2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
80 #include <asm/cacheflush.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
94 #include <net/inet_common.h>
99 - if device has no dev->hard_header routine, it adds and removes ll header
100 inside itself. In this case ll header is invisible outside of device,
101 but higher levels still should reserve dev->hard_header_len.
102 Some devices are enough clever to reallocate skb, when header
103 will not fit to reserved space (tunnel), another ones are silly
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
111 Incoming, dev->hard_header!=NULL
112 mac_header -> ll header
115 Outgoing, dev->hard_header!=NULL
116 mac_header -> ll header
119 Incoming, dev->hard_header==NULL
120 mac_header -> UNKNOWN position. It is very likely, that it points to ll
121 header. PPP makes it, that is wrong, because introduce
122 assymetry between rx and tx paths.
125 Outgoing, dev->hard_header==NULL
126 mac_header -> data. ll header is still not built!
130 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
136 dev->hard_header != NULL
137 mac_header -> ll header
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144 We should set nh.raw on output to correct posistion,
145 packet classifier depends on it.
148 /* Private packet socket structures. */
150 struct packet_mclist {
151 struct packet_mclist *next;
156 unsigned char addr[MAX_ADDR_LEN];
158 /* identical to struct packet_mreq except it has
159 * a longer address field.
161 struct packet_mreq_max {
163 unsigned short mr_type;
164 unsigned short mr_alen;
165 unsigned char mr_address[MAX_ADDR_LEN];
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
182 unsigned int feature_req_word;
184 unsigned char reset_pending_on_curr_blk;
185 unsigned char delete_blk_timer;
186 unsigned short kactive_blk_num;
187 unsigned short blk_sizeof_priv;
189 /* last_kactive_blk_num:
190 * trick to see if user-space has caught up
191 * in order to avoid refreshing timer when every single pkt arrives.
193 unsigned short last_kactive_blk_num;
198 unsigned int max_frame_len;
199 unsigned int knum_blocks;
200 uint64_t knxt_seq_num;
205 atomic_t blk_fill_in_prog;
207 /* Default is set to 8ms */
208 #define DEFAULT_PRB_RETIRE_TOV (8)
210 unsigned short retire_blk_tov;
211 unsigned short version;
212 unsigned long tov_in_jiffies;
214 /* timer to retire an outstanding block */
215 struct timer_list retire_blk_timer;
218 #define PGV_FROM_VMALLOC 1
223 struct packet_ring_buffer {
226 unsigned int frames_per_block;
227 unsigned int frame_size;
228 unsigned int frame_max;
230 unsigned int pg_vec_order;
231 unsigned int pg_vec_pages;
232 unsigned int pg_vec_len;
234 struct tpacket_kbdq_core prb_bdqc;
238 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
239 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
240 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
241 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
242 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
243 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
244 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
247 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
249 static void *packet_previous_frame(struct packet_sock *po,
250 struct packet_ring_buffer *rb,
252 static void packet_increment_head(struct packet_ring_buffer *buff);
253 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
254 struct tpacket_block_desc *);
255 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
256 struct packet_sock *);
257 static void prb_retire_current_block(struct tpacket_kbdq_core *,
258 struct packet_sock *, unsigned int status);
259 static int prb_queue_frozen(struct tpacket_kbdq_core *);
260 static void prb_open_block(struct tpacket_kbdq_core *,
261 struct tpacket_block_desc *);
262 static void prb_retire_rx_blk_timer_expired(unsigned long);
263 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
264 static void prb_init_blk_timer(struct packet_sock *,
265 struct tpacket_kbdq_core *,
266 void (*func) (unsigned long));
267 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
268 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
269 struct tpacket3_hdr *);
270 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
271 struct tpacket3_hdr *);
272 static void packet_flush_mclist(struct sock *sk);
274 struct packet_fanout;
276 /* struct sock has to be the first member of packet_sock */
278 struct packet_fanout *fanout;
279 struct tpacket_stats stats;
280 union tpacket_stats_u stats_u;
281 struct packet_ring_buffer rx_ring;
282 struct packet_ring_buffer tx_ring;
284 spinlock_t bind_lock;
285 struct mutex pg_vec_lock;
286 unsigned int running:1, /* prot_hook is attached*/
290 int ifindex; /* bound device */
292 struct packet_mclist *mclist;
294 enum tpacket_versions tp_version;
295 unsigned int tp_hdrlen;
296 unsigned int tp_reserve;
297 unsigned int tp_loss:1;
298 unsigned int tp_tstamp;
299 struct net_device __rcu *cached_dev;
300 struct packet_type prot_hook ____cacheline_aligned_in_smp;
303 #define PACKET_FANOUT_MAX 256
305 struct packet_fanout {
309 unsigned int num_members;
314 struct list_head list;
315 struct sock *arr[PACKET_FANOUT_MAX];
318 struct packet_type prot_hook ____cacheline_aligned_in_smp;
321 struct packet_skb_cb {
322 unsigned int origlen;
324 struct sockaddr_pkt pkt;
325 struct sockaddr_ll ll;
329 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
331 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
332 #define GET_PBLOCK_DESC(x, bid) \
333 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
334 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
335 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
336 #define GET_NEXT_PRB_BLK_NUM(x) \
337 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
338 ((x)->kactive_blk_num+1) : 0)
340 static struct packet_sock *pkt_sk(struct sock *sk)
342 return (struct packet_sock *)sk;
345 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
346 static void __fanout_link(struct sock *sk, struct packet_sock *po);
348 /* register_prot_hook must be invoked with the po->bind_lock held,
349 * or from a context in which asynchronous accesses to the packet
350 * socket is not possible (packet_create()).
352 static void register_prot_hook(struct sock *sk)
354 struct packet_sock *po = pkt_sk(sk);
358 __fanout_link(sk, po);
360 dev_add_pack(&po->prot_hook);
361 rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
369 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
370 * held. If the sync parameter is true, we will temporarily drop
371 * the po->bind_lock and do a synchronize_net to make sure no
372 * asynchronous packet processing paths still refer to the elements
373 * of po->prot_hook. If the sync parameter is false, it is the
374 * callers responsibility to take care of this.
376 static void __unregister_prot_hook(struct sock *sk, bool sync)
378 struct packet_sock *po = pkt_sk(sk);
382 __fanout_unlink(sk, po);
384 __dev_remove_pack(&po->prot_hook);
385 RCU_INIT_POINTER(po->cached_dev, NULL);
391 spin_unlock(&po->bind_lock);
393 spin_lock(&po->bind_lock);
397 static void unregister_prot_hook(struct sock *sk, bool sync)
399 struct packet_sock *po = pkt_sk(sk);
402 __unregister_prot_hook(sk, sync);
405 static inline __pure struct page *pgv_to_page(void *addr)
407 if (is_vmalloc_addr(addr))
408 return vmalloc_to_page(addr);
409 return virt_to_page(addr);
412 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
415 struct tpacket_hdr *h1;
416 struct tpacket2_hdr *h2;
421 switch (po->tp_version) {
423 h.h1->tp_status = status;
424 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
427 h.h2->tp_status = status;
428 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
432 WARN(1, "TPACKET version not supported.\n");
439 static int __packet_get_status(struct packet_sock *po, void *frame)
442 struct tpacket_hdr *h1;
443 struct tpacket2_hdr *h2;
450 switch (po->tp_version) {
452 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
453 return h.h1->tp_status;
455 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
456 return h.h2->tp_status;
459 WARN(1, "TPACKET version not supported.\n");
465 static void *packet_lookup_frame(struct packet_sock *po,
466 struct packet_ring_buffer *rb,
467 unsigned int position,
470 unsigned int pg_vec_pos, frame_offset;
472 struct tpacket_hdr *h1;
473 struct tpacket2_hdr *h2;
477 pg_vec_pos = position / rb->frames_per_block;
478 frame_offset = position % rb->frames_per_block;
480 h.raw = rb->pg_vec[pg_vec_pos].buffer +
481 (frame_offset * rb->frame_size);
483 if (status != __packet_get_status(po, h.raw))
489 static void *packet_current_frame(struct packet_sock *po,
490 struct packet_ring_buffer *rb,
493 return packet_lookup_frame(po, rb, rb->head, status);
496 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
498 del_timer_sync(&pkc->retire_blk_timer);
501 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
503 struct sk_buff_head *rb_queue)
505 struct tpacket_kbdq_core *pkc;
507 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
509 spin_lock_bh(&rb_queue->lock);
510 pkc->delete_blk_timer = 1;
511 spin_unlock_bh(&rb_queue->lock);
513 prb_del_retire_blk_timer(pkc);
516 static void prb_init_blk_timer(struct packet_sock *po,
517 struct tpacket_kbdq_core *pkc,
518 void (*func) (unsigned long))
520 init_timer(&pkc->retire_blk_timer);
521 pkc->retire_blk_timer.data = (long)po;
522 pkc->retire_blk_timer.function = func;
523 pkc->retire_blk_timer.expires = jiffies;
526 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
528 struct tpacket_kbdq_core *pkc;
533 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
534 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
537 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
538 int blk_size_in_bytes)
540 struct net_device *dev;
541 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
542 struct ethtool_cmd ecmd;
546 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
547 if (unlikely(!dev)) {
549 return DEFAULT_PRB_RETIRE_TOV;
551 err = __ethtool_get_settings(dev, &ecmd);
554 switch (ecmd.speed) {
564 * If the link speed is so slow you don't really
565 * need to worry about perf anyways
570 return DEFAULT_PRB_RETIRE_TOV;
574 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
586 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
587 union tpacket_req_u *req_u)
589 p1->feature_req_word = req_u->req3.tp_feature_req_word;
592 static void init_prb_bdqc(struct packet_sock *po,
593 struct packet_ring_buffer *rb,
595 union tpacket_req_u *req_u, int tx_ring)
597 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
598 struct tpacket_block_desc *pbd;
600 memset(p1, 0x0, sizeof(*p1));
602 p1->knxt_seq_num = 1;
604 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
605 p1->pkblk_start = (char *)pg_vec[0].buffer;
606 p1->kblk_size = req_u->req3.tp_block_size;
607 p1->knum_blocks = req_u->req3.tp_block_nr;
608 p1->hdrlen = po->tp_hdrlen;
609 p1->version = po->tp_version;
610 p1->last_kactive_blk_num = 0;
611 po->stats_u.stats3.tp_freeze_q_cnt = 0;
612 if (req_u->req3.tp_retire_blk_tov)
613 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
615 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
616 req_u->req3.tp_block_size);
617 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
618 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
620 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
621 prb_init_ft_ops(p1, req_u);
622 prb_setup_retire_blk_timer(po, tx_ring);
623 prb_open_block(p1, pbd);
626 /* Do NOT update the last_blk_num first.
627 * Assumes sk_buff_head lock is held.
629 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
631 mod_timer(&pkc->retire_blk_timer,
632 jiffies + pkc->tov_in_jiffies);
633 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
638 * 1) We refresh the timer only when we open a block.
639 * By doing this we don't waste cycles refreshing the timer
640 * on packet-by-packet basis.
642 * With a 1MB block-size, on a 1Gbps line, it will take
643 * i) ~8 ms to fill a block + ii) memcpy etc.
644 * In this cut we are not accounting for the memcpy time.
646 * So, if the user sets the 'tmo' to 10ms then the timer
647 * will never fire while the block is still getting filled
648 * (which is what we want). However, the user could choose
649 * to close a block early and that's fine.
651 * But when the timer does fire, we check whether or not to refresh it.
652 * Since the tmo granularity is in msecs, it is not too expensive
653 * to refresh the timer, lets say every '8' msecs.
654 * Either the user can set the 'tmo' or we can derive it based on
655 * a) line-speed and b) block-size.
656 * prb_calc_retire_blk_tmo() calculates the tmo.
659 static void prb_retire_rx_blk_timer_expired(unsigned long data)
661 struct packet_sock *po = (struct packet_sock *)data;
662 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
664 struct tpacket_block_desc *pbd;
666 spin_lock(&po->sk.sk_receive_queue.lock);
668 frozen = prb_queue_frozen(pkc);
669 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
671 if (unlikely(pkc->delete_blk_timer))
674 /* We only need to plug the race when the block is partially filled.
676 * lock(); increment BLOCK_NUM_PKTS; unlock()
677 * copy_bits() is in progress ...
678 * timer fires on other cpu:
679 * we can't retire the current block because copy_bits
683 if (BLOCK_NUM_PKTS(pbd)) {
684 while (atomic_read(&pkc->blk_fill_in_prog)) {
685 /* Waiting for skb_copy_bits to finish... */
690 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
692 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
693 if (!prb_dispatch_next_block(pkc, po))
698 /* Case 1. Queue was frozen because user-space was
701 if (prb_curr_blk_in_use(pkc, pbd)) {
703 * Ok, user-space is still behind.
704 * So just refresh the timer.
708 /* Case 2. queue was frozen,user-space caught up,
709 * now the link went idle && the timer fired.
710 * We don't have a block to close.So we open this
711 * block and restart the timer.
712 * opening a block thaws the queue,restarts timer
713 * Thawing/timer-refresh is a side effect.
715 prb_open_block(pkc, pbd);
722 _prb_refresh_rx_retire_blk_timer(pkc);
725 spin_unlock(&po->sk.sk_receive_queue.lock);
728 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
729 struct tpacket_block_desc *pbd1, __u32 status)
731 /* Flush everything minus the block header */
733 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
738 /* Skip the block header(we know header WILL fit in 4K) */
741 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
742 for (; start < end; start += PAGE_SIZE)
743 flush_dcache_page(pgv_to_page(start));
748 /* Now update the block status. */
750 BLOCK_STATUS(pbd1) = status;
752 /* Flush the block header */
754 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
756 flush_dcache_page(pgv_to_page(start));
766 * 2) Increment active_blk_num
768 * Note:We DONT refresh the timer on purpose.
769 * Because almost always the next block will be opened.
771 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
772 struct tpacket_block_desc *pbd1,
773 struct packet_sock *po, unsigned int stat)
775 __u32 status = TP_STATUS_USER | stat;
777 struct tpacket3_hdr *last_pkt;
778 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
780 if (po->stats.tp_drops)
781 status |= TP_STATUS_LOSING;
783 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
784 last_pkt->tp_next_offset = 0;
786 /* Get the ts of the last pkt */
787 if (BLOCK_NUM_PKTS(pbd1)) {
788 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
789 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
791 /* Ok, we tmo'd - so get the current time */
794 h1->ts_last_pkt.ts_sec = ts.tv_sec;
795 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
800 /* Flush the block */
801 prb_flush_block(pkc1, pbd1, status);
803 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
806 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
808 pkc->reset_pending_on_curr_blk = 0;
812 * Side effect of opening a block:
814 * 1) prb_queue is thawed.
815 * 2) retire_blk_timer is refreshed.
818 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
819 struct tpacket_block_desc *pbd1)
822 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
826 /* We could have just memset this but we will lose the
827 * flexibility of making the priv area sticky
829 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
830 BLOCK_NUM_PKTS(pbd1) = 0;
831 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
833 h1->ts_first_pkt.ts_sec = ts.tv_sec;
834 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
835 pkc1->pkblk_start = (char *)pbd1;
836 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
837 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
838 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
839 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
840 pbd1->version = pkc1->version;
841 pkc1->prev = pkc1->nxt_offset;
842 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
843 prb_thaw_queue(pkc1);
844 _prb_refresh_rx_retire_blk_timer(pkc1);
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats_u.stats3.tp_freeze_q_cnt++;
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
890 struct tpacket_block_desc *pbd;
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
905 * open this block and return the offset where the first packet
906 * needs to get stored.
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 while (atomic_read(&pkc->blk_fill_in_prog)) {
930 /* Waiting for skb_copy_bits to finish... */
934 prb_close_block(pkc, pbd, po, status);
939 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
940 struct tpacket_block_desc *pbd)
942 return TP_STATUS_USER & BLOCK_STATUS(pbd);
945 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
947 return pkc->reset_pending_on_curr_blk;
950 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
952 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
953 atomic_dec(&pkc->blk_fill_in_prog);
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957 struct tpacket3_hdr *ppd)
959 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963 struct tpacket3_hdr *ppd)
965 ppd->hv1.tp_rxhash = 0;
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969 struct tpacket3_hdr *ppd)
971 if (vlan_tx_tag_present(pkc->skb)) {
972 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
973 ppd->tp_status = TP_STATUS_VLAN_VALID;
975 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
979 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
980 struct tpacket3_hdr *ppd)
982 prb_fill_vlan_info(pkc, ppd);
984 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
985 prb_fill_rxhash(pkc, ppd);
987 prb_clear_rxhash(pkc, ppd);
990 static void prb_fill_curr_block(char *curr,
991 struct tpacket_kbdq_core *pkc,
992 struct tpacket_block_desc *pbd,
995 struct tpacket3_hdr *ppd;
997 ppd = (struct tpacket3_hdr *)curr;
998 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1000 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002 BLOCK_NUM_PKTS(pbd) += 1;
1003 atomic_inc(&pkc->blk_fill_in_prog);
1004 prb_run_all_ft_ops(pkc, ppd);
1007 /* Assumes caller has the sk->rx_queue.lock */
1008 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009 struct sk_buff *skb,
1014 struct tpacket_kbdq_core *pkc;
1015 struct tpacket_block_desc *pbd;
1018 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1019 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1021 /* Queue is frozen when user space is lagging behind */
1022 if (prb_queue_frozen(pkc)) {
1024 * Check if that last block which caused the queue to freeze,
1025 * is still in_use by user-space.
1027 if (prb_curr_blk_in_use(pkc, pbd)) {
1028 /* Can't record this packet */
1032 * Ok, the block was released by user-space.
1033 * Now let's open that block.
1034 * opening a block also thaws the queue.
1035 * Thawing is a side effect.
1037 prb_open_block(pkc, pbd);
1042 curr = pkc->nxt_offset;
1044 end = (char *) ((char *)pbd + pkc->kblk_size);
1046 /* first try the current block */
1047 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1048 prb_fill_curr_block(curr, pkc, pbd, len);
1049 return (void *)curr;
1052 /* Ok, close the current block */
1053 prb_retire_current_block(pkc, po, 0);
1055 /* Now, try to dispatch the next block */
1056 curr = (char *)prb_dispatch_next_block(pkc, po);
1058 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1059 prb_fill_curr_block(curr, pkc, pbd, len);
1060 return (void *)curr;
1064 * No free blocks are available.user_space hasn't caught up yet.
1065 * Queue was just frozen and now this packet will get dropped.
1070 static void *packet_current_rx_frame(struct packet_sock *po,
1071 struct sk_buff *skb,
1072 int status, unsigned int len)
1075 switch (po->tp_version) {
1078 curr = packet_lookup_frame(po, &po->rx_ring,
1079 po->rx_ring.head, status);
1082 return __packet_lookup_frame_in_block(po, skb, status, len);
1084 WARN(1, "TPACKET version not supported\n");
1090 static void *prb_lookup_block(struct packet_sock *po,
1091 struct packet_ring_buffer *rb,
1092 unsigned int previous,
1095 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1096 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1098 if (status != BLOCK_STATUS(pbd))
1103 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1106 if (rb->prb_bdqc.kactive_blk_num)
1107 prev = rb->prb_bdqc.kactive_blk_num-1;
1109 prev = rb->prb_bdqc.knum_blocks-1;
1113 /* Assumes caller has held the rx_queue.lock */
1114 static void *__prb_previous_block(struct packet_sock *po,
1115 struct packet_ring_buffer *rb,
1118 unsigned int previous = prb_previous_blk_num(rb);
1119 return prb_lookup_block(po, rb, previous, status);
1122 static void *packet_previous_rx_frame(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1126 if (po->tp_version <= TPACKET_V2)
1127 return packet_previous_frame(po, rb, status);
1129 return __prb_previous_block(po, rb, status);
1132 static void packet_increment_rx_head(struct packet_sock *po,
1133 struct packet_ring_buffer *rb)
1135 switch (po->tp_version) {
1138 return packet_increment_head(rb);
1141 WARN(1, "TPACKET version not supported.\n");
1147 static void *packet_previous_frame(struct packet_sock *po,
1148 struct packet_ring_buffer *rb,
1151 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1152 return packet_lookup_frame(po, rb, previous, status);
1155 static void packet_increment_head(struct packet_ring_buffer *buff)
1157 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1160 static void packet_sock_destruct(struct sock *sk)
1162 skb_queue_purge(&sk->sk_error_queue);
1164 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1165 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1167 if (!sock_flag(sk, SOCK_DEAD)) {
1168 pr_err("Attempt to release alive packet socket: %p\n", sk);
1172 sk_refcnt_debug_dec(sk);
1175 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1177 u32 idx, hash = skb->rxhash;
1179 idx = ((u64)hash * num) >> 32;
1184 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1186 unsigned int val = atomic_inc_return(&f->rr_cur);
1188 return f->arr[val % num];
1191 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1193 unsigned int cpu = smp_processor_id();
1195 return f->arr[cpu % num];
1198 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1199 struct packet_type *pt, struct net_device *orig_dev)
1201 struct packet_fanout *f = pt->af_packet_priv;
1202 unsigned int num = ACCESS_ONCE(f->num_members);
1203 struct packet_sock *po;
1206 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1213 case PACKET_FANOUT_HASH:
1216 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1220 skb_get_rxhash(skb);
1221 sk = fanout_demux_hash(f, skb, num);
1223 case PACKET_FANOUT_LB:
1224 sk = fanout_demux_lb(f, skb, num);
1226 case PACKET_FANOUT_CPU:
1227 sk = fanout_demux_cpu(f, skb, num);
1233 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1236 static DEFINE_MUTEX(fanout_mutex);
1237 static LIST_HEAD(fanout_list);
1239 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1241 struct packet_fanout *f = po->fanout;
1243 spin_lock(&f->lock);
1244 f->arr[f->num_members] = sk;
1247 if (f->num_members == 1)
1248 dev_add_pack(&f->prot_hook);
1249 spin_unlock(&f->lock);
1252 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1254 struct packet_fanout *f = po->fanout;
1257 spin_lock(&f->lock);
1258 for (i = 0; i < f->num_members; i++) {
1259 if (f->arr[i] == sk)
1262 BUG_ON(i >= f->num_members);
1263 f->arr[i] = f->arr[f->num_members - 1];
1265 if (f->num_members == 0)
1266 __dev_remove_pack(&f->prot_hook);
1267 spin_unlock(&f->lock);
1270 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1272 if (sk->sk_family != PF_PACKET)
1275 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1278 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1280 struct packet_sock *po = pkt_sk(sk);
1281 struct packet_fanout *f, *match;
1282 u8 type = type_flags & 0xff;
1283 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1287 case PACKET_FANOUT_HASH:
1288 case PACKET_FANOUT_LB:
1289 case PACKET_FANOUT_CPU:
1295 mutex_lock(&fanout_mutex);
1306 list_for_each_entry(f, &fanout_list, list) {
1308 read_pnet(&f->net) == sock_net(sk)) {
1314 if (match && match->defrag != defrag)
1318 match = kzalloc(sizeof(*match), GFP_KERNEL);
1321 write_pnet(&match->net, sock_net(sk));
1324 match->defrag = defrag;
1325 atomic_set(&match->rr_cur, 0);
1326 INIT_LIST_HEAD(&match->list);
1327 spin_lock_init(&match->lock);
1328 atomic_set(&match->sk_ref, 0);
1329 match->prot_hook.type = po->prot_hook.type;
1330 match->prot_hook.dev = po->prot_hook.dev;
1331 match->prot_hook.func = packet_rcv_fanout;
1332 match->prot_hook.af_packet_priv = match;
1333 match->prot_hook.id_match = match_fanout_group;
1334 list_add(&match->list, &fanout_list);
1337 if (match->type == type &&
1338 match->prot_hook.type == po->prot_hook.type &&
1339 match->prot_hook.dev == po->prot_hook.dev) {
1341 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1342 __dev_remove_pack(&po->prot_hook);
1344 atomic_inc(&match->sk_ref);
1345 __fanout_link(sk, po);
1350 mutex_unlock(&fanout_mutex);
1354 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1355 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1356 * It is the responsibility of the caller to call fanout_release_data() and
1357 * free the returned packet_fanout (after synchronize_net())
1359 static struct packet_fanout *fanout_release(struct sock *sk)
1361 struct packet_sock *po = pkt_sk(sk);
1362 struct packet_fanout *f;
1364 mutex_lock(&fanout_mutex);
1369 if (atomic_dec_and_test(&f->sk_ref))
1374 mutex_unlock(&fanout_mutex);
1379 static const struct proto_ops packet_ops;
1381 static const struct proto_ops packet_ops_spkt;
1383 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1384 struct packet_type *pt, struct net_device *orig_dev)
1387 struct sockaddr_pkt *spkt;
1390 * When we registered the protocol we saved the socket in the data
1391 * field for just this event.
1394 sk = pt->af_packet_priv;
1397 * Yank back the headers [hope the device set this
1398 * right or kerboom...]
1400 * Incoming packets have ll header pulled,
1403 * For outgoing ones skb->data == skb_mac_header(skb)
1404 * so that this procedure is noop.
1407 if (skb->pkt_type == PACKET_LOOPBACK)
1410 if (!net_eq(dev_net(dev), sock_net(sk)))
1413 skb = skb_share_check(skb, GFP_ATOMIC);
1417 /* drop any routing info */
1420 /* drop conntrack reference */
1423 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1425 skb_push(skb, skb->data - skb_mac_header(skb));
1428 * The SOCK_PACKET socket receives _all_ frames.
1431 spkt->spkt_family = dev->type;
1432 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1433 spkt->spkt_protocol = skb->protocol;
1436 * Charge the memory to the socket. This is done specifically
1437 * to prevent sockets using all the memory up.
1440 if (sock_queue_rcv_skb(sk, skb) == 0)
1451 * Output a raw packet to a device layer. This bypasses all the other
1452 * protocol layers and you must therefore supply it with a complete frame
1455 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1456 struct msghdr *msg, size_t len)
1458 struct sock *sk = sock->sk;
1459 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1460 struct sk_buff *skb = NULL;
1461 struct net_device *dev;
1466 * Get and verify the address.
1470 if (msg->msg_namelen < sizeof(struct sockaddr))
1472 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1473 proto = saddr->spkt_protocol;
1475 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1478 * Find the device first to size check it
1481 saddr->spkt_device[13] = 0;
1484 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1490 if (!(dev->flags & IFF_UP))
1494 * You may not queue a frame bigger than the mtu. This is the lowest level
1495 * raw protocol and you must do your own fragmentation at this level.
1499 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1503 size_t reserved = LL_RESERVED_SPACE(dev);
1504 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1507 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1510 /* FIXME: Save some space for broken drivers that write a hard
1511 * header at transmission time by themselves. PPP is the notable
1512 * one here. This should really be fixed at the driver level.
1514 skb_reserve(skb, reserved);
1515 skb_reset_network_header(skb);
1517 /* Try to align data part correctly */
1522 skb_reset_network_header(skb);
1524 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1530 if (len > (dev->mtu + dev->hard_header_len)) {
1531 /* Earlier code assumed this would be a VLAN pkt,
1532 * double-check this now that we have the actual
1535 struct ethhdr *ehdr;
1536 skb_reset_mac_header(skb);
1537 ehdr = eth_hdr(skb);
1538 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1544 skb->protocol = proto;
1546 skb->priority = sk->sk_priority;
1547 skb->mark = sk->sk_mark;
1548 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1552 dev_queue_xmit(skb);
1563 static unsigned int run_filter(const struct sk_buff *skb,
1564 const struct sock *sk,
1567 struct sk_filter *filter;
1570 filter = rcu_dereference(sk->sk_filter);
1572 res = SK_RUN_FILTER(filter, skb);
1579 * This function makes lazy skb cloning in hope that most of packets
1580 * are discarded by BPF.
1582 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1583 * and skb->cb are mangled. It works because (and until) packets
1584 * falling here are owned by current CPU. Output packets are cloned
1585 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1586 * sequencially, so that if we return skb to original state on exit,
1587 * we will not harm anyone.
1590 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1591 struct packet_type *pt, struct net_device *orig_dev)
1594 struct sockaddr_ll *sll;
1595 struct packet_sock *po;
1596 u8 *skb_head = skb->data;
1597 int skb_len = skb->len;
1598 unsigned int snaplen, res;
1600 if (skb->pkt_type == PACKET_LOOPBACK)
1603 sk = pt->af_packet_priv;
1606 if (!net_eq(dev_net(dev), sock_net(sk)))
1611 if (dev->header_ops) {
1612 /* The device has an explicit notion of ll header,
1613 * exported to higher levels.
1615 * Otherwise, the device hides details of its frame
1616 * structure, so that corresponding packet head is
1617 * never delivered to user.
1619 if (sk->sk_type != SOCK_DGRAM)
1620 skb_push(skb, skb->data - skb_mac_header(skb));
1621 else if (skb->pkt_type == PACKET_OUTGOING) {
1622 /* Special case: outgoing packets have ll header at head */
1623 skb_pull(skb, skb_network_offset(skb));
1629 res = run_filter(skb, sk, snaplen);
1631 goto drop_n_restore;
1635 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1638 if (skb_shared(skb)) {
1639 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1643 if (skb_head != skb->data) {
1644 skb->data = skb_head;
1651 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1654 sll = &PACKET_SKB_CB(skb)->sa.ll;
1655 sll->sll_family = AF_PACKET;
1656 sll->sll_hatype = dev->type;
1657 sll->sll_protocol = skb->protocol;
1658 sll->sll_pkttype = skb->pkt_type;
1659 if (unlikely(po->origdev))
1660 sll->sll_ifindex = orig_dev->ifindex;
1662 sll->sll_ifindex = dev->ifindex;
1664 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1666 PACKET_SKB_CB(skb)->origlen = skb->len;
1668 if (pskb_trim(skb, snaplen))
1671 skb_set_owner_r(skb, sk);
1675 /* drop conntrack reference */
1678 spin_lock(&sk->sk_receive_queue.lock);
1679 po->stats.tp_packets++;
1680 skb->dropcount = atomic_read(&sk->sk_drops);
1681 __skb_queue_tail(&sk->sk_receive_queue, skb);
1682 spin_unlock(&sk->sk_receive_queue.lock);
1683 sk->sk_data_ready(sk, skb->len);
1687 spin_lock(&sk->sk_receive_queue.lock);
1688 po->stats.tp_drops++;
1689 atomic_inc(&sk->sk_drops);
1690 spin_unlock(&sk->sk_receive_queue.lock);
1693 if (skb_head != skb->data && skb_shared(skb)) {
1694 skb->data = skb_head;
1702 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1703 struct packet_type *pt, struct net_device *orig_dev)
1706 struct packet_sock *po;
1707 struct sockaddr_ll *sll;
1709 struct tpacket_hdr *h1;
1710 struct tpacket2_hdr *h2;
1711 struct tpacket3_hdr *h3;
1714 u8 *skb_head = skb->data;
1715 int skb_len = skb->len;
1716 unsigned int snaplen, res;
1717 unsigned long status = TP_STATUS_USER;
1718 unsigned short macoff, netoff, hdrlen;
1719 struct sk_buff *copy_skb = NULL;
1722 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1724 if (skb->pkt_type == PACKET_LOOPBACK)
1727 sk = pt->af_packet_priv;
1730 if (!net_eq(dev_net(dev), sock_net(sk)))
1733 if (dev->header_ops) {
1734 if (sk->sk_type != SOCK_DGRAM)
1735 skb_push(skb, skb->data - skb_mac_header(skb));
1736 else if (skb->pkt_type == PACKET_OUTGOING) {
1737 /* Special case: outgoing packets have ll header at head */
1738 skb_pull(skb, skb_network_offset(skb));
1742 if (skb->ip_summed == CHECKSUM_PARTIAL)
1743 status |= TP_STATUS_CSUMNOTREADY;
1747 res = run_filter(skb, sk, snaplen);
1749 goto drop_n_restore;
1753 if (sk->sk_type == SOCK_DGRAM) {
1754 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1757 unsigned maclen = skb_network_offset(skb);
1758 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1759 (maclen < 16 ? 16 : maclen)) +
1761 macoff = netoff - maclen;
1763 if (po->tp_version <= TPACKET_V2) {
1764 if (macoff + snaplen > po->rx_ring.frame_size) {
1765 if (po->copy_thresh &&
1766 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1767 if (skb_shared(skb)) {
1768 copy_skb = skb_clone(skb, GFP_ATOMIC);
1770 copy_skb = skb_get(skb);
1771 skb_head = skb->data;
1774 skb_set_owner_r(copy_skb, sk);
1776 snaplen = po->rx_ring.frame_size - macoff;
1777 if ((int)snaplen < 0)
1780 } else if (unlikely(macoff + snaplen >
1781 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1784 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1785 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1786 snaplen, nval, macoff);
1788 if (unlikely((int)snaplen < 0)) {
1790 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1793 spin_lock(&sk->sk_receive_queue.lock);
1794 h.raw = packet_current_rx_frame(po, skb,
1795 TP_STATUS_KERNEL, (macoff+snaplen));
1798 if (po->tp_version <= TPACKET_V2) {
1799 packet_increment_rx_head(po, &po->rx_ring);
1801 * LOSING will be reported till you read the stats,
1802 * because it's COR - Clear On Read.
1803 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1806 if (po->stats.tp_drops)
1807 status |= TP_STATUS_LOSING;
1809 po->stats.tp_packets++;
1811 status |= TP_STATUS_COPY;
1812 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1814 spin_unlock(&sk->sk_receive_queue.lock);
1816 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1818 switch (po->tp_version) {
1820 h.h1->tp_len = skb->len;
1821 h.h1->tp_snaplen = snaplen;
1822 h.h1->tp_mac = macoff;
1823 h.h1->tp_net = netoff;
1824 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1825 && shhwtstamps->syststamp.tv64)
1826 tv = ktime_to_timeval(shhwtstamps->syststamp);
1827 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1828 && shhwtstamps->hwtstamp.tv64)
1829 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1830 else if (skb->tstamp.tv64)
1831 tv = ktime_to_timeval(skb->tstamp);
1833 do_gettimeofday(&tv);
1834 h.h1->tp_sec = tv.tv_sec;
1835 h.h1->tp_usec = tv.tv_usec;
1836 hdrlen = sizeof(*h.h1);
1839 h.h2->tp_len = skb->len;
1840 h.h2->tp_snaplen = snaplen;
1841 h.h2->tp_mac = macoff;
1842 h.h2->tp_net = netoff;
1843 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1844 && shhwtstamps->syststamp.tv64)
1845 ts = ktime_to_timespec(shhwtstamps->syststamp);
1846 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1847 && shhwtstamps->hwtstamp.tv64)
1848 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1849 else if (skb->tstamp.tv64)
1850 ts = ktime_to_timespec(skb->tstamp);
1852 getnstimeofday(&ts);
1853 h.h2->tp_sec = ts.tv_sec;
1854 h.h2->tp_nsec = ts.tv_nsec;
1855 if (vlan_tx_tag_present(skb)) {
1856 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1857 status |= TP_STATUS_VLAN_VALID;
1859 h.h2->tp_vlan_tci = 0;
1861 h.h2->tp_padding = 0;
1862 hdrlen = sizeof(*h.h2);
1865 /* tp_nxt_offset,vlan are already populated above.
1866 * So DONT clear those fields here
1868 h.h3->tp_status |= status;
1869 h.h3->tp_len = skb->len;
1870 h.h3->tp_snaplen = snaplen;
1871 h.h3->tp_mac = macoff;
1872 h.h3->tp_net = netoff;
1873 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1874 && shhwtstamps->syststamp.tv64)
1875 ts = ktime_to_timespec(shhwtstamps->syststamp);
1876 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1877 && shhwtstamps->hwtstamp.tv64)
1878 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1879 else if (skb->tstamp.tv64)
1880 ts = ktime_to_timespec(skb->tstamp);
1882 getnstimeofday(&ts);
1883 h.h3->tp_sec = ts.tv_sec;
1884 h.h3->tp_nsec = ts.tv_nsec;
1885 hdrlen = sizeof(*h.h3);
1891 sll = h.raw + TPACKET_ALIGN(hdrlen);
1892 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1893 sll->sll_family = AF_PACKET;
1894 sll->sll_hatype = dev->type;
1895 sll->sll_protocol = skb->protocol;
1896 sll->sll_pkttype = skb->pkt_type;
1897 if (unlikely(po->origdev))
1898 sll->sll_ifindex = orig_dev->ifindex;
1900 sll->sll_ifindex = dev->ifindex;
1903 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1907 if (po->tp_version <= TPACKET_V2) {
1908 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1909 + macoff + snaplen);
1910 for (start = h.raw; start < end; start += PAGE_SIZE)
1911 flush_dcache_page(pgv_to_page(start));
1916 if (po->tp_version <= TPACKET_V2)
1917 __packet_set_status(po, h.raw, status);
1919 prb_clear_blk_fill_status(&po->rx_ring);
1921 sk->sk_data_ready(sk, 0);
1924 if (skb_head != skb->data && skb_shared(skb)) {
1925 skb->data = skb_head;
1933 po->stats.tp_drops++;
1934 spin_unlock(&sk->sk_receive_queue.lock);
1936 sk->sk_data_ready(sk, 0);
1937 kfree_skb(copy_skb);
1938 goto drop_n_restore;
1941 static void tpacket_destruct_skb(struct sk_buff *skb)
1943 struct packet_sock *po = pkt_sk(skb->sk);
1946 if (likely(po->tx_ring.pg_vec)) {
1947 ph = skb_shinfo(skb)->destructor_arg;
1948 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1949 atomic_dec(&po->tx_ring.pending);
1950 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1956 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1957 void *frame, struct net_device *dev, int size_max,
1958 __be16 proto, unsigned char *addr)
1961 struct tpacket_hdr *h1;
1962 struct tpacket2_hdr *h2;
1965 int to_write, offset, len, tp_len, nr_frags, len_max;
1966 struct socket *sock = po->sk.sk_socket;
1973 skb->protocol = proto;
1975 skb->priority = po->sk.sk_priority;
1976 skb->mark = po->sk.sk_mark;
1977 skb_shinfo(skb)->destructor_arg = ph.raw;
1979 switch (po->tp_version) {
1981 tp_len = ph.h2->tp_len;
1984 tp_len = ph.h1->tp_len;
1987 if (unlikely(tp_len > size_max)) {
1988 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1992 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1993 skb_reset_network_header(skb);
1995 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1998 if (sock->type == SOCK_DGRAM) {
1999 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2001 if (unlikely(err < 0))
2003 } else if (dev->hard_header_len) {
2004 /* net device doesn't like empty head */
2005 if (unlikely(tp_len <= dev->hard_header_len)) {
2006 pr_err("packet size is too short (%d < %d)\n",
2007 tp_len, dev->hard_header_len);
2011 skb_push(skb, dev->hard_header_len);
2012 err = skb_store_bits(skb, 0, data,
2013 dev->hard_header_len);
2017 data += dev->hard_header_len;
2018 to_write -= dev->hard_header_len;
2022 offset = offset_in_page(data);
2023 len_max = PAGE_SIZE - offset;
2024 len = ((to_write > len_max) ? len_max : to_write);
2026 skb->data_len = to_write;
2027 skb->len += to_write;
2028 skb->truesize += to_write;
2029 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2031 while (likely(to_write)) {
2032 nr_frags = skb_shinfo(skb)->nr_frags;
2034 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2035 pr_err("Packet exceed the number of skb frags(%lu)\n",
2040 page = pgv_to_page(data);
2042 flush_dcache_page(page);
2044 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2047 len_max = PAGE_SIZE;
2048 len = ((to_write > len_max) ? len_max : to_write);
2054 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2056 struct net_device *dev;
2059 dev = rcu_dereference(po->cached_dev);
2067 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2069 struct sk_buff *skb;
2070 struct net_device *dev;
2072 int err, reserve = 0;
2074 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2075 int tp_len, size_max;
2076 unsigned char *addr;
2080 mutex_lock(&po->pg_vec_lock);
2083 if (saddr == NULL) {
2084 dev = packet_cached_dev_get(po);
2089 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2091 if (msg->msg_namelen < (saddr->sll_halen
2092 + offsetof(struct sockaddr_ll,
2095 proto = saddr->sll_protocol;
2096 addr = saddr->sll_addr;
2097 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2101 if (unlikely(dev == NULL))
2104 if (unlikely(!(dev->flags & IFF_UP)))
2107 reserve = dev->hard_header_len;
2109 size_max = po->tx_ring.frame_size
2110 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2112 if (size_max > dev->mtu + reserve)
2113 size_max = dev->mtu + reserve;
2116 ph = packet_current_frame(po, &po->tx_ring,
2117 TP_STATUS_SEND_REQUEST);
2119 if (unlikely(ph == NULL)) {
2124 status = TP_STATUS_SEND_REQUEST;
2125 skb = sock_alloc_send_skb(&po->sk,
2126 LL_ALLOCATED_SPACE(dev)
2127 + sizeof(struct sockaddr_ll),
2130 if (unlikely(skb == NULL))
2133 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2136 if (unlikely(tp_len < 0)) {
2138 __packet_set_status(po, ph,
2139 TP_STATUS_AVAILABLE);
2140 packet_increment_head(&po->tx_ring);
2144 status = TP_STATUS_WRONG_FORMAT;
2150 skb->destructor = tpacket_destruct_skb;
2151 __packet_set_status(po, ph, TP_STATUS_SENDING);
2152 atomic_inc(&po->tx_ring.pending);
2154 status = TP_STATUS_SEND_REQUEST;
2155 err = dev_queue_xmit(skb);
2156 if (unlikely(err > 0)) {
2157 err = net_xmit_errno(err);
2158 if (err && __packet_get_status(po, ph) ==
2159 TP_STATUS_AVAILABLE) {
2160 /* skb was destructed already */
2165 * skb was dropped but not destructed yet;
2166 * let's treat it like congestion or err < 0
2170 packet_increment_head(&po->tx_ring);
2172 } while (likely((ph != NULL) ||
2173 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2174 (atomic_read(&po->tx_ring.pending))))
2181 __packet_set_status(po, ph, status);
2186 mutex_unlock(&po->pg_vec_lock);
2190 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2191 size_t reserve, size_t len,
2192 size_t linear, int noblock,
2195 struct sk_buff *skb;
2197 /* Under a page? Don't bother with paged skb. */
2198 if (prepad + len < PAGE_SIZE || !linear)
2201 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2206 skb_reserve(skb, reserve);
2207 skb_put(skb, linear);
2208 skb->data_len = len - linear;
2209 skb->len += len - linear;
2214 static int packet_snd(struct socket *sock,
2215 struct msghdr *msg, size_t len)
2217 struct sock *sk = sock->sk;
2218 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2219 struct sk_buff *skb;
2220 struct net_device *dev;
2222 unsigned char *addr;
2223 int err, reserve = 0;
2224 struct virtio_net_hdr vnet_hdr = { 0 };
2227 struct packet_sock *po = pkt_sk(sk);
2228 unsigned short gso_type = 0;
2231 * Get and verify the address.
2234 if (saddr == NULL) {
2235 dev = packet_cached_dev_get(po);
2240 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2242 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2244 proto = saddr->sll_protocol;
2245 addr = saddr->sll_addr;
2246 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2250 if (unlikely(dev == NULL))
2253 if (unlikely(!(dev->flags & IFF_UP)))
2256 if (sock->type == SOCK_RAW)
2257 reserve = dev->hard_header_len;
2258 if (po->has_vnet_hdr) {
2259 vnet_hdr_len = sizeof(vnet_hdr);
2262 if (len < vnet_hdr_len)
2265 len -= vnet_hdr_len;
2267 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2272 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2273 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2275 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2276 vnet_hdr.csum_offset + 2;
2279 if (vnet_hdr.hdr_len > len)
2282 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2283 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2284 case VIRTIO_NET_HDR_GSO_TCPV4:
2285 gso_type = SKB_GSO_TCPV4;
2287 case VIRTIO_NET_HDR_GSO_TCPV6:
2288 gso_type = SKB_GSO_TCPV6;
2290 case VIRTIO_NET_HDR_GSO_UDP:
2291 gso_type = SKB_GSO_UDP;
2297 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2298 gso_type |= SKB_GSO_TCP_ECN;
2300 if (vnet_hdr.gso_size == 0)
2307 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2311 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2312 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2313 msg->msg_flags & MSG_DONTWAIT, &err);
2317 skb_set_network_header(skb, reserve);
2320 if (sock->type == SOCK_DGRAM &&
2321 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2324 /* Returns -EFAULT on error */
2325 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2328 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2332 if (!gso_type && (len > dev->mtu + reserve)) {
2333 /* Earlier code assumed this would be a VLAN pkt,
2334 * double-check this now that we have the actual
2337 struct ethhdr *ehdr;
2338 skb_reset_mac_header(skb);
2339 ehdr = eth_hdr(skb);
2340 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2346 skb->protocol = proto;
2348 skb->priority = sk->sk_priority;
2349 skb->mark = sk->sk_mark;
2351 if (po->has_vnet_hdr) {
2352 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2353 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2354 vnet_hdr.csum_offset)) {
2360 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2361 skb_shinfo(skb)->gso_type = gso_type;
2363 /* Header must be checked, and gso_segs computed. */
2364 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2365 skb_shinfo(skb)->gso_segs = 0;
2367 len += vnet_hdr_len;
2374 err = dev_queue_xmit(skb);
2375 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2391 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2392 struct msghdr *msg, size_t len)
2394 struct sock *sk = sock->sk;
2395 struct packet_sock *po = pkt_sk(sk);
2396 if (po->tx_ring.pg_vec)
2397 return tpacket_snd(po, msg);
2399 return packet_snd(sock, msg, len);
2403 * Close a PACKET socket. This is fairly simple. We immediately go
2404 * to 'closed' state and remove our protocol entry in the device list.
2407 static int packet_release(struct socket *sock)
2409 struct sock *sk = sock->sk;
2410 struct packet_sock *po;
2411 struct packet_fanout *f;
2413 union tpacket_req_u req_u;
2421 spin_lock_bh(&net->packet.sklist_lock);
2422 sk_del_node_init_rcu(sk);
2423 sock_prot_inuse_add(net, sk->sk_prot, -1);
2424 spin_unlock_bh(&net->packet.sklist_lock);
2426 spin_lock(&po->bind_lock);
2427 unregister_prot_hook(sk, false);
2428 if (po->prot_hook.dev) {
2429 dev_put(po->prot_hook.dev);
2430 po->prot_hook.dev = NULL;
2432 spin_unlock(&po->bind_lock);
2434 packet_flush_mclist(sk);
2436 if (po->rx_ring.pg_vec) {
2437 memset(&req_u, 0, sizeof(req_u));
2438 packet_set_ring(sk, &req_u, 1, 0);
2441 if (po->tx_ring.pg_vec) {
2442 memset(&req_u, 0, sizeof(req_u));
2443 packet_set_ring(sk, &req_u, 1, 1);
2446 f = fanout_release(sk);
2453 * Now the socket is dead. No more input will appear.
2460 skb_queue_purge(&sk->sk_receive_queue);
2461 sk_refcnt_debug_release(sk);
2468 * Attach a packet hook.
2471 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2474 struct packet_sock *po = pkt_sk(sk);
2475 struct net_device *dev_curr;
2476 struct net_device *dev = NULL;
2478 bool unlisted = false;
2485 spin_lock(&po->bind_lock);
2489 dev = dev_get_by_name_rcu(sock_net(sk), name);
2494 } else if (ifindex) {
2495 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2505 dev_curr = po->prot_hook.dev;
2509 __unregister_prot_hook(sk, true);
2511 dev_curr = po->prot_hook.dev;
2513 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2517 po->prot_hook.type = protocol;
2519 if (unlikely(unlisted)) {
2521 po->prot_hook.dev = NULL;
2524 po->prot_hook.dev = dev;
2525 po->ifindex = dev ? dev->ifindex : 0;
2534 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2535 register_prot_hook(sk);
2537 sk->sk_err = ENETDOWN;
2538 if (!sock_flag(sk, SOCK_DEAD))
2539 sk->sk_error_report(sk);
2544 spin_unlock(&po->bind_lock);
2550 * Bind a packet socket to a device
2553 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2556 struct sock *sk = sock->sk;
2563 if (addr_len != sizeof(struct sockaddr))
2565 strlcpy(name, uaddr->sa_data, sizeof(name));
2567 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
2570 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2572 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2573 struct sock *sk = sock->sk;
2579 if (addr_len < sizeof(struct sockaddr_ll))
2581 if (sll->sll_family != AF_PACKET)
2584 return packet_do_bind(sk, NULL, sll->sll_ifindex,
2585 sll->sll_protocol ? : pkt_sk(sk)->num);
2588 static struct proto packet_proto = {
2590 .owner = THIS_MODULE,
2591 .obj_size = sizeof(struct packet_sock),
2595 * Create a packet of type SOCK_PACKET.
2598 static int packet_create(struct net *net, struct socket *sock, int protocol,
2602 struct packet_sock *po;
2603 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2606 if (!capable(CAP_NET_RAW))
2608 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2609 sock->type != SOCK_PACKET)
2610 return -ESOCKTNOSUPPORT;
2612 sock->state = SS_UNCONNECTED;
2615 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2619 sock->ops = &packet_ops;
2620 if (sock->type == SOCK_PACKET)
2621 sock->ops = &packet_ops_spkt;
2623 sock_init_data(sock, sk);
2626 sk->sk_family = PF_PACKET;
2628 RCU_INIT_POINTER(po->cached_dev, NULL);
2630 sk->sk_destruct = packet_sock_destruct;
2631 sk_refcnt_debug_inc(sk);
2634 * Attach a protocol block
2637 spin_lock_init(&po->bind_lock);
2638 mutex_init(&po->pg_vec_lock);
2639 po->prot_hook.func = packet_rcv;
2641 if (sock->type == SOCK_PACKET)
2642 po->prot_hook.func = packet_rcv_spkt;
2644 po->prot_hook.af_packet_priv = sk;
2647 po->prot_hook.type = proto;
2648 register_prot_hook(sk);
2651 spin_lock_bh(&net->packet.sklist_lock);
2652 sk_add_node_rcu(sk, &net->packet.sklist);
2653 sock_prot_inuse_add(net, &packet_proto, 1);
2654 spin_unlock_bh(&net->packet.sklist_lock);
2661 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2663 struct sock_exterr_skb *serr;
2664 struct sk_buff *skb, *skb2;
2668 skb = skb_dequeue(&sk->sk_error_queue);
2674 msg->msg_flags |= MSG_TRUNC;
2677 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2681 sock_recv_timestamp(msg, sk, skb);
2683 serr = SKB_EXT_ERR(skb);
2684 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2685 sizeof(serr->ee), &serr->ee);
2687 msg->msg_flags |= MSG_ERRQUEUE;
2690 /* Reset and regenerate socket error */
2691 spin_lock_bh(&sk->sk_error_queue.lock);
2693 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2694 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2695 spin_unlock_bh(&sk->sk_error_queue.lock);
2696 sk->sk_error_report(sk);
2698 spin_unlock_bh(&sk->sk_error_queue.lock);
2707 * Pull a packet from our receive queue and hand it to the user.
2708 * If necessary we block.
2711 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2712 struct msghdr *msg, size_t len, int flags)
2714 struct sock *sk = sock->sk;
2715 struct sk_buff *skb;
2717 int vnet_hdr_len = 0;
2720 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2724 /* What error should we return now? EUNATTACH? */
2725 if (pkt_sk(sk)->ifindex < 0)
2729 if (flags & MSG_ERRQUEUE) {
2730 err = packet_recv_error(sk, msg, len);
2735 * Call the generic datagram receiver. This handles all sorts
2736 * of horrible races and re-entrancy so we can forget about it
2737 * in the protocol layers.
2739 * Now it will return ENETDOWN, if device have just gone down,
2740 * but then it will block.
2743 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2746 * An error occurred so return it. Because skb_recv_datagram()
2747 * handles the blocking we don't see and worry about blocking
2754 if (pkt_sk(sk)->has_vnet_hdr) {
2755 struct virtio_net_hdr vnet_hdr = { 0 };
2758 vnet_hdr_len = sizeof(vnet_hdr);
2759 if (len < vnet_hdr_len)
2762 len -= vnet_hdr_len;
2764 if (skb_is_gso(skb)) {
2765 struct skb_shared_info *sinfo = skb_shinfo(skb);
2767 /* This is a hint as to how much should be linear. */
2768 vnet_hdr.hdr_len = skb_headlen(skb);
2769 vnet_hdr.gso_size = sinfo->gso_size;
2770 if (sinfo->gso_type & SKB_GSO_TCPV4)
2771 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2772 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2773 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2774 else if (sinfo->gso_type & SKB_GSO_UDP)
2775 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2776 else if (sinfo->gso_type & SKB_GSO_FCOE)
2780 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2781 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2783 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2785 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2786 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2787 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2788 vnet_hdr.csum_offset = skb->csum_offset;
2789 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2790 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2791 } /* else everything is zero */
2793 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2799 /* You lose any data beyond the buffer you gave. If it worries
2800 * a user program they can ask the device for its MTU
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);
2815 if (msg->msg_name) {
2816 /* If the address length field is there to be filled
2817 * in, we fill it in now.
2819 if (sock->type == SOCK_PACKET) {
2820 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2822 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2823 msg->msg_namelen = sll->sll_halen +
2824 offsetof(struct sockaddr_ll, sll_addr);
2826 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2830 if (pkt_sk(sk)->auxdata) {
2831 struct tpacket_auxdata aux;
2833 aux.tp_status = TP_STATUS_USER;
2834 if (skb->ip_summed == CHECKSUM_PARTIAL)
2835 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2836 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2837 aux.tp_snaplen = skb->len;
2839 aux.tp_net = skb_network_offset(skb);
2840 if (vlan_tx_tag_present(skb)) {
2841 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2842 aux.tp_status |= TP_STATUS_VLAN_VALID;
2844 aux.tp_vlan_tci = 0;
2847 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2851 * Free or return the buffer as appropriate. Again this
2852 * hides all the races and re-entrancy issues from us.
2854 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2857 skb_free_datagram(sk, skb);
2862 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2863 int *uaddr_len, int peer)
2865 struct net_device *dev;
2866 struct sock *sk = sock->sk;
2871 uaddr->sa_family = AF_PACKET;
2872 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2874 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2876 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2878 *uaddr_len = sizeof(*uaddr);
2883 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2884 int *uaddr_len, int peer)
2886 struct net_device *dev;
2887 struct sock *sk = sock->sk;
2888 struct packet_sock *po = pkt_sk(sk);
2889 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2894 sll->sll_family = AF_PACKET;
2895 sll->sll_ifindex = po->ifindex;
2896 sll->sll_protocol = po->num;
2897 sll->sll_pkttype = 0;
2899 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2901 sll->sll_hatype = dev->type;
2902 sll->sll_halen = dev->addr_len;
2903 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2905 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2909 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2914 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2918 case PACKET_MR_MULTICAST:
2919 if (i->alen != dev->addr_len)
2922 return dev_mc_add(dev, i->addr);
2924 return dev_mc_del(dev, i->addr);
2926 case PACKET_MR_PROMISC:
2927 return dev_set_promiscuity(dev, what);
2929 case PACKET_MR_ALLMULTI:
2930 return dev_set_allmulti(dev, what);
2932 case PACKET_MR_UNICAST:
2933 if (i->alen != dev->addr_len)
2936 return dev_uc_add(dev, i->addr);
2938 return dev_uc_del(dev, i->addr);
2946 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2948 for ( ; i; i = i->next) {
2949 if (i->ifindex == dev->ifindex)
2950 packet_dev_mc(dev, i, what);
2954 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2956 struct packet_sock *po = pkt_sk(sk);
2957 struct packet_mclist *ml, *i;
2958 struct net_device *dev;
2964 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2969 if (mreq->mr_alen > dev->addr_len)
2973 i = kmalloc(sizeof(*i), GFP_KERNEL);
2978 for (ml = po->mclist; ml; ml = ml->next) {
2979 if (ml->ifindex == mreq->mr_ifindex &&
2980 ml->type == mreq->mr_type &&
2981 ml->alen == mreq->mr_alen &&
2982 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2984 /* Free the new element ... */
2990 i->type = mreq->mr_type;
2991 i->ifindex = mreq->mr_ifindex;
2992 i->alen = mreq->mr_alen;
2993 memcpy(i->addr, mreq->mr_address, i->alen);
2995 i->next = po->mclist;
2997 err = packet_dev_mc(dev, i, 1);
2999 po->mclist = i->next;
3008 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3010 struct packet_mclist *ml, **mlp;
3014 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3015 if (ml->ifindex == mreq->mr_ifindex &&
3016 ml->type == mreq->mr_type &&
3017 ml->alen == mreq->mr_alen &&
3018 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3019 if (--ml->count == 0) {
3020 struct net_device *dev;
3022 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3024 packet_dev_mc(dev, ml, -1);
3032 return -EADDRNOTAVAIL;
3035 static void packet_flush_mclist(struct sock *sk)
3037 struct packet_sock *po = pkt_sk(sk);
3038 struct packet_mclist *ml;
3044 while ((ml = po->mclist) != NULL) {
3045 struct net_device *dev;
3047 po->mclist = ml->next;
3048 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3050 packet_dev_mc(dev, ml, -1);
3057 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3059 struct sock *sk = sock->sk;
3060 struct packet_sock *po = pkt_sk(sk);
3063 if (level != SOL_PACKET)
3064 return -ENOPROTOOPT;
3067 case PACKET_ADD_MEMBERSHIP:
3068 case PACKET_DROP_MEMBERSHIP:
3070 struct packet_mreq_max mreq;
3072 memset(&mreq, 0, sizeof(mreq));
3073 if (len < sizeof(struct packet_mreq))
3075 if (len > sizeof(mreq))
3077 if (copy_from_user(&mreq, optval, len))
3079 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3081 if (optname == PACKET_ADD_MEMBERSHIP)
3082 ret = packet_mc_add(sk, &mreq);
3084 ret = packet_mc_drop(sk, &mreq);
3088 case PACKET_RX_RING:
3089 case PACKET_TX_RING:
3091 union tpacket_req_u req_u;
3094 switch (po->tp_version) {
3097 len = sizeof(req_u.req);
3101 len = sizeof(req_u.req3);
3106 if (pkt_sk(sk)->has_vnet_hdr)
3108 if (copy_from_user(&req_u.req, optval, len))
3110 return packet_set_ring(sk, &req_u, 0,
3111 optname == PACKET_TX_RING);
3113 case PACKET_COPY_THRESH:
3117 if (optlen != sizeof(val))
3119 if (copy_from_user(&val, optval, sizeof(val)))
3122 pkt_sk(sk)->copy_thresh = val;
3125 case PACKET_VERSION:
3129 if (optlen != sizeof(val))
3131 if (copy_from_user(&val, optval, sizeof(val)))
3142 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3145 po->tp_version = val;
3151 case PACKET_RESERVE:
3155 if (optlen != sizeof(val))
3157 if (copy_from_user(&val, optval, sizeof(val)))
3162 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3165 po->tp_reserve = val;
3175 if (optlen != sizeof(val))
3177 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3179 if (copy_from_user(&val, optval, sizeof(val)))
3181 po->tp_loss = !!val;
3184 case PACKET_AUXDATA:
3188 if (optlen < sizeof(val))
3190 if (copy_from_user(&val, optval, sizeof(val)))
3193 po->auxdata = !!val;
3196 case PACKET_ORIGDEV:
3200 if (optlen < sizeof(val))
3202 if (copy_from_user(&val, optval, sizeof(val)))
3205 po->origdev = !!val;
3208 case PACKET_VNET_HDR:
3212 if (sock->type != SOCK_RAW)
3214 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3216 if (optlen < sizeof(val))
3218 if (copy_from_user(&val, optval, sizeof(val)))
3221 po->has_vnet_hdr = !!val;
3224 case PACKET_TIMESTAMP:
3228 if (optlen != sizeof(val))
3230 if (copy_from_user(&val, optval, sizeof(val)))
3233 po->tp_tstamp = val;
3240 if (optlen != sizeof(val))
3242 if (copy_from_user(&val, optval, sizeof(val)))
3245 return fanout_add(sk, val & 0xffff, val >> 16);
3248 return -ENOPROTOOPT;
3252 static int packet_getsockopt(struct socket *sock, int level, int optname,
3253 char __user *optval, int __user *optlen)
3257 struct sock *sk = sock->sk;
3258 struct packet_sock *po = pkt_sk(sk);
3260 struct tpacket_stats st;
3261 union tpacket_stats_u st_u;
3263 if (level != SOL_PACKET)
3264 return -ENOPROTOOPT;
3266 if (get_user(len, optlen))
3273 case PACKET_STATISTICS:
3274 if (po->tp_version == TPACKET_V3) {
3275 len = sizeof(struct tpacket_stats_v3);
3277 if (len > sizeof(struct tpacket_stats))
3278 len = sizeof(struct tpacket_stats);
3280 spin_lock_bh(&sk->sk_receive_queue.lock);
3281 if (po->tp_version == TPACKET_V3) {
3282 memcpy(&st_u.stats3, &po->stats,
3283 sizeof(struct tpacket_stats));
3284 st_u.stats3.tp_freeze_q_cnt =
3285 po->stats_u.stats3.tp_freeze_q_cnt;
3286 st_u.stats3.tp_packets += po->stats.tp_drops;
3287 data = &st_u.stats3;
3290 st.tp_packets += st.tp_drops;
3293 memset(&po->stats, 0, sizeof(st));
3294 spin_unlock_bh(&sk->sk_receive_queue.lock);
3296 case PACKET_AUXDATA:
3297 if (len > sizeof(int))
3303 case PACKET_ORIGDEV:
3304 if (len > sizeof(int))
3310 case PACKET_VNET_HDR:
3311 if (len > sizeof(int))
3313 val = po->has_vnet_hdr;
3317 case PACKET_VERSION:
3318 if (len > sizeof(int))
3320 val = po->tp_version;
3324 if (len > sizeof(int))
3326 if (copy_from_user(&val, optval, len))
3330 val = sizeof(struct tpacket_hdr);
3333 val = sizeof(struct tpacket2_hdr);
3336 val = sizeof(struct tpacket3_hdr);
3343 case PACKET_RESERVE:
3344 if (len > sizeof(unsigned int))
3345 len = sizeof(unsigned int);
3346 val = po->tp_reserve;
3350 if (len > sizeof(unsigned int))
3351 len = sizeof(unsigned int);
3355 case PACKET_TIMESTAMP:
3356 if (len > sizeof(int))
3358 val = po->tp_tstamp;
3362 if (len > sizeof(int))
3365 ((u32)po->fanout->id |
3366 ((u32)po->fanout->type << 16)) :
3371 return -ENOPROTOOPT;
3374 if (put_user(len, optlen))
3376 if (copy_to_user(optval, data, len))
3382 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3385 struct hlist_node *node;
3386 struct net_device *dev = data;
3387 struct net *net = dev_net(dev);
3390 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3391 struct packet_sock *po = pkt_sk(sk);
3394 case NETDEV_UNREGISTER:
3396 packet_dev_mclist(dev, po->mclist, -1);
3400 if (dev->ifindex == po->ifindex) {
3401 spin_lock(&po->bind_lock);
3403 __unregister_prot_hook(sk, false);
3404 sk->sk_err = ENETDOWN;
3405 if (!sock_flag(sk, SOCK_DEAD))
3406 sk->sk_error_report(sk);
3408 if (msg == NETDEV_UNREGISTER) {
3410 if (po->prot_hook.dev)
3411 dev_put(po->prot_hook.dev);
3412 po->prot_hook.dev = NULL;
3414 spin_unlock(&po->bind_lock);
3418 if (dev->ifindex == po->ifindex) {
3419 spin_lock(&po->bind_lock);
3421 register_prot_hook(sk);
3422 spin_unlock(&po->bind_lock);
3432 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3435 struct sock *sk = sock->sk;
3440 int amount = sk_wmem_alloc_get(sk);
3442 return put_user(amount, (int __user *)arg);
3446 struct sk_buff *skb;
3449 spin_lock_bh(&sk->sk_receive_queue.lock);
3450 skb = skb_peek(&sk->sk_receive_queue);
3453 spin_unlock_bh(&sk->sk_receive_queue.lock);
3454 return put_user(amount, (int __user *)arg);
3457 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3459 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3469 case SIOCGIFBRDADDR:
3470 case SIOCSIFBRDADDR:
3471 case SIOCGIFNETMASK:
3472 case SIOCSIFNETMASK:
3473 case SIOCGIFDSTADDR:
3474 case SIOCSIFDSTADDR:
3476 return inet_dgram_ops.ioctl(sock, cmd, arg);
3480 return -ENOIOCTLCMD;
3485 static unsigned int packet_poll(struct file *file, struct socket *sock,
3488 struct sock *sk = sock->sk;
3489 struct packet_sock *po = pkt_sk(sk);
3490 unsigned int mask = datagram_poll(file, sock, wait);
3492 spin_lock_bh(&sk->sk_receive_queue.lock);
3493 if (po->rx_ring.pg_vec) {
3494 if (!packet_previous_rx_frame(po, &po->rx_ring,
3496 mask |= POLLIN | POLLRDNORM;
3498 spin_unlock_bh(&sk->sk_receive_queue.lock);
3499 spin_lock_bh(&sk->sk_write_queue.lock);
3500 if (po->tx_ring.pg_vec) {
3501 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3502 mask |= POLLOUT | POLLWRNORM;
3504 spin_unlock_bh(&sk->sk_write_queue.lock);
3509 /* Dirty? Well, I still did not learn better way to account
3513 static void packet_mm_open(struct vm_area_struct *vma)
3515 struct file *file = vma->vm_file;
3516 struct socket *sock = file->private_data;
3517 struct sock *sk = sock->sk;
3520 atomic_inc(&pkt_sk(sk)->mapped);
3523 static void packet_mm_close(struct vm_area_struct *vma)
3525 struct file *file = vma->vm_file;
3526 struct socket *sock = file->private_data;
3527 struct sock *sk = sock->sk;
3530 atomic_dec(&pkt_sk(sk)->mapped);
3533 static const struct vm_operations_struct packet_mmap_ops = {
3534 .open = packet_mm_open,
3535 .close = packet_mm_close,
3538 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3543 for (i = 0; i < len; i++) {
3544 if (likely(pg_vec[i].buffer)) {
3545 if (is_vmalloc_addr(pg_vec[i].buffer))
3546 vfree(pg_vec[i].buffer);
3548 free_pages((unsigned long)pg_vec[i].buffer,
3550 pg_vec[i].buffer = NULL;
3556 static char *alloc_one_pg_vec_page(unsigned long order)
3558 char *buffer = NULL;
3559 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3560 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3562 buffer = (char *) __get_free_pages(gfp_flags, order);
3568 * __get_free_pages failed, fall back to vmalloc
3570 buffer = vzalloc((1 << order) * PAGE_SIZE);
3576 * vmalloc failed, lets dig into swap here
3578 gfp_flags &= ~__GFP_NORETRY;
3579 buffer = (char *)__get_free_pages(gfp_flags, order);
3584 * complete and utter failure
3589 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3591 unsigned int block_nr = req->tp_block_nr;
3595 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3596 if (unlikely(!pg_vec))
3599 for (i = 0; i < block_nr; i++) {
3600 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3601 if (unlikely(!pg_vec[i].buffer))
3602 goto out_free_pgvec;
3609 free_pg_vec(pg_vec, order, block_nr);
3614 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3615 int closing, int tx_ring)
3617 struct pgv *pg_vec = NULL;
3618 struct packet_sock *po = pkt_sk(sk);
3619 int was_running, order = 0;
3620 struct packet_ring_buffer *rb;
3621 struct sk_buff_head *rb_queue;
3624 /* Added to avoid minimal code churn */
3625 struct tpacket_req *req = &req_u->req;
3628 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3629 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3630 WARN(1, "Tx-ring is not supported.\n");
3634 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3635 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3639 if (atomic_read(&po->mapped))
3641 if (atomic_read(&rb->pending))
3645 if (req->tp_block_nr) {
3646 /* Sanity tests and some calculations */
3648 if (unlikely(rb->pg_vec))
3651 switch (po->tp_version) {
3653 po->tp_hdrlen = TPACKET_HDRLEN;
3656 po->tp_hdrlen = TPACKET2_HDRLEN;
3659 po->tp_hdrlen = TPACKET3_HDRLEN;
3664 if (unlikely((int)req->tp_block_size <= 0))
3666 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3668 if (po->tp_version >= TPACKET_V3 &&
3669 req->tp_block_size <=
3670 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
3672 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3675 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3678 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3679 if (unlikely(rb->frames_per_block <= 0))
3681 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
3683 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3688 order = get_order(req->tp_block_size);
3689 pg_vec = alloc_pg_vec(req, order);
3690 if (unlikely(!pg_vec))
3692 switch (po->tp_version) {
3694 /* Transmit path is not supported. We checked
3695 * it above but just being paranoid
3698 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3707 if (unlikely(req->tp_frame_nr))
3712 /* Detach socket from network */
3713 spin_lock(&po->bind_lock);
3714 was_running = po->running;
3718 __unregister_prot_hook(sk, false);
3720 spin_unlock(&po->bind_lock);
3725 mutex_lock(&po->pg_vec_lock);
3726 if (closing || atomic_read(&po->mapped) == 0) {
3728 spin_lock_bh(&rb_queue->lock);
3729 swap(rb->pg_vec, pg_vec);
3730 rb->frame_max = (req->tp_frame_nr - 1);
3732 rb->frame_size = req->tp_frame_size;
3733 spin_unlock_bh(&rb_queue->lock);
3735 swap(rb->pg_vec_order, order);
3736 swap(rb->pg_vec_len, req->tp_block_nr);
3738 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3739 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3740 tpacket_rcv : packet_rcv;
3741 skb_queue_purge(rb_queue);
3742 if (atomic_read(&po->mapped))
3743 pr_err("packet_mmap: vma is busy: %d\n",
3744 atomic_read(&po->mapped));
3746 mutex_unlock(&po->pg_vec_lock);
3748 spin_lock(&po->bind_lock);
3751 register_prot_hook(sk);
3753 spin_unlock(&po->bind_lock);
3754 if (closing && (po->tp_version > TPACKET_V2)) {
3755 /* Because we don't support block-based V3 on tx-ring */
3757 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3761 free_pg_vec(pg_vec, order, req->tp_block_nr);
3767 static int packet_mmap(struct file *file, struct socket *sock,
3768 struct vm_area_struct *vma)
3770 struct sock *sk = sock->sk;
3771 struct packet_sock *po = pkt_sk(sk);
3772 unsigned long size, expected_size;
3773 struct packet_ring_buffer *rb;
3774 unsigned long start;
3781 mutex_lock(&po->pg_vec_lock);
3784 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3786 expected_size += rb->pg_vec_len
3792 if (expected_size == 0)
3795 size = vma->vm_end - vma->vm_start;
3796 if (size != expected_size)
3799 start = vma->vm_start;
3800 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3801 if (rb->pg_vec == NULL)
3804 for (i = 0; i < rb->pg_vec_len; i++) {
3806 void *kaddr = rb->pg_vec[i].buffer;
3809 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3810 page = pgv_to_page(kaddr);
3811 err = vm_insert_page(vma, start, page);
3820 atomic_inc(&po->mapped);
3821 vma->vm_ops = &packet_mmap_ops;
3825 mutex_unlock(&po->pg_vec_lock);
3829 static const struct proto_ops packet_ops_spkt = {
3830 .family = PF_PACKET,
3831 .owner = THIS_MODULE,
3832 .release = packet_release,
3833 .bind = packet_bind_spkt,
3834 .connect = sock_no_connect,
3835 .socketpair = sock_no_socketpair,
3836 .accept = sock_no_accept,
3837 .getname = packet_getname_spkt,
3838 .poll = datagram_poll,
3839 .ioctl = packet_ioctl,
3840 .listen = sock_no_listen,
3841 .shutdown = sock_no_shutdown,
3842 .setsockopt = sock_no_setsockopt,
3843 .getsockopt = sock_no_getsockopt,
3844 .sendmsg = packet_sendmsg_spkt,
3845 .recvmsg = packet_recvmsg,
3846 .mmap = sock_no_mmap,
3847 .sendpage = sock_no_sendpage,
3850 static const struct proto_ops packet_ops = {
3851 .family = PF_PACKET,
3852 .owner = THIS_MODULE,
3853 .release = packet_release,
3854 .bind = packet_bind,
3855 .connect = sock_no_connect,
3856 .socketpair = sock_no_socketpair,
3857 .accept = sock_no_accept,
3858 .getname = packet_getname,
3859 .poll = packet_poll,
3860 .ioctl = packet_ioctl,
3861 .listen = sock_no_listen,
3862 .shutdown = sock_no_shutdown,
3863 .setsockopt = packet_setsockopt,
3864 .getsockopt = packet_getsockopt,
3865 .sendmsg = packet_sendmsg,
3866 .recvmsg = packet_recvmsg,
3867 .mmap = packet_mmap,
3868 .sendpage = sock_no_sendpage,
3871 static const struct net_proto_family packet_family_ops = {
3872 .family = PF_PACKET,
3873 .create = packet_create,
3874 .owner = THIS_MODULE,
3877 static struct notifier_block packet_netdev_notifier = {
3878 .notifier_call = packet_notifier,
3881 #ifdef CONFIG_PROC_FS
3883 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3886 struct net *net = seq_file_net(seq);
3889 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3892 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3894 struct net *net = seq_file_net(seq);
3895 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3898 static void packet_seq_stop(struct seq_file *seq, void *v)
3904 static int packet_seq_show(struct seq_file *seq, void *v)
3906 if (v == SEQ_START_TOKEN)
3907 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3909 struct sock *s = sk_entry(v);
3910 const struct packet_sock *po = pkt_sk(s);
3913 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3915 atomic_read(&s->sk_refcnt),
3920 atomic_read(&s->sk_rmem_alloc),
3928 static const struct seq_operations packet_seq_ops = {
3929 .start = packet_seq_start,
3930 .next = packet_seq_next,
3931 .stop = packet_seq_stop,
3932 .show = packet_seq_show,
3935 static int packet_seq_open(struct inode *inode, struct file *file)
3937 return seq_open_net(inode, file, &packet_seq_ops,
3938 sizeof(struct seq_net_private));
3941 static const struct file_operations packet_seq_fops = {
3942 .owner = THIS_MODULE,
3943 .open = packet_seq_open,
3945 .llseek = seq_lseek,
3946 .release = seq_release_net,
3951 static int __net_init packet_net_init(struct net *net)
3953 spin_lock_init(&net->packet.sklist_lock);
3954 INIT_HLIST_HEAD(&net->packet.sklist);
3956 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3962 static void __net_exit packet_net_exit(struct net *net)
3964 proc_net_remove(net, "packet");
3967 static struct pernet_operations packet_net_ops = {
3968 .init = packet_net_init,
3969 .exit = packet_net_exit,
3973 static void __exit packet_exit(void)
3975 unregister_netdevice_notifier(&packet_netdev_notifier);
3976 unregister_pernet_subsys(&packet_net_ops);
3977 sock_unregister(PF_PACKET);
3978 proto_unregister(&packet_proto);
3981 static int __init packet_init(void)
3983 int rc = proto_register(&packet_proto, 0);
3988 sock_register(&packet_family_ops);
3989 register_pernet_subsys(&packet_net_ops);
3990 register_netdevice_notifier(&packet_netdev_notifier);
3995 module_init(packet_init);
3996 module_exit(packet_exit);
3997 MODULE_LICENSE("GPL");
3998 MODULE_ALIAS_NETPROTO(PF_PACKET);