2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
80 #include <asm/cacheflush.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
94 #include <net/inet_common.h>
99 - if device has no dev->hard_header routine, it adds and removes ll header
100 inside itself. In this case ll header is invisible outside of device,
101 but higher levels still should reserve dev->hard_header_len.
102 Some devices are enough clever to reallocate skb, when header
103 will not fit to reserved space (tunnel), another ones are silly
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
111 Incoming, dev->hard_header!=NULL
112 mac_header -> ll header
115 Outgoing, dev->hard_header!=NULL
116 mac_header -> ll header
119 Incoming, dev->hard_header==NULL
120 mac_header -> UNKNOWN position. It is very likely, that it points to ll
121 header. PPP makes it, that is wrong, because introduce
122 assymetry between rx and tx paths.
125 Outgoing, dev->hard_header==NULL
126 mac_header -> data. ll header is still not built!
130 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
136 dev->hard_header != NULL
137 mac_header -> ll header
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144 We should set nh.raw on output to correct posistion,
145 packet classifier depends on it.
148 /* Private packet socket structures. */
150 struct packet_mclist {
151 struct packet_mclist *next;
156 unsigned char addr[MAX_ADDR_LEN];
158 /* identical to struct packet_mreq except it has
159 * a longer address field.
161 struct packet_mreq_max {
163 unsigned short mr_type;
164 unsigned short mr_alen;
165 unsigned char mr_address[MAX_ADDR_LEN];
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
182 unsigned int feature_req_word;
184 unsigned char reset_pending_on_curr_blk;
185 unsigned char delete_blk_timer;
186 unsigned short kactive_blk_num;
187 unsigned short blk_sizeof_priv;
189 /* last_kactive_blk_num:
190 * trick to see if user-space has caught up
191 * in order to avoid refreshing timer when every single pkt arrives.
193 unsigned short last_kactive_blk_num;
198 unsigned int knum_blocks;
199 uint64_t knxt_seq_num;
204 atomic_t blk_fill_in_prog;
206 /* Default is set to 8ms */
207 #define DEFAULT_PRB_RETIRE_TOV (8)
209 unsigned short retire_blk_tov;
210 unsigned short version;
211 unsigned long tov_in_jiffies;
213 /* timer to retire an outstanding block */
214 struct timer_list retire_blk_timer;
217 #define PGV_FROM_VMALLOC 1
222 struct packet_ring_buffer {
225 unsigned int frames_per_block;
226 unsigned int frame_size;
227 unsigned int frame_max;
229 unsigned int pg_vec_order;
230 unsigned int pg_vec_pages;
231 unsigned int pg_vec_len;
233 struct tpacket_kbdq_core prb_bdqc;
237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
239 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
240 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
241 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
248 static void *packet_previous_frame(struct packet_sock *po,
249 struct packet_ring_buffer *rb,
251 static void packet_increment_head(struct packet_ring_buffer *buff);
252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253 struct tpacket_block_desc *);
254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
255 struct packet_sock *);
256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
257 struct packet_sock *, unsigned int status);
258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
259 static void prb_open_block(struct tpacket_kbdq_core *,
260 struct tpacket_block_desc *);
261 static void prb_retire_rx_blk_timer_expired(unsigned long);
262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263 static void prb_init_blk_timer(struct packet_sock *,
264 struct tpacket_kbdq_core *,
265 void (*func) (unsigned long));
266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268 struct tpacket3_hdr *);
269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270 struct tpacket3_hdr *);
271 static void packet_flush_mclist(struct sock *sk);
273 struct packet_fanout;
275 /* struct sock has to be the first member of packet_sock */
277 struct packet_fanout *fanout;
278 struct tpacket_stats stats;
279 union tpacket_stats_u stats_u;
280 struct packet_ring_buffer rx_ring;
281 struct packet_ring_buffer tx_ring;
283 spinlock_t bind_lock;
284 struct mutex pg_vec_lock;
285 unsigned int running:1, /* prot_hook is attached*/
289 int ifindex; /* bound device */
291 struct packet_mclist *mclist;
293 enum tpacket_versions tp_version;
294 unsigned int tp_hdrlen;
295 unsigned int tp_reserve;
296 unsigned int tp_loss:1;
297 unsigned int tp_tstamp;
298 struct packet_type prot_hook ____cacheline_aligned_in_smp;
301 #define PACKET_FANOUT_MAX 256
303 struct packet_fanout {
307 unsigned int num_members;
312 struct list_head list;
313 struct sock *arr[PACKET_FANOUT_MAX];
316 struct packet_type prot_hook ____cacheline_aligned_in_smp;
319 struct packet_skb_cb {
320 unsigned int origlen;
322 struct sockaddr_pkt pkt;
323 struct sockaddr_ll ll;
327 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
329 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
330 #define GET_PBLOCK_DESC(x, bid) \
331 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
333 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
334 #define GET_NEXT_PRB_BLK_NUM(x) \
335 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336 ((x)->kactive_blk_num+1) : 0)
338 static struct packet_sock *pkt_sk(struct sock *sk)
340 return (struct packet_sock *)sk;
343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
346 /* register_prot_hook must be invoked with the po->bind_lock held,
347 * or from a context in which asynchronous accesses to the packet
348 * socket is not possible (packet_create()).
350 static void register_prot_hook(struct sock *sk)
352 struct packet_sock *po = pkt_sk(sk);
355 __fanout_link(sk, po);
357 dev_add_pack(&po->prot_hook);
363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364 * held. If the sync parameter is true, we will temporarily drop
365 * the po->bind_lock and do a synchronize_net to make sure no
366 * asynchronous packet processing paths still refer to the elements
367 * of po->prot_hook. If the sync parameter is false, it is the
368 * callers responsibility to take care of this.
370 static void __unregister_prot_hook(struct sock *sk, bool sync)
372 struct packet_sock *po = pkt_sk(sk);
376 __fanout_unlink(sk, po);
378 __dev_remove_pack(&po->prot_hook);
382 spin_unlock(&po->bind_lock);
384 spin_lock(&po->bind_lock);
388 static void unregister_prot_hook(struct sock *sk, bool sync)
390 struct packet_sock *po = pkt_sk(sk);
393 __unregister_prot_hook(sk, sync);
396 static inline __pure struct page *pgv_to_page(void *addr)
398 if (is_vmalloc_addr(addr))
399 return vmalloc_to_page(addr);
400 return virt_to_page(addr);
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
406 struct tpacket_hdr *h1;
407 struct tpacket2_hdr *h2;
412 switch (po->tp_version) {
414 h.h1->tp_status = status;
415 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
418 h.h2->tp_status = status;
419 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
423 WARN(1, "TPACKET version not supported.\n");
430 static int __packet_get_status(struct packet_sock *po, void *frame)
433 struct tpacket_hdr *h1;
434 struct tpacket2_hdr *h2;
441 switch (po->tp_version) {
443 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
444 return h.h1->tp_status;
446 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
447 return h.h2->tp_status;
450 WARN(1, "TPACKET version not supported.\n");
456 static void *packet_lookup_frame(struct packet_sock *po,
457 struct packet_ring_buffer *rb,
458 unsigned int position,
461 unsigned int pg_vec_pos, frame_offset;
463 struct tpacket_hdr *h1;
464 struct tpacket2_hdr *h2;
468 pg_vec_pos = position / rb->frames_per_block;
469 frame_offset = position % rb->frames_per_block;
471 h.raw = rb->pg_vec[pg_vec_pos].buffer +
472 (frame_offset * rb->frame_size);
474 if (status != __packet_get_status(po, h.raw))
480 static void *packet_current_frame(struct packet_sock *po,
481 struct packet_ring_buffer *rb,
484 return packet_lookup_frame(po, rb, rb->head, status);
487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
489 del_timer_sync(&pkc->retire_blk_timer);
492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
494 struct sk_buff_head *rb_queue)
496 struct tpacket_kbdq_core *pkc;
498 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
500 spin_lock(&rb_queue->lock);
501 pkc->delete_blk_timer = 1;
502 spin_unlock(&rb_queue->lock);
504 prb_del_retire_blk_timer(pkc);
507 static void prb_init_blk_timer(struct packet_sock *po,
508 struct tpacket_kbdq_core *pkc,
509 void (*func) (unsigned long))
511 init_timer(&pkc->retire_blk_timer);
512 pkc->retire_blk_timer.data = (long)po;
513 pkc->retire_blk_timer.function = func;
514 pkc->retire_blk_timer.expires = jiffies;
517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
519 struct tpacket_kbdq_core *pkc;
524 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529 int blk_size_in_bytes)
531 struct net_device *dev;
532 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
533 struct ethtool_cmd ecmd;
537 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
538 if (unlikely(!dev)) {
540 return DEFAULT_PRB_RETIRE_TOV;
542 err = __ethtool_get_settings(dev, &ecmd);
545 switch (ecmd.speed) {
555 * If the link speed is so slow you don't really
556 * need to worry about perf anyways
561 return DEFAULT_PRB_RETIRE_TOV;
565 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
578 union tpacket_req_u *req_u)
580 p1->feature_req_word = req_u->req3.tp_feature_req_word;
583 static void init_prb_bdqc(struct packet_sock *po,
584 struct packet_ring_buffer *rb,
586 union tpacket_req_u *req_u, int tx_ring)
588 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
589 struct tpacket_block_desc *pbd;
591 memset(p1, 0x0, sizeof(*p1));
593 p1->knxt_seq_num = 1;
595 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
596 p1->pkblk_start = (char *)pg_vec[0].buffer;
597 p1->kblk_size = req_u->req3.tp_block_size;
598 p1->knum_blocks = req_u->req3.tp_block_nr;
599 p1->hdrlen = po->tp_hdrlen;
600 p1->version = po->tp_version;
601 p1->last_kactive_blk_num = 0;
602 po->stats_u.stats3.tp_freeze_q_cnt = 0;
603 if (req_u->req3.tp_retire_blk_tov)
604 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
606 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
607 req_u->req3.tp_block_size);
608 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
609 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
611 prb_init_ft_ops(p1, req_u);
612 prb_setup_retire_blk_timer(po, tx_ring);
613 prb_open_block(p1, pbd);
616 /* Do NOT update the last_blk_num first.
617 * Assumes sk_buff_head lock is held.
619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
621 mod_timer(&pkc->retire_blk_timer,
622 jiffies + pkc->tov_in_jiffies);
623 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
628 * 1) We refresh the timer only when we open a block.
629 * By doing this we don't waste cycles refreshing the timer
630 * on packet-by-packet basis.
632 * With a 1MB block-size, on a 1Gbps line, it will take
633 * i) ~8 ms to fill a block + ii) memcpy etc.
634 * In this cut we are not accounting for the memcpy time.
636 * So, if the user sets the 'tmo' to 10ms then the timer
637 * will never fire while the block is still getting filled
638 * (which is what we want). However, the user could choose
639 * to close a block early and that's fine.
641 * But when the timer does fire, we check whether or not to refresh it.
642 * Since the tmo granularity is in msecs, it is not too expensive
643 * to refresh the timer, lets say every '8' msecs.
644 * Either the user can set the 'tmo' or we can derive it based on
645 * a) line-speed and b) block-size.
646 * prb_calc_retire_blk_tmo() calculates the tmo.
649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
651 struct packet_sock *po = (struct packet_sock *)data;
652 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
654 struct tpacket_block_desc *pbd;
656 spin_lock(&po->sk.sk_receive_queue.lock);
658 frozen = prb_queue_frozen(pkc);
659 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
661 if (unlikely(pkc->delete_blk_timer))
664 /* We only need to plug the race when the block is partially filled.
666 * lock(); increment BLOCK_NUM_PKTS; unlock()
667 * copy_bits() is in progress ...
668 * timer fires on other cpu:
669 * we can't retire the current block because copy_bits
673 if (BLOCK_NUM_PKTS(pbd)) {
674 while (atomic_read(&pkc->blk_fill_in_prog)) {
675 /* Waiting for skb_copy_bits to finish... */
680 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
682 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
683 if (!prb_dispatch_next_block(pkc, po))
688 /* Case 1. Queue was frozen because user-space was
691 if (prb_curr_blk_in_use(pkc, pbd)) {
693 * Ok, user-space is still behind.
694 * So just refresh the timer.
698 /* Case 2. queue was frozen,user-space caught up,
699 * now the link went idle && the timer fired.
700 * We don't have a block to close.So we open this
701 * block and restart the timer.
702 * opening a block thaws the queue,restarts timer
703 * Thawing/timer-refresh is a side effect.
705 prb_open_block(pkc, pbd);
712 _prb_refresh_rx_retire_blk_timer(pkc);
715 spin_unlock(&po->sk.sk_receive_queue.lock);
718 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
719 struct tpacket_block_desc *pbd1, __u32 status)
721 /* Flush everything minus the block header */
723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
728 /* Skip the block header(we know header WILL fit in 4K) */
731 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
732 for (; start < end; start += PAGE_SIZE)
733 flush_dcache_page(pgv_to_page(start));
738 /* Now update the block status. */
740 BLOCK_STATUS(pbd1) = status;
742 /* Flush the block header */
744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
746 flush_dcache_page(pgv_to_page(start));
756 * 2) Increment active_blk_num
758 * Note:We DONT refresh the timer on purpose.
759 * Because almost always the next block will be opened.
761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
762 struct tpacket_block_desc *pbd1,
763 struct packet_sock *po, unsigned int stat)
765 __u32 status = TP_STATUS_USER | stat;
767 struct tpacket3_hdr *last_pkt;
768 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
770 if (po->stats.tp_drops)
771 status |= TP_STATUS_LOSING;
773 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
774 last_pkt->tp_next_offset = 0;
776 /* Get the ts of the last pkt */
777 if (BLOCK_NUM_PKTS(pbd1)) {
778 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
779 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
781 /* Ok, we tmo'd - so get the current time */
784 h1->ts_last_pkt.ts_sec = ts.tv_sec;
785 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
790 /* Flush the block */
791 prb_flush_block(pkc1, pbd1, status);
793 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
796 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
798 pkc->reset_pending_on_curr_blk = 0;
802 * Side effect of opening a block:
804 * 1) prb_queue is thawed.
805 * 2) retire_blk_timer is refreshed.
808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
809 struct tpacket_block_desc *pbd1)
812 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
816 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
818 /* We could have just memset this but we will lose the
819 * flexibility of making the priv area sticky
821 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
822 BLOCK_NUM_PKTS(pbd1) = 0;
823 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827 pkc1->pkblk_start = (char *)pbd1;
828 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
829 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
830 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835 prb_thaw_queue(pkc1);
836 _prb_refresh_rx_retire_blk_timer(pkc1);
843 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
844 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats_u.stats3.tp_freeze_q_cnt++;
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
890 struct tpacket_block_desc *pbd;
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
905 * open this block and return the offset where the first packet
906 * needs to get stored.
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 while (atomic_read(&pkc->blk_fill_in_prog)) {
930 /* Waiting for skb_copy_bits to finish... */
934 prb_close_block(pkc, pbd, po, status);
938 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
943 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
944 struct tpacket_block_desc *pbd)
946 return TP_STATUS_USER & BLOCK_STATUS(pbd);
949 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
951 return pkc->reset_pending_on_curr_blk;
954 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
956 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
957 atomic_dec(&pkc->blk_fill_in_prog);
960 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
963 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
966 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
967 struct tpacket3_hdr *ppd)
969 ppd->hv1.tp_rxhash = 0;
972 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
973 struct tpacket3_hdr *ppd)
975 if (vlan_tx_tag_present(pkc->skb)) {
976 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
977 ppd->tp_status = TP_STATUS_VLAN_VALID;
979 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
984 struct tpacket3_hdr *ppd)
986 prb_fill_vlan_info(pkc, ppd);
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
991 prb_clear_rxhash(pkc, ppd);
994 static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
999 struct tpacket3_hdr *ppd;
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 atomic_inc(&pkc->blk_fill_in_prog);
1008 prb_run_all_ft_ops(pkc, ppd);
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1022 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1031 if (prb_curr_blk_in_use(pkc, pbd)) {
1032 /* Can't record this packet */
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1041 prb_open_block(pkc, pbd);
1046 curr = pkc->nxt_offset;
1048 end = (char *) ((char *)pbd + pkc->kblk_size);
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1079 switch (po->tp_version) {
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1086 return __packet_lookup_frame_in_block(po, skb, status, len);
1088 WARN(1, "TPACKET version not supported\n");
1094 static void *prb_lookup_block(struct packet_sock *po,
1095 struct packet_ring_buffer *rb,
1096 unsigned int previous,
1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1102 if (status != BLOCK_STATUS(pbd))
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1113 prev = rb->prb_bdqc.knum_blocks-1;
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1133 return __prb_previous_block(po, rb, status);
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137 struct packet_ring_buffer *rb)
1139 switch (po->tp_version) {
1142 return packet_increment_head(rb);
1145 WARN(1, "TPACKET version not supported.\n");
1151 static void *packet_previous_frame(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1164 static void packet_sock_destruct(struct sock *sk)
1166 skb_queue_purge(&sk->sk_error_queue);
1168 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1171 if (!sock_flag(sk, SOCK_DEAD)) {
1172 pr_err("Attempt to release alive packet socket: %p\n", sk);
1176 sk_refcnt_debug_dec(sk);
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1181 int x = atomic_read(&f->rr_cur) + 1;
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1191 u32 idx, hash = skb->rxhash;
1193 idx = ((u64)hash * num) >> 32;
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1202 cur = atomic_read(&f->rr_cur);
1203 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204 fanout_rr_next(f, num))) != cur)
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1211 unsigned int cpu = smp_processor_id();
1213 return f->arr[cpu % num];
1216 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217 struct packet_type *pt, struct net_device *orig_dev)
1219 struct packet_fanout *f = pt->af_packet_priv;
1220 unsigned int num = f->num_members;
1221 struct packet_sock *po;
1224 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1231 case PACKET_FANOUT_HASH:
1234 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1238 skb_get_rxhash(skb);
1239 sk = fanout_demux_hash(f, skb, num);
1241 case PACKET_FANOUT_LB:
1242 sk = fanout_demux_lb(f, skb, num);
1244 case PACKET_FANOUT_CPU:
1245 sk = fanout_demux_cpu(f, skb, num);
1251 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1254 static DEFINE_MUTEX(fanout_mutex);
1255 static LIST_HEAD(fanout_list);
1257 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1259 struct packet_fanout *f = po->fanout;
1261 spin_lock(&f->lock);
1262 f->arr[f->num_members] = sk;
1265 spin_unlock(&f->lock);
1268 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1270 struct packet_fanout *f = po->fanout;
1273 spin_lock(&f->lock);
1274 for (i = 0; i < f->num_members; i++) {
1275 if (f->arr[i] == sk)
1278 BUG_ON(i >= f->num_members);
1279 f->arr[i] = f->arr[f->num_members - 1];
1281 spin_unlock(&f->lock);
1284 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1286 struct packet_sock *po = pkt_sk(sk);
1287 struct packet_fanout *f, *match;
1288 u8 type = type_flags & 0xff;
1289 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1293 case PACKET_FANOUT_HASH:
1294 case PACKET_FANOUT_LB:
1295 case PACKET_FANOUT_CPU:
1307 mutex_lock(&fanout_mutex);
1309 list_for_each_entry(f, &fanout_list, list) {
1311 read_pnet(&f->net) == sock_net(sk)) {
1317 if (match && match->defrag != defrag)
1321 match = kzalloc(sizeof(*match), GFP_KERNEL);
1324 write_pnet(&match->net, sock_net(sk));
1327 match->defrag = defrag;
1328 atomic_set(&match->rr_cur, 0);
1329 INIT_LIST_HEAD(&match->list);
1330 spin_lock_init(&match->lock);
1331 atomic_set(&match->sk_ref, 0);
1332 match->prot_hook.type = po->prot_hook.type;
1333 match->prot_hook.dev = po->prot_hook.dev;
1334 match->prot_hook.func = packet_rcv_fanout;
1335 match->prot_hook.af_packet_priv = match;
1336 dev_add_pack(&match->prot_hook);
1337 list_add(&match->list, &fanout_list);
1340 if (match->type == type &&
1341 match->prot_hook.type == po->prot_hook.type &&
1342 match->prot_hook.dev == po->prot_hook.dev) {
1344 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1345 __dev_remove_pack(&po->prot_hook);
1347 atomic_inc(&match->sk_ref);
1348 __fanout_link(sk, po);
1353 mutex_unlock(&fanout_mutex);
1357 static void fanout_release(struct sock *sk)
1359 struct packet_sock *po = pkt_sk(sk);
1360 struct packet_fanout *f;
1368 mutex_lock(&fanout_mutex);
1369 if (atomic_dec_and_test(&f->sk_ref)) {
1371 dev_remove_pack(&f->prot_hook);
1374 mutex_unlock(&fanout_mutex);
1377 static const struct proto_ops packet_ops;
1379 static const struct proto_ops packet_ops_spkt;
1381 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382 struct packet_type *pt, struct net_device *orig_dev)
1385 struct sockaddr_pkt *spkt;
1388 * When we registered the protocol we saved the socket in the data
1389 * field for just this event.
1392 sk = pt->af_packet_priv;
1395 * Yank back the headers [hope the device set this
1396 * right or kerboom...]
1398 * Incoming packets have ll header pulled,
1401 * For outgoing ones skb->data == skb_mac_header(skb)
1402 * so that this procedure is noop.
1405 if (skb->pkt_type == PACKET_LOOPBACK)
1408 if (!net_eq(dev_net(dev), sock_net(sk)))
1411 skb = skb_share_check(skb, GFP_ATOMIC);
1415 /* drop any routing info */
1418 /* drop conntrack reference */
1421 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1423 skb_push(skb, skb->data - skb_mac_header(skb));
1426 * The SOCK_PACKET socket receives _all_ frames.
1429 spkt->spkt_family = dev->type;
1430 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431 spkt->spkt_protocol = skb->protocol;
1434 * Charge the memory to the socket. This is done specifically
1435 * to prevent sockets using all the memory up.
1438 if (sock_queue_rcv_skb(sk, skb) == 0)
1449 * Output a raw packet to a device layer. This bypasses all the other
1450 * protocol layers and you must therefore supply it with a complete frame
1453 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454 struct msghdr *msg, size_t len)
1456 struct sock *sk = sock->sk;
1457 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1458 struct sk_buff *skb = NULL;
1459 struct net_device *dev;
1465 * Get and verify the address.
1469 if (msg->msg_namelen < sizeof(struct sockaddr))
1471 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1472 proto = saddr->spkt_protocol;
1474 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1477 * Find the device first to size check it
1480 saddr->spkt_device[13] = 0;
1483 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1489 if (!(dev->flags & IFF_UP))
1493 * You may not queue a frame bigger than the mtu. This is the lowest level
1494 * raw protocol and you must do your own fragmentation at this level.
1497 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1498 if (!netif_supports_nofcs(dev)) {
1499 err = -EPROTONOSUPPORT;
1502 extra_len = 4; /* We're doing our own CRC */
1506 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1510 size_t reserved = LL_RESERVED_SPACE(dev);
1511 int tlen = dev->needed_tailroom;
1512 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1515 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1518 /* FIXME: Save some space for broken drivers that write a hard
1519 * header at transmission time by themselves. PPP is the notable
1520 * one here. This should really be fixed at the driver level.
1522 skb_reserve(skb, reserved);
1523 skb_reset_network_header(skb);
1525 /* Try to align data part correctly */
1530 skb_reset_network_header(skb);
1532 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1538 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1539 /* Earlier code assumed this would be a VLAN pkt,
1540 * double-check this now that we have the actual
1543 struct ethhdr *ehdr;
1544 skb_reset_mac_header(skb);
1545 ehdr = eth_hdr(skb);
1546 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1552 skb->protocol = proto;
1554 skb->priority = sk->sk_priority;
1555 skb->mark = sk->sk_mark;
1556 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1560 if (unlikely(extra_len == 4))
1563 dev_queue_xmit(skb);
1574 static unsigned int run_filter(const struct sk_buff *skb,
1575 const struct sock *sk,
1578 struct sk_filter *filter;
1581 filter = rcu_dereference(sk->sk_filter);
1583 res = SK_RUN_FILTER(filter, skb);
1590 * This function makes lazy skb cloning in hope that most of packets
1591 * are discarded by BPF.
1593 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1594 * and skb->cb are mangled. It works because (and until) packets
1595 * falling here are owned by current CPU. Output packets are cloned
1596 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1597 * sequencially, so that if we return skb to original state on exit,
1598 * we will not harm anyone.
1601 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1602 struct packet_type *pt, struct net_device *orig_dev)
1605 struct sockaddr_ll *sll;
1606 struct packet_sock *po;
1607 u8 *skb_head = skb->data;
1608 int skb_len = skb->len;
1609 unsigned int snaplen, res;
1611 if (skb->pkt_type == PACKET_LOOPBACK)
1614 sk = pt->af_packet_priv;
1617 if (!net_eq(dev_net(dev), sock_net(sk)))
1622 if (dev->header_ops) {
1623 /* The device has an explicit notion of ll header,
1624 * exported to higher levels.
1626 * Otherwise, the device hides details of its frame
1627 * structure, so that corresponding packet head is
1628 * never delivered to user.
1630 if (sk->sk_type != SOCK_DGRAM)
1631 skb_push(skb, skb->data - skb_mac_header(skb));
1632 else if (skb->pkt_type == PACKET_OUTGOING) {
1633 /* Special case: outgoing packets have ll header at head */
1634 skb_pull(skb, skb_network_offset(skb));
1640 res = run_filter(skb, sk, snaplen);
1642 goto drop_n_restore;
1646 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1649 if (skb_shared(skb)) {
1650 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1654 if (skb_head != skb->data) {
1655 skb->data = skb_head;
1662 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1665 sll = &PACKET_SKB_CB(skb)->sa.ll;
1666 sll->sll_family = AF_PACKET;
1667 sll->sll_hatype = dev->type;
1668 sll->sll_protocol = skb->protocol;
1669 sll->sll_pkttype = skb->pkt_type;
1670 if (unlikely(po->origdev))
1671 sll->sll_ifindex = orig_dev->ifindex;
1673 sll->sll_ifindex = dev->ifindex;
1675 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1677 PACKET_SKB_CB(skb)->origlen = skb->len;
1679 if (pskb_trim(skb, snaplen))
1682 skb_set_owner_r(skb, sk);
1686 /* drop conntrack reference */
1689 spin_lock(&sk->sk_receive_queue.lock);
1690 po->stats.tp_packets++;
1691 skb->dropcount = atomic_read(&sk->sk_drops);
1692 __skb_queue_tail(&sk->sk_receive_queue, skb);
1693 spin_unlock(&sk->sk_receive_queue.lock);
1694 sk->sk_data_ready(sk, skb->len);
1698 spin_lock(&sk->sk_receive_queue.lock);
1699 po->stats.tp_drops++;
1700 atomic_inc(&sk->sk_drops);
1701 spin_unlock(&sk->sk_receive_queue.lock);
1704 if (skb_head != skb->data && skb_shared(skb)) {
1705 skb->data = skb_head;
1713 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1714 struct packet_type *pt, struct net_device *orig_dev)
1717 struct packet_sock *po;
1718 struct sockaddr_ll *sll;
1720 struct tpacket_hdr *h1;
1721 struct tpacket2_hdr *h2;
1722 struct tpacket3_hdr *h3;
1725 u8 *skb_head = skb->data;
1726 int skb_len = skb->len;
1727 unsigned int snaplen, res;
1728 unsigned long status = TP_STATUS_USER;
1729 unsigned short macoff, netoff, hdrlen;
1730 struct sk_buff *copy_skb = NULL;
1733 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1735 if (skb->pkt_type == PACKET_LOOPBACK)
1738 sk = pt->af_packet_priv;
1741 if (!net_eq(dev_net(dev), sock_net(sk)))
1744 if (dev->header_ops) {
1745 if (sk->sk_type != SOCK_DGRAM)
1746 skb_push(skb, skb->data - skb_mac_header(skb));
1747 else if (skb->pkt_type == PACKET_OUTGOING) {
1748 /* Special case: outgoing packets have ll header at head */
1749 skb_pull(skb, skb_network_offset(skb));
1753 if (skb->ip_summed == CHECKSUM_PARTIAL)
1754 status |= TP_STATUS_CSUMNOTREADY;
1758 res = run_filter(skb, sk, snaplen);
1760 goto drop_n_restore;
1764 if (sk->sk_type == SOCK_DGRAM) {
1765 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1768 unsigned maclen = skb_network_offset(skb);
1769 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1770 (maclen < 16 ? 16 : maclen)) +
1772 macoff = netoff - maclen;
1774 if (po->tp_version <= TPACKET_V2) {
1775 if (macoff + snaplen > po->rx_ring.frame_size) {
1776 if (po->copy_thresh &&
1777 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1778 if (skb_shared(skb)) {
1779 copy_skb = skb_clone(skb, GFP_ATOMIC);
1781 copy_skb = skb_get(skb);
1782 skb_head = skb->data;
1785 skb_set_owner_r(copy_skb, sk);
1787 snaplen = po->rx_ring.frame_size - macoff;
1788 if ((int)snaplen < 0)
1792 spin_lock(&sk->sk_receive_queue.lock);
1793 h.raw = packet_current_rx_frame(po, skb,
1794 TP_STATUS_KERNEL, (macoff+snaplen));
1797 if (po->tp_version <= TPACKET_V2) {
1798 packet_increment_rx_head(po, &po->rx_ring);
1800 * LOSING will be reported till you read the stats,
1801 * because it's COR - Clear On Read.
1802 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1805 if (po->stats.tp_drops)
1806 status |= TP_STATUS_LOSING;
1808 po->stats.tp_packets++;
1810 status |= TP_STATUS_COPY;
1811 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1813 spin_unlock(&sk->sk_receive_queue.lock);
1815 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1817 switch (po->tp_version) {
1819 h.h1->tp_len = skb->len;
1820 h.h1->tp_snaplen = snaplen;
1821 h.h1->tp_mac = macoff;
1822 h.h1->tp_net = netoff;
1823 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1824 && shhwtstamps->syststamp.tv64)
1825 tv = ktime_to_timeval(shhwtstamps->syststamp);
1826 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1827 && shhwtstamps->hwtstamp.tv64)
1828 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1829 else if (skb->tstamp.tv64)
1830 tv = ktime_to_timeval(skb->tstamp);
1832 do_gettimeofday(&tv);
1833 h.h1->tp_sec = tv.tv_sec;
1834 h.h1->tp_usec = tv.tv_usec;
1835 hdrlen = sizeof(*h.h1);
1838 h.h2->tp_len = skb->len;
1839 h.h2->tp_snaplen = snaplen;
1840 h.h2->tp_mac = macoff;
1841 h.h2->tp_net = netoff;
1842 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1843 && shhwtstamps->syststamp.tv64)
1844 ts = ktime_to_timespec(shhwtstamps->syststamp);
1845 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1846 && shhwtstamps->hwtstamp.tv64)
1847 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1848 else if (skb->tstamp.tv64)
1849 ts = ktime_to_timespec(skb->tstamp);
1851 getnstimeofday(&ts);
1852 h.h2->tp_sec = ts.tv_sec;
1853 h.h2->tp_nsec = ts.tv_nsec;
1854 if (vlan_tx_tag_present(skb)) {
1855 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1856 status |= TP_STATUS_VLAN_VALID;
1858 h.h2->tp_vlan_tci = 0;
1860 h.h2->tp_padding = 0;
1861 hdrlen = sizeof(*h.h2);
1864 /* tp_nxt_offset,vlan are already populated above.
1865 * So DONT clear those fields here
1867 h.h3->tp_status |= status;
1868 h.h3->tp_len = skb->len;
1869 h.h3->tp_snaplen = snaplen;
1870 h.h3->tp_mac = macoff;
1871 h.h3->tp_net = netoff;
1872 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1873 && shhwtstamps->syststamp.tv64)
1874 ts = ktime_to_timespec(shhwtstamps->syststamp);
1875 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1876 && shhwtstamps->hwtstamp.tv64)
1877 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1878 else if (skb->tstamp.tv64)
1879 ts = ktime_to_timespec(skb->tstamp);
1881 getnstimeofday(&ts);
1882 h.h3->tp_sec = ts.tv_sec;
1883 h.h3->tp_nsec = ts.tv_nsec;
1884 hdrlen = sizeof(*h.h3);
1890 sll = h.raw + TPACKET_ALIGN(hdrlen);
1891 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1892 sll->sll_family = AF_PACKET;
1893 sll->sll_hatype = dev->type;
1894 sll->sll_protocol = skb->protocol;
1895 sll->sll_pkttype = skb->pkt_type;
1896 if (unlikely(po->origdev))
1897 sll->sll_ifindex = orig_dev->ifindex;
1899 sll->sll_ifindex = dev->ifindex;
1902 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1906 if (po->tp_version <= TPACKET_V2) {
1907 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1908 + macoff + snaplen);
1909 for (start = h.raw; start < end; start += PAGE_SIZE)
1910 flush_dcache_page(pgv_to_page(start));
1915 if (po->tp_version <= TPACKET_V2)
1916 __packet_set_status(po, h.raw, status);
1918 prb_clear_blk_fill_status(&po->rx_ring);
1920 sk->sk_data_ready(sk, 0);
1923 if (skb_head != skb->data && skb_shared(skb)) {
1924 skb->data = skb_head;
1932 po->stats.tp_drops++;
1933 spin_unlock(&sk->sk_receive_queue.lock);
1935 sk->sk_data_ready(sk, 0);
1936 kfree_skb(copy_skb);
1937 goto drop_n_restore;
1940 static void tpacket_destruct_skb(struct sk_buff *skb)
1942 struct packet_sock *po = pkt_sk(skb->sk);
1945 if (likely(po->tx_ring.pg_vec)) {
1946 ph = skb_shinfo(skb)->destructor_arg;
1947 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
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, int hlen)
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, hlen);
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 int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2056 struct sk_buff *skb;
2057 struct net_device *dev;
2059 bool need_rls_dev = false;
2060 int err, reserve = 0;
2062 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2063 int tp_len, size_max;
2064 unsigned char *addr;
2069 mutex_lock(&po->pg_vec_lock);
2072 if (saddr == NULL) {
2073 dev = po->prot_hook.dev;
2078 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2080 if (msg->msg_namelen < (saddr->sll_halen
2081 + offsetof(struct sockaddr_ll,
2084 proto = saddr->sll_protocol;
2085 addr = saddr->sll_addr;
2086 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2087 need_rls_dev = true;
2091 if (unlikely(dev == NULL))
2094 reserve = dev->hard_header_len;
2097 if (unlikely(!(dev->flags & IFF_UP)))
2100 size_max = po->tx_ring.frame_size
2101 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2103 if (size_max > dev->mtu + reserve)
2104 size_max = dev->mtu + reserve;
2107 ph = packet_current_frame(po, &po->tx_ring,
2108 TP_STATUS_SEND_REQUEST);
2110 if (unlikely(ph == NULL)) {
2115 status = TP_STATUS_SEND_REQUEST;
2116 hlen = LL_RESERVED_SPACE(dev);
2117 tlen = dev->needed_tailroom;
2118 skb = sock_alloc_send_skb(&po->sk,
2119 hlen + tlen + sizeof(struct sockaddr_ll),
2122 if (unlikely(skb == NULL))
2125 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2128 if (unlikely(tp_len < 0)) {
2130 __packet_set_status(po, ph,
2131 TP_STATUS_AVAILABLE);
2132 packet_increment_head(&po->tx_ring);
2136 status = TP_STATUS_WRONG_FORMAT;
2142 skb->destructor = tpacket_destruct_skb;
2143 __packet_set_status(po, ph, TP_STATUS_SENDING);
2144 atomic_inc(&po->tx_ring.pending);
2146 status = TP_STATUS_SEND_REQUEST;
2147 err = dev_queue_xmit(skb);
2148 if (unlikely(err > 0)) {
2149 err = net_xmit_errno(err);
2150 if (err && __packet_get_status(po, ph) ==
2151 TP_STATUS_AVAILABLE) {
2152 /* skb was destructed already */
2157 * skb was dropped but not destructed yet;
2158 * let's treat it like congestion or err < 0
2162 packet_increment_head(&po->tx_ring);
2164 } while (likely((ph != NULL) ||
2165 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2166 (atomic_read(&po->tx_ring.pending))))
2173 __packet_set_status(po, ph, status);
2179 mutex_unlock(&po->pg_vec_lock);
2183 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2184 size_t reserve, size_t len,
2185 size_t linear, int noblock,
2188 struct sk_buff *skb;
2190 /* Under a page? Don't bother with paged skb. */
2191 if (prepad + len < PAGE_SIZE || !linear)
2194 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2199 skb_reserve(skb, reserve);
2200 skb_put(skb, linear);
2201 skb->data_len = len - linear;
2202 skb->len += len - linear;
2207 static int packet_snd(struct socket *sock,
2208 struct msghdr *msg, size_t len)
2210 struct sock *sk = sock->sk;
2211 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2212 struct sk_buff *skb;
2213 struct net_device *dev;
2215 bool need_rls_dev = false;
2216 unsigned char *addr;
2217 int err, reserve = 0;
2218 struct virtio_net_hdr vnet_hdr = { 0 };
2221 struct packet_sock *po = pkt_sk(sk);
2222 unsigned short gso_type = 0;
2227 * Get and verify the address.
2230 if (saddr == NULL) {
2231 dev = po->prot_hook.dev;
2236 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2238 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2240 proto = saddr->sll_protocol;
2241 addr = saddr->sll_addr;
2242 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2243 need_rls_dev = true;
2249 if (sock->type == SOCK_RAW)
2250 reserve = dev->hard_header_len;
2253 if (!(dev->flags & IFF_UP))
2256 if (po->has_vnet_hdr) {
2257 vnet_hdr_len = sizeof(vnet_hdr);
2260 if (len < vnet_hdr_len)
2263 len -= vnet_hdr_len;
2265 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2270 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2271 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2273 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2274 vnet_hdr.csum_offset + 2;
2277 if (vnet_hdr.hdr_len > len)
2280 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2281 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2282 case VIRTIO_NET_HDR_GSO_TCPV4:
2283 gso_type = SKB_GSO_TCPV4;
2285 case VIRTIO_NET_HDR_GSO_TCPV6:
2286 gso_type = SKB_GSO_TCPV6;
2288 case VIRTIO_NET_HDR_GSO_UDP:
2289 gso_type = SKB_GSO_UDP;
2295 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2296 gso_type |= SKB_GSO_TCP_ECN;
2298 if (vnet_hdr.gso_size == 0)
2304 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2305 if (!netif_supports_nofcs(dev)) {
2306 err = -EPROTONOSUPPORT;
2309 extra_len = 4; /* We're doing our own CRC */
2313 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2317 hlen = LL_RESERVED_SPACE(dev);
2318 tlen = dev->needed_tailroom;
2319 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2320 msg->msg_flags & MSG_DONTWAIT, &err);
2324 skb_set_network_header(skb, reserve);
2327 if (sock->type == SOCK_DGRAM &&
2328 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2331 /* Returns -EFAULT on error */
2332 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2335 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2339 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2340 /* Earlier code assumed this would be a VLAN pkt,
2341 * double-check this now that we have the actual
2344 struct ethhdr *ehdr;
2345 skb_reset_mac_header(skb);
2346 ehdr = eth_hdr(skb);
2347 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2353 skb->protocol = proto;
2355 skb->priority = sk->sk_priority;
2356 skb->mark = sk->sk_mark;
2358 if (po->has_vnet_hdr) {
2359 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2360 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2361 vnet_hdr.csum_offset)) {
2367 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2368 skb_shinfo(skb)->gso_type = gso_type;
2370 /* Header must be checked, and gso_segs computed. */
2371 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2372 skb_shinfo(skb)->gso_segs = 0;
2374 len += vnet_hdr_len;
2377 if (unlikely(extra_len == 4))
2384 err = dev_queue_xmit(skb);
2385 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2396 if (dev && need_rls_dev)
2402 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2403 struct msghdr *msg, size_t len)
2405 struct sock *sk = sock->sk;
2406 struct packet_sock *po = pkt_sk(sk);
2407 if (po->tx_ring.pg_vec)
2408 return tpacket_snd(po, msg);
2410 return packet_snd(sock, msg, len);
2414 * Close a PACKET socket. This is fairly simple. We immediately go
2415 * to 'closed' state and remove our protocol entry in the device list.
2418 static int packet_release(struct socket *sock)
2420 struct sock *sk = sock->sk;
2421 struct packet_sock *po;
2423 union tpacket_req_u req_u;
2431 spin_lock_bh(&net->packet.sklist_lock);
2432 sk_del_node_init_rcu(sk);
2433 sock_prot_inuse_add(net, sk->sk_prot, -1);
2434 spin_unlock_bh(&net->packet.sklist_lock);
2436 spin_lock(&po->bind_lock);
2437 unregister_prot_hook(sk, false);
2438 if (po->prot_hook.dev) {
2439 dev_put(po->prot_hook.dev);
2440 po->prot_hook.dev = NULL;
2442 spin_unlock(&po->bind_lock);
2444 packet_flush_mclist(sk);
2446 memset(&req_u, 0, sizeof(req_u));
2448 if (po->rx_ring.pg_vec)
2449 packet_set_ring(sk, &req_u, 1, 0);
2451 if (po->tx_ring.pg_vec)
2452 packet_set_ring(sk, &req_u, 1, 1);
2458 * Now the socket is dead. No more input will appear.
2465 skb_queue_purge(&sk->sk_receive_queue);
2466 sk_refcnt_debug_release(sk);
2473 * Attach a packet hook.
2476 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2478 struct packet_sock *po = pkt_sk(sk);
2489 spin_lock(&po->bind_lock);
2490 unregister_prot_hook(sk, true);
2492 po->prot_hook.type = protocol;
2493 if (po->prot_hook.dev)
2494 dev_put(po->prot_hook.dev);
2495 po->prot_hook.dev = dev;
2497 po->ifindex = dev ? dev->ifindex : 0;
2502 if (!dev || (dev->flags & IFF_UP)) {
2503 register_prot_hook(sk);
2505 sk->sk_err = ENETDOWN;
2506 if (!sock_flag(sk, SOCK_DEAD))
2507 sk->sk_error_report(sk);
2511 spin_unlock(&po->bind_lock);
2517 * Bind a packet socket to a device
2520 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2523 struct sock *sk = sock->sk;
2525 struct net_device *dev;
2532 if (addr_len != sizeof(struct sockaddr))
2534 strlcpy(name, uaddr->sa_data, sizeof(name));
2536 dev = dev_get_by_name(sock_net(sk), name);
2538 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2542 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2544 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2545 struct sock *sk = sock->sk;
2546 struct net_device *dev = NULL;
2554 if (addr_len < sizeof(struct sockaddr_ll))
2556 if (sll->sll_family != AF_PACKET)
2559 if (sll->sll_ifindex) {
2561 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2565 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2571 static struct proto packet_proto = {
2573 .owner = THIS_MODULE,
2574 .obj_size = sizeof(struct packet_sock),
2578 * Create a packet of type SOCK_PACKET.
2581 static int packet_create(struct net *net, struct socket *sock, int protocol,
2585 struct packet_sock *po;
2586 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2589 if (!capable(CAP_NET_RAW))
2591 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2592 sock->type != SOCK_PACKET)
2593 return -ESOCKTNOSUPPORT;
2595 sock->state = SS_UNCONNECTED;
2598 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2602 sock->ops = &packet_ops;
2603 if (sock->type == SOCK_PACKET)
2604 sock->ops = &packet_ops_spkt;
2606 sock_init_data(sock, sk);
2609 sk->sk_family = PF_PACKET;
2612 sk->sk_destruct = packet_sock_destruct;
2613 sk_refcnt_debug_inc(sk);
2616 * Attach a protocol block
2619 spin_lock_init(&po->bind_lock);
2620 mutex_init(&po->pg_vec_lock);
2621 po->prot_hook.func = packet_rcv;
2623 if (sock->type == SOCK_PACKET)
2624 po->prot_hook.func = packet_rcv_spkt;
2626 po->prot_hook.af_packet_priv = sk;
2629 po->prot_hook.type = proto;
2630 register_prot_hook(sk);
2633 spin_lock_bh(&net->packet.sklist_lock);
2634 sk_add_node_rcu(sk, &net->packet.sklist);
2635 sock_prot_inuse_add(net, &packet_proto, 1);
2636 spin_unlock_bh(&net->packet.sklist_lock);
2643 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2645 struct sock_exterr_skb *serr;
2646 struct sk_buff *skb, *skb2;
2650 skb = skb_dequeue(&sk->sk_error_queue);
2656 msg->msg_flags |= MSG_TRUNC;
2659 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2663 sock_recv_timestamp(msg, sk, skb);
2665 serr = SKB_EXT_ERR(skb);
2666 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2667 sizeof(serr->ee), &serr->ee);
2669 msg->msg_flags |= MSG_ERRQUEUE;
2672 /* Reset and regenerate socket error */
2673 spin_lock_bh(&sk->sk_error_queue.lock);
2675 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2676 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2677 spin_unlock_bh(&sk->sk_error_queue.lock);
2678 sk->sk_error_report(sk);
2680 spin_unlock_bh(&sk->sk_error_queue.lock);
2689 * Pull a packet from our receive queue and hand it to the user.
2690 * If necessary we block.
2693 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2694 struct msghdr *msg, size_t len, int flags)
2696 struct sock *sk = sock->sk;
2697 struct sk_buff *skb;
2699 struct sockaddr_ll *sll;
2700 int vnet_hdr_len = 0;
2703 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2707 /* What error should we return now? EUNATTACH? */
2708 if (pkt_sk(sk)->ifindex < 0)
2712 if (flags & MSG_ERRQUEUE) {
2713 err = packet_recv_error(sk, msg, len);
2718 * Call the generic datagram receiver. This handles all sorts
2719 * of horrible races and re-entrancy so we can forget about it
2720 * in the protocol layers.
2722 * Now it will return ENETDOWN, if device have just gone down,
2723 * but then it will block.
2726 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2729 * An error occurred so return it. Because skb_recv_datagram()
2730 * handles the blocking we don't see and worry about blocking
2737 if (pkt_sk(sk)->has_vnet_hdr) {
2738 struct virtio_net_hdr vnet_hdr = { 0 };
2741 vnet_hdr_len = sizeof(vnet_hdr);
2742 if (len < vnet_hdr_len)
2745 len -= vnet_hdr_len;
2747 if (skb_is_gso(skb)) {
2748 struct skb_shared_info *sinfo = skb_shinfo(skb);
2750 /* This is a hint as to how much should be linear. */
2751 vnet_hdr.hdr_len = skb_headlen(skb);
2752 vnet_hdr.gso_size = sinfo->gso_size;
2753 if (sinfo->gso_type & SKB_GSO_TCPV4)
2754 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2755 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2756 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2757 else if (sinfo->gso_type & SKB_GSO_UDP)
2758 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2759 else if (sinfo->gso_type & SKB_GSO_FCOE)
2763 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2764 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2766 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2768 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2769 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2770 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2771 vnet_hdr.csum_offset = skb->csum_offset;
2772 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2773 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2774 } /* else everything is zero */
2776 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2783 * If the address length field is there to be filled in, we fill
2787 sll = &PACKET_SKB_CB(skb)->sa.ll;
2788 if (sock->type == SOCK_PACKET)
2789 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2791 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2794 * You lose any data beyond the buffer you gave. If it worries a
2795 * user program they can ask the device for its MTU anyway.
2801 msg->msg_flags |= MSG_TRUNC;
2804 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2808 sock_recv_ts_and_drops(msg, sk, skb);
2811 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2814 if (pkt_sk(sk)->auxdata) {
2815 struct tpacket_auxdata aux;
2817 aux.tp_status = TP_STATUS_USER;
2818 if (skb->ip_summed == CHECKSUM_PARTIAL)
2819 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2820 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2821 aux.tp_snaplen = skb->len;
2823 aux.tp_net = skb_network_offset(skb);
2824 if (vlan_tx_tag_present(skb)) {
2825 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2826 aux.tp_status |= TP_STATUS_VLAN_VALID;
2828 aux.tp_vlan_tci = 0;
2831 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2835 * Free or return the buffer as appropriate. Again this
2836 * hides all the races and re-entrancy issues from us.
2838 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2841 skb_free_datagram(sk, skb);
2846 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2847 int *uaddr_len, int peer)
2849 struct net_device *dev;
2850 struct sock *sk = sock->sk;
2855 uaddr->sa_family = AF_PACKET;
2857 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2859 strncpy(uaddr->sa_data, dev->name, 14);
2861 memset(uaddr->sa_data, 0, 14);
2863 *uaddr_len = sizeof(*uaddr);
2868 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2869 int *uaddr_len, int peer)
2871 struct net_device *dev;
2872 struct sock *sk = sock->sk;
2873 struct packet_sock *po = pkt_sk(sk);
2874 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2879 sll->sll_family = AF_PACKET;
2880 sll->sll_ifindex = po->ifindex;
2881 sll->sll_protocol = po->num;
2882 sll->sll_pkttype = 0;
2884 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2886 sll->sll_hatype = dev->type;
2887 sll->sll_halen = dev->addr_len;
2888 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2890 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2894 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2899 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2903 case PACKET_MR_MULTICAST:
2904 if (i->alen != dev->addr_len)
2907 return dev_mc_add(dev, i->addr);
2909 return dev_mc_del(dev, i->addr);
2911 case PACKET_MR_PROMISC:
2912 return dev_set_promiscuity(dev, what);
2914 case PACKET_MR_ALLMULTI:
2915 return dev_set_allmulti(dev, what);
2917 case PACKET_MR_UNICAST:
2918 if (i->alen != dev->addr_len)
2921 return dev_uc_add(dev, i->addr);
2923 return dev_uc_del(dev, i->addr);
2931 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2933 for ( ; i; i = i->next) {
2934 if (i->ifindex == dev->ifindex)
2935 packet_dev_mc(dev, i, what);
2939 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2941 struct packet_sock *po = pkt_sk(sk);
2942 struct packet_mclist *ml, *i;
2943 struct net_device *dev;
2949 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2954 if (mreq->mr_alen > dev->addr_len)
2958 i = kmalloc(sizeof(*i), GFP_KERNEL);
2963 for (ml = po->mclist; ml; ml = ml->next) {
2964 if (ml->ifindex == mreq->mr_ifindex &&
2965 ml->type == mreq->mr_type &&
2966 ml->alen == mreq->mr_alen &&
2967 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2969 /* Free the new element ... */
2975 i->type = mreq->mr_type;
2976 i->ifindex = mreq->mr_ifindex;
2977 i->alen = mreq->mr_alen;
2978 memcpy(i->addr, mreq->mr_address, i->alen);
2980 i->next = po->mclist;
2982 err = packet_dev_mc(dev, i, 1);
2984 po->mclist = i->next;
2993 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2995 struct packet_mclist *ml, **mlp;
2999 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3000 if (ml->ifindex == mreq->mr_ifindex &&
3001 ml->type == mreq->mr_type &&
3002 ml->alen == mreq->mr_alen &&
3003 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3004 if (--ml->count == 0) {
3005 struct net_device *dev;
3007 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3009 packet_dev_mc(dev, ml, -1);
3017 return -EADDRNOTAVAIL;
3020 static void packet_flush_mclist(struct sock *sk)
3022 struct packet_sock *po = pkt_sk(sk);
3023 struct packet_mclist *ml;
3029 while ((ml = po->mclist) != NULL) {
3030 struct net_device *dev;
3032 po->mclist = ml->next;
3033 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3035 packet_dev_mc(dev, ml, -1);
3042 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3044 struct sock *sk = sock->sk;
3045 struct packet_sock *po = pkt_sk(sk);
3048 if (level != SOL_PACKET)
3049 return -ENOPROTOOPT;
3052 case PACKET_ADD_MEMBERSHIP:
3053 case PACKET_DROP_MEMBERSHIP:
3055 struct packet_mreq_max mreq;
3057 memset(&mreq, 0, sizeof(mreq));
3058 if (len < sizeof(struct packet_mreq))
3060 if (len > sizeof(mreq))
3062 if (copy_from_user(&mreq, optval, len))
3064 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3066 if (optname == PACKET_ADD_MEMBERSHIP)
3067 ret = packet_mc_add(sk, &mreq);
3069 ret = packet_mc_drop(sk, &mreq);
3073 case PACKET_RX_RING:
3074 case PACKET_TX_RING:
3076 union tpacket_req_u req_u;
3079 switch (po->tp_version) {
3082 len = sizeof(req_u.req);
3086 len = sizeof(req_u.req3);
3091 if (pkt_sk(sk)->has_vnet_hdr)
3093 if (copy_from_user(&req_u.req, optval, len))
3095 return packet_set_ring(sk, &req_u, 0,
3096 optname == PACKET_TX_RING);
3098 case PACKET_COPY_THRESH:
3102 if (optlen != sizeof(val))
3104 if (copy_from_user(&val, optval, sizeof(val)))
3107 pkt_sk(sk)->copy_thresh = val;
3110 case PACKET_VERSION:
3114 if (optlen != sizeof(val))
3116 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3118 if (copy_from_user(&val, optval, sizeof(val)))
3124 po->tp_version = val;
3130 case PACKET_RESERVE:
3134 if (optlen != sizeof(val))
3136 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3138 if (copy_from_user(&val, optval, sizeof(val)))
3140 po->tp_reserve = val;
3147 if (optlen != sizeof(val))
3149 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3151 if (copy_from_user(&val, optval, sizeof(val)))
3153 po->tp_loss = !!val;
3156 case PACKET_AUXDATA:
3160 if (optlen < sizeof(val))
3162 if (copy_from_user(&val, optval, sizeof(val)))
3165 po->auxdata = !!val;
3168 case PACKET_ORIGDEV:
3172 if (optlen < sizeof(val))
3174 if (copy_from_user(&val, optval, sizeof(val)))
3177 po->origdev = !!val;
3180 case PACKET_VNET_HDR:
3184 if (sock->type != SOCK_RAW)
3186 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3188 if (optlen < sizeof(val))
3190 if (copy_from_user(&val, optval, sizeof(val)))
3193 po->has_vnet_hdr = !!val;
3196 case PACKET_TIMESTAMP:
3200 if (optlen != sizeof(val))
3202 if (copy_from_user(&val, optval, sizeof(val)))
3205 po->tp_tstamp = val;
3212 if (optlen != sizeof(val))
3214 if (copy_from_user(&val, optval, sizeof(val)))
3217 return fanout_add(sk, val & 0xffff, val >> 16);
3220 return -ENOPROTOOPT;
3224 static int packet_getsockopt(struct socket *sock, int level, int optname,
3225 char __user *optval, int __user *optlen)
3229 struct sock *sk = sock->sk;
3230 struct packet_sock *po = pkt_sk(sk);
3232 struct tpacket_stats st;
3233 union tpacket_stats_u st_u;
3235 if (level != SOL_PACKET)
3236 return -ENOPROTOOPT;
3238 if (get_user(len, optlen))
3245 case PACKET_STATISTICS:
3246 if (po->tp_version == TPACKET_V3) {
3247 len = sizeof(struct tpacket_stats_v3);
3249 if (len > sizeof(struct tpacket_stats))
3250 len = sizeof(struct tpacket_stats);
3252 spin_lock_bh(&sk->sk_receive_queue.lock);
3253 if (po->tp_version == TPACKET_V3) {
3254 memcpy(&st_u.stats3, &po->stats,
3255 sizeof(struct tpacket_stats));
3256 st_u.stats3.tp_freeze_q_cnt =
3257 po->stats_u.stats3.tp_freeze_q_cnt;
3258 st_u.stats3.tp_packets += po->stats.tp_drops;
3259 data = &st_u.stats3;
3262 st.tp_packets += st.tp_drops;
3265 memset(&po->stats, 0, sizeof(st));
3266 spin_unlock_bh(&sk->sk_receive_queue.lock);
3268 case PACKET_AUXDATA:
3269 if (len > sizeof(int))
3275 case PACKET_ORIGDEV:
3276 if (len > sizeof(int))
3282 case PACKET_VNET_HDR:
3283 if (len > sizeof(int))
3285 val = po->has_vnet_hdr;
3289 case PACKET_VERSION:
3290 if (len > sizeof(int))
3292 val = po->tp_version;
3296 if (len > sizeof(int))
3298 if (copy_from_user(&val, optval, len))
3302 val = sizeof(struct tpacket_hdr);
3305 val = sizeof(struct tpacket2_hdr);
3308 val = sizeof(struct tpacket3_hdr);
3315 case PACKET_RESERVE:
3316 if (len > sizeof(unsigned int))
3317 len = sizeof(unsigned int);
3318 val = po->tp_reserve;
3322 if (len > sizeof(unsigned int))
3323 len = sizeof(unsigned int);
3327 case PACKET_TIMESTAMP:
3328 if (len > sizeof(int))
3330 val = po->tp_tstamp;
3334 if (len > sizeof(int))
3337 ((u32)po->fanout->id |
3338 ((u32)po->fanout->type << 16)) :
3343 return -ENOPROTOOPT;
3346 if (put_user(len, optlen))
3348 if (copy_to_user(optval, data, len))
3354 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3357 struct hlist_node *node;
3358 struct net_device *dev = data;
3359 struct net *net = dev_net(dev);
3362 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3363 struct packet_sock *po = pkt_sk(sk);
3366 case NETDEV_UNREGISTER:
3368 packet_dev_mclist(dev, po->mclist, -1);
3372 if (dev->ifindex == po->ifindex) {
3373 spin_lock(&po->bind_lock);
3375 __unregister_prot_hook(sk, false);
3376 sk->sk_err = ENETDOWN;
3377 if (!sock_flag(sk, SOCK_DEAD))
3378 sk->sk_error_report(sk);
3380 if (msg == NETDEV_UNREGISTER) {
3382 if (po->prot_hook.dev)
3383 dev_put(po->prot_hook.dev);
3384 po->prot_hook.dev = NULL;
3386 spin_unlock(&po->bind_lock);
3390 if (dev->ifindex == po->ifindex) {
3391 spin_lock(&po->bind_lock);
3393 register_prot_hook(sk);
3394 spin_unlock(&po->bind_lock);
3404 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3407 struct sock *sk = sock->sk;
3412 int amount = sk_wmem_alloc_get(sk);
3414 return put_user(amount, (int __user *)arg);
3418 struct sk_buff *skb;
3421 spin_lock_bh(&sk->sk_receive_queue.lock);
3422 skb = skb_peek(&sk->sk_receive_queue);
3425 spin_unlock_bh(&sk->sk_receive_queue.lock);
3426 return put_user(amount, (int __user *)arg);
3429 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3431 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3441 case SIOCGIFBRDADDR:
3442 case SIOCSIFBRDADDR:
3443 case SIOCGIFNETMASK:
3444 case SIOCSIFNETMASK:
3445 case SIOCGIFDSTADDR:
3446 case SIOCSIFDSTADDR:
3448 return inet_dgram_ops.ioctl(sock, cmd, arg);
3452 return -ENOIOCTLCMD;
3457 static unsigned int packet_poll(struct file *file, struct socket *sock,
3460 struct sock *sk = sock->sk;
3461 struct packet_sock *po = pkt_sk(sk);
3462 unsigned int mask = datagram_poll(file, sock, wait);
3464 spin_lock_bh(&sk->sk_receive_queue.lock);
3465 if (po->rx_ring.pg_vec) {
3466 if (!packet_previous_rx_frame(po, &po->rx_ring,
3468 mask |= POLLIN | POLLRDNORM;
3470 spin_unlock_bh(&sk->sk_receive_queue.lock);
3471 spin_lock_bh(&sk->sk_write_queue.lock);
3472 if (po->tx_ring.pg_vec) {
3473 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3474 mask |= POLLOUT | POLLWRNORM;
3476 spin_unlock_bh(&sk->sk_write_queue.lock);
3481 /* Dirty? Well, I still did not learn better way to account
3485 static void packet_mm_open(struct vm_area_struct *vma)
3487 struct file *file = vma->vm_file;
3488 struct socket *sock = file->private_data;
3489 struct sock *sk = sock->sk;
3492 atomic_inc(&pkt_sk(sk)->mapped);
3495 static void packet_mm_close(struct vm_area_struct *vma)
3497 struct file *file = vma->vm_file;
3498 struct socket *sock = file->private_data;
3499 struct sock *sk = sock->sk;
3502 atomic_dec(&pkt_sk(sk)->mapped);
3505 static const struct vm_operations_struct packet_mmap_ops = {
3506 .open = packet_mm_open,
3507 .close = packet_mm_close,
3510 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3515 for (i = 0; i < len; i++) {
3516 if (likely(pg_vec[i].buffer)) {
3517 if (is_vmalloc_addr(pg_vec[i].buffer))
3518 vfree(pg_vec[i].buffer);
3520 free_pages((unsigned long)pg_vec[i].buffer,
3522 pg_vec[i].buffer = NULL;
3528 static char *alloc_one_pg_vec_page(unsigned long order)
3530 char *buffer = NULL;
3531 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3532 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3534 buffer = (char *) __get_free_pages(gfp_flags, order);
3540 * __get_free_pages failed, fall back to vmalloc
3542 buffer = vzalloc((1 << order) * PAGE_SIZE);
3548 * vmalloc failed, lets dig into swap here
3550 gfp_flags &= ~__GFP_NORETRY;
3551 buffer = (char *)__get_free_pages(gfp_flags, order);
3556 * complete and utter failure
3561 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3563 unsigned int block_nr = req->tp_block_nr;
3567 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3568 if (unlikely(!pg_vec))
3571 for (i = 0; i < block_nr; i++) {
3572 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3573 if (unlikely(!pg_vec[i].buffer))
3574 goto out_free_pgvec;
3581 free_pg_vec(pg_vec, order, block_nr);
3586 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3587 int closing, int tx_ring)
3589 struct pgv *pg_vec = NULL;
3590 struct packet_sock *po = pkt_sk(sk);
3591 int was_running, order = 0;
3592 struct packet_ring_buffer *rb;
3593 struct sk_buff_head *rb_queue;
3596 /* Added to avoid minimal code churn */
3597 struct tpacket_req *req = &req_u->req;
3599 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3600 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3601 WARN(1, "Tx-ring is not supported.\n");
3605 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3606 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3610 if (atomic_read(&po->mapped))
3612 if (atomic_read(&rb->pending))
3616 if (req->tp_block_nr) {
3617 /* Sanity tests and some calculations */
3619 if (unlikely(rb->pg_vec))
3622 switch (po->tp_version) {
3624 po->tp_hdrlen = TPACKET_HDRLEN;
3627 po->tp_hdrlen = TPACKET2_HDRLEN;
3630 po->tp_hdrlen = TPACKET3_HDRLEN;
3635 if (unlikely((int)req->tp_block_size <= 0))
3637 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3639 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3642 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3645 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3646 if (unlikely(rb->frames_per_block <= 0))
3648 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3653 order = get_order(req->tp_block_size);
3654 pg_vec = alloc_pg_vec(req, order);
3655 if (unlikely(!pg_vec))
3657 switch (po->tp_version) {
3659 /* Transmit path is not supported. We checked
3660 * it above but just being paranoid
3663 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3672 if (unlikely(req->tp_frame_nr))
3678 /* Detach socket from network */
3679 spin_lock(&po->bind_lock);
3680 was_running = po->running;
3684 __unregister_prot_hook(sk, false);
3686 spin_unlock(&po->bind_lock);
3691 mutex_lock(&po->pg_vec_lock);
3692 if (closing || atomic_read(&po->mapped) == 0) {
3694 spin_lock_bh(&rb_queue->lock);
3695 swap(rb->pg_vec, pg_vec);
3696 rb->frame_max = (req->tp_frame_nr - 1);
3698 rb->frame_size = req->tp_frame_size;
3699 spin_unlock_bh(&rb_queue->lock);
3701 swap(rb->pg_vec_order, order);
3702 swap(rb->pg_vec_len, req->tp_block_nr);
3704 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3705 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3706 tpacket_rcv : packet_rcv;
3707 skb_queue_purge(rb_queue);
3708 if (atomic_read(&po->mapped))
3709 pr_err("packet_mmap: vma is busy: %d\n",
3710 atomic_read(&po->mapped));
3712 mutex_unlock(&po->pg_vec_lock);
3714 spin_lock(&po->bind_lock);
3717 register_prot_hook(sk);
3719 spin_unlock(&po->bind_lock);
3720 if (closing && (po->tp_version > TPACKET_V2)) {
3721 /* Because we don't support block-based V3 on tx-ring */
3723 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3728 free_pg_vec(pg_vec, order, req->tp_block_nr);
3733 static int packet_mmap(struct file *file, struct socket *sock,
3734 struct vm_area_struct *vma)
3736 struct sock *sk = sock->sk;
3737 struct packet_sock *po = pkt_sk(sk);
3738 unsigned long size, expected_size;
3739 struct packet_ring_buffer *rb;
3740 unsigned long start;
3747 mutex_lock(&po->pg_vec_lock);
3750 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3752 expected_size += rb->pg_vec_len
3758 if (expected_size == 0)
3761 size = vma->vm_end - vma->vm_start;
3762 if (size != expected_size)
3765 start = vma->vm_start;
3766 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3767 if (rb->pg_vec == NULL)
3770 for (i = 0; i < rb->pg_vec_len; i++) {
3772 void *kaddr = rb->pg_vec[i].buffer;
3775 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3776 page = pgv_to_page(kaddr);
3777 err = vm_insert_page(vma, start, page);
3786 atomic_inc(&po->mapped);
3787 vma->vm_ops = &packet_mmap_ops;
3791 mutex_unlock(&po->pg_vec_lock);
3795 static const struct proto_ops packet_ops_spkt = {
3796 .family = PF_PACKET,
3797 .owner = THIS_MODULE,
3798 .release = packet_release,
3799 .bind = packet_bind_spkt,
3800 .connect = sock_no_connect,
3801 .socketpair = sock_no_socketpair,
3802 .accept = sock_no_accept,
3803 .getname = packet_getname_spkt,
3804 .poll = datagram_poll,
3805 .ioctl = packet_ioctl,
3806 .listen = sock_no_listen,
3807 .shutdown = sock_no_shutdown,
3808 .setsockopt = sock_no_setsockopt,
3809 .getsockopt = sock_no_getsockopt,
3810 .sendmsg = packet_sendmsg_spkt,
3811 .recvmsg = packet_recvmsg,
3812 .mmap = sock_no_mmap,
3813 .sendpage = sock_no_sendpage,
3816 static const struct proto_ops packet_ops = {
3817 .family = PF_PACKET,
3818 .owner = THIS_MODULE,
3819 .release = packet_release,
3820 .bind = packet_bind,
3821 .connect = sock_no_connect,
3822 .socketpair = sock_no_socketpair,
3823 .accept = sock_no_accept,
3824 .getname = packet_getname,
3825 .poll = packet_poll,
3826 .ioctl = packet_ioctl,
3827 .listen = sock_no_listen,
3828 .shutdown = sock_no_shutdown,
3829 .setsockopt = packet_setsockopt,
3830 .getsockopt = packet_getsockopt,
3831 .sendmsg = packet_sendmsg,
3832 .recvmsg = packet_recvmsg,
3833 .mmap = packet_mmap,
3834 .sendpage = sock_no_sendpage,
3837 static const struct net_proto_family packet_family_ops = {
3838 .family = PF_PACKET,
3839 .create = packet_create,
3840 .owner = THIS_MODULE,
3843 static struct notifier_block packet_netdev_notifier = {
3844 .notifier_call = packet_notifier,
3847 #ifdef CONFIG_PROC_FS
3849 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3852 struct net *net = seq_file_net(seq);
3855 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3858 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3860 struct net *net = seq_file_net(seq);
3861 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3864 static void packet_seq_stop(struct seq_file *seq, void *v)
3870 static int packet_seq_show(struct seq_file *seq, void *v)
3872 if (v == SEQ_START_TOKEN)
3873 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3875 struct sock *s = sk_entry(v);
3876 const struct packet_sock *po = pkt_sk(s);
3879 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3881 atomic_read(&s->sk_refcnt),
3886 atomic_read(&s->sk_rmem_alloc),
3894 static const struct seq_operations packet_seq_ops = {
3895 .start = packet_seq_start,
3896 .next = packet_seq_next,
3897 .stop = packet_seq_stop,
3898 .show = packet_seq_show,
3901 static int packet_seq_open(struct inode *inode, struct file *file)
3903 return seq_open_net(inode, file, &packet_seq_ops,
3904 sizeof(struct seq_net_private));
3907 static const struct file_operations packet_seq_fops = {
3908 .owner = THIS_MODULE,
3909 .open = packet_seq_open,
3911 .llseek = seq_lseek,
3912 .release = seq_release_net,
3917 static int __net_init packet_net_init(struct net *net)
3919 spin_lock_init(&net->packet.sklist_lock);
3920 INIT_HLIST_HEAD(&net->packet.sklist);
3922 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3928 static void __net_exit packet_net_exit(struct net *net)
3930 proc_net_remove(net, "packet");
3933 static struct pernet_operations packet_net_ops = {
3934 .init = packet_net_init,
3935 .exit = packet_net_exit,
3939 static void __exit packet_exit(void)
3941 unregister_netdevice_notifier(&packet_netdev_notifier);
3942 unregister_pernet_subsys(&packet_net_ops);
3943 sock_unregister(PF_PACKET);
3944 proto_unregister(&packet_proto);
3947 static int __init packet_init(void)
3949 int rc = proto_register(&packet_proto, 0);
3954 sock_register(&packet_family_ops);
3955 register_pernet_subsys(&packet_net_ops);
3956 register_netdevice_notifier(&packet_netdev_notifier);
3961 module_init(packet_init);
3962 module_exit(packet_exit);
3963 MODULE_LICENSE("GPL");
3964 MODULE_ALIAS_NETPROTO(PF_PACKET);