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);
1302 list_for_each_entry(f, &fanout_list, list) {
1304 read_pnet(&f->net) == sock_net(sk)) {
1310 if (match && match->defrag != defrag)
1314 match = kzalloc(sizeof(*match), GFP_KERNEL);
1317 write_pnet(&match->net, sock_net(sk));
1320 match->defrag = defrag;
1321 atomic_set(&match->rr_cur, 0);
1322 INIT_LIST_HEAD(&match->list);
1323 spin_lock_init(&match->lock);
1324 atomic_set(&match->sk_ref, 0);
1325 match->prot_hook.type = po->prot_hook.type;
1326 match->prot_hook.dev = po->prot_hook.dev;
1327 match->prot_hook.func = packet_rcv_fanout;
1328 match->prot_hook.af_packet_priv = match;
1329 match->prot_hook.id_match = match_fanout_group;
1330 list_add(&match->list, &fanout_list);
1334 spin_lock(&po->bind_lock);
1336 match->type == type &&
1337 match->prot_hook.type == po->prot_hook.type &&
1338 match->prot_hook.dev == po->prot_hook.dev) {
1340 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1341 __dev_remove_pack(&po->prot_hook);
1343 atomic_inc(&match->sk_ref);
1344 __fanout_link(sk, po);
1348 spin_unlock(&po->bind_lock);
1350 if (err && !atomic_read(&match->sk_ref)) {
1351 list_del(&match->list);
1356 mutex_unlock(&fanout_mutex);
1360 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1361 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1362 * It is the responsibility of the caller to call fanout_release_data() and
1363 * free the returned packet_fanout (after synchronize_net())
1365 static struct packet_fanout *fanout_release(struct sock *sk)
1367 struct packet_sock *po = pkt_sk(sk);
1368 struct packet_fanout *f;
1370 mutex_lock(&fanout_mutex);
1375 if (atomic_dec_and_test(&f->sk_ref))
1380 mutex_unlock(&fanout_mutex);
1385 static const struct proto_ops packet_ops;
1387 static const struct proto_ops packet_ops_spkt;
1389 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1390 struct packet_type *pt, struct net_device *orig_dev)
1393 struct sockaddr_pkt *spkt;
1396 * When we registered the protocol we saved the socket in the data
1397 * field for just this event.
1400 sk = pt->af_packet_priv;
1403 * Yank back the headers [hope the device set this
1404 * right or kerboom...]
1406 * Incoming packets have ll header pulled,
1409 * For outgoing ones skb->data == skb_mac_header(skb)
1410 * so that this procedure is noop.
1413 if (skb->pkt_type == PACKET_LOOPBACK)
1416 if (!net_eq(dev_net(dev), sock_net(sk)))
1419 skb = skb_share_check(skb, GFP_ATOMIC);
1423 /* drop any routing info */
1426 /* drop conntrack reference */
1429 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1431 skb_push(skb, skb->data - skb_mac_header(skb));
1434 * The SOCK_PACKET socket receives _all_ frames.
1437 spkt->spkt_family = dev->type;
1438 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1439 spkt->spkt_protocol = skb->protocol;
1442 * Charge the memory to the socket. This is done specifically
1443 * to prevent sockets using all the memory up.
1446 if (sock_queue_rcv_skb(sk, skb) == 0)
1457 * Output a raw packet to a device layer. This bypasses all the other
1458 * protocol layers and you must therefore supply it with a complete frame
1461 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1462 struct msghdr *msg, size_t len)
1464 struct sock *sk = sock->sk;
1465 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1466 struct sk_buff *skb = NULL;
1467 struct net_device *dev;
1472 * Get and verify the address.
1476 if (msg->msg_namelen < sizeof(struct sockaddr))
1478 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1479 proto = saddr->spkt_protocol;
1481 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1484 * Find the device first to size check it
1487 saddr->spkt_device[13] = 0;
1490 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1496 if (!(dev->flags & IFF_UP))
1500 * You may not queue a frame bigger than the mtu. This is the lowest level
1501 * raw protocol and you must do your own fragmentation at this level.
1505 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1509 size_t reserved = LL_RESERVED_SPACE(dev);
1510 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1513 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1516 /* FIXME: Save some space for broken drivers that write a hard
1517 * header at transmission time by themselves. PPP is the notable
1518 * one here. This should really be fixed at the driver level.
1520 skb_reserve(skb, reserved);
1521 skb_reset_network_header(skb);
1523 /* Try to align data part correctly */
1528 skb_reset_network_header(skb);
1530 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1536 if (len > (dev->mtu + dev->hard_header_len)) {
1537 /* Earlier code assumed this would be a VLAN pkt,
1538 * double-check this now that we have the actual
1541 struct ethhdr *ehdr;
1542 skb_reset_mac_header(skb);
1543 ehdr = eth_hdr(skb);
1544 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1550 skb->protocol = proto;
1552 skb->priority = sk->sk_priority;
1553 skb->mark = sk->sk_mark;
1554 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1558 dev_queue_xmit(skb);
1569 static unsigned int run_filter(const struct sk_buff *skb,
1570 const struct sock *sk,
1573 struct sk_filter *filter;
1576 filter = rcu_dereference(sk->sk_filter);
1578 res = SK_RUN_FILTER(filter, skb);
1585 * This function makes lazy skb cloning in hope that most of packets
1586 * are discarded by BPF.
1588 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1589 * and skb->cb are mangled. It works because (and until) packets
1590 * falling here are owned by current CPU. Output packets are cloned
1591 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1592 * sequencially, so that if we return skb to original state on exit,
1593 * we will not harm anyone.
1596 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1597 struct packet_type *pt, struct net_device *orig_dev)
1600 struct sockaddr_ll *sll;
1601 struct packet_sock *po;
1602 u8 *skb_head = skb->data;
1603 int skb_len = skb->len;
1604 unsigned int snaplen, res;
1606 if (skb->pkt_type == PACKET_LOOPBACK)
1609 sk = pt->af_packet_priv;
1612 if (!net_eq(dev_net(dev), sock_net(sk)))
1617 if (dev->header_ops) {
1618 /* The device has an explicit notion of ll header,
1619 * exported to higher levels.
1621 * Otherwise, the device hides details of its frame
1622 * structure, so that corresponding packet head is
1623 * never delivered to user.
1625 if (sk->sk_type != SOCK_DGRAM)
1626 skb_push(skb, skb->data - skb_mac_header(skb));
1627 else if (skb->pkt_type == PACKET_OUTGOING) {
1628 /* Special case: outgoing packets have ll header at head */
1629 skb_pull(skb, skb_network_offset(skb));
1635 res = run_filter(skb, sk, snaplen);
1637 goto drop_n_restore;
1641 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1644 if (skb_shared(skb)) {
1645 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1649 if (skb_head != skb->data) {
1650 skb->data = skb_head;
1657 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1660 sll = &PACKET_SKB_CB(skb)->sa.ll;
1661 sll->sll_family = AF_PACKET;
1662 sll->sll_hatype = dev->type;
1663 sll->sll_protocol = skb->protocol;
1664 sll->sll_pkttype = skb->pkt_type;
1665 if (unlikely(po->origdev))
1666 sll->sll_ifindex = orig_dev->ifindex;
1668 sll->sll_ifindex = dev->ifindex;
1670 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1672 PACKET_SKB_CB(skb)->origlen = skb->len;
1674 if (pskb_trim(skb, snaplen))
1677 skb_set_owner_r(skb, sk);
1681 /* drop conntrack reference */
1684 spin_lock(&sk->sk_receive_queue.lock);
1685 po->stats.tp_packets++;
1686 skb->dropcount = atomic_read(&sk->sk_drops);
1687 __skb_queue_tail(&sk->sk_receive_queue, skb);
1688 spin_unlock(&sk->sk_receive_queue.lock);
1689 sk->sk_data_ready(sk, skb->len);
1693 spin_lock(&sk->sk_receive_queue.lock);
1694 po->stats.tp_drops++;
1695 atomic_inc(&sk->sk_drops);
1696 spin_unlock(&sk->sk_receive_queue.lock);
1699 if (skb_head != skb->data && skb_shared(skb)) {
1700 skb->data = skb_head;
1708 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1709 struct packet_type *pt, struct net_device *orig_dev)
1712 struct packet_sock *po;
1713 struct sockaddr_ll *sll;
1715 struct tpacket_hdr *h1;
1716 struct tpacket2_hdr *h2;
1717 struct tpacket3_hdr *h3;
1720 u8 *skb_head = skb->data;
1721 int skb_len = skb->len;
1722 unsigned int snaplen, res;
1723 unsigned long status = TP_STATUS_USER;
1724 unsigned short macoff, netoff, hdrlen;
1725 struct sk_buff *copy_skb = NULL;
1728 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1730 if (skb->pkt_type == PACKET_LOOPBACK)
1733 sk = pt->af_packet_priv;
1736 if (!net_eq(dev_net(dev), sock_net(sk)))
1739 if (dev->header_ops) {
1740 if (sk->sk_type != SOCK_DGRAM)
1741 skb_push(skb, skb->data - skb_mac_header(skb));
1742 else if (skb->pkt_type == PACKET_OUTGOING) {
1743 /* Special case: outgoing packets have ll header at head */
1744 skb_pull(skb, skb_network_offset(skb));
1748 if (skb->ip_summed == CHECKSUM_PARTIAL)
1749 status |= TP_STATUS_CSUMNOTREADY;
1753 res = run_filter(skb, sk, snaplen);
1755 goto drop_n_restore;
1759 if (sk->sk_type == SOCK_DGRAM) {
1760 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1763 unsigned maclen = skb_network_offset(skb);
1764 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1765 (maclen < 16 ? 16 : maclen)) +
1767 macoff = netoff - maclen;
1769 if (po->tp_version <= TPACKET_V2) {
1770 if (macoff + snaplen > po->rx_ring.frame_size) {
1771 if (po->copy_thresh &&
1772 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1773 if (skb_shared(skb)) {
1774 copy_skb = skb_clone(skb, GFP_ATOMIC);
1776 copy_skb = skb_get(skb);
1777 skb_head = skb->data;
1780 skb_set_owner_r(copy_skb, sk);
1782 snaplen = po->rx_ring.frame_size - macoff;
1783 if ((int)snaplen < 0)
1786 } else if (unlikely(macoff + snaplen >
1787 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1790 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1791 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1792 snaplen, nval, macoff);
1794 if (unlikely((int)snaplen < 0)) {
1796 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1799 spin_lock(&sk->sk_receive_queue.lock);
1800 h.raw = packet_current_rx_frame(po, skb,
1801 TP_STATUS_KERNEL, (macoff+snaplen));
1804 if (po->tp_version <= TPACKET_V2) {
1805 packet_increment_rx_head(po, &po->rx_ring);
1807 * LOSING will be reported till you read the stats,
1808 * because it's COR - Clear On Read.
1809 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1812 if (po->stats.tp_drops)
1813 status |= TP_STATUS_LOSING;
1815 po->stats.tp_packets++;
1817 status |= TP_STATUS_COPY;
1818 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1820 spin_unlock(&sk->sk_receive_queue.lock);
1822 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1824 switch (po->tp_version) {
1826 h.h1->tp_len = skb->len;
1827 h.h1->tp_snaplen = snaplen;
1828 h.h1->tp_mac = macoff;
1829 h.h1->tp_net = netoff;
1830 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1831 && shhwtstamps->syststamp.tv64)
1832 tv = ktime_to_timeval(shhwtstamps->syststamp);
1833 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1834 && shhwtstamps->hwtstamp.tv64)
1835 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1836 else if (skb->tstamp.tv64)
1837 tv = ktime_to_timeval(skb->tstamp);
1839 do_gettimeofday(&tv);
1840 h.h1->tp_sec = tv.tv_sec;
1841 h.h1->tp_usec = tv.tv_usec;
1842 hdrlen = sizeof(*h.h1);
1845 h.h2->tp_len = skb->len;
1846 h.h2->tp_snaplen = snaplen;
1847 h.h2->tp_mac = macoff;
1848 h.h2->tp_net = netoff;
1849 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1850 && shhwtstamps->syststamp.tv64)
1851 ts = ktime_to_timespec(shhwtstamps->syststamp);
1852 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1853 && shhwtstamps->hwtstamp.tv64)
1854 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1855 else if (skb->tstamp.tv64)
1856 ts = ktime_to_timespec(skb->tstamp);
1858 getnstimeofday(&ts);
1859 h.h2->tp_sec = ts.tv_sec;
1860 h.h2->tp_nsec = ts.tv_nsec;
1861 if (vlan_tx_tag_present(skb)) {
1862 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1863 status |= TP_STATUS_VLAN_VALID;
1865 h.h2->tp_vlan_tci = 0;
1867 h.h2->tp_padding = 0;
1868 hdrlen = sizeof(*h.h2);
1871 /* tp_nxt_offset,vlan are already populated above.
1872 * So DONT clear those fields here
1874 h.h3->tp_status |= status;
1875 h.h3->tp_len = skb->len;
1876 h.h3->tp_snaplen = snaplen;
1877 h.h3->tp_mac = macoff;
1878 h.h3->tp_net = netoff;
1879 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1880 && shhwtstamps->syststamp.tv64)
1881 ts = ktime_to_timespec(shhwtstamps->syststamp);
1882 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1883 && shhwtstamps->hwtstamp.tv64)
1884 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1885 else if (skb->tstamp.tv64)
1886 ts = ktime_to_timespec(skb->tstamp);
1888 getnstimeofday(&ts);
1889 h.h3->tp_sec = ts.tv_sec;
1890 h.h3->tp_nsec = ts.tv_nsec;
1891 hdrlen = sizeof(*h.h3);
1897 sll = h.raw + TPACKET_ALIGN(hdrlen);
1898 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1899 sll->sll_family = AF_PACKET;
1900 sll->sll_hatype = dev->type;
1901 sll->sll_protocol = skb->protocol;
1902 sll->sll_pkttype = skb->pkt_type;
1903 if (unlikely(po->origdev))
1904 sll->sll_ifindex = orig_dev->ifindex;
1906 sll->sll_ifindex = dev->ifindex;
1909 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1913 if (po->tp_version <= TPACKET_V2) {
1914 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1915 + macoff + snaplen);
1916 for (start = h.raw; start < end; start += PAGE_SIZE)
1917 flush_dcache_page(pgv_to_page(start));
1922 if (po->tp_version <= TPACKET_V2)
1923 __packet_set_status(po, h.raw, status);
1925 prb_clear_blk_fill_status(&po->rx_ring);
1927 sk->sk_data_ready(sk, 0);
1930 if (skb_head != skb->data && skb_shared(skb)) {
1931 skb->data = skb_head;
1939 po->stats.tp_drops++;
1940 spin_unlock(&sk->sk_receive_queue.lock);
1942 sk->sk_data_ready(sk, 0);
1943 kfree_skb(copy_skb);
1944 goto drop_n_restore;
1947 static void tpacket_destruct_skb(struct sk_buff *skb)
1949 struct packet_sock *po = pkt_sk(skb->sk);
1952 if (likely(po->tx_ring.pg_vec)) {
1953 ph = skb_shinfo(skb)->destructor_arg;
1954 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1955 atomic_dec(&po->tx_ring.pending);
1956 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1962 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1963 void *frame, struct net_device *dev, int size_max,
1964 __be16 proto, unsigned char *addr)
1967 struct tpacket_hdr *h1;
1968 struct tpacket2_hdr *h2;
1971 int to_write, offset, len, tp_len, nr_frags, len_max;
1972 struct socket *sock = po->sk.sk_socket;
1979 skb->protocol = proto;
1981 skb->priority = po->sk.sk_priority;
1982 skb->mark = po->sk.sk_mark;
1983 skb_shinfo(skb)->destructor_arg = ph.raw;
1985 switch (po->tp_version) {
1987 tp_len = ph.h2->tp_len;
1990 tp_len = ph.h1->tp_len;
1993 if (unlikely(tp_len > size_max)) {
1994 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1998 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1999 skb_reset_network_header(skb);
2001 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2004 if (sock->type == SOCK_DGRAM) {
2005 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2007 if (unlikely(err < 0))
2009 } else if (dev->hard_header_len) {
2010 /* net device doesn't like empty head */
2011 if (unlikely(tp_len <= dev->hard_header_len)) {
2012 pr_err("packet size is too short (%d < %d)\n",
2013 tp_len, dev->hard_header_len);
2017 skb_push(skb, dev->hard_header_len);
2018 err = skb_store_bits(skb, 0, data,
2019 dev->hard_header_len);
2023 data += dev->hard_header_len;
2024 to_write -= dev->hard_header_len;
2028 offset = offset_in_page(data);
2029 len_max = PAGE_SIZE - offset;
2030 len = ((to_write > len_max) ? len_max : to_write);
2032 skb->data_len = to_write;
2033 skb->len += to_write;
2034 skb->truesize += to_write;
2035 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2037 while (likely(to_write)) {
2038 nr_frags = skb_shinfo(skb)->nr_frags;
2040 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2041 pr_err("Packet exceed the number of skb frags(%lu)\n",
2046 page = pgv_to_page(data);
2048 flush_dcache_page(page);
2050 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2053 len_max = PAGE_SIZE;
2054 len = ((to_write > len_max) ? len_max : to_write);
2060 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2062 struct net_device *dev;
2065 dev = rcu_dereference(po->cached_dev);
2073 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2075 struct sk_buff *skb;
2076 struct net_device *dev;
2078 int err, reserve = 0;
2080 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2081 int tp_len, size_max;
2082 unsigned char *addr;
2086 mutex_lock(&po->pg_vec_lock);
2089 if (saddr == NULL) {
2090 dev = packet_cached_dev_get(po);
2095 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2097 if (msg->msg_namelen < (saddr->sll_halen
2098 + offsetof(struct sockaddr_ll,
2101 proto = saddr->sll_protocol;
2102 addr = saddr->sll_addr;
2103 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2107 if (unlikely(dev == NULL))
2110 if (unlikely(!(dev->flags & IFF_UP)))
2113 reserve = dev->hard_header_len;
2115 size_max = po->tx_ring.frame_size
2116 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2118 if (size_max > dev->mtu + reserve)
2119 size_max = dev->mtu + reserve;
2122 ph = packet_current_frame(po, &po->tx_ring,
2123 TP_STATUS_SEND_REQUEST);
2125 if (unlikely(ph == NULL)) {
2130 status = TP_STATUS_SEND_REQUEST;
2131 skb = sock_alloc_send_skb(&po->sk,
2132 LL_ALLOCATED_SPACE(dev)
2133 + sizeof(struct sockaddr_ll),
2136 if (unlikely(skb == NULL))
2139 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2142 if (unlikely(tp_len < 0)) {
2144 __packet_set_status(po, ph,
2145 TP_STATUS_AVAILABLE);
2146 packet_increment_head(&po->tx_ring);
2150 status = TP_STATUS_WRONG_FORMAT;
2156 skb->destructor = tpacket_destruct_skb;
2157 __packet_set_status(po, ph, TP_STATUS_SENDING);
2158 atomic_inc(&po->tx_ring.pending);
2160 status = TP_STATUS_SEND_REQUEST;
2161 err = dev_queue_xmit(skb);
2162 if (unlikely(err > 0)) {
2163 err = net_xmit_errno(err);
2164 if (err && __packet_get_status(po, ph) ==
2165 TP_STATUS_AVAILABLE) {
2166 /* skb was destructed already */
2171 * skb was dropped but not destructed yet;
2172 * let's treat it like congestion or err < 0
2176 packet_increment_head(&po->tx_ring);
2178 } while (likely((ph != NULL) ||
2179 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2180 (atomic_read(&po->tx_ring.pending))))
2187 __packet_set_status(po, ph, status);
2192 mutex_unlock(&po->pg_vec_lock);
2196 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2197 size_t reserve, size_t len,
2198 size_t linear, int noblock,
2201 struct sk_buff *skb;
2203 /* Under a page? Don't bother with paged skb. */
2204 if (prepad + len < PAGE_SIZE || !linear)
2207 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2212 skb_reserve(skb, reserve);
2213 skb_put(skb, linear);
2214 skb->data_len = len - linear;
2215 skb->len += len - linear;
2220 static int packet_snd(struct socket *sock,
2221 struct msghdr *msg, size_t len)
2223 struct sock *sk = sock->sk;
2224 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2225 struct sk_buff *skb;
2226 struct net_device *dev;
2228 unsigned char *addr;
2229 int err, reserve = 0;
2230 struct virtio_net_hdr vnet_hdr = { 0 };
2233 struct packet_sock *po = pkt_sk(sk);
2234 bool has_vnet_hdr = false;
2235 unsigned short gso_type = 0;
2238 * Get and verify the address.
2241 if (saddr == NULL) {
2242 dev = packet_cached_dev_get(po);
2247 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2249 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2251 proto = saddr->sll_protocol;
2252 addr = saddr->sll_addr;
2253 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2257 if (unlikely(dev == NULL))
2260 if (unlikely(!(dev->flags & IFF_UP)))
2263 if (sock->type == SOCK_RAW)
2264 reserve = dev->hard_header_len;
2265 if (po->has_vnet_hdr) {
2266 vnet_hdr_len = sizeof(vnet_hdr);
2267 has_vnet_hdr = true;
2270 if (len < vnet_hdr_len)
2273 len -= vnet_hdr_len;
2275 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2280 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2281 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2283 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2284 vnet_hdr.csum_offset + 2;
2287 if (vnet_hdr.hdr_len > len)
2290 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2291 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2292 case VIRTIO_NET_HDR_GSO_TCPV4:
2293 gso_type = SKB_GSO_TCPV4;
2295 case VIRTIO_NET_HDR_GSO_TCPV6:
2296 gso_type = SKB_GSO_TCPV6;
2298 case VIRTIO_NET_HDR_GSO_UDP:
2299 gso_type = SKB_GSO_UDP;
2305 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2306 gso_type |= SKB_GSO_TCP_ECN;
2308 if (vnet_hdr.gso_size == 0)
2315 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2319 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2320 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2321 msg->msg_flags & MSG_DONTWAIT, &err);
2325 skb_set_network_header(skb, reserve);
2328 if (sock->type == SOCK_DGRAM &&
2329 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2332 /* Returns -EFAULT on error */
2333 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2336 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2340 if (!gso_type && (len > dev->mtu + reserve)) {
2341 /* Earlier code assumed this would be a VLAN pkt,
2342 * double-check this now that we have the actual
2345 struct ethhdr *ehdr;
2346 skb_reset_mac_header(skb);
2347 ehdr = eth_hdr(skb);
2348 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2354 skb->protocol = proto;
2356 skb->priority = sk->sk_priority;
2357 skb->mark = sk->sk_mark;
2360 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2361 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2362 vnet_hdr.csum_offset)) {
2368 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2369 skb_shinfo(skb)->gso_type = gso_type;
2371 /* Header must be checked, and gso_segs computed. */
2372 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2373 skb_shinfo(skb)->gso_segs = 0;
2375 len += vnet_hdr_len;
2382 err = dev_queue_xmit(skb);
2383 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2399 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2400 struct msghdr *msg, size_t len)
2402 struct sock *sk = sock->sk;
2403 struct packet_sock *po = pkt_sk(sk);
2404 if (po->tx_ring.pg_vec)
2405 return tpacket_snd(po, msg);
2407 return packet_snd(sock, msg, len);
2411 * Close a PACKET socket. This is fairly simple. We immediately go
2412 * to 'closed' state and remove our protocol entry in the device list.
2415 static int packet_release(struct socket *sock)
2417 struct sock *sk = sock->sk;
2418 struct packet_sock *po;
2419 struct packet_fanout *f;
2421 union tpacket_req_u req_u;
2429 spin_lock_bh(&net->packet.sklist_lock);
2430 sk_del_node_init_rcu(sk);
2431 sock_prot_inuse_add(net, sk->sk_prot, -1);
2432 spin_unlock_bh(&net->packet.sklist_lock);
2434 spin_lock(&po->bind_lock);
2435 unregister_prot_hook(sk, false);
2436 if (po->prot_hook.dev) {
2437 dev_put(po->prot_hook.dev);
2438 po->prot_hook.dev = NULL;
2440 spin_unlock(&po->bind_lock);
2442 packet_flush_mclist(sk);
2444 if (po->rx_ring.pg_vec) {
2445 memset(&req_u, 0, sizeof(req_u));
2446 packet_set_ring(sk, &req_u, 1, 0);
2449 if (po->tx_ring.pg_vec) {
2450 memset(&req_u, 0, sizeof(req_u));
2451 packet_set_ring(sk, &req_u, 1, 1);
2454 f = fanout_release(sk);
2461 * Now the socket is dead. No more input will appear.
2468 skb_queue_purge(&sk->sk_receive_queue);
2469 sk_refcnt_debug_release(sk);
2476 * Attach a packet hook.
2479 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2482 struct packet_sock *po = pkt_sk(sk);
2483 struct net_device *dev_curr;
2484 struct net_device *dev = NULL;
2486 bool unlisted = false;
2490 spin_lock(&po->bind_lock);
2499 dev = dev_get_by_name_rcu(sock_net(sk), name);
2504 } else if (ifindex) {
2505 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2515 dev_curr = po->prot_hook.dev;
2519 /* prevents packet_notifier() from calling
2520 * register_prot_hook()
2523 __unregister_prot_hook(sk, true);
2525 dev_curr = po->prot_hook.dev;
2527 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2530 BUG_ON(po->running);
2532 po->prot_hook.type = protocol;
2534 if (unlikely(unlisted)) {
2536 po->prot_hook.dev = NULL;
2539 po->prot_hook.dev = dev;
2540 po->ifindex = dev ? dev->ifindex : 0;
2549 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2550 register_prot_hook(sk);
2552 sk->sk_err = ENETDOWN;
2553 if (!sock_flag(sk, SOCK_DEAD))
2554 sk->sk_error_report(sk);
2559 spin_unlock(&po->bind_lock);
2565 * Bind a packet socket to a device
2568 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2571 struct sock *sk = sock->sk;
2578 if (addr_len != sizeof(struct sockaddr))
2580 strlcpy(name, uaddr->sa_data, sizeof(name));
2582 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
2585 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2587 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2588 struct sock *sk = sock->sk;
2594 if (addr_len < sizeof(struct sockaddr_ll))
2596 if (sll->sll_family != AF_PACKET)
2599 return packet_do_bind(sk, NULL, sll->sll_ifindex,
2600 sll->sll_protocol ? : pkt_sk(sk)->num);
2603 static struct proto packet_proto = {
2605 .owner = THIS_MODULE,
2606 .obj_size = sizeof(struct packet_sock),
2610 * Create a packet of type SOCK_PACKET.
2613 static int packet_create(struct net *net, struct socket *sock, int protocol,
2617 struct packet_sock *po;
2618 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2621 if (!capable(CAP_NET_RAW))
2623 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2624 sock->type != SOCK_PACKET)
2625 return -ESOCKTNOSUPPORT;
2627 sock->state = SS_UNCONNECTED;
2630 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2634 sock->ops = &packet_ops;
2635 if (sock->type == SOCK_PACKET)
2636 sock->ops = &packet_ops_spkt;
2638 sock_init_data(sock, sk);
2641 sk->sk_family = PF_PACKET;
2643 RCU_INIT_POINTER(po->cached_dev, NULL);
2645 sk->sk_destruct = packet_sock_destruct;
2646 sk_refcnt_debug_inc(sk);
2649 * Attach a protocol block
2652 spin_lock_init(&po->bind_lock);
2653 mutex_init(&po->pg_vec_lock);
2654 po->prot_hook.func = packet_rcv;
2656 if (sock->type == SOCK_PACKET)
2657 po->prot_hook.func = packet_rcv_spkt;
2659 po->prot_hook.af_packet_priv = sk;
2662 po->prot_hook.type = proto;
2663 register_prot_hook(sk);
2666 spin_lock_bh(&net->packet.sklist_lock);
2667 sk_add_node_rcu(sk, &net->packet.sklist);
2668 sock_prot_inuse_add(net, &packet_proto, 1);
2669 spin_unlock_bh(&net->packet.sklist_lock);
2676 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2678 struct sock_exterr_skb *serr;
2679 struct sk_buff *skb, *skb2;
2683 skb = skb_dequeue(&sk->sk_error_queue);
2689 msg->msg_flags |= MSG_TRUNC;
2692 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2696 sock_recv_timestamp(msg, sk, skb);
2698 serr = SKB_EXT_ERR(skb);
2699 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2700 sizeof(serr->ee), &serr->ee);
2702 msg->msg_flags |= MSG_ERRQUEUE;
2705 /* Reset and regenerate socket error */
2706 spin_lock_bh(&sk->sk_error_queue.lock);
2708 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2709 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2710 spin_unlock_bh(&sk->sk_error_queue.lock);
2711 sk->sk_error_report(sk);
2713 spin_unlock_bh(&sk->sk_error_queue.lock);
2722 * Pull a packet from our receive queue and hand it to the user.
2723 * If necessary we block.
2726 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2727 struct msghdr *msg, size_t len, int flags)
2729 struct sock *sk = sock->sk;
2730 struct sk_buff *skb;
2732 int vnet_hdr_len = 0;
2735 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2739 /* What error should we return now? EUNATTACH? */
2740 if (pkt_sk(sk)->ifindex < 0)
2744 if (flags & MSG_ERRQUEUE) {
2745 err = packet_recv_error(sk, msg, len);
2750 * Call the generic datagram receiver. This handles all sorts
2751 * of horrible races and re-entrancy so we can forget about it
2752 * in the protocol layers.
2754 * Now it will return ENETDOWN, if device have just gone down,
2755 * but then it will block.
2758 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2761 * An error occurred so return it. Because skb_recv_datagram()
2762 * handles the blocking we don't see and worry about blocking
2769 if (pkt_sk(sk)->has_vnet_hdr) {
2770 struct virtio_net_hdr vnet_hdr = { 0 };
2773 vnet_hdr_len = sizeof(vnet_hdr);
2774 if (len < vnet_hdr_len)
2777 len -= vnet_hdr_len;
2779 if (skb_is_gso(skb)) {
2780 struct skb_shared_info *sinfo = skb_shinfo(skb);
2782 /* This is a hint as to how much should be linear. */
2783 vnet_hdr.hdr_len = skb_headlen(skb);
2784 vnet_hdr.gso_size = sinfo->gso_size;
2785 if (sinfo->gso_type & SKB_GSO_TCPV4)
2786 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2787 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2788 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2789 else if (sinfo->gso_type & SKB_GSO_UDP)
2790 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2791 else if (sinfo->gso_type & SKB_GSO_FCOE)
2795 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2796 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2798 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2800 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2801 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2802 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2803 vnet_hdr.csum_offset = skb->csum_offset;
2804 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2805 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2806 } /* else everything is zero */
2808 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2814 /* You lose any data beyond the buffer you gave. If it worries
2815 * a user program they can ask the device for its MTU
2821 msg->msg_flags |= MSG_TRUNC;
2824 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2828 sock_recv_ts_and_drops(msg, sk, skb);
2830 if (msg->msg_name) {
2831 /* If the address length field is there to be filled
2832 * in, we fill it in now.
2834 if (sock->type == SOCK_PACKET) {
2835 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2837 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2838 msg->msg_namelen = sll->sll_halen +
2839 offsetof(struct sockaddr_ll, sll_addr);
2841 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2845 if (pkt_sk(sk)->auxdata) {
2846 struct tpacket_auxdata aux;
2848 aux.tp_status = TP_STATUS_USER;
2849 if (skb->ip_summed == CHECKSUM_PARTIAL)
2850 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2851 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2852 aux.tp_snaplen = skb->len;
2854 aux.tp_net = skb_network_offset(skb);
2855 if (vlan_tx_tag_present(skb)) {
2856 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2857 aux.tp_status |= TP_STATUS_VLAN_VALID;
2859 aux.tp_vlan_tci = 0;
2862 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2866 * Free or return the buffer as appropriate. Again this
2867 * hides all the races and re-entrancy issues from us.
2869 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2872 skb_free_datagram(sk, skb);
2877 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2878 int *uaddr_len, int peer)
2880 struct net_device *dev;
2881 struct sock *sk = sock->sk;
2886 uaddr->sa_family = AF_PACKET;
2887 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2889 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2891 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2893 *uaddr_len = sizeof(*uaddr);
2898 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2899 int *uaddr_len, int peer)
2901 struct net_device *dev;
2902 struct sock *sk = sock->sk;
2903 struct packet_sock *po = pkt_sk(sk);
2904 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2909 sll->sll_family = AF_PACKET;
2910 sll->sll_ifindex = po->ifindex;
2911 sll->sll_protocol = po->num;
2912 sll->sll_pkttype = 0;
2914 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2916 sll->sll_hatype = dev->type;
2917 sll->sll_halen = dev->addr_len;
2918 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2920 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2924 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2929 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2933 case PACKET_MR_MULTICAST:
2934 if (i->alen != dev->addr_len)
2937 return dev_mc_add(dev, i->addr);
2939 return dev_mc_del(dev, i->addr);
2941 case PACKET_MR_PROMISC:
2942 return dev_set_promiscuity(dev, what);
2944 case PACKET_MR_ALLMULTI:
2945 return dev_set_allmulti(dev, what);
2947 case PACKET_MR_UNICAST:
2948 if (i->alen != dev->addr_len)
2951 return dev_uc_add(dev, i->addr);
2953 return dev_uc_del(dev, i->addr);
2961 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2963 for ( ; i; i = i->next) {
2964 if (i->ifindex == dev->ifindex)
2965 packet_dev_mc(dev, i, what);
2969 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2971 struct packet_sock *po = pkt_sk(sk);
2972 struct packet_mclist *ml, *i;
2973 struct net_device *dev;
2979 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2984 if (mreq->mr_alen > dev->addr_len)
2988 i = kmalloc(sizeof(*i), GFP_KERNEL);
2993 for (ml = po->mclist; ml; ml = ml->next) {
2994 if (ml->ifindex == mreq->mr_ifindex &&
2995 ml->type == mreq->mr_type &&
2996 ml->alen == mreq->mr_alen &&
2997 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2999 /* Free the new element ... */
3005 i->type = mreq->mr_type;
3006 i->ifindex = mreq->mr_ifindex;
3007 i->alen = mreq->mr_alen;
3008 memcpy(i->addr, mreq->mr_address, i->alen);
3010 i->next = po->mclist;
3012 err = packet_dev_mc(dev, i, 1);
3014 po->mclist = i->next;
3023 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3025 struct packet_mclist *ml, **mlp;
3029 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3030 if (ml->ifindex == mreq->mr_ifindex &&
3031 ml->type == mreq->mr_type &&
3032 ml->alen == mreq->mr_alen &&
3033 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3034 if (--ml->count == 0) {
3035 struct net_device *dev;
3037 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3039 packet_dev_mc(dev, ml, -1);
3047 return -EADDRNOTAVAIL;
3050 static void packet_flush_mclist(struct sock *sk)
3052 struct packet_sock *po = pkt_sk(sk);
3053 struct packet_mclist *ml;
3059 while ((ml = po->mclist) != NULL) {
3060 struct net_device *dev;
3062 po->mclist = ml->next;
3063 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3065 packet_dev_mc(dev, ml, -1);
3072 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3074 struct sock *sk = sock->sk;
3075 struct packet_sock *po = pkt_sk(sk);
3078 if (level != SOL_PACKET)
3079 return -ENOPROTOOPT;
3082 case PACKET_ADD_MEMBERSHIP:
3083 case PACKET_DROP_MEMBERSHIP:
3085 struct packet_mreq_max mreq;
3087 memset(&mreq, 0, sizeof(mreq));
3088 if (len < sizeof(struct packet_mreq))
3090 if (len > sizeof(mreq))
3092 if (copy_from_user(&mreq, optval, len))
3094 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3096 if (optname == PACKET_ADD_MEMBERSHIP)
3097 ret = packet_mc_add(sk, &mreq);
3099 ret = packet_mc_drop(sk, &mreq);
3103 case PACKET_RX_RING:
3104 case PACKET_TX_RING:
3106 union tpacket_req_u req_u;
3109 switch (po->tp_version) {
3112 len = sizeof(req_u.req);
3116 len = sizeof(req_u.req3);
3121 if (pkt_sk(sk)->has_vnet_hdr)
3123 if (copy_from_user(&req_u.req, optval, len))
3125 return packet_set_ring(sk, &req_u, 0,
3126 optname == PACKET_TX_RING);
3128 case PACKET_COPY_THRESH:
3132 if (optlen != sizeof(val))
3134 if (copy_from_user(&val, optval, sizeof(val)))
3137 pkt_sk(sk)->copy_thresh = val;
3140 case PACKET_VERSION:
3144 if (optlen != sizeof(val))
3146 if (copy_from_user(&val, optval, sizeof(val)))
3157 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3160 po->tp_version = val;
3166 case PACKET_RESERVE:
3170 if (optlen != sizeof(val))
3172 if (copy_from_user(&val, optval, sizeof(val)))
3177 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3180 po->tp_reserve = val;
3190 if (optlen != sizeof(val))
3192 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3194 if (copy_from_user(&val, optval, sizeof(val)))
3196 po->tp_loss = !!val;
3199 case PACKET_AUXDATA:
3203 if (optlen < sizeof(val))
3205 if (copy_from_user(&val, optval, sizeof(val)))
3208 po->auxdata = !!val;
3211 case PACKET_ORIGDEV:
3215 if (optlen < sizeof(val))
3217 if (copy_from_user(&val, optval, sizeof(val)))
3220 po->origdev = !!val;
3223 case PACKET_VNET_HDR:
3227 if (sock->type != SOCK_RAW)
3229 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3231 if (optlen < sizeof(val))
3233 if (copy_from_user(&val, optval, sizeof(val)))
3236 po->has_vnet_hdr = !!val;
3239 case PACKET_TIMESTAMP:
3243 if (optlen != sizeof(val))
3245 if (copy_from_user(&val, optval, sizeof(val)))
3248 po->tp_tstamp = val;
3255 if (optlen != sizeof(val))
3257 if (copy_from_user(&val, optval, sizeof(val)))
3260 return fanout_add(sk, val & 0xffff, val >> 16);
3263 return -ENOPROTOOPT;
3267 static int packet_getsockopt(struct socket *sock, int level, int optname,
3268 char __user *optval, int __user *optlen)
3272 struct sock *sk = sock->sk;
3273 struct packet_sock *po = pkt_sk(sk);
3275 struct tpacket_stats st;
3276 union tpacket_stats_u st_u;
3278 if (level != SOL_PACKET)
3279 return -ENOPROTOOPT;
3281 if (get_user(len, optlen))
3288 case PACKET_STATISTICS:
3289 if (po->tp_version == TPACKET_V3) {
3290 len = sizeof(struct tpacket_stats_v3);
3292 if (len > sizeof(struct tpacket_stats))
3293 len = sizeof(struct tpacket_stats);
3295 spin_lock_bh(&sk->sk_receive_queue.lock);
3296 if (po->tp_version == TPACKET_V3) {
3297 memcpy(&st_u.stats3, &po->stats,
3298 sizeof(struct tpacket_stats));
3299 st_u.stats3.tp_freeze_q_cnt =
3300 po->stats_u.stats3.tp_freeze_q_cnt;
3301 st_u.stats3.tp_packets += po->stats.tp_drops;
3302 data = &st_u.stats3;
3305 st.tp_packets += st.tp_drops;
3308 memset(&po->stats, 0, sizeof(st));
3309 spin_unlock_bh(&sk->sk_receive_queue.lock);
3311 case PACKET_AUXDATA:
3312 if (len > sizeof(int))
3318 case PACKET_ORIGDEV:
3319 if (len > sizeof(int))
3325 case PACKET_VNET_HDR:
3326 if (len > sizeof(int))
3328 val = po->has_vnet_hdr;
3332 case PACKET_VERSION:
3333 if (len > sizeof(int))
3335 val = po->tp_version;
3339 if (len > sizeof(int))
3341 if (copy_from_user(&val, optval, len))
3345 val = sizeof(struct tpacket_hdr);
3348 val = sizeof(struct tpacket2_hdr);
3351 val = sizeof(struct tpacket3_hdr);
3358 case PACKET_RESERVE:
3359 if (len > sizeof(unsigned int))
3360 len = sizeof(unsigned int);
3361 val = po->tp_reserve;
3365 if (len > sizeof(unsigned int))
3366 len = sizeof(unsigned int);
3370 case PACKET_TIMESTAMP:
3371 if (len > sizeof(int))
3373 val = po->tp_tstamp;
3377 if (len > sizeof(int))
3380 ((u32)po->fanout->id |
3381 ((u32)po->fanout->type << 16)) :
3386 return -ENOPROTOOPT;
3389 if (put_user(len, optlen))
3391 if (copy_to_user(optval, data, len))
3397 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3400 struct hlist_node *node;
3401 struct net_device *dev = data;
3402 struct net *net = dev_net(dev);
3405 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3406 struct packet_sock *po = pkt_sk(sk);
3409 case NETDEV_UNREGISTER:
3411 packet_dev_mclist(dev, po->mclist, -1);
3415 if (dev->ifindex == po->ifindex) {
3416 spin_lock(&po->bind_lock);
3418 __unregister_prot_hook(sk, false);
3419 sk->sk_err = ENETDOWN;
3420 if (!sock_flag(sk, SOCK_DEAD))
3421 sk->sk_error_report(sk);
3423 if (msg == NETDEV_UNREGISTER) {
3425 if (po->prot_hook.dev)
3426 dev_put(po->prot_hook.dev);
3427 po->prot_hook.dev = NULL;
3429 spin_unlock(&po->bind_lock);
3433 if (dev->ifindex == po->ifindex) {
3434 spin_lock(&po->bind_lock);
3436 register_prot_hook(sk);
3437 spin_unlock(&po->bind_lock);
3447 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3450 struct sock *sk = sock->sk;
3455 int amount = sk_wmem_alloc_get(sk);
3457 return put_user(amount, (int __user *)arg);
3461 struct sk_buff *skb;
3464 spin_lock_bh(&sk->sk_receive_queue.lock);
3465 skb = skb_peek(&sk->sk_receive_queue);
3468 spin_unlock_bh(&sk->sk_receive_queue.lock);
3469 return put_user(amount, (int __user *)arg);
3472 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3474 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3484 case SIOCGIFBRDADDR:
3485 case SIOCSIFBRDADDR:
3486 case SIOCGIFNETMASK:
3487 case SIOCSIFNETMASK:
3488 case SIOCGIFDSTADDR:
3489 case SIOCSIFDSTADDR:
3491 return inet_dgram_ops.ioctl(sock, cmd, arg);
3495 return -ENOIOCTLCMD;
3500 static unsigned int packet_poll(struct file *file, struct socket *sock,
3503 struct sock *sk = sock->sk;
3504 struct packet_sock *po = pkt_sk(sk);
3505 unsigned int mask = datagram_poll(file, sock, wait);
3507 spin_lock_bh(&sk->sk_receive_queue.lock);
3508 if (po->rx_ring.pg_vec) {
3509 if (!packet_previous_rx_frame(po, &po->rx_ring,
3511 mask |= POLLIN | POLLRDNORM;
3513 spin_unlock_bh(&sk->sk_receive_queue.lock);
3514 spin_lock_bh(&sk->sk_write_queue.lock);
3515 if (po->tx_ring.pg_vec) {
3516 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3517 mask |= POLLOUT | POLLWRNORM;
3519 spin_unlock_bh(&sk->sk_write_queue.lock);
3524 /* Dirty? Well, I still did not learn better way to account
3528 static void packet_mm_open(struct vm_area_struct *vma)
3530 struct file *file = vma->vm_file;
3531 struct socket *sock = file->private_data;
3532 struct sock *sk = sock->sk;
3535 atomic_inc(&pkt_sk(sk)->mapped);
3538 static void packet_mm_close(struct vm_area_struct *vma)
3540 struct file *file = vma->vm_file;
3541 struct socket *sock = file->private_data;
3542 struct sock *sk = sock->sk;
3545 atomic_dec(&pkt_sk(sk)->mapped);
3548 static const struct vm_operations_struct packet_mmap_ops = {
3549 .open = packet_mm_open,
3550 .close = packet_mm_close,
3553 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3558 for (i = 0; i < len; i++) {
3559 if (likely(pg_vec[i].buffer)) {
3560 if (is_vmalloc_addr(pg_vec[i].buffer))
3561 vfree(pg_vec[i].buffer);
3563 free_pages((unsigned long)pg_vec[i].buffer,
3565 pg_vec[i].buffer = NULL;
3571 static char *alloc_one_pg_vec_page(unsigned long order)
3573 char *buffer = NULL;
3574 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3575 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3577 buffer = (char *) __get_free_pages(gfp_flags, order);
3583 * __get_free_pages failed, fall back to vmalloc
3585 buffer = vzalloc((1 << order) * PAGE_SIZE);
3591 * vmalloc failed, lets dig into swap here
3593 gfp_flags &= ~__GFP_NORETRY;
3594 buffer = (char *)__get_free_pages(gfp_flags, order);
3599 * complete and utter failure
3604 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3606 unsigned int block_nr = req->tp_block_nr;
3610 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3611 if (unlikely(!pg_vec))
3614 for (i = 0; i < block_nr; i++) {
3615 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3616 if (unlikely(!pg_vec[i].buffer))
3617 goto out_free_pgvec;
3624 free_pg_vec(pg_vec, order, block_nr);
3629 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3630 int closing, int tx_ring)
3632 struct pgv *pg_vec = NULL;
3633 struct packet_sock *po = pkt_sk(sk);
3634 int was_running, order = 0;
3635 struct packet_ring_buffer *rb;
3636 struct sk_buff_head *rb_queue;
3639 /* Added to avoid minimal code churn */
3640 struct tpacket_req *req = &req_u->req;
3643 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3644 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3645 WARN(1, "Tx-ring is not supported.\n");
3649 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3650 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3654 if (atomic_read(&po->mapped))
3656 if (atomic_read(&rb->pending))
3660 if (req->tp_block_nr) {
3661 /* Sanity tests and some calculations */
3663 if (unlikely(rb->pg_vec))
3666 switch (po->tp_version) {
3668 po->tp_hdrlen = TPACKET_HDRLEN;
3671 po->tp_hdrlen = TPACKET2_HDRLEN;
3674 po->tp_hdrlen = TPACKET3_HDRLEN;
3679 if (unlikely((int)req->tp_block_size <= 0))
3681 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3683 if (po->tp_version >= TPACKET_V3 &&
3684 req->tp_block_size <=
3685 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
3687 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3690 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3693 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3694 if (unlikely(rb->frames_per_block <= 0))
3696 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
3698 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3703 order = get_order(req->tp_block_size);
3704 pg_vec = alloc_pg_vec(req, order);
3705 if (unlikely(!pg_vec))
3707 switch (po->tp_version) {
3709 /* Transmit path is not supported. We checked
3710 * it above but just being paranoid
3713 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3722 if (unlikely(req->tp_frame_nr))
3727 /* Detach socket from network */
3728 spin_lock(&po->bind_lock);
3729 was_running = po->running;
3733 __unregister_prot_hook(sk, false);
3735 spin_unlock(&po->bind_lock);
3740 mutex_lock(&po->pg_vec_lock);
3741 if (closing || atomic_read(&po->mapped) == 0) {
3743 spin_lock_bh(&rb_queue->lock);
3744 swap(rb->pg_vec, pg_vec);
3745 rb->frame_max = (req->tp_frame_nr - 1);
3747 rb->frame_size = req->tp_frame_size;
3748 spin_unlock_bh(&rb_queue->lock);
3750 swap(rb->pg_vec_order, order);
3751 swap(rb->pg_vec_len, req->tp_block_nr);
3753 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3754 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3755 tpacket_rcv : packet_rcv;
3756 skb_queue_purge(rb_queue);
3757 if (atomic_read(&po->mapped))
3758 pr_err("packet_mmap: vma is busy: %d\n",
3759 atomic_read(&po->mapped));
3761 mutex_unlock(&po->pg_vec_lock);
3763 spin_lock(&po->bind_lock);
3766 register_prot_hook(sk);
3768 spin_unlock(&po->bind_lock);
3769 if (closing && (po->tp_version > TPACKET_V2)) {
3770 /* Because we don't support block-based V3 on tx-ring */
3772 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3776 free_pg_vec(pg_vec, order, req->tp_block_nr);
3782 static int packet_mmap(struct file *file, struct socket *sock,
3783 struct vm_area_struct *vma)
3785 struct sock *sk = sock->sk;
3786 struct packet_sock *po = pkt_sk(sk);
3787 unsigned long size, expected_size;
3788 struct packet_ring_buffer *rb;
3789 unsigned long start;
3796 mutex_lock(&po->pg_vec_lock);
3799 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3801 expected_size += rb->pg_vec_len
3807 if (expected_size == 0)
3810 size = vma->vm_end - vma->vm_start;
3811 if (size != expected_size)
3814 start = vma->vm_start;
3815 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3816 if (rb->pg_vec == NULL)
3819 for (i = 0; i < rb->pg_vec_len; i++) {
3821 void *kaddr = rb->pg_vec[i].buffer;
3824 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3825 page = pgv_to_page(kaddr);
3826 err = vm_insert_page(vma, start, page);
3835 atomic_inc(&po->mapped);
3836 vma->vm_ops = &packet_mmap_ops;
3840 mutex_unlock(&po->pg_vec_lock);
3844 static const struct proto_ops packet_ops_spkt = {
3845 .family = PF_PACKET,
3846 .owner = THIS_MODULE,
3847 .release = packet_release,
3848 .bind = packet_bind_spkt,
3849 .connect = sock_no_connect,
3850 .socketpair = sock_no_socketpair,
3851 .accept = sock_no_accept,
3852 .getname = packet_getname_spkt,
3853 .poll = datagram_poll,
3854 .ioctl = packet_ioctl,
3855 .listen = sock_no_listen,
3856 .shutdown = sock_no_shutdown,
3857 .setsockopt = sock_no_setsockopt,
3858 .getsockopt = sock_no_getsockopt,
3859 .sendmsg = packet_sendmsg_spkt,
3860 .recvmsg = packet_recvmsg,
3861 .mmap = sock_no_mmap,
3862 .sendpage = sock_no_sendpage,
3865 static const struct proto_ops packet_ops = {
3866 .family = PF_PACKET,
3867 .owner = THIS_MODULE,
3868 .release = packet_release,
3869 .bind = packet_bind,
3870 .connect = sock_no_connect,
3871 .socketpair = sock_no_socketpair,
3872 .accept = sock_no_accept,
3873 .getname = packet_getname,
3874 .poll = packet_poll,
3875 .ioctl = packet_ioctl,
3876 .listen = sock_no_listen,
3877 .shutdown = sock_no_shutdown,
3878 .setsockopt = packet_setsockopt,
3879 .getsockopt = packet_getsockopt,
3880 .sendmsg = packet_sendmsg,
3881 .recvmsg = packet_recvmsg,
3882 .mmap = packet_mmap,
3883 .sendpage = sock_no_sendpage,
3886 static const struct net_proto_family packet_family_ops = {
3887 .family = PF_PACKET,
3888 .create = packet_create,
3889 .owner = THIS_MODULE,
3892 static struct notifier_block packet_netdev_notifier = {
3893 .notifier_call = packet_notifier,
3896 #ifdef CONFIG_PROC_FS
3898 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3901 struct net *net = seq_file_net(seq);
3904 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3907 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3909 struct net *net = seq_file_net(seq);
3910 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3913 static void packet_seq_stop(struct seq_file *seq, void *v)
3919 static int packet_seq_show(struct seq_file *seq, void *v)
3921 if (v == SEQ_START_TOKEN)
3922 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3924 struct sock *s = sk_entry(v);
3925 const struct packet_sock *po = pkt_sk(s);
3928 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3930 atomic_read(&s->sk_refcnt),
3935 atomic_read(&s->sk_rmem_alloc),
3943 static const struct seq_operations packet_seq_ops = {
3944 .start = packet_seq_start,
3945 .next = packet_seq_next,
3946 .stop = packet_seq_stop,
3947 .show = packet_seq_show,
3950 static int packet_seq_open(struct inode *inode, struct file *file)
3952 return seq_open_net(inode, file, &packet_seq_ops,
3953 sizeof(struct seq_net_private));
3956 static const struct file_operations packet_seq_fops = {
3957 .owner = THIS_MODULE,
3958 .open = packet_seq_open,
3960 .llseek = seq_lseek,
3961 .release = seq_release_net,
3966 static int __net_init packet_net_init(struct net *net)
3968 spin_lock_init(&net->packet.sklist_lock);
3969 INIT_HLIST_HEAD(&net->packet.sklist);
3971 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3977 static void __net_exit packet_net_exit(struct net *net)
3979 proc_net_remove(net, "packet");
3982 static struct pernet_operations packet_net_ops = {
3983 .init = packet_net_init,
3984 .exit = packet_net_exit,
3988 static void __exit packet_exit(void)
3990 unregister_netdevice_notifier(&packet_netdev_notifier);
3991 unregister_pernet_subsys(&packet_net_ops);
3992 sock_unregister(PF_PACKET);
3993 proto_unregister(&packet_proto);
3996 static int __init packet_init(void)
3998 int rc = proto_register(&packet_proto, 0);
4003 sock_register(&packet_family_ops);
4004 register_pernet_subsys(&packet_net_ops);
4005 register_netdevice_notifier(&packet_netdev_notifier);
4010 module_init(packet_init);
4011 module_exit(packet_exit);
4012 MODULE_LICENSE("GPL");
4013 MODULE_ALIAS_NETPROTO(PF_PACKET);