket *sock = po->sk.sk_socket;
1941 skb->protocol = proto;
1943 skb->priority = po->sk.sk_priority;
1944 skb->mark = po->sk.sk_mark;
1945 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1946 skb_shinfo(skb)->destructor_arg = ph.raw;
1948 switch (po->tp_version) {
1950 tp_len = ph.h2->tp_len;
1953 tp_len = ph.h1->tp_len;
1956 if (unlikely(tp_len > size_max)) {
1957 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1961 skb_reserve(skb, hlen);
1962 skb_reset_network_header(skb);
1963 skb_probe_transport_header(skb, 0);
1965 if (po->tp_tx_has_off) {
1966 int off_min, off_max, off;
1967 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1968 off_max = po->tx_ring.frame_size - tp_len;
1969 if (sock->type == SOCK_DGRAM) {
1970 switch (po->tp_version) {
1972 off = ph.h2->tp_net;
1975 off = ph.h1->tp_net;
1979 switch (po->tp_version) {
1981 off = ph.h2->tp_mac;
1984 off = ph.h1->tp_mac;
1988 if (unlikely((off < off_min) || (off_max < off)))
1990 data = ph.raw + off;
1992 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1996 if (sock->type == SOCK_DGRAM) {
1997 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1999 if (unlikely(err < 0))
2001 } else if (dev->hard_header_len) {
2002 /* net device doesn't like empty head */
2003 if (unlikely(tp_len <= dev->hard_header_len)) {
2004 pr_err("packet size is too short (%d < %d)\n",
2005 tp_len, dev->hard_header_len);
2009 skb_push(skb, dev->hard_header_len);
2010 err = skb_store_bits(skb, 0, data,
2011 dev->hard_header_len);
2015 data += dev->hard_header_len;
2016 to_write -= dev->hard_header_len;
2019 offset = offset_in_page(data);
2020 len_max = PAGE_SIZE - offset;
2021 len = ((to_write > len_max) ? len_max : to_write);
2023 skb->data_len = to_write;
2024 skb->len += to_write;
2025 skb->truesize += to_write;
2026 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2028 while (likely(to_write)) {
2029 nr_frags = skb_shinfo(skb)->nr_frags;
2031 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2032 pr_err("Packet exceed the number of skb frags(%lu)\n",
2037 page = pgv_to_page(data);
2039 flush_dcache_page(page);
2041 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2044 len_max = PAGE_SIZE;
2045 len = ((to_write > len_max) ? len_max : to_write);
2051 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2053 struct sk_buff *skb;
2054 struct net_device *dev;
2056 bool need_rls_dev = false;
2057 int err, reserve = 0;
2059 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2060 int tp_len, size_max;
2061 unsigned char *addr;
2063 int status = TP_STATUS_AVAILABLE;
2066 mutex_lock(&po->pg_vec_lock);
2068 if (saddr == NULL) {
2069 dev = po->prot_hook.dev;
2074 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2076 if (msg->msg_namelen < (saddr->sll_halen
2077 + offsetof(struct sockaddr_ll,
2080 proto = saddr->sll_protocol;
2081 addr = saddr->sll_addr;
2082 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2083 need_rls_dev = true;
2087 if (unlikely(dev == NULL))
2090 reserve = dev->hard_header_len;
2093 if (unlikely(!(dev->flags & IFF_UP)))
2096 size_max = po->tx_ring.frame_size
2097 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2099 if (size_max > dev->mtu + reserve)
2100 size_max = dev->mtu + reserve;
2103 ph = packet_current_frame(po, &po->tx_ring,
2104 TP_STATUS_SEND_REQUEST);
2106 if (unlikely(ph == NULL)) {
2111 status = TP_STATUS_SEND_REQUEST;
2112 hlen = LL_RESERVED_SPACE(dev);
2113 tlen = dev->needed_tailroom;
2114 skb = sock_alloc_send_skb(&po->sk,
2115 hlen + tlen + sizeof(struct sockaddr_ll),
2118 if (unlikely(skb == NULL))
2121 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2124 if (unlikely(tp_len < 0)) {
2126 __packet_set_status(po, ph,
2127 TP_STATUS_AVAILABLE);
2128 packet_increment_head(&po->tx_ring);
2132 status = TP_STATUS_WRONG_FORMAT;
2138 skb->destructor = tpacket_destruct_skb;
2139 __packet_set_status(po, ph, TP_STATUS_SENDING);
2140 atomic_inc(&po->tx_ring.pending);
2142 status = TP_STATUS_SEND_REQUEST;
2143 err = dev_queue_xmit(skb);
2144 if (unlikely(err > 0)) {
2145 err = net_xmit_errno(err);
2146 if (err && __packet_get_status(po, ph) ==
2147 TP_STATUS_AVAILABLE) {
2148 /* skb was destructed already */
2153 * skb was dropped but not destructed yet;
2154 * let's treat it like congestion or err < 0
2158 packet_increment_head(&po->tx_ring);
2160 } while (likely((ph != NULL) ||
2161 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2162 (atomic_read(&po->tx_ring.pending))))
2169 __packet_set_status(po, ph, status);
2175 mutex_unlock(&po->pg_vec_lock);
2179 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2180 size_t reserve, size_t len,
2181 size_t linear, int noblock,
2184 struct sk_buff *skb;
2186 /* Under a page? Don't bother with paged skb. */
2187 if (prepad + len < PAGE_SIZE || !linear)
2190 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2195 skb_reserve(skb, reserve);
2196 skb_put(skb, linear);
2197 skb->data_len = len - linear;
2198 skb->len += len - linear;
2203 static int packet_snd(struct socket *sock,
2204 struct msghdr *msg, size_t len)
2206 struct sock *sk = sock->sk;
2207 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2208 struct sk_buff *skb;
2209 struct net_device *dev;
2211 bool need_rls_dev = false;
2212 unsigned char *addr;
2213 int err, reserve = 0;
2214 struct virtio_net_hdr vnet_hdr = { 0 };
2217 struct packet_sock *po = pkt_sk(sk);
2218 unsigned short gso_type = 0;
2223 * Get and verify the address.
2226 if (saddr == NULL) {
2227 dev = po->prot_hook.dev;
2232 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2234 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2236 proto = saddr->sll_protocol;
2237 addr = saddr->sll_addr;
2238 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2239 need_rls_dev = true;
2245 if (sock->type == SOCK_RAW)
2246 reserve = dev->hard_header_len;
2249 if (!(dev->flags & IFF_UP))
2252 if (po->has_vnet_hdr) {
2253 vnet_hdr_len = sizeof(vnet_hdr);
2256 if (len < vnet_hdr_len)
2259 len -= vnet_hdr_len;
2261 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2266 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2267 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2269 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2270 vnet_hdr.csum_offset + 2;
2273 if (vnet_hdr.hdr_len > len)
2276 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2277 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2278 case VIRTIO_NET_HDR_GSO_TCPV4:
2279 gso_type = SKB_GSO_TCPV4;
2281 case VIRTIO_NET_HDR_GSO_TCPV6:
2282 gso_type = SKB_GSO_TCPV6;
2284 case VIRTIO_NET_HDR_GSO_UDP:
2285 gso_type = SKB_GSO_UDP;
2291 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2292 gso_type |= SKB_GSO_TCP_ECN;
2294 if (vnet_hdr.gso_size == 0)
2300 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2301 if (!netif_supports_nofcs(dev)) {
2302 err = -EPROTONOSUPPORT;
2305 extra_len = 4; /* We're doing our own CRC */
2309 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2313 hlen = LL_RESERVED_SPACE(dev);
2314 tlen = dev->needed_tailroom;
2315 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2316 msg->msg_flags & MSG_DONTWAIT, &err);
2320 skb_set_network_header(skb, reserve);
2323 if (sock->type == SOCK_DGRAM &&
2324 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2327 /* Returns -EFAULT on error */
2328 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2332 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2334 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2335 /* Earlier code assumed this would be a VLAN pkt,
2336 * double-check this now that we have the actual
2339 struct ethhdr *ehdr;
2340 skb_reset_mac_header(skb);
2341 ehdr = eth_hdr(skb);
2342 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2348 skb->protocol = proto;
2350 skb->priority = sk->sk_priority;
2351 skb->mark = sk->sk_mark;
2353 if (po->has_vnet_hdr) {
2354 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2355 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2356 vnet_hdr.csum_offset)) {
2362 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2363 skb_shinfo(skb)->gso_type = gso_type;
2365 /* Header must be checked, and gso_segs computed. */
2366 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2367 skb_shinfo(skb)->gso_segs = 0;
2369 len += vnet_hdr_len;
2372 skb_probe_transport_header(skb, reserve);
2374 if (unlikely(extra_len == 4))
2381 err = dev_queue_xmit(skb);
2382 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2393 if (dev && need_rls_dev)
2399 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2400 struct msghdr *msg, size_t len)
2402 struct sock *sk = sock->sk;
2403 struct packet_sock *po = pkt_sk(sk);
2404 if (po->tx_ring.pg_vec)
2405 return tpacket_snd(po, msg);
2407 return packet_snd(sock, msg, len);
2411 * Close a PACKET socket. This is fairly simple. We immediately go
2412 * to 'closed' state and remove our protocol entry in the device list.
2415 static int packet_release(struct socket *sock)
2417 struct sock *sk = sock->sk;
2418 struct packet_sock *po;
2420 union tpacket_req_u req_u;
2428 mutex_lock(&net->packet.sklist_lock);
2429 sk_del_node_init_rcu(sk);
2430 mutex_unlock(&net->packet.sklist_lock);
2433 sock_prot_inuse_add(net, sk->sk_prot, -1);
2436 spin_lock(&po->bind_lock);
2437 unregister_prot_hook(sk, false);
2438 if (po->prot_hook.dev) {
2439 dev_put(po->prot_hook.dev);
2440 po->prot_hook.dev = NULL;
2442 spin_unlock(&po->bind_lock);
2444 packet_flush_mclist(sk);
2446 if (po->rx_ring.pg_vec) {
2447 memset(&req_u, 0, sizeof(req_u));
2448 packet_set_ring(sk, &req_u, 1, 0);
2451 if (po->tx_ring.pg_vec) {
2452 memset(&req_u, 0, sizeof(req_u));
2453 packet_set_ring(sk, &req_u, 1, 1);
2460 * Now the socket is dead. No more input will appear.
2467 skb_queue_purge(&sk->sk_receive_queue);
2468 sk_refcnt_debug_release(sk);
2475 * Attach a packet hook.
2478 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2480 struct packet_sock *po = pkt_sk(sk);
2491 spin_lock(&po->bind_lock);
2492 unregister_prot_hook(sk, true);
2494 po->prot_hook.type = protocol;
2495 if (po->prot_hook.dev)
2496 dev_put(po->prot_hook.dev);
2497 po->prot_hook.dev = dev;
2499 po->ifindex = dev ? dev->ifindex : 0;
2504 if (!dev || (dev->flags & IFF_UP)) {
2505 register_prot_hook(sk);
2507 sk->sk_err = ENETDOWN;
2508 if (!sock_flag(sk, SOCK_DEAD))
2509 sk->sk_error_report(sk);
2513 spin_unlock(&po->bind_lock);
2519 * Bind a packet socket to a device
2522 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2525 struct sock *sk = sock->sk;
2527 struct net_device *dev;
2534 if (addr_len != sizeof(struct sockaddr))
2536 strlcpy(name, uaddr->sa_data, sizeof(name));
2538 dev = dev_get_by_name(sock_net(sk), name);
2540 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2544 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2546 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2547 struct sock *sk = sock->sk;
2548 struct net_device *dev = NULL;
2556 if (addr_len < sizeof(struct sockaddr_ll))
2558 if (sll->sll_family != AF_PACKET)
2561 if (sll->sll_ifindex) {
2563 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2567 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2573 static struct proto packet_proto = {
2575 .owner = THIS_MODULE,
2576 .obj_size = sizeof(struct packet_sock),
2580 * Create a packet of type SOCK_PACKET.
2583 static int packet_create(struct net *net, struct socket *sock, int protocol,
2587 struct packet_sock *po;
2588 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2591 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2593 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2594 sock->type != SOCK_PACKET)
2595 return -ESOCKTNOSUPPORT;
2597 sock->state = SS_UNCONNECTED;
2600 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2604 sock->ops = &packet_ops;
2605 if (sock->type == SOCK_PACKET)
2606 sock->ops = &packet_ops_spkt;
2608 sock_init_data(sock, sk);
2611 sk->sk_family = PF_PACKET;
2614 sk->sk_destruct = packet_sock_destruct;
2615 sk_refcnt_debug_inc(sk);
2618 * Attach a protocol block
2621 spin_lock_init(&po->bind_lock);
2622 mutex_init(&po->pg_vec_lock);
2623 po->prot_hook.func = packet_rcv;
2625 if (sock->type == SOCK_PACKET)
2626 po->prot_hook.func = packet_rcv_spkt;
2628 po->prot_hook.af_packet_priv = sk;
2631 po->prot_hook.type = proto;
2632 register_prot_hook(sk);
2635 mutex_lock(&net->packet.sklist_lock);
2636 sk_add_node_rcu(sk, &net->packet.sklist);
2637 mutex_unlock(&net->packet.sklist_lock);
2640 sock_prot_inuse_add(net, &packet_proto, 1);
2648 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2650 struct sock_exterr_skb *serr;
2651 struct sk_buff *skb, *skb2;
2655 skb = skb_dequeue(&sk->sk_error_queue);
2661 msg->msg_flags |= MSG_TRUNC;
2664 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2668 sock_recv_timestamp(msg, sk, skb);
2670 serr = SKB_EXT_ERR(skb);
2671 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2672 sizeof(serr->ee), &serr->ee);
2674 msg->msg_flags |= MSG_ERRQUEUE;
2677 /* Reset and regenerate socket error */
2678 spin_lock_bh(&sk->sk_error_queue.lock);
2680 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2681 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2682 spin_unlock_bh(&sk->sk_error_queue.lock);
2683 sk->sk_error_report(sk);
2685 spin_unlock_bh(&sk->sk_error_queue.lock);
2694 * Pull a packet from our receive queue and hand it to the user.
2695 * If necessary we block.
2698 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2699 struct msghdr *msg, size_t len, int flags)
2701 struct sock *sk = sock->sk;
2702 struct sk_buff *skb;
2704 struct sockaddr_ll *sll;
2705 int vnet_hdr_len = 0;
2708 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2712 /* What error should we return now? EUNATTACH? */
2713 if (pkt_sk(sk)->ifindex < 0)
2717 if (flags & MSG_ERRQUEUE) {
2718 err = packet_recv_error(sk, msg, len);
2723 * Call the generic datagram receiver. This handles all sorts
2724 * of horrible races and re-entrancy so we can forget about it
2725 * in the protocol layers.
2727 * Now it will return ENETDOWN, if device have just gone down,
2728 * but then it will block.
2731 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2734 * An error occurred so return it. Because skb_recv_datagram()
2735 * handles the blocking we don't see and worry about blocking
2742 if (pkt_sk(sk)->has_vnet_hdr) {
2743 struct virtio_net_hdr vnet_hdr = { 0 };
2746 vnet_hdr_len = sizeof(vnet_hdr);
2747 if (len < vnet_hdr_len)
2750 len -= vnet_hdr_len;
2752 if (skb_is_gso(skb)) {
2753 struct skb_shared_info *sinfo = skb_shinfo(skb);
2755 /* This is a hint as to how much should be linear. */
2756 vnet_hdr.hdr_len = skb_headlen(skb);
2757 vnet_hdr.gso_size = sinfo->gso_size;
2758 if (sinfo->gso_type & SKB_GSO_TCPV4)
2759 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2760 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2761 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2762 else if (sinfo->gso_type & SKB_GSO_UDP)
2763 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2764 else if (sinfo->gso_type & SKB_GSO_FCOE)
2768 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2769 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2771 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2773 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2774 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2775 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2776 vnet_hdr.csum_offset = skb->csum_offset;
2777 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2778 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2779 } /* else everything is zero */
2781 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2788 * If the address length field is there to be filled in, we fill
2792 sll = &PACKET_SKB_CB(skb)->sa.ll;
2793 if (sock->type == SOCK_PACKET)
2794 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2796 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2799 * You lose any data beyond the buffer you gave. If it worries a
2800 * user program they can ask the device for its MTU anyway.
2806 msg->msg_flags |= MSG_TRUNC;
2809 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2813 sock_recv_ts_and_drops(msg, sk, skb);
2816 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2819 if (pkt_sk(sk)->auxdata) {
2820 struct tpacket_auxdata aux;
2822 aux.tp_status = TP_STATUS_USER;
2823 if (skb->ip_summed == CHECKSUM_PARTIAL)
2824 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2825 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2826 aux.tp_snaplen = skb->len;
2828 aux.tp_net = skb_network_offset(skb);
2829 if (vlan_tx_tag_present(skb)) {
2830 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2831 aux.tp_status |= TP_STATUS_VLAN_VALID;
2833 aux.tp_vlan_tci = 0;
2836 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2840 * Free or return the buffer as appropriate. Again this
2841 * hides all the races and re-entrancy issues from us.
2843 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2846 skb_free_datagram(sk, skb);
2851 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2852 int *uaddr_len, int peer)
2854 struct net_device *dev;
2855 struct sock *sk = sock->sk;
2860 uaddr->sa_family = AF_PACKET;
2862 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2864 strncpy(uaddr->sa_data, dev->name, 14);
2866 memset(uaddr->sa_data, 0, 14);
2868 *uaddr_len = sizeof(*uaddr);
2873 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2874 int *uaddr_len, int peer)
2876 struct net_device *dev;
2877 struct sock *sk = sock->sk;
2878 struct packet_sock *po = pkt_sk(sk);
2879 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2884 sll->sll_family = AF_PACKET;
2885 sll->sll_ifindex = po->ifindex;
2886 sll->sll_protocol = po->num;
2887 sll->sll_pkttype = 0;
2889 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2891 sll->sll_hatype = dev->type;
2892 sll->sll_halen = dev->addr_len;
2893 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2895 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2899 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2904 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2908 case PACKET_MR_MULTICAST:
2909 if (i->alen != dev->addr_len)
2912 return dev_mc_add(dev, i->addr);
2914 return dev_mc_del(dev, i->addr);
2916 case PACKET_MR_PROMISC:
2917 return dev_set_promiscuity(dev, what);
2919 case PACKET_MR_ALLMULTI:
2920 return dev_set_allmulti(dev, what);
2922 case PACKET_MR_UNICAST:
2923 if (i->alen != dev->addr_len)
2926 return dev_uc_add(dev, i->addr);
2928 return dev_uc_del(dev, i->addr);
2936 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2938 for ( ; i; i = i->next) {
2939 if (i->ifindex == dev->ifindex)
2940 packet_dev_mc(dev, i, what);
2944 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2946 struct packet_sock *po = pkt_sk(sk);
2947 struct packet_mclist *ml, *i;
2948 struct net_device *dev;
2954 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2959 if (mreq->mr_alen > dev->addr_len)
2963 i = kmalloc(sizeof(*i), GFP_KERNEL);
2968 for (ml = po->mclist; ml; ml = ml->next) {
2969 if (ml->ifindex == mreq->mr_ifindex &&
2970 ml->type == mreq->mr_type &&
2971 ml->alen == mreq->mr_alen &&
2972 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2974 /* Free the new element ... */
2980 i->type = mreq->mr_type;
2981 i->ifindex = mreq->mr_ifindex;
2982 i->alen = mreq->mr_alen;
2983 memcpy(i->addr, mreq->mr_address, i->alen);
2985 i->next = po->mclist;
2987 err = packet_dev_mc(dev, i, 1);
2989 po->mclist = i->next;
2998 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3000 struct packet_mclist *ml, **mlp;
3004 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3005 if (ml->ifindex == mreq->mr_ifindex &&
3006 ml->type == mreq->mr_type &&
3007 ml->alen == mreq->mr_alen &&
3008 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3009 if (--ml->count == 0) {
3010 struct net_device *dev;
3012 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3014 packet_dev_mc(dev, ml, -1);
3022 return -EADDRNOTAVAIL;
3025 static void packet_flush_mclist(struct sock *sk)
3027 struct packet_sock *po = pkt_sk(sk);
3028 struct packet_mclist *ml;
3034 while ((ml = po->mclist) != NULL) {
3035 struct net_device *dev;
3037 po->mclist = ml->next;
3038 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3040 packet_dev_mc(dev, ml, -1);
3047 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3049 struct sock *sk = sock->sk;
3050 struct packet_sock *po = pkt_sk(sk);
3053 if (level != SOL_PACKET)
3054 return -ENOPROTOOPT;
3057 case PACKET_ADD_MEMBERSHIP:
3058 case PACKET_DROP_MEMBERSHIP:
3060 struct packet_mreq_max mreq;
3062 memset(&mreq, 0, sizeof(mreq));
3063 if (len < sizeof(struct packet_mreq))
3065 if (len > sizeof(mreq))
3067 if (copy_from_user(&mreq, optval, len))
3069 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3071 if (optname == PACKET_ADD_MEMBERSHIP)
3072 ret = packet_mc_add(sk, &mreq);
3074 ret = packet_mc_drop(sk, &mreq);
3078 case PACKET_RX_RING:
3079 case PACKET_TX_RING:
3081 union tpacket_req_u req_u;
3084 switch (po->tp_version) {
3087 len = sizeof(req_u.req);
3091 len = sizeof(req_u.req3);
3096 if (pkt_sk(sk)->has_vnet_hdr)
3098 if (copy_from_user(&req_u.req, optval, len))
3100 return packet_set_ring(sk, &req_u, 0,
3101 optname == PACKET_TX_RING);
3103 case PACKET_COPY_THRESH:
3107 if (optlen != sizeof(val))
3109 if (copy_from_user(&val, optval, sizeof(val)))
3112 pkt_sk(sk)->copy_thresh = val;
3115 case PACKET_VERSION:
3119 if (optlen != sizeof(val))
3121 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3123 if (copy_from_user(&val, optval, sizeof(val)))
3129 po->tp_version = val;
3135 case PACKET_RESERVE:
3139 if (optlen != sizeof(val))
3141 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3143 if (copy_from_user(&val, optval, sizeof(val)))
3145 po->tp_reserve = val;
3152 if (optlen != sizeof(val))
3154 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3156 if (copy_from_user(&val, optval, sizeof(val)))
3158 po->tp_loss = !!val;
3161 case PACKET_AUXDATA:
3165 if (optlen < sizeof(val))
3167 if (copy_from_user(&val, optval, sizeof(val)))
3170 po->auxdata = !!val;
3173 case PACKET_ORIGDEV:
3177 if (optlen < sizeof(val))
3179 if (copy_from_user(&val, optval, sizeof(val)))
3182 po->origdev = !!val;
3185 case PACKET_VNET_HDR:
3189 if (sock->type != SOCK_RAW)
3191 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3193 if (optlen < sizeof(val))
3195 if (copy_from_user(&val, optval, sizeof(val)))
3198 po->has_vnet_hdr = !!val;
3201 case PACKET_TIMESTAMP:
3205 if (optlen != sizeof(val))
3207 if (copy_from_user(&val, optval, sizeof(val)))
3210 po->tp_tstamp = val;
3217 if (optlen != sizeof(val))
3219 if (copy_from_user(&val, optval, sizeof(val)))
3222 return fanout_add(sk, val & 0xffff, val >> 16);
3224 case PACKET_TX_HAS_OFF:
3228 if (optlen != sizeof(val))
3230 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3232 if (copy_from_user(&val, optval, sizeof(val)))
3234 po->tp_tx_has_off = !!val;
3238 return -ENOPROTOOPT;
3242 static int packet_getsockopt(struct socket *sock, int level, int optname,
3243 char __user *optval, int __user *optlen)
3246 int val, lv = sizeof(val);
3247 struct sock *sk = sock->sk;
3248 struct packet_sock *po = pkt_sk(sk);
3250 union tpacket_stats_u st;
3252 if (level != SOL_PACKET)
3253 return -ENOPROTOOPT;
3255 if (get_user(len, optlen))
3262 case PACKET_STATISTICS:
3263 spin_lock_bh(&sk->sk_receive_queue.lock);
3264 memcpy(&st, &po->stats, sizeof(st));
3265 memset(&po->stats, 0, sizeof(po->stats));
3266 spin_unlock_bh(&sk->sk_receive_queue.lock);
3268 if (po->tp_version == TPACKET_V3) {
3269 lv = sizeof(struct tpacket_stats_v3);
3272 lv = sizeof(struct tpacket_stats);
3277 case PACKET_AUXDATA:
3280 case PACKET_ORIGDEV:
3283 case PACKET_VNET_HDR:
3284 val = po->has_vnet_hdr;
3286 case PACKET_VERSION:
3287 val = po->tp_version;
3290 if (len > sizeof(int))
3292 if (copy_from_user(&val, optval, len))
3296 val = sizeof(struct tpacket_hdr);
3299 val = sizeof(struct tpacket2_hdr);
3302 val = sizeof(struct tpacket3_hdr);
3308 case PACKET_RESERVE:
3309 val = po->tp_reserve;
3314 case PACKET_TIMESTAMP:
3315 val = po->tp_tstamp;
3319 ((u32)po->fanout->id |
3320 ((u32)po->fanout->type << 16) |
3321 ((u32)po->fanout->flags << 24)) :
3324 case PACKET_TX_HAS_OFF:
3325 val = po->tp_tx_has_off;
3328 return -ENOPROTOOPT;
3333 if (put_user(len, optlen))
3335 if (copy_to_user(optval, data, len))
3341 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3344 struct net_device *dev = data;
3345 struct net *net = dev_net(dev);
3348 sk_for_each_rcu(sk, &net->packet.sklist) {
3349 struct packet_sock *po = pkt_sk(sk);
3352 case NETDEV_UNREGISTER:
3354 packet_dev_mclist(dev, po->mclist, -1);
3358 if (dev->ifindex == po->ifindex) {
3359 spin_lock(&po->bind_lock);
3361 __unregister_prot_hook(sk, false);
3362 sk->sk_err = ENETDOWN;
3363 if (!sock_flag(sk, SOCK_DEAD))
3364 sk->sk_error_report(sk);
3366 if (msg == NETDEV_UNREGISTER) {
3368 if (po->prot_hook.dev)
3369 dev_put(po->prot_hook.dev);
3370 po->prot_hook.dev = NULL;
3372 spin_unlock(&po->bind_lock);
3376 if (dev->ifindex == po->ifindex) {
3377 spin_lock(&po->bind_lock);
3379 register_prot_hook(sk);
3380 spin_unlock(&po->bind_lock);
3390 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3393 struct sock *sk = sock->sk;
3398 int amount = sk_wmem_alloc_get(sk);
3400 return put_user(amount, (int __user *)arg);
3404 struct sk_buff *skb;
3407 spin_lock_bh(&sk->sk_receive_queue.lock);
3408 skb = skb_peek(&sk->sk_receive_queue);
3411 spin_unlock_bh(&sk->sk_receive_queue.lock);
3412 return put_user(amount, (int __user *)arg);
3415 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3417 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3427 case SIOCGIFBRDADDR:
3428 case SIOCSIFBRDADDR:
3429 case SIOCGIFNETMASK:
3430 case SIOCSIFNETMASK:
3431 case SIOCGIFDSTADDR:
3432 case SIOCSIFDSTADDR:
3434 return inet_dgram_ops.ioctl(sock, cmd, arg);
3438 return -ENOIOCTLCMD;
3443 static unsigned int packet_poll(struct file *file, struct socket *sock,
3446 struct sock *sk = sock->sk;
3447 struct packet_sock *po = pkt_sk(sk);
3448 unsigned int mask = datagram_poll(file, sock, wait);
3450 spin_lock_bh(&sk->sk_receive_queue.lock);
3451 if (po->rx_ring.pg_vec) {
3452 if (!packet_previous_rx_frame(po, &po->rx_ring,
3454 mask |= POLLIN | POLLRDNORM;
3456 spin_unlock_bh(&sk->sk_receive_queue.lock);
3457 spin_lock_bh(&sk->sk_write_queue.lock);
3458 if (po->tx_ring.pg_vec) {
3459 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3460 mask |= POLLOUT | POLLWRNORM;
3462 spin_unlock_bh(&sk->sk_write_queue.lock);
3467 /* Dirty? Well, I still did not learn better way to account
3471 static void packet_mm_open(struct vm_area_struct *vma)
3473 struct file *file = vma->vm_file;
3474 struct socket *sock = file->private_data;
3475 struct sock *sk = sock->sk;
3478 atomic_inc(&pkt_sk(sk)->mapped);
3481 static void packet_mm_close(struct vm_area_struct *vma)
3483 struct file *file = vma->vm_file;
3484 struct socket *sock = file->private_data;
3485 struct sock *sk = sock->sk;
3488 atomic_dec(&pkt_sk(sk)->mapped);
3491 static const struct vm_operations_struct packet_mmap_ops = {
3492 .open = packet_mm_open,
3493 .close = packet_mm_close,
3496 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3501 for (i = 0; i < len; i++) {
3502 if (likely(pg_vec[i].buffer)) {
3503 if (is_vmalloc_addr(pg_vec[i].buffer))
3504 vfree(pg_vec[i].buffer);
3506 free_pages((unsigned long)pg_vec[i].buffer,
3508 pg_vec[i].buffer = NULL;
3514 static char *alloc_one_pg_vec_page(unsigned long order)
3516 char *buffer = NULL;
3517 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3518 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3520 buffer = (char *) __get_free_pages(gfp_flags, order);
3526 * __get_free_pages failed, fall back to vmalloc
3528 buffer = vzalloc((1 << order) * PAGE_SIZE);
3534 * vmalloc failed, lets dig into swap here
3536 gfp_flags &= ~__GFP_NORETRY;
3537 buffer = (char *)__get_free_pages(gfp_flags, order);
3542 * complete and utter failure
3547 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3549 unsigned int block_nr = req->tp_block_nr;
3553 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3554 if (unlikely(!pg_vec))
3557 for (i = 0; i < block_nr; i++) {
3558 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3559 if (unlikely(!pg_vec[i].buffer))
3560 goto out_free_pgvec;
3567 free_pg_vec(pg_vec, order, block_nr);
3572 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3573 int closing, int tx_ring)
3575 struct pgv *pg_vec = NULL;
3576 struct packet_sock *po = pkt_sk(sk);
3577 int was_running, order = 0;
3578 struct packet_ring_buffer *rb;
3579 struct sk_buff_head *rb_queue;
3582 /* Added to avoid minimal code churn */
3583 struct tpacket_req *req = &req_u->req;
3585 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3586 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3587 WARN(1, "Tx-ring is not supported.\n");
3591 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3592 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3596 if (atomic_read(&po->mapped))
3598 if (atomic_read(&rb->pending))
3602 if (req->tp_block_nr) {
3603 /* Sanity tests and some calculations */
3605 if (unlikely(rb->pg_vec))
3608 switch (po->tp_version) {
3610 po->tp_hdrlen = TPACKET_HDRLEN;
3613 po->tp_hdrlen = TPACKET2_HDRLEN;
3616 po->tp_hdrlen = TPACKET3_HDRLEN;
3621 if (unlikely((int)req->tp_block_size <= 0))
3623 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3625 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3628 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3631 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3632 if (unlikely(rb->frames_per_block <= 0))
3634 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3639 order = get_order(req->tp_block_size);
3640 pg_vec = alloc_pg_vec(req, order);
3641 if (unlikely(!pg_vec))
3643 switch (po->tp_version) {
3645 /* Transmit path is not supported. We checked
3646 * it above but just being paranoid
3649 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3658 if (unlikely(req->tp_frame_nr))
3664 /* Detach socket from network */
3665 spin_lock(&po->bind_lock);
3666 was_running = po->running;
3670 __unregister_prot_hook(sk, false);
3672 spin_unlock(&po->bind_lock);
3677 mutex_lock(&po->pg_vec_lock);
3678 if (closing || atomic_read(&po->mapped) == 0) {
3680 spin_lock_bh(&rb_queue->lock);
3681 swap(rb->pg_vec, pg_vec);
3682 rb->frame_max = (req->tp_frame_nr - 1);
3684 rb->frame_size = req->tp_frame_size;
3685 spin_unlock_bh(&rb_queue->lock);
3687 swap(rb->pg_vec_order, order);
3688 swap(rb->pg_vec_len, req->tp_block_nr);
3690 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3691 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3692 tpacket_rcv : packet_rcv;
3693 skb_queue_purge(rb_queue);
3694 if (atomic_read(&po->mapped))
3695 pr_err("packet_mmap: vma is busy: %d\n",
3696 atomic_read(&po->mapped));
3698 mutex_unlock(&po->pg_vec_lock);
3700 spin_lock(&po->bind_lock);
3703 register_prot_hook(sk);
3705 spin_unlock(&po->bind_lock);
3706 if (closing && (po->tp_version > TPACKET_V2)) {
3707 /* Because we don't support block-based V3 on tx-ring */
3709 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3714 free_pg_vec(pg_vec, order, req->tp_block_nr);
3719 static int packet_mmap(struct file *file, struct socket *sock,
3720 struct vm_area_struct *vma)
3722 struct sock *sk = sock->sk;
3723 struct packet_sock *po = pkt_sk(sk);
3724 unsigned long size, expected_size;
3725 struct packet_ring_buffer *rb;
3726 unsigned long start;
3733 mutex_lock(&po->pg_vec_lock);
3736 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3738 expected_size += rb->pg_vec_len
3744 if (expected_size == 0)
3747 size = vma->vm_end - vma->vm_start;
3748 if (size != expected_size)
3751 start = vma->vm_start;
3752 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3753 if (rb->pg_vec == NULL)
3756 for (i = 0; i < rb->pg_vec_len; i++) {
3758 void *kaddr = rb->pg_vec[i].buffer;
3761 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3762 page = pgv_to_page(kaddr);
3763 err = vm_insert_page(vma, start, page);
3772 atomic_inc(&po->mapped);
3773 vma->vm_ops = &packet_mmap_ops;
3777 mutex_unlock(&po->pg_vec_lock);
3781 static const struct proto_ops packet_ops_spkt = {
3782 .family = PF_PACKET,
3783 .owner = THIS_MODULE,
3784 .release = packet_release,
3785 .bind = packet_bind_spkt,
3786 .connect = sock_no_connect,
3787 .socketpair = sock_no_socketpair,
3788 .accept = sock_no_accept,
3789 .getname = packet_getname_spkt,
3790 .poll = datagram_poll,
3791 .ioctl = packet_ioctl,
3792 .listen = sock_no_listen,
3793 .shutdown = sock_no_shutdown,
3794 .setsockopt = sock_no_setsockopt,
3795 .getsockopt = sock_no_getsockopt,
3796 .sendmsg = packet_sendmsg_spkt,
3797 .recvmsg = packet_recvmsg,
3798 .mmap = sock_no_mmap,
3799 .sendpage = sock_no_sendpage,
3802 static const struct proto_ops packet_ops = {
3803 .family = PF_PACKET,
3804 .owner = THIS_MODULE,
3805 .release = packet_release,
3806 .bind = packet_bind,
3807 .connect = sock_no_connect,
3808 .socketpair = sock_no_socketpair,
3809 .accept = sock_no_accept,
3810 .getname = packet_getname,
3811 .poll = packet_poll,
3812 .ioctl = packet_ioctl,
3813 .listen = sock_no_listen,
3814 .shutdown = sock_no_shutdown,
3815 .setsockopt = packet_setsockopt,
3816 .getsockopt = packet_getsockopt,
3817 .sendmsg = packet_sendmsg,
3818 .recvmsg = packet_recvmsg,
3819 .mmap = packet_mmap,
3820 .sendpage = sock_no_sendpage,
3823 static const struct net_proto_family packet_family_ops = {
3824 .family = PF_PACKET,
3825 .create = packet_create,
3826 .owner = THIS_MODULE,
3829 static struct notifier_block packet_netdev_notifier = {
3830 .notifier_call = packet_notifier,
3833 #ifdef CONFIG_PROC_FS
3835 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3838 struct net *net = seq_file_net(seq);
3841 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3844 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3846 struct net *net = seq_file_net(seq);
3847 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3850 static void packet_seq_stop(struct seq_file *seq, void *v)
3856 static int packet_seq_show(struct seq_file *seq, void *v)
3858 if (v == SEQ_START_TOKEN)
3859 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3861 struct sock *s = sk_entry(v);
3862 const struct packet_sock *po = pkt_sk(s);
3865 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3867 atomic_read(&s->sk_refcnt),
3872 atomic_read(&s->sk_rmem_alloc),
3873 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3880 static const struct seq_operations packet_seq_ops = {
3881 .start = packet_seq_start,
3882 .next = packet_seq_next,
3883 .stop = packet_seq_stop,
3884 .show = packet_seq_show,
3887 static int packet_seq_open(struct inode *inode, struct file *file)
3889 return seq_open_net(inode, file, &packet_seq_ops,
3890 sizeof(struct seq_net_private));
3893 static const struct file_operations packet_seq_fops = {
3894 .owner = THIS_MODULE,
3895 .open = packet_seq_open,
3897 .llseek = seq_lseek,
3898 .release = seq_release_net,
3903 static int __net_init packet_net_init(struct net *net)
3905 mutex_init(&net->packet.sklist_lock);
3906 INIT_HLIST_HEAD(&net->packet.sklist);
3908 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3914 static void __net_exit packet_net_exit(struct net *net)
3916 remove_proc_entry("packet", net->proc_net);
3919 static struct pernet_operations packet_net_ops = {
3920 .init = packet_net_init,
3921 .exit = packet_net_exit,
3925 static void __exit packet_exit(void)
3927 unregister_netdevice_notifier(&packet_netdev_notifier);
3928 unregister_pernet_subsys(&packet_net_ops);
3929 sock_unregister(PF_PACKET);
3930 proto_unregister(&packet_proto);
3933 static int __init packet_init(void)
3935 int rc = proto_register(&packet_proto, 0);
3940 sock_register(&packet_family_ops);
3941 register_pernet_subsys(&packet_net_ops);
3942 register_netdevice_notifier(&packet_netdev_notifier);
3947 module_init(packet_init);
3948 module_exit(packet_exit);
3949 MODULE_LICENSE("GPL");
3950 MODULE_ALIAS_NETPROTO(PF_PACKET);