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 void *packet_lookup_frame(struct packet_sock *po,
343 struct packet_ring_buffer *rb,
344 unsigned int position,
347 unsigned int pg_vec_pos, frame_offset;
348 union tpacket_uhdr h;
350 pg_vec_pos = position / rb->frames_per_block;
351 frame_offset = position % rb->frames_per_block;
353 h.raw = rb->pg_vec[pg_vec_pos].buffer +
354 (frame_offset * rb->frame_size);
356 if (status != __packet_get_status(po, h.raw))
362 static void *packet_current_frame(struct packet_sock *po,
363 struct packet_ring_buffer *rb,
366 return packet_lookup_frame(po, rb, rb->head, status);
369 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
371 del_timer_sync(&pkc->retire_blk_timer);
374 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
376 struct sk_buff_head *rb_queue)
378 struct tpacket_kbdq_core *pkc;
380 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
382 spin_lock(&rb_queue->lock);
383 pkc->delete_blk_timer = 1;
384 spin_unlock(&rb_queue->lock);
386 prb_del_retire_blk_timer(pkc);
389 static void prb_init_blk_timer(struct packet_sock *po,
390 struct tpacket_kbdq_core *pkc,
391 void (*func) (unsigned long))
393 init_timer(&pkc->retire_blk_timer);
394 pkc->retire_blk_timer.data = (long)po;
395 pkc->retire_blk_timer.function = func;
396 pkc->retire_blk_timer.expires = jiffies;
399 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
401 struct tpacket_kbdq_core *pkc;
406 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
407 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
410 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
411 int blk_size_in_bytes)
413 struct net_device *dev;
414 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
415 struct ethtool_cmd ecmd;
420 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
421 if (unlikely(!dev)) {
423 return DEFAULT_PRB_RETIRE_TOV;
425 err = __ethtool_get_settings(dev, &ecmd);
426 speed = ethtool_cmd_speed(&ecmd);
430 * If the link speed is so slow you don't really
431 * need to worry about perf anyways
433 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
434 return DEFAULT_PRB_RETIRE_TOV;
441 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
453 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
454 union tpacket_req_u *req_u)
456 p1->feature_req_word = req_u->req3.tp_feature_req_word;
459 static void init_prb_bdqc(struct packet_sock *po,
460 struct packet_ring_buffer *rb,
462 union tpacket_req_u *req_u, int tx_ring)
464 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
465 struct tpacket_block_desc *pbd;
467 memset(p1, 0x0, sizeof(*p1));
469 p1->knxt_seq_num = 1;
471 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
472 p1->pkblk_start = pg_vec[0].buffer;
473 p1->kblk_size = req_u->req3.tp_block_size;
474 p1->knum_blocks = req_u->req3.tp_block_nr;
475 p1->hdrlen = po->tp_hdrlen;
476 p1->version = po->tp_version;
477 p1->last_kactive_blk_num = 0;
478 po->stats_u.stats3.tp_freeze_q_cnt = 0;
479 if (req_u->req3.tp_retire_blk_tov)
480 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
482 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
483 req_u->req3.tp_block_size);
484 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
485 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
487 prb_init_ft_ops(p1, req_u);
488 prb_setup_retire_blk_timer(po, tx_ring);
489 prb_open_block(p1, pbd);
492 /* Do NOT update the last_blk_num first.
493 * Assumes sk_buff_head lock is held.
495 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
497 mod_timer(&pkc->retire_blk_timer,
498 jiffies + pkc->tov_in_jiffies);
499 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
504 * 1) We refresh the timer only when we open a block.
505 * By doing this we don't waste cycles refreshing the timer
506 * on packet-by-packet basis.
508 * With a 1MB block-size, on a 1Gbps line, it will take
509 * i) ~8 ms to fill a block + ii) memcpy etc.
510 * In this cut we are not accounting for the memcpy time.
512 * So, if the user sets the 'tmo' to 10ms then the timer
513 * will never fire while the block is still getting filled
514 * (which is what we want). However, the user could choose
515 * to close a block early and that's fine.
517 * But when the timer does fire, we check whether or not to refresh it.
518 * Since the tmo granularity is in msecs, it is not too expensive
519 * to refresh the timer, lets say every '8' msecs.
520 * Either the user can set the 'tmo' or we can derive it based on
521 * a) line-speed and b) block-size.
522 * prb_calc_retire_blk_tmo() calculates the tmo.
525 static void prb_retire_rx_blk_timer_expired(unsigned long data)
527 struct packet_sock *po = (struct packet_sock *)data;
528 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
530 struct tpacket_block_desc *pbd;
532 spin_lock(&po->sk.sk_receive_queue.lock);
534 frozen = prb_queue_frozen(pkc);
535 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
537 if (unlikely(pkc->delete_blk_timer))
540 /* We only need to plug the race when the block is partially filled.
542 * lock(); increment BLOCK_NUM_PKTS; unlock()
543 * copy_bits() is in progress ...
544 * timer fires on other cpu:
545 * we can't retire the current block because copy_bits
549 if (BLOCK_NUM_PKTS(pbd)) {
550 while (atomic_read(&pkc->blk_fill_in_prog)) {
551 /* Waiting for skb_copy_bits to finish... */
556 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
558 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
559 if (!prb_dispatch_next_block(pkc, po))
564 /* Case 1. Queue was frozen because user-space was
567 if (prb_curr_blk_in_use(pkc, pbd)) {
569 * Ok, user-space is still behind.
570 * So just refresh the timer.
574 /* Case 2. queue was frozen,user-space caught up,
575 * now the link went idle && the timer fired.
576 * We don't have a block to close.So we open this
577 * block and restart the timer.
578 * opening a block thaws the queue,restarts timer
579 * Thawing/timer-refresh is a side effect.
581 prb_open_block(pkc, pbd);
588 _prb_refresh_rx_retire_blk_timer(pkc);
591 spin_unlock(&po->sk.sk_receive_queue.lock);
594 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
595 struct tpacket_block_desc *pbd1, __u32 status)
597 /* Flush everything minus the block header */
599 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
604 /* Skip the block header(we know header WILL fit in 4K) */
607 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
608 for (; start < end; start += PAGE_SIZE)
609 flush_dcache_page(pgv_to_page(start));
614 /* Now update the block status. */
616 BLOCK_STATUS(pbd1) = status;
618 /* Flush the block header */
620 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
622 flush_dcache_page(pgv_to_page(start));
632 * 2) Increment active_blk_num
634 * Note:We DONT refresh the timer on purpose.
635 * Because almost always the next block will be opened.
637 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
638 struct tpacket_block_desc *pbd1,
639 struct packet_sock *po, unsigned int stat)
641 __u32 status = TP_STATUS_USER | stat;
643 struct tpacket3_hdr *last_pkt;
644 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
646 if (po->stats.tp_drops)
647 status |= TP_STATUS_LOSING;
649 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
650 last_pkt->tp_next_offset = 0;
652 /* Get the ts of the last pkt */
653 if (BLOCK_NUM_PKTS(pbd1)) {
654 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
655 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
657 /* Ok, we tmo'd - so get the current time */
660 h1->ts_last_pkt.ts_sec = ts.tv_sec;
661 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
666 /* Flush the block */
667 prb_flush_block(pkc1, pbd1, status);
669 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
672 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
674 pkc->reset_pending_on_curr_blk = 0;
678 * Side effect of opening a block:
680 * 1) prb_queue is thawed.
681 * 2) retire_blk_timer is refreshed.
684 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
685 struct tpacket_block_desc *pbd1)
688 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
692 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
694 /* We could have just memset this but we will lose the
695 * flexibility of making the priv area sticky
697 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
698 BLOCK_NUM_PKTS(pbd1) = 0;
699 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
701 h1->ts_first_pkt.ts_sec = ts.tv_sec;
702 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
703 pkc1->pkblk_start = (char *)pbd1;
704 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
705 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
706 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
707 pbd1->version = pkc1->version;
708 pkc1->prev = pkc1->nxt_offset;
709 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
710 prb_thaw_queue(pkc1);
711 _prb_refresh_rx_retire_blk_timer(pkc1);
718 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
719 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
725 * Queue freeze logic:
726 * 1) Assume tp_block_nr = 8 blocks.
727 * 2) At time 't0', user opens Rx ring.
728 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
729 * 4) user-space is either sleeping or processing block '0'.
730 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
731 * it will close block-7,loop around and try to fill block '0'.
733 * __packet_lookup_frame_in_block
734 * prb_retire_current_block()
735 * prb_dispatch_next_block()
736 * |->(BLOCK_STATUS == USER) evaluates to true
737 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
738 * 6) Now there are two cases:
739 * 6.1) Link goes idle right after the queue is frozen.
740 * But remember, the last open_block() refreshed the timer.
741 * When this timer expires,it will refresh itself so that we can
742 * re-open block-0 in near future.
743 * 6.2) Link is busy and keeps on receiving packets. This is a simple
744 * case and __packet_lookup_frame_in_block will check if block-0
745 * is free and can now be re-used.
747 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
748 struct packet_sock *po)
750 pkc->reset_pending_on_curr_blk = 1;
751 po->stats_u.stats3.tp_freeze_q_cnt++;
754 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
757 * If the next block is free then we will dispatch it
758 * and return a good offset.
759 * Else, we will freeze the queue.
760 * So, caller must check the return value.
762 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
763 struct packet_sock *po)
765 struct tpacket_block_desc *pbd;
769 /* 1. Get current block num */
770 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
772 /* 2. If this block is currently in_use then freeze the queue */
773 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
774 prb_freeze_queue(pkc, po);
780 * open this block and return the offset where the first packet
781 * needs to get stored.
783 prb_open_block(pkc, pbd);
784 return (void *)pkc->nxt_offset;
787 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
788 struct packet_sock *po, unsigned int status)
790 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
792 /* retire/close the current block */
793 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
795 * Plug the case where copy_bits() is in progress on
796 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
797 * have space to copy the pkt in the current block and
798 * called prb_retire_current_block()
800 * We don't need to worry about the TMO case because
801 * the timer-handler already handled this case.
803 if (!(status & TP_STATUS_BLK_TMO)) {
804 while (atomic_read(&pkc->blk_fill_in_prog)) {
805 /* Waiting for skb_copy_bits to finish... */
809 prb_close_block(pkc, pbd, po, status);
813 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
818 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
819 struct tpacket_block_desc *pbd)
821 return TP_STATUS_USER & BLOCK_STATUS(pbd);
824 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
826 return pkc->reset_pending_on_curr_blk;
829 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
831 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
832 atomic_dec(&pkc->blk_fill_in_prog);
835 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
836 struct tpacket3_hdr *ppd)
838 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
841 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
842 struct tpacket3_hdr *ppd)
844 ppd->hv1.tp_rxhash = 0;
847 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
848 struct tpacket3_hdr *ppd)
850 if (vlan_tx_tag_present(pkc->skb)) {
851 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
852 ppd->tp_status = TP_STATUS_VLAN_VALID;
854 ppd->hv1.tp_vlan_tci = 0;
855 ppd->tp_status = TP_STATUS_AVAILABLE;
859 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
860 struct tpacket3_hdr *ppd)
862 prb_fill_vlan_info(pkc, ppd);
864 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
865 prb_fill_rxhash(pkc, ppd);
867 prb_clear_rxhash(pkc, ppd);
870 static void prb_fill_curr_block(char *curr,
871 struct tpacket_kbdq_core *pkc,
872 struct tpacket_block_desc *pbd,
875 struct tpacket3_hdr *ppd;
877 ppd = (struct tpacket3_hdr *)curr;
878 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
880 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
881 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
882 BLOCK_NUM_PKTS(pbd) += 1;
883 atomic_inc(&pkc->blk_fill_in_prog);
884 prb_run_all_ft_ops(pkc, ppd);
887 /* Assumes caller has the sk->rx_queue.lock */
888 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
894 struct tpacket_kbdq_core *pkc;
895 struct tpacket_block_desc *pbd;
898 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
899 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
901 /* Queue is frozen when user space is lagging behind */
902 if (prb_queue_frozen(pkc)) {
904 * Check if that last block which caused the queue to freeze,
905 * is still in_use by user-space.
907 if (prb_curr_blk_in_use(pkc, pbd)) {
908 /* Can't record this packet */
912 * Ok, the block was released by user-space.
913 * Now let's open that block.
914 * opening a block also thaws the queue.
915 * Thawing is a side effect.
917 prb_open_block(pkc, pbd);
922 curr = pkc->nxt_offset;
924 end = (char *)pbd + pkc->kblk_size;
926 /* first try the current block */
927 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
928 prb_fill_curr_block(curr, pkc, pbd, len);
932 /* Ok, close the current block */
933 prb_retire_current_block(pkc, po, 0);
935 /* Now, try to dispatch the next block */
936 curr = (char *)prb_dispatch_next_block(pkc, po);
938 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
939 prb_fill_curr_block(curr, pkc, pbd, len);
944 * No free blocks are available.user_space hasn't caught up yet.
945 * Queue was just frozen and now this packet will get dropped.
950 static void *packet_current_rx_frame(struct packet_sock *po,
952 int status, unsigned int len)
955 switch (po->tp_version) {
958 curr = packet_lookup_frame(po, &po->rx_ring,
959 po->rx_ring.head, status);
962 return __packet_lookup_frame_in_block(po, skb, status, len);
964 WARN(1, "TPACKET version not supported\n");
970 static void *prb_lookup_block(struct packet_sock *po,
971 struct packet_ring_buffer *rb,
975 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
976 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
978 if (status != BLOCK_STATUS(pbd))
983 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
986 if (rb->prb_bdqc.kactive_blk_num)
987 prev = rb->prb_bdqc.kactive_blk_num-1;
989 prev = rb->prb_bdqc.knum_blocks-1;
993 /* Assumes caller has held the rx_queue.lock */
994 static void *__prb_previous_block(struct packet_sock *po,
995 struct packet_ring_buffer *rb,
998 unsigned int previous = prb_previous_blk_num(rb);
999 return prb_lookup_block(po, rb, previous, status);
1002 static void *packet_previous_rx_frame(struct packet_sock *po,
1003 struct packet_ring_buffer *rb,
1006 if (po->tp_version <= TPACKET_V2)
1007 return packet_previous_frame(po, rb, status);
1009 return __prb_previous_block(po, rb, status);
1012 static void packet_increment_rx_head(struct packet_sock *po,
1013 struct packet_ring_buffer *rb)
1015 switch (po->tp_version) {
1018 return packet_increment_head(rb);
1021 WARN(1, "TPACKET version not supported.\n");
1027 static void *packet_previous_frame(struct packet_sock *po,
1028 struct packet_ring_buffer *rb,
1031 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1032 return packet_lookup_frame(po, rb, previous, status);
1035 static void packet_increment_head(struct packet_ring_buffer *buff)
1037 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1040 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1042 struct sock *sk = &po->sk;
1045 if (po->prot_hook.func != tpacket_rcv)
1046 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1049 spin_lock(&sk->sk_receive_queue.lock);
1050 if (po->tp_version == TPACKET_V3)
1051 has_room = prb_lookup_block(po, &po->rx_ring,
1052 po->rx_ring.prb_bdqc.kactive_blk_num,
1055 has_room = packet_lookup_frame(po, &po->rx_ring,
1058 spin_unlock(&sk->sk_receive_queue.lock);
1063 static void packet_sock_destruct(struct sock *sk)
1065 skb_queue_purge(&sk->sk_error_queue);
1067 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1068 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1070 if (!sock_flag(sk, SOCK_DEAD)) {
1071 pr_err("Attempt to release alive packet socket: %p\n", sk);
1075 sk_refcnt_debug_dec(sk);
1078 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1080 int x = atomic_read(&f->rr_cur) + 1;
1088 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1089 struct sk_buff *skb,
1092 return (((u64)skb->rxhash) * num) >> 32;
1095 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1096 struct sk_buff *skb,
1101 cur = atomic_read(&f->rr_cur);
1102 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1103 fanout_rr_next(f, num))) != cur)
1108 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1109 struct sk_buff *skb,
1112 return smp_processor_id() % num;
1115 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1116 struct sk_buff *skb,
1117 unsigned int idx, unsigned int skip,
1122 i = j = min_t(int, f->next[idx], num - 1);
1124 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1136 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1138 return f->flags & (flag >> 8);
1141 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1142 struct packet_type *pt, struct net_device *orig_dev)
1144 struct packet_fanout *f = pt->af_packet_priv;
1145 unsigned int num = f->num_members;
1146 struct packet_sock *po;
1149 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1156 case PACKET_FANOUT_HASH:
1158 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1159 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1163 skb_get_rxhash(skb);
1164 idx = fanout_demux_hash(f, skb, num);
1166 case PACKET_FANOUT_LB:
1167 idx = fanout_demux_lb(f, skb, num);
1169 case PACKET_FANOUT_CPU:
1170 idx = fanout_demux_cpu(f, skb, num);
1172 case PACKET_FANOUT_ROLLOVER:
1173 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1177 po = pkt_sk(f->arr[idx]);
1178 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1179 unlikely(!packet_rcv_has_room(po, skb))) {
1180 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1181 po = pkt_sk(f->arr[idx]);
1184 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1187 DEFINE_MUTEX(fanout_mutex);
1188 EXPORT_SYMBOL_GPL(fanout_mutex);
1189 static LIST_HEAD(fanout_list);
1191 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1193 struct packet_fanout *f = po->fanout;
1195 spin_lock(&f->lock);
1196 f->arr[f->num_members] = sk;
1199 spin_unlock(&f->lock);
1202 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1204 struct packet_fanout *f = po->fanout;
1207 spin_lock(&f->lock);
1208 for (i = 0; i < f->num_members; i++) {
1209 if (f->arr[i] == sk)
1212 BUG_ON(i >= f->num_members);
1213 f->arr[i] = f->arr[f->num_members - 1];
1215 spin_unlock(&f->lock);
1218 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1220 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1226 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1228 struct packet_sock *po = pkt_sk(sk);
1229 struct packet_fanout *f, *match;
1230 u8 type = type_flags & 0xff;
1231 u8 flags = type_flags >> 8;
1235 case PACKET_FANOUT_ROLLOVER:
1236 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1238 case PACKET_FANOUT_HASH:
1239 case PACKET_FANOUT_LB:
1240 case PACKET_FANOUT_CPU:
1252 mutex_lock(&fanout_mutex);
1254 list_for_each_entry(f, &fanout_list, list) {
1256 read_pnet(&f->net) == sock_net(sk)) {
1262 if (match && match->flags != flags)
1266 match = kzalloc(sizeof(*match), GFP_KERNEL);
1269 write_pnet(&match->net, sock_net(sk));
1272 match->flags = flags;
1273 atomic_set(&match->rr_cur, 0);
1274 INIT_LIST_HEAD(&match->list);
1275 spin_lock_init(&match->lock);
1276 atomic_set(&match->sk_ref, 0);
1277 match->prot_hook.type = po->prot_hook.type;
1278 match->prot_hook.dev = po->prot_hook.dev;
1279 match->prot_hook.func = packet_rcv_fanout;
1280 match->prot_hook.af_packet_priv = match;
1281 match->prot_hook.id_match = match_fanout_group;
1282 dev_add_pack(&match->prot_hook);
1283 list_add(&match->list, &fanout_list);
1286 if (match->type == type &&
1287 match->prot_hook.type == po->prot_hook.type &&
1288 match->prot_hook.dev == po->prot_hook.dev) {
1290 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1291 __dev_remove_pack(&po->prot_hook);
1293 atomic_inc(&match->sk_ref);
1294 __fanout_link(sk, po);
1299 mutex_unlock(&fanout_mutex);
1303 static void fanout_release(struct sock *sk)
1305 struct packet_sock *po = pkt_sk(sk);
1306 struct packet_fanout *f;
1312 mutex_lock(&fanout_mutex);
1315 if (atomic_dec_and_test(&f->sk_ref)) {
1317 dev_remove_pack(&f->prot_hook);
1320 mutex_unlock(&fanout_mutex);
1323 static const struct proto_ops packet_ops;
1325 static const struct proto_ops packet_ops_spkt;
1327 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1328 struct packet_type *pt, struct net_device *orig_dev)
1331 struct sockaddr_pkt *spkt;
1334 * When we registered the protocol we saved the socket in the data
1335 * field for just this event.
1338 sk = pt->af_packet_priv;
1341 * Yank back the headers [hope the device set this
1342 * right or kerboom...]
1344 * Incoming packets have ll header pulled,
1347 * For outgoing ones skb->data == skb_mac_header(skb)
1348 * so that this procedure is noop.
1351 if (skb->pkt_type == PACKET_LOOPBACK)
1354 if (!net_eq(dev_net(dev), sock_net(sk)))
1357 skb = skb_share_check(skb, GFP_ATOMIC);
1361 /* drop any routing info */
1364 /* drop conntrack reference */
1367 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1369 skb_push(skb, skb->data - skb_mac_header(skb));
1372 * The SOCK_PACKET socket receives _all_ frames.
1375 spkt->spkt_family = dev->type;
1376 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1377 spkt->spkt_protocol = skb->protocol;
1380 * Charge the memory to the socket. This is done specifically
1381 * to prevent sockets using all the memory up.
1384 if (sock_queue_rcv_skb(sk, skb) == 0)
1395 * Output a raw packet to a device layer. This bypasses all the other
1396 * protocol layers and you must therefore supply it with a complete frame
1399 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1400 struct msghdr *msg, size_t len)
1402 struct sock *sk = sock->sk;
1403 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1404 struct sk_buff *skb = NULL;
1405 struct net_device *dev;
1411 * Get and verify the address.
1415 if (msg->msg_namelen < sizeof(struct sockaddr))
1417 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1418 proto = saddr->spkt_protocol;
1420 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1423 * Find the device first to size check it
1426 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1429 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1435 if (!(dev->flags & IFF_UP))
1439 * You may not queue a frame bigger than the mtu. This is the lowest level
1440 * raw protocol and you must do your own fragmentation at this level.
1443 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1444 if (!netif_supports_nofcs(dev)) {
1445 err = -EPROTONOSUPPORT;
1448 extra_len = 4; /* We're doing our own CRC */
1452 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1456 size_t reserved = LL_RESERVED_SPACE(dev);
1457 int tlen = dev->needed_tailroom;
1458 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1461 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1464 /* FIXME: Save some space for broken drivers that write a hard
1465 * header at transmission time by themselves. PPP is the notable
1466 * one here. This should really be fixed at the driver level.
1468 skb_reserve(skb, reserved);
1469 skb_reset_network_header(skb);
1471 /* Try to align data part correctly */
1476 skb_reset_network_header(skb);
1478 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1484 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1485 /* Earlier code assumed this would be a VLAN pkt,
1486 * double-check this now that we have the actual
1489 struct ethhdr *ehdr;
1490 skb_reset_mac_header(skb);
1491 ehdr = eth_hdr(skb);
1492 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1498 skb->protocol = proto;
1500 skb->priority = sk->sk_priority;
1501 skb->mark = sk->sk_mark;
1503 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1505 if (unlikely(extra_len == 4))
1508 skb_probe_transport_header(skb, 0);
1510 dev_queue_xmit(skb);
1521 static unsigned int run_filter(const struct sk_buff *skb,
1522 const struct sock *sk,
1525 struct sk_filter *filter;
1528 filter = rcu_dereference(sk->sk_filter);
1530 res = SK_RUN_FILTER(filter, skb);
1537 * This function makes lazy skb cloning in hope that most of packets
1538 * are discarded by BPF.
1540 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1541 * and skb->cb are mangled. It works because (and until) packets
1542 * falling here are owned by current CPU. Output packets are cloned
1543 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1544 * sequencially, so that if we return skb to original state on exit,
1545 * we will not harm anyone.
1548 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1549 struct packet_type *pt, struct net_device *orig_dev)
1552 struct sockaddr_ll *sll;
1553 struct packet_sock *po;
1554 u8 *skb_head = skb->data;
1555 int skb_len = skb->len;
1556 unsigned int snaplen, res;
1558 if (skb->pkt_type == PACKET_LOOPBACK)
1561 sk = pt->af_packet_priv;
1564 if (!net_eq(dev_net(dev), sock_net(sk)))
1569 if (dev->header_ops) {
1570 /* The device has an explicit notion of ll header,
1571 * exported to higher levels.
1573 * Otherwise, the device hides details of its frame
1574 * structure, so that corresponding packet head is
1575 * never delivered to user.
1577 if (sk->sk_type != SOCK_DGRAM)
1578 skb_push(skb, skb->data - skb_mac_header(skb));
1579 else if (skb->pkt_type == PACKET_OUTGOING) {
1580 /* Special case: outgoing packets have ll header at head */
1581 skb_pull(skb, skb_network_offset(skb));
1587 res = run_filter(skb, sk, snaplen);
1589 goto drop_n_restore;
1593 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1596 if (skb_shared(skb)) {
1597 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1601 if (skb_head != skb->data) {
1602 skb->data = skb_head;
1609 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1612 sll = &PACKET_SKB_CB(skb)->sa.ll;
1613 sll->sll_family = AF_PACKET;
1614 sll->sll_hatype = dev->type;
1615 sll->sll_protocol = skb->protocol;
1616 sll->sll_pkttype = skb->pkt_type;
1617 if (unlikely(po->origdev))
1618 sll->sll_ifindex = orig_dev->ifindex;
1620 sll->sll_ifindex = dev->ifindex;
1622 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1624 PACKET_SKB_CB(skb)->origlen = skb->len;
1626 if (pskb_trim(skb, snaplen))
1629 skb_set_owner_r(skb, sk);
1633 /* drop conntrack reference */
1636 spin_lock(&sk->sk_receive_queue.lock);
1637 po->stats.tp_packets++;
1638 skb->dropcount = atomic_read(&sk->sk_drops);
1639 __skb_queue_tail(&sk->sk_receive_queue, skb);
1640 spin_unlock(&sk->sk_receive_queue.lock);
1641 sk->sk_data_ready(sk, skb->len);
1645 spin_lock(&sk->sk_receive_queue.lock);
1646 po->stats.tp_drops++;
1647 atomic_inc(&sk->sk_drops);
1648 spin_unlock(&sk->sk_receive_queue.lock);
1651 if (skb_head != skb->data && skb_shared(skb)) {
1652 skb->data = skb_head;
1660 static void tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
1663 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1666 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
1667 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
1669 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
1670 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
1674 if (ktime_to_timespec_cond(skb->tstamp, ts))
1680 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1681 struct packet_type *pt, struct net_device *orig_dev)
1684 struct packet_sock *po;
1685 struct sockaddr_ll *sll;
1686 union tpacket_uhdr h;
1687 u8 *skb_head = skb->data;
1688 int skb_len = skb->len;
1689 unsigned int snaplen, res;
1690 unsigned long status = TP_STATUS_USER;
1691 unsigned short macoff, netoff, hdrlen;
1692 struct sk_buff *copy_skb = NULL;
1695 if (skb->pkt_type == PACKET_LOOPBACK)
1698 sk = pt->af_packet_priv;
1701 if (!net_eq(dev_net(dev), sock_net(sk)))
1704 if (dev->header_ops) {
1705 if (sk->sk_type != SOCK_DGRAM)
1706 skb_push(skb, skb->data - skb_mac_header(skb));
1707 else if (skb->pkt_type == PACKET_OUTGOING) {
1708 /* Special case: outgoing packets have ll header at head */
1709 skb_pull(skb, skb_network_offset(skb));
1713 if (skb->ip_summed == CHECKSUM_PARTIAL)
1714 status |= TP_STATUS_CSUMNOTREADY;
1718 res = run_filter(skb, sk, snaplen);
1720 goto drop_n_restore;
1724 if (sk->sk_type == SOCK_DGRAM) {
1725 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1728 unsigned int maclen = skb_network_offset(skb);
1729 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1730 (maclen < 16 ? 16 : maclen)) +
1732 macoff = netoff - maclen;
1734 if (po->tp_version <= TPACKET_V2) {
1735 if (macoff + snaplen > po->rx_ring.frame_size) {
1736 if (po->copy_thresh &&
1737 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1738 if (skb_shared(skb)) {
1739 copy_skb = skb_clone(skb, GFP_ATOMIC);
1741 copy_skb = skb_get(skb);
1742 skb_head = skb->data;
1745 skb_set_owner_r(copy_skb, sk);
1747 snaplen = po->rx_ring.frame_size - macoff;
1748 if ((int)snaplen < 0)
1752 spin_lock(&sk->sk_receive_queue.lock);
1753 h.raw = packet_current_rx_frame(po, skb,
1754 TP_STATUS_KERNEL, (macoff+snaplen));
1757 if (po->tp_version <= TPACKET_V2) {
1758 packet_increment_rx_head(po, &po->rx_ring);
1760 * LOSING will be reported till you read the stats,
1761 * because it's COR - Clear On Read.
1762 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1765 if (po->stats.tp_drops)
1766 status |= TP_STATUS_LOSING;
1768 po->stats.tp_packets++;
1770 status |= TP_STATUS_COPY;
1771 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1773 spin_unlock(&sk->sk_receive_queue.lock);
1775 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1776 tpacket_get_timestamp(skb, &ts, po->tp_tstamp);
1778 switch (po->tp_version) {
1780 h.h1->tp_len = skb->len;
1781 h.h1->tp_snaplen = snaplen;
1782 h.h1->tp_mac = macoff;
1783 h.h1->tp_net = netoff;
1784 h.h1->tp_sec = ts.tv_sec;
1785 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1786 hdrlen = sizeof(*h.h1);
1789 h.h2->tp_len = skb->len;
1790 h.h2->tp_snaplen = snaplen;
1791 h.h2->tp_mac = macoff;
1792 h.h2->tp_net = netoff;
1793 h.h2->tp_sec = ts.tv_sec;
1794 h.h2->tp_nsec = ts.tv_nsec;
1795 if (vlan_tx_tag_present(skb)) {
1796 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1797 status |= TP_STATUS_VLAN_VALID;
1799 h.h2->tp_vlan_tci = 0;
1801 h.h2->tp_padding = 0;
1802 hdrlen = sizeof(*h.h2);
1805 /* tp_nxt_offset,vlan are already populated above.
1806 * So DONT clear those fields here
1808 h.h3->tp_status |= status;
1809 h.h3->tp_len = skb->len;
1810 h.h3->tp_snaplen = snaplen;
1811 h.h3->tp_mac = macoff;
1812 h.h3->tp_net = netoff;
1813 h.h3->tp_sec = ts.tv_sec;
1814 h.h3->tp_nsec = ts.tv_nsec;
1815 hdrlen = sizeof(*h.h3);
1821 sll = h.raw + TPACKET_ALIGN(hdrlen);
1822 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1823 sll->sll_family = AF_PACKET;
1824 sll->sll_hatype = dev->type;
1825 sll->sll_protocol = skb->protocol;
1826 sll->sll_pkttype = skb->pkt_type;
1827 if (unlikely(po->origdev))
1828 sll->sll_ifindex = orig_dev->ifindex;
1830 sll->sll_ifindex = dev->ifindex;
1833 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1837 if (po->tp_version <= TPACKET_V2) {
1838 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1839 + macoff + snaplen);
1840 for (start = h.raw; start < end; start += PAGE_SIZE)
1841 flush_dcache_page(pgv_to_page(start));
1846 if (po->tp_version <= TPACKET_V2)
1847 __packet_set_status(po, h.raw, status);
1849 prb_clear_blk_fill_status(&po->rx_ring);
1851 sk->sk_data_ready(sk, 0);
1854 if (skb_head != skb->data && skb_shared(skb)) {
1855 skb->data = skb_head;
1863 po->stats.tp_drops++;
1864 spin_unlock(&sk->sk_receive_queue.lock);
1866 sk->sk_data_ready(sk, 0);
1867 kfree_skb(copy_skb);
1868 goto drop_n_restore;
1871 static void tpacket_destruct_skb(struct sk_buff *skb)
1873 struct packet_sock *po = pkt_sk(skb->sk);
1876 if (likely(po->tx_ring.pg_vec)) {
1877 ph = skb_shinfo(skb)->destructor_arg;
1878 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1879 atomic_dec(&po->tx_ring.pending);
1880 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1886 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1887 void *frame, struct net_device *dev, int size_max,
1888 __be16 proto, unsigned char *addr, int hlen)
1890 union tpacket_uhdr ph;
1891 int to_write, offset, len, tp_len, nr_frags, len_max;
1892 struct socket *sock = po->sk.sk_socket;
1899 skb->protocol = proto;
1901 skb->priority = po->sk.sk_priority;
1902 skb->mark = po->sk.sk_mark;
1903 skb_shinfo(skb)->destructor_arg = ph.raw;
1905 switch (po->tp_version) {
1907 tp_len = ph.h2->tp_len;
1910 tp_len = ph.h1->tp_len;
1913 if (unlikely(tp_len > size_max)) {
1914 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1918 skb_reserve(skb, hlen);
1919 skb_reset_network_header(skb);
1920 skb_probe_transport_header(skb, 0);
1922 if (po->tp_tx_has_off) {
1923 int off_min, off_max, off;
1924 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1925 off_max = po->tx_ring.frame_size - tp_len;
1926 if (sock->type == SOCK_DGRAM) {
1927 switch (po->tp_version) {
1929 off = ph.h2->tp_net;
1932 off = ph.h1->tp_net;
1936 switch (po->tp_version) {
1938 off = ph.h2->tp_mac;
1941 off = ph.h1->tp_mac;
1945 if (unlikely((off < off_min) || (off_max < off)))
1947 data = ph.raw + off;
1949 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1953 if (sock->type == SOCK_DGRAM) {
1954 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1956 if (unlikely(err < 0))
1958 } else if (dev->hard_header_len) {
1959 /* net device doesn't like empty head */
1960 if (unlikely(tp_len <= dev->hard_header_len)) {
1961 pr_err("packet size is too short (%d < %d)\n",
1962 tp_len, dev->hard_header_len);
1966 skb_push(skb, dev->hard_header_len);
1967 err = skb_store_bits(skb, 0, data,
1968 dev->hard_header_len);
1972 data += dev->hard_header_len;
1973 to_write -= dev->hard_header_len;
1976 offset = offset_in_page(data);
1977 len_max = PAGE_SIZE - offset;
1978 len = ((to_write > len_max) ? len_max : to_write);
1980 skb->data_len = to_write;
1981 skb->len += to_write;
1982 skb->truesize += to_write;
1983 atomic_add(to_write, &po->sk.sk_wmem_alloc);
1985 while (likely(to_write)) {
1986 nr_frags = skb_shinfo(skb)->nr_frags;
1988 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
1989 pr_err("Packet exceed the number of skb frags(%lu)\n",
1994 page = pgv_to_page(data);
1996 flush_dcache_page(page);
1998 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2001 len_max = PAGE_SIZE;
2002 len = ((to_write > len_max) ? len_max : to_write);
2008 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2010 struct sk_buff *skb;
2011 struct net_device *dev;
2013 bool need_rls_dev = false;
2014 int err, reserve = 0;
2016 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2017 int tp_len, size_max;
2018 unsigned char *addr;
2020 int status = TP_STATUS_AVAILABLE;
2023 mutex_lock(&po->pg_vec_lock);
2025 if (saddr == NULL) {
2026 dev = po->prot_hook.dev;
2031 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2033 if (msg->msg_namelen < (saddr->sll_halen
2034 + offsetof(struct sockaddr_ll,
2037 proto = saddr->sll_protocol;
2038 addr = saddr->sll_addr;
2039 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2040 need_rls_dev = true;
2044 if (unlikely(dev == NULL))
2047 reserve = dev->hard_header_len;
2050 if (unlikely(!(dev->flags & IFF_UP)))
2053 size_max = po->tx_ring.frame_size
2054 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2056 if (size_max > dev->mtu + reserve)
2057 size_max = dev->mtu + reserve;
2060 ph = packet_current_frame(po, &po->tx_ring,
2061 TP_STATUS_SEND_REQUEST);
2063 if (unlikely(ph == NULL)) {
2068 status = TP_STATUS_SEND_REQUEST;
2069 hlen = LL_RESERVED_SPACE(dev);
2070 tlen = dev->needed_tailroom;
2071 skb = sock_alloc_send_skb(&po->sk,
2072 hlen + tlen + sizeof(struct sockaddr_ll),
2075 if (unlikely(skb == NULL))
2078 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2081 if (unlikely(tp_len < 0)) {
2083 __packet_set_status(po, ph,
2084 TP_STATUS_AVAILABLE);
2085 packet_increment_head(&po->tx_ring);
2089 status = TP_STATUS_WRONG_FORMAT;
2095 skb->destructor = tpacket_destruct_skb;
2096 __packet_set_status(po, ph, TP_STATUS_SENDING);
2097 atomic_inc(&po->tx_ring.pending);
2099 status = TP_STATUS_SEND_REQUEST;
2100 err = dev_queue_xmit(skb);
2101 if (unlikely(err > 0)) {
2102 err = net_xmit_errno(err);
2103 if (err && __packet_get_status(po, ph) ==
2104 TP_STATUS_AVAILABLE) {
2105 /* skb was destructed already */
2110 * skb was dropped but not destructed yet;
2111 * let's treat it like congestion or err < 0
2115 packet_increment_head(&po->tx_ring);
2117 } while (likely((ph != NULL) ||
2118 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2119 (atomic_read(&po->tx_ring.pending))))
2126 __packet_set_status(po, ph, status);
2132 mutex_unlock(&po->pg_vec_lock);
2136 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2137 size_t reserve, size_t len,
2138 size_t linear, int noblock,
2141 struct sk_buff *skb;
2143 /* Under a page? Don't bother with paged skb. */
2144 if (prepad + len < PAGE_SIZE || !linear)
2147 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2152 skb_reserve(skb, reserve);
2153 skb_put(skb, linear);
2154 skb->data_len = len - linear;
2155 skb->len += len - linear;
2160 static int packet_snd(struct socket *sock,
2161 struct msghdr *msg, size_t len)
2163 struct sock *sk = sock->sk;
2164 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2165 struct sk_buff *skb;
2166 struct net_device *dev;
2168 bool need_rls_dev = false;
2169 unsigned char *addr;
2170 int err, reserve = 0;
2171 struct virtio_net_hdr vnet_hdr = { 0 };
2174 struct packet_sock *po = pkt_sk(sk);
2175 unsigned short gso_type = 0;
2180 * Get and verify the address.
2183 if (saddr == NULL) {
2184 dev = po->prot_hook.dev;
2189 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2191 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2193 proto = saddr->sll_protocol;
2194 addr = saddr->sll_addr;
2195 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2196 need_rls_dev = true;
2202 if (sock->type == SOCK_RAW)
2203 reserve = dev->hard_header_len;
2206 if (!(dev->flags & IFF_UP))
2209 if (po->has_vnet_hdr) {
2210 vnet_hdr_len = sizeof(vnet_hdr);
2213 if (len < vnet_hdr_len)
2216 len -= vnet_hdr_len;
2218 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2223 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2224 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2226 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2227 vnet_hdr.csum_offset + 2;
2230 if (vnet_hdr.hdr_len > len)
2233 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2234 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2235 case VIRTIO_NET_HDR_GSO_TCPV4:
2236 gso_type = SKB_GSO_TCPV4;
2238 case VIRTIO_NET_HDR_GSO_TCPV6:
2239 gso_type = SKB_GSO_TCPV6;
2241 case VIRTIO_NET_HDR_GSO_UDP:
2242 gso_type = SKB_GSO_UDP;
2248 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2249 gso_type |= SKB_GSO_TCP_ECN;
2251 if (vnet_hdr.gso_size == 0)
2257 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2258 if (!netif_supports_nofcs(dev)) {
2259 err = -EPROTONOSUPPORT;
2262 extra_len = 4; /* We're doing our own CRC */
2266 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2270 hlen = LL_RESERVED_SPACE(dev);
2271 tlen = dev->needed_tailroom;
2272 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2273 msg->msg_flags & MSG_DONTWAIT, &err);
2277 skb_set_network_header(skb, reserve);
2280 if (sock->type == SOCK_DGRAM &&
2281 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2284 /* Returns -EFAULT on error */
2285 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2289 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2291 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2292 /* Earlier code assumed this would be a VLAN pkt,
2293 * double-check this now that we have the actual
2296 struct ethhdr *ehdr;
2297 skb_reset_mac_header(skb);
2298 ehdr = eth_hdr(skb);
2299 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2305 skb->protocol = proto;
2307 skb->priority = sk->sk_priority;
2308 skb->mark = sk->sk_mark;
2310 if (po->has_vnet_hdr) {
2311 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2312 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2313 vnet_hdr.csum_offset)) {
2319 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2320 skb_shinfo(skb)->gso_type = gso_type;
2322 /* Header must be checked, and gso_segs computed. */
2323 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2324 skb_shinfo(skb)->gso_segs = 0;
2326 len += vnet_hdr_len;
2329 skb_probe_transport_header(skb, reserve);
2331 if (unlikely(extra_len == 4))
2338 err = dev_queue_xmit(skb);
2339 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2350 if (dev && need_rls_dev)
2356 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2357 struct msghdr *msg, size_t len)
2359 struct sock *sk = sock->sk;
2360 struct packet_sock *po = pkt_sk(sk);
2361 if (po->tx_ring.pg_vec)
2362 return tpacket_snd(po, msg);
2364 return packet_snd(sock, msg, len);
2368 * Close a PACKET socket. This is fairly simple. We immediately go
2369 * to 'closed' state and remove our protocol entry in the device list.
2372 static int packet_release(struct socket *sock)
2374 struct sock *sk = sock->sk;
2375 struct packet_sock *po;
2377 union tpacket_req_u req_u;
2385 mutex_lock(&net->packet.sklist_lock);
2386 sk_del_node_init_rcu(sk);
2387 mutex_unlock(&net->packet.sklist_lock);
2390 sock_prot_inuse_add(net, sk->sk_prot, -1);
2393 spin_lock(&po->bind_lock);
2394 unregister_prot_hook(sk, false);
2395 if (po->prot_hook.dev) {
2396 dev_put(po->prot_hook.dev);
2397 po->prot_hook.dev = NULL;
2399 spin_unlock(&po->bind_lock);
2401 packet_flush_mclist(sk);
2403 if (po->rx_ring.pg_vec) {
2404 memset(&req_u, 0, sizeof(req_u));
2405 packet_set_ring(sk, &req_u, 1, 0);
2408 if (po->tx_ring.pg_vec) {
2409 memset(&req_u, 0, sizeof(req_u));
2410 packet_set_ring(sk, &req_u, 1, 1);
2417 * Now the socket is dead. No more input will appear.
2424 skb_queue_purge(&sk->sk_receive_queue);
2425 sk_refcnt_debug_release(sk);
2432 * Attach a packet hook.
2435 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2437 struct packet_sock *po = pkt_sk(sk);
2448 spin_lock(&po->bind_lock);
2449 unregister_prot_hook(sk, true);
2451 po->prot_hook.type = protocol;
2452 if (po->prot_hook.dev)
2453 dev_put(po->prot_hook.dev);
2454 po->prot_hook.dev = dev;
2456 po->ifindex = dev ? dev->ifindex : 0;
2461 if (!dev || (dev->flags & IFF_UP)) {
2462 register_prot_hook(sk);
2464 sk->sk_err = ENETDOWN;
2465 if (!sock_flag(sk, SOCK_DEAD))
2466 sk->sk_error_report(sk);
2470 spin_unlock(&po->bind_lock);
2476 * Bind a packet socket to a device
2479 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2482 struct sock *sk = sock->sk;
2484 struct net_device *dev;
2491 if (addr_len != sizeof(struct sockaddr))
2493 strlcpy(name, uaddr->sa_data, sizeof(name));
2495 dev = dev_get_by_name(sock_net(sk), name);
2497 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2501 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2503 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2504 struct sock *sk = sock->sk;
2505 struct net_device *dev = NULL;
2513 if (addr_len < sizeof(struct sockaddr_ll))
2515 if (sll->sll_family != AF_PACKET)
2518 if (sll->sll_ifindex) {
2520 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2524 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2530 static struct proto packet_proto = {
2532 .owner = THIS_MODULE,
2533 .obj_size = sizeof(struct packet_sock),
2537 * Create a packet of type SOCK_PACKET.
2540 static int packet_create(struct net *net, struct socket *sock, int protocol,
2544 struct packet_sock *po;
2545 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2548 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2550 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2551 sock->type != SOCK_PACKET)
2552 return -ESOCKTNOSUPPORT;
2554 sock->state = SS_UNCONNECTED;
2557 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2561 sock->ops = &packet_ops;
2562 if (sock->type == SOCK_PACKET)
2563 sock->ops = &packet_ops_spkt;
2565 sock_init_data(sock, sk);
2568 sk->sk_family = PF_PACKET;
2571 sk->sk_destruct = packet_sock_destruct;
2572 sk_refcnt_debug_inc(sk);
2575 * Attach a protocol block
2578 spin_lock_init(&po->bind_lock);
2579 mutex_init(&po->pg_vec_lock);
2580 po->prot_hook.func = packet_rcv;
2582 if (sock->type == SOCK_PACKET)
2583 po->prot_hook.func = packet_rcv_spkt;
2585 po->prot_hook.af_packet_priv = sk;
2588 po->prot_hook.type = proto;
2589 register_prot_hook(sk);
2592 mutex_lock(&net->packet.sklist_lock);
2593 sk_add_node_rcu(sk, &net->packet.sklist);
2594 mutex_unlock(&net->packet.sklist_lock);
2597 sock_prot_inuse_add(net, &packet_proto, 1);
2605 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2607 struct sock_exterr_skb *serr;
2608 struct sk_buff *skb, *skb2;
2612 skb = skb_dequeue(&sk->sk_error_queue);
2618 msg->msg_flags |= MSG_TRUNC;
2621 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2625 sock_recv_timestamp(msg, sk, skb);
2627 serr = SKB_EXT_ERR(skb);
2628 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2629 sizeof(serr->ee), &serr->ee);
2631 msg->msg_flags |= MSG_ERRQUEUE;
2634 /* Reset and regenerate socket error */
2635 spin_lock_bh(&sk->sk_error_queue.lock);
2637 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2638 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2639 spin_unlock_bh(&sk->sk_error_queue.lock);
2640 sk->sk_error_report(sk);
2642 spin_unlock_bh(&sk->sk_error_queue.lock);
2651 * Pull a packet from our receive queue and hand it to the user.
2652 * If necessary we block.
2655 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2656 struct msghdr *msg, size_t len, int flags)
2658 struct sock *sk = sock->sk;
2659 struct sk_buff *skb;
2661 struct sockaddr_ll *sll;
2662 int vnet_hdr_len = 0;
2665 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2669 /* What error should we return now? EUNATTACH? */
2670 if (pkt_sk(sk)->ifindex < 0)
2674 if (flags & MSG_ERRQUEUE) {
2675 err = packet_recv_error(sk, msg, len);
2680 * Call the generic datagram receiver. This handles all sorts
2681 * of horrible races and re-entrancy so we can forget about it
2682 * in the protocol layers.
2684 * Now it will return ENETDOWN, if device have just gone down,
2685 * but then it will block.
2688 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2691 * An error occurred so return it. Because skb_recv_datagram()
2692 * handles the blocking we don't see and worry about blocking
2699 if (pkt_sk(sk)->has_vnet_hdr) {
2700 struct virtio_net_hdr vnet_hdr = { 0 };
2703 vnet_hdr_len = sizeof(vnet_hdr);
2704 if (len < vnet_hdr_len)
2707 len -= vnet_hdr_len;
2709 if (skb_is_gso(skb)) {
2710 struct skb_shared_info *sinfo = skb_shinfo(skb);
2712 /* This is a hint as to how much should be linear. */
2713 vnet_hdr.hdr_len = skb_headlen(skb);
2714 vnet_hdr.gso_size = sinfo->gso_size;
2715 if (sinfo->gso_type & SKB_GSO_TCPV4)
2716 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2717 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2718 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2719 else if (sinfo->gso_type & SKB_GSO_UDP)
2720 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2721 else if (sinfo->gso_type & SKB_GSO_FCOE)
2725 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2726 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2728 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2730 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2731 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2732 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2733 vnet_hdr.csum_offset = skb->csum_offset;
2734 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2735 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2736 } /* else everything is zero */
2738 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2745 * If the address length field is there to be filled in, we fill
2749 sll = &PACKET_SKB_CB(skb)->sa.ll;
2750 if (sock->type == SOCK_PACKET)
2751 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2753 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2756 * You lose any data beyond the buffer you gave. If it worries a
2757 * user program they can ask the device for its MTU anyway.
2763 msg->msg_flags |= MSG_TRUNC;
2766 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2770 sock_recv_ts_and_drops(msg, sk, skb);
2773 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2776 if (pkt_sk(sk)->auxdata) {
2777 struct tpacket_auxdata aux;
2779 aux.tp_status = TP_STATUS_USER;
2780 if (skb->ip_summed == CHECKSUM_PARTIAL)
2781 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2782 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2783 aux.tp_snaplen = skb->len;
2785 aux.tp_net = skb_network_offset(skb);
2786 if (vlan_tx_tag_present(skb)) {
2787 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2788 aux.tp_status |= TP_STATUS_VLAN_VALID;
2790 aux.tp_vlan_tci = 0;
2793 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2797 * Free or return the buffer as appropriate. Again this
2798 * hides all the races and re-entrancy issues from us.
2800 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2803 skb_free_datagram(sk, skb);
2808 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2809 int *uaddr_len, int peer)
2811 struct net_device *dev;
2812 struct sock *sk = sock->sk;
2817 uaddr->sa_family = AF_PACKET;
2819 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2821 strncpy(uaddr->sa_data, dev->name, 14);
2823 memset(uaddr->sa_data, 0, 14);
2825 *uaddr_len = sizeof(*uaddr);
2830 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2831 int *uaddr_len, int peer)
2833 struct net_device *dev;
2834 struct sock *sk = sock->sk;
2835 struct packet_sock *po = pkt_sk(sk);
2836 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2841 sll->sll_family = AF_PACKET;
2842 sll->sll_ifindex = po->ifindex;
2843 sll->sll_protocol = po->num;
2844 sll->sll_pkttype = 0;
2846 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2848 sll->sll_hatype = dev->type;
2849 sll->sll_halen = dev->addr_len;
2850 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2852 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2856 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2861 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2865 case PACKET_MR_MULTICAST:
2866 if (i->alen != dev->addr_len)
2869 return dev_mc_add(dev, i->addr);
2871 return dev_mc_del(dev, i->addr);
2873 case PACKET_MR_PROMISC:
2874 return dev_set_promiscuity(dev, what);
2876 case PACKET_MR_ALLMULTI:
2877 return dev_set_allmulti(dev, what);
2879 case PACKET_MR_UNICAST:
2880 if (i->alen != dev->addr_len)
2883 return dev_uc_add(dev, i->addr);
2885 return dev_uc_del(dev, i->addr);
2893 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2895 for ( ; i; i = i->next) {
2896 if (i->ifindex == dev->ifindex)
2897 packet_dev_mc(dev, i, what);
2901 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2903 struct packet_sock *po = pkt_sk(sk);
2904 struct packet_mclist *ml, *i;
2905 struct net_device *dev;
2911 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2916 if (mreq->mr_alen > dev->addr_len)
2920 i = kmalloc(sizeof(*i), GFP_KERNEL);
2925 for (ml = po->mclist; ml; ml = ml->next) {
2926 if (ml->ifindex == mreq->mr_ifindex &&
2927 ml->type == mreq->mr_type &&
2928 ml->alen == mreq->mr_alen &&
2929 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2931 /* Free the new element ... */
2937 i->type = mreq->mr_type;
2938 i->ifindex = mreq->mr_ifindex;
2939 i->alen = mreq->mr_alen;
2940 memcpy(i->addr, mreq->mr_address, i->alen);
2942 i->next = po->mclist;
2944 err = packet_dev_mc(dev, i, 1);
2946 po->mclist = i->next;
2955 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2957 struct packet_mclist *ml, **mlp;
2961 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2962 if (ml->ifindex == mreq->mr_ifindex &&
2963 ml->type == mreq->mr_type &&
2964 ml->alen == mreq->mr_alen &&
2965 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2966 if (--ml->count == 0) {
2967 struct net_device *dev;
2969 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2971 packet_dev_mc(dev, ml, -1);
2979 return -EADDRNOTAVAIL;
2982 static void packet_flush_mclist(struct sock *sk)
2984 struct packet_sock *po = pkt_sk(sk);
2985 struct packet_mclist *ml;
2991 while ((ml = po->mclist) != NULL) {
2992 struct net_device *dev;
2994 po->mclist = ml->next;
2995 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2997 packet_dev_mc(dev, ml, -1);
3004 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3006 struct sock *sk = sock->sk;
3007 struct packet_sock *po = pkt_sk(sk);
3010 if (level != SOL_PACKET)
3011 return -ENOPROTOOPT;
3014 case PACKET_ADD_MEMBERSHIP:
3015 case PACKET_DROP_MEMBERSHIP:
3017 struct packet_mreq_max mreq;
3019 memset(&mreq, 0, sizeof(mreq));
3020 if (len < sizeof(struct packet_mreq))
3022 if (len > sizeof(mreq))
3024 if (copy_from_user(&mreq, optval, len))
3026 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3028 if (optname == PACKET_ADD_MEMBERSHIP)
3029 ret = packet_mc_add(sk, &mreq);
3031 ret = packet_mc_drop(sk, &mreq);
3035 case PACKET_RX_RING:
3036 case PACKET_TX_RING:
3038 union tpacket_req_u req_u;
3041 switch (po->tp_version) {
3044 len = sizeof(req_u.req);
3048 len = sizeof(req_u.req3);
3053 if (pkt_sk(sk)->has_vnet_hdr)
3055 if (copy_from_user(&req_u.req, optval, len))
3057 return packet_set_ring(sk, &req_u, 0,
3058 optname == PACKET_TX_RING);
3060 case PACKET_COPY_THRESH:
3064 if (optlen != sizeof(val))
3066 if (copy_from_user(&val, optval, sizeof(val)))
3069 pkt_sk(sk)->copy_thresh = val;
3072 case PACKET_VERSION:
3076 if (optlen != sizeof(val))
3078 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3080 if (copy_from_user(&val, optval, sizeof(val)))
3086 po->tp_version = val;
3092 case PACKET_RESERVE:
3096 if (optlen != sizeof(val))
3098 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3100 if (copy_from_user(&val, optval, sizeof(val)))
3102 po->tp_reserve = val;
3109 if (optlen != sizeof(val))
3111 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3113 if (copy_from_user(&val, optval, sizeof(val)))
3115 po->tp_loss = !!val;
3118 case PACKET_AUXDATA:
3122 if (optlen < sizeof(val))
3124 if (copy_from_user(&val, optval, sizeof(val)))
3127 po->auxdata = !!val;
3130 case PACKET_ORIGDEV:
3134 if (optlen < sizeof(val))
3136 if (copy_from_user(&val, optval, sizeof(val)))
3139 po->origdev = !!val;
3142 case PACKET_VNET_HDR:
3146 if (sock->type != SOCK_RAW)
3148 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3150 if (optlen < sizeof(val))
3152 if (copy_from_user(&val, optval, sizeof(val)))
3155 po->has_vnet_hdr = !!val;
3158 case PACKET_TIMESTAMP:
3162 if (optlen != sizeof(val))
3164 if (copy_from_user(&val, optval, sizeof(val)))
3167 po->tp_tstamp = val;
3174 if (optlen != sizeof(val))
3176 if (copy_from_user(&val, optval, sizeof(val)))
3179 return fanout_add(sk, val & 0xffff, val >> 16);
3181 case PACKET_TX_HAS_OFF:
3185 if (optlen != sizeof(val))
3187 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3189 if (copy_from_user(&val, optval, sizeof(val)))
3191 po->tp_tx_has_off = !!val;
3195 return -ENOPROTOOPT;
3199 static int packet_getsockopt(struct socket *sock, int level, int optname,
3200 char __user *optval, int __user *optlen)
3203 int val, lv = sizeof(val);
3204 struct sock *sk = sock->sk;
3205 struct packet_sock *po = pkt_sk(sk);
3207 struct tpacket_stats st;
3208 union tpacket_stats_u st_u;
3210 if (level != SOL_PACKET)
3211 return -ENOPROTOOPT;
3213 if (get_user(len, optlen))
3220 case PACKET_STATISTICS:
3221 spin_lock_bh(&sk->sk_receive_queue.lock);
3222 if (po->tp_version == TPACKET_V3) {
3223 lv = sizeof(struct tpacket_stats_v3);
3224 memcpy(&st_u.stats3, &po->stats,
3225 sizeof(struct tpacket_stats));
3226 st_u.stats3.tp_freeze_q_cnt =
3227 po->stats_u.stats3.tp_freeze_q_cnt;
3228 st_u.stats3.tp_packets += po->stats.tp_drops;
3229 data = &st_u.stats3;
3231 lv = sizeof(struct tpacket_stats);
3233 st.tp_packets += st.tp_drops;
3236 memset(&po->stats, 0, sizeof(st));
3237 spin_unlock_bh(&sk->sk_receive_queue.lock);
3239 case PACKET_AUXDATA:
3242 case PACKET_ORIGDEV:
3245 case PACKET_VNET_HDR:
3246 val = po->has_vnet_hdr;
3248 case PACKET_VERSION:
3249 val = po->tp_version;
3252 if (len > sizeof(int))
3254 if (copy_from_user(&val, optval, len))
3258 val = sizeof(struct tpacket_hdr);
3261 val = sizeof(struct tpacket2_hdr);
3264 val = sizeof(struct tpacket3_hdr);
3270 case PACKET_RESERVE:
3271 val = po->tp_reserve;
3276 case PACKET_TIMESTAMP:
3277 val = po->tp_tstamp;
3281 ((u32)po->fanout->id |
3282 ((u32)po->fanout->type << 16) |
3283 ((u32)po->fanout->flags << 24)) :
3286 case PACKET_TX_HAS_OFF:
3287 val = po->tp_tx_has_off;
3290 return -ENOPROTOOPT;
3295 if (put_user(len, optlen))
3297 if (copy_to_user(optval, data, len))
3303 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3306 struct net_device *dev = data;
3307 struct net *net = dev_net(dev);
3310 sk_for_each_rcu(sk, &net->packet.sklist) {
3311 struct packet_sock *po = pkt_sk(sk);
3314 case NETDEV_UNREGISTER:
3316 packet_dev_mclist(dev, po->mclist, -1);
3320 if (dev->ifindex == po->ifindex) {
3321 spin_lock(&po->bind_lock);
3323 __unregister_prot_hook(sk, false);
3324 sk->sk_err = ENETDOWN;
3325 if (!sock_flag(sk, SOCK_DEAD))
3326 sk->sk_error_report(sk);
3328 if (msg == NETDEV_UNREGISTER) {
3330 if (po->prot_hook.dev)
3331 dev_put(po->prot_hook.dev);
3332 po->prot_hook.dev = NULL;
3334 spin_unlock(&po->bind_lock);
3338 if (dev->ifindex == po->ifindex) {
3339 spin_lock(&po->bind_lock);
3341 register_prot_hook(sk);
3342 spin_unlock(&po->bind_lock);
3352 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3355 struct sock *sk = sock->sk;
3360 int amount = sk_wmem_alloc_get(sk);
3362 return put_user(amount, (int __user *)arg);
3366 struct sk_buff *skb;
3369 spin_lock_bh(&sk->sk_receive_queue.lock);
3370 skb = skb_peek(&sk->sk_receive_queue);
3373 spin_unlock_bh(&sk->sk_receive_queue.lock);
3374 return put_user(amount, (int __user *)arg);
3377 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3379 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3389 case SIOCGIFBRDADDR:
3390 case SIOCSIFBRDADDR:
3391 case SIOCGIFNETMASK:
3392 case SIOCSIFNETMASK:
3393 case SIOCGIFDSTADDR:
3394 case SIOCSIFDSTADDR:
3396 return inet_dgram_ops.ioctl(sock, cmd, arg);
3400 return -ENOIOCTLCMD;
3405 static unsigned int packet_poll(struct file *file, struct socket *sock,
3408 struct sock *sk = sock->sk;
3409 struct packet_sock *po = pkt_sk(sk);
3410 unsigned int mask = datagram_poll(file, sock, wait);
3412 spin_lock_bh(&sk->sk_receive_queue.lock);
3413 if (po->rx_ring.pg_vec) {
3414 if (!packet_previous_rx_frame(po, &po->rx_ring,
3416 mask |= POLLIN | POLLRDNORM;
3418 spin_unlock_bh(&sk->sk_receive_queue.lock);
3419 spin_lock_bh(&sk->sk_write_queue.lock);
3420 if (po->tx_ring.pg_vec) {
3421 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3422 mask |= POLLOUT | POLLWRNORM;
3424 spin_unlock_bh(&sk->sk_write_queue.lock);
3429 /* Dirty? Well, I still did not learn better way to account
3433 static void packet_mm_open(struct vm_area_struct *vma)
3435 struct file *file = vma->vm_file;
3436 struct socket *sock = file->private_data;
3437 struct sock *sk = sock->sk;
3440 atomic_inc(&pkt_sk(sk)->mapped);
3443 static void packet_mm_close(struct vm_area_struct *vma)
3445 struct file *file = vma->vm_file;
3446 struct socket *sock = file->private_data;
3447 struct sock *sk = sock->sk;
3450 atomic_dec(&pkt_sk(sk)->mapped);
3453 static const struct vm_operations_struct packet_mmap_ops = {
3454 .open = packet_mm_open,
3455 .close = packet_mm_close,
3458 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3463 for (i = 0; i < len; i++) {
3464 if (likely(pg_vec[i].buffer)) {
3465 if (is_vmalloc_addr(pg_vec[i].buffer))
3466 vfree(pg_vec[i].buffer);
3468 free_pages((unsigned long)pg_vec[i].buffer,
3470 pg_vec[i].buffer = NULL;
3476 static char *alloc_one_pg_vec_page(unsigned long order)
3478 char *buffer = NULL;
3479 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3480 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3482 buffer = (char *) __get_free_pages(gfp_flags, order);
3488 * __get_free_pages failed, fall back to vmalloc
3490 buffer = vzalloc((1 << order) * PAGE_SIZE);
3496 * vmalloc failed, lets dig into swap here
3498 gfp_flags &= ~__GFP_NORETRY;
3499 buffer = (char *)__get_free_pages(gfp_flags, order);
3504 * complete and utter failure
3509 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3511 unsigned int block_nr = req->tp_block_nr;
3515 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3516 if (unlikely(!pg_vec))
3519 for (i = 0; i < block_nr; i++) {
3520 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3521 if (unlikely(!pg_vec[i].buffer))
3522 goto out_free_pgvec;
3529 free_pg_vec(pg_vec, order, block_nr);
3534 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3535 int closing, int tx_ring)
3537 struct pgv *pg_vec = NULL;
3538 struct packet_sock *po = pkt_sk(sk);
3539 int was_running, order = 0;
3540 struct packet_ring_buffer *rb;
3541 struct sk_buff_head *rb_queue;
3544 /* Added to avoid minimal code churn */
3545 struct tpacket_req *req = &req_u->req;
3547 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3548 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3549 WARN(1, "Tx-ring is not supported.\n");
3553 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3554 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3558 if (atomic_read(&po->mapped))
3560 if (atomic_read(&rb->pending))
3564 if (req->tp_block_nr) {
3565 /* Sanity tests and some calculations */
3567 if (unlikely(rb->pg_vec))
3570 switch (po->tp_version) {
3572 po->tp_hdrlen = TPACKET_HDRLEN;
3575 po->tp_hdrlen = TPACKET2_HDRLEN;
3578 po->tp_hdrlen = TPACKET3_HDRLEN;
3583 if (unlikely((int)req->tp_block_size <= 0))
3585 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3587 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3590 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3593 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3594 if (unlikely(rb->frames_per_block <= 0))
3596 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3601 order = get_order(req->tp_block_size);
3602 pg_vec = alloc_pg_vec(req, order);
3603 if (unlikely(!pg_vec))
3605 switch (po->tp_version) {
3607 /* Transmit path is not supported. We checked
3608 * it above but just being paranoid
3611 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3620 if (unlikely(req->tp_frame_nr))
3626 /* Detach socket from network */
3627 spin_lock(&po->bind_lock);
3628 was_running = po->running;
3632 __unregister_prot_hook(sk, false);
3634 spin_unlock(&po->bind_lock);
3639 mutex_lock(&po->pg_vec_lock);
3640 if (closing || atomic_read(&po->mapped) == 0) {
3642 spin_lock_bh(&rb_queue->lock);
3643 swap(rb->pg_vec, pg_vec);
3644 rb->frame_max = (req->tp_frame_nr - 1);
3646 rb->frame_size = req->tp_frame_size;
3647 spin_unlock_bh(&rb_queue->lock);
3649 swap(rb->pg_vec_order, order);
3650 swap(rb->pg_vec_len, req->tp_block_nr);
3652 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3653 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3654 tpacket_rcv : packet_rcv;
3655 skb_queue_purge(rb_queue);
3656 if (atomic_read(&po->mapped))
3657 pr_err("packet_mmap: vma is busy: %d\n",
3658 atomic_read(&po->mapped));
3660 mutex_unlock(&po->pg_vec_lock);
3662 spin_lock(&po->bind_lock);
3665 register_prot_hook(sk);
3667 spin_unlock(&po->bind_lock);
3668 if (closing && (po->tp_version > TPACKET_V2)) {
3669 /* Because we don't support block-based V3 on tx-ring */
3671 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3676 free_pg_vec(pg_vec, order, req->tp_block_nr);
3681 static int packet_mmap(struct file *file, struct socket *sock,
3682 struct vm_area_struct *vma)
3684 struct sock *sk = sock->sk;
3685 struct packet_sock *po = pkt_sk(sk);
3686 unsigned long size, expected_size;
3687 struct packet_ring_buffer *rb;
3688 unsigned long start;
3695 mutex_lock(&po->pg_vec_lock);
3698 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3700 expected_size += rb->pg_vec_len
3706 if (expected_size == 0)
3709 size = vma->vm_end - vma->vm_start;
3710 if (size != expected_size)
3713 start = vma->vm_start;
3714 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3715 if (rb->pg_vec == NULL)
3718 for (i = 0; i < rb->pg_vec_len; i++) {
3720 void *kaddr = rb->pg_vec[i].buffer;
3723 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3724 page = pgv_to_page(kaddr);
3725 err = vm_insert_page(vma, start, page);
3734 atomic_inc(&po->mapped);
3735 vma->vm_ops = &packet_mmap_ops;
3739 mutex_unlock(&po->pg_vec_lock);
3743 static const struct proto_ops packet_ops_spkt = {
3744 .family = PF_PACKET,
3745 .owner = THIS_MODULE,
3746 .release = packet_release,
3747 .bind = packet_bind_spkt,
3748 .connect = sock_no_connect,
3749 .socketpair = sock_no_socketpair,
3750 .accept = sock_no_accept,
3751 .getname = packet_getname_spkt,
3752 .poll = datagram_poll,
3753 .ioctl = packet_ioctl,
3754 .listen = sock_no_listen,
3755 .shutdown = sock_no_shutdown,
3756 .setsockopt = sock_no_setsockopt,
3757 .getsockopt = sock_no_getsockopt,
3758 .sendmsg = packet_sendmsg_spkt,
3759 .recvmsg = packet_recvmsg,
3760 .mmap = sock_no_mmap,
3761 .sendpage = sock_no_sendpage,
3764 static const struct proto_ops packet_ops = {
3765 .family = PF_PACKET,
3766 .owner = THIS_MODULE,
3767 .release = packet_release,
3768 .bind = packet_bind,
3769 .connect = sock_no_connect,
3770 .socketpair = sock_no_socketpair,
3771 .accept = sock_no_accept,
3772 .getname = packet_getname,
3773 .poll = packet_poll,
3774 .ioctl = packet_ioctl,
3775 .listen = sock_no_listen,
3776 .shutdown = sock_no_shutdown,
3777 .setsockopt = packet_setsockopt,
3778 .getsockopt = packet_getsockopt,
3779 .sendmsg = packet_sendmsg,
3780 .recvmsg = packet_recvmsg,
3781 .mmap = packet_mmap,
3782 .sendpage = sock_no_sendpage,
3785 static const struct net_proto_family packet_family_ops = {
3786 .family = PF_PACKET,
3787 .create = packet_create,
3788 .owner = THIS_MODULE,
3791 static struct notifier_block packet_netdev_notifier = {
3792 .notifier_call = packet_notifier,
3795 #ifdef CONFIG_PROC_FS
3797 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3800 struct net *net = seq_file_net(seq);
3803 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3806 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3808 struct net *net = seq_file_net(seq);
3809 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3812 static void packet_seq_stop(struct seq_file *seq, void *v)
3818 static int packet_seq_show(struct seq_file *seq, void *v)
3820 if (v == SEQ_START_TOKEN)
3821 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3823 struct sock *s = sk_entry(v);
3824 const struct packet_sock *po = pkt_sk(s);
3827 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3829 atomic_read(&s->sk_refcnt),
3834 atomic_read(&s->sk_rmem_alloc),
3835 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3842 static const struct seq_operations packet_seq_ops = {
3843 .start = packet_seq_start,
3844 .next = packet_seq_next,
3845 .stop = packet_seq_stop,
3846 .show = packet_seq_show,
3849 static int packet_seq_open(struct inode *inode, struct file *file)
3851 return seq_open_net(inode, file, &packet_seq_ops,
3852 sizeof(struct seq_net_private));
3855 static const struct file_operations packet_seq_fops = {
3856 .owner = THIS_MODULE,
3857 .open = packet_seq_open,
3859 .llseek = seq_lseek,
3860 .release = seq_release_net,
3865 static int __net_init packet_net_init(struct net *net)
3867 mutex_init(&net->packet.sklist_lock);
3868 INIT_HLIST_HEAD(&net->packet.sklist);
3870 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3876 static void __net_exit packet_net_exit(struct net *net)
3878 remove_proc_entry("packet", net->proc_net);
3881 static struct pernet_operations packet_net_ops = {
3882 .init = packet_net_init,
3883 .exit = packet_net_exit,
3887 static void __exit packet_exit(void)
3889 unregister_netdevice_notifier(&packet_netdev_notifier);
3890 unregister_pernet_subsys(&packet_net_ops);
3891 sock_unregister(PF_PACKET);
3892 proto_unregister(&packet_proto);
3895 static int __init packet_init(void)
3897 int rc = proto_register(&packet_proto, 0);
3902 sock_register(&packet_family_ops);
3903 register_pernet_subsys(&packet_net_ops);
3904 register_netdevice_notifier(&packet_netdev_notifier);
3909 module_init(packet_init);
3910 module_exit(packet_exit);
3911 MODULE_LICENSE("GPL");
3912 MODULE_ALIAS_NETPROTO(PF_PACKET);