2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
113 Incoming, dev->hard_header!=NULL
114 mac_header -> ll header
117 Outgoing, dev->hard_header!=NULL
118 mac_header -> ll header
121 Incoming, dev->hard_header==NULL
122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
124 assymetry between rx and tx paths.
127 Outgoing, dev->hard_header==NULL
128 mac_header -> data. ll header is still not built!
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
138 dev->hard_header != NULL
139 mac_header -> ll header
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
150 /* Private packet socket structures. */
152 /* identical to struct packet_mreq except it has
153 * a longer address field.
155 struct packet_mreq_max {
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 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 #define PGV_FROM_VMALLOC 1
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
199 struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206 struct tpacket_block_desc *);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209 static void prb_init_blk_timer(struct packet_sock *,
210 struct tpacket_kbdq_core *,
211 void (*func) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 struct tpacket3_hdr *);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 struct tpacket3_hdr *);
217 static void packet_flush_mclist(struct sock *sk);
219 struct packet_skb_cb {
221 struct sockaddr_pkt pkt;
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
227 unsigned int origlen;
228 struct sockaddr_ll ll;
233 #define vio_le() virtio_legacy_is_little_endian()
235 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
237 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
249 static int packet_direct_xmit(struct sk_buff *skb)
251 struct net_device *dev = skb->dev;
252 struct sk_buff *orig_skb = skb;
253 struct netdev_queue *txq;
254 int ret = NETDEV_TX_BUSY;
256 if (unlikely(!netif_running(dev) ||
257 !netif_carrier_ok(dev)))
260 skb = validate_xmit_skb_list(skb, dev);
264 txq = skb_get_tx_queue(dev, skb);
268 HARD_TX_LOCK(dev, txq, smp_processor_id());
269 if (!netif_xmit_frozen_or_drv_stopped(txq))
270 ret = netdev_start_xmit(skb, dev, txq, false);
271 HARD_TX_UNLOCK(dev, txq);
275 if (!dev_xmit_complete(ret))
280 atomic_long_inc(&dev->tx_dropped);
282 return NET_XMIT_DROP;
285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 struct net_device *dev;
290 dev = rcu_dereference(po->cached_dev);
298 static void packet_cached_dev_assign(struct packet_sock *po,
299 struct net_device *dev)
301 rcu_assign_pointer(po->cached_dev, dev);
304 static void packet_cached_dev_reset(struct packet_sock *po)
306 RCU_INIT_POINTER(po->cached_dev, NULL);
309 static bool packet_use_direct_xmit(const struct packet_sock *po)
311 return po->xmit == packet_direct_xmit;
314 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
316 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
319 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321 const struct net_device_ops *ops = dev->netdev_ops;
324 if (ops->ndo_select_queue) {
325 queue_index = ops->ndo_select_queue(dev, skb, NULL,
326 __packet_pick_tx_queue);
327 queue_index = netdev_cap_txqueue(dev, queue_index);
329 queue_index = __packet_pick_tx_queue(dev, skb);
332 skb_set_queue_mapping(skb, queue_index);
335 /* __register_prot_hook must be invoked through register_prot_hook
336 * or from a context in which asynchronous accesses to the packet
337 * socket is not possible (packet_create()).
339 static void __register_prot_hook(struct sock *sk)
341 struct packet_sock *po = pkt_sk(sk);
345 __fanout_link(sk, po);
347 dev_add_pack(&po->prot_hook);
354 static void register_prot_hook(struct sock *sk)
356 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
357 __register_prot_hook(sk);
360 /* If the sync parameter is true, we will temporarily drop
361 * the po->bind_lock and do a synchronize_net to make sure no
362 * asynchronous packet processing paths still refer to the elements
363 * of po->prot_hook. If the sync parameter is false, it is the
364 * callers responsibility to take care of this.
366 static void __unregister_prot_hook(struct sock *sk, bool sync)
368 struct packet_sock *po = pkt_sk(sk);
370 lockdep_assert_held_once(&po->bind_lock);
375 __fanout_unlink(sk, po);
377 __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 struct page * __pure 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)
405 union tpacket_uhdr h;
408 switch (po->tp_version) {
410 h.h1->tp_status = status;
411 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
414 h.h2->tp_status = status;
415 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
419 WARN(1, "TPACKET version not supported.\n");
426 static int __packet_get_status(struct packet_sock *po, void *frame)
428 union tpacket_uhdr h;
433 switch (po->tp_version) {
435 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
436 return h.h1->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
439 return h.h2->tp_status;
442 WARN(1, "TPACKET version not supported.\n");
448 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
451 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
454 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
455 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
456 return TP_STATUS_TS_RAW_HARDWARE;
458 if (ktime_to_timespec_cond(skb->tstamp, ts))
459 return TP_STATUS_TS_SOFTWARE;
464 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
467 union tpacket_uhdr h;
471 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
475 switch (po->tp_version) {
477 h.h1->tp_sec = ts.tv_sec;
478 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
481 h.h2->tp_sec = ts.tv_sec;
482 h.h2->tp_nsec = ts.tv_nsec;
486 WARN(1, "TPACKET version not supported.\n");
490 /* one flush is safe, as both fields always lie on the same cacheline */
491 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
497 static void *packet_lookup_frame(struct packet_sock *po,
498 struct packet_ring_buffer *rb,
499 unsigned int position,
502 unsigned int pg_vec_pos, frame_offset;
503 union tpacket_uhdr h;
505 pg_vec_pos = position / rb->frames_per_block;
506 frame_offset = position % rb->frames_per_block;
508 h.raw = rb->pg_vec[pg_vec_pos].buffer +
509 (frame_offset * rb->frame_size);
511 if (status != __packet_get_status(po, h.raw))
517 static void *packet_current_frame(struct packet_sock *po,
518 struct packet_ring_buffer *rb,
521 return packet_lookup_frame(po, rb, rb->head, status);
524 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
526 del_timer_sync(&pkc->retire_blk_timer);
529 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
530 struct sk_buff_head *rb_queue)
532 struct tpacket_kbdq_core *pkc;
534 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
536 spin_lock_bh(&rb_queue->lock);
537 pkc->delete_blk_timer = 1;
538 spin_unlock_bh(&rb_queue->lock);
540 prb_del_retire_blk_timer(pkc);
543 static void prb_init_blk_timer(struct packet_sock *po,
544 struct tpacket_kbdq_core *pkc,
545 void (*func) (unsigned long))
547 init_timer(&pkc->retire_blk_timer);
548 pkc->retire_blk_timer.data = (long)po;
549 pkc->retire_blk_timer.function = func;
550 pkc->retire_blk_timer.expires = jiffies;
553 static void prb_setup_retire_blk_timer(struct packet_sock *po)
555 struct tpacket_kbdq_core *pkc;
557 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
558 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
561 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
562 int blk_size_in_bytes)
564 struct net_device *dev;
565 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
566 struct ethtool_cmd ecmd;
571 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
572 if (unlikely(!dev)) {
574 return DEFAULT_PRB_RETIRE_TOV;
576 err = __ethtool_get_settings(dev, &ecmd);
577 speed = ethtool_cmd_speed(&ecmd);
581 * If the link speed is so slow you don't really
582 * need to worry about perf anyways
584 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
585 return DEFAULT_PRB_RETIRE_TOV;
592 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
604 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
605 union tpacket_req_u *req_u)
607 p1->feature_req_word = req_u->req3.tp_feature_req_word;
610 static void init_prb_bdqc(struct packet_sock *po,
611 struct packet_ring_buffer *rb,
613 union tpacket_req_u *req_u)
615 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
616 struct tpacket_block_desc *pbd;
618 memset(p1, 0x0, sizeof(*p1));
620 p1->knxt_seq_num = 1;
622 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
623 p1->pkblk_start = pg_vec[0].buffer;
624 p1->kblk_size = req_u->req3.tp_block_size;
625 p1->knum_blocks = req_u->req3.tp_block_nr;
626 p1->hdrlen = po->tp_hdrlen;
627 p1->version = po->tp_version;
628 p1->last_kactive_blk_num = 0;
629 po->stats.stats3.tp_freeze_q_cnt = 0;
630 if (req_u->req3.tp_retire_blk_tov)
631 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
633 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
634 req_u->req3.tp_block_size);
635 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
636 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
638 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
639 prb_init_ft_ops(p1, req_u);
640 prb_setup_retire_blk_timer(po);
641 prb_open_block(p1, pbd);
644 /* Do NOT update the last_blk_num first.
645 * Assumes sk_buff_head lock is held.
647 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
649 mod_timer(&pkc->retire_blk_timer,
650 jiffies + pkc->tov_in_jiffies);
651 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
656 * 1) We refresh the timer only when we open a block.
657 * By doing this we don't waste cycles refreshing the timer
658 * on packet-by-packet basis.
660 * With a 1MB block-size, on a 1Gbps line, it will take
661 * i) ~8 ms to fill a block + ii) memcpy etc.
662 * In this cut we are not accounting for the memcpy time.
664 * So, if the user sets the 'tmo' to 10ms then the timer
665 * will never fire while the block is still getting filled
666 * (which is what we want). However, the user could choose
667 * to close a block early and that's fine.
669 * But when the timer does fire, we check whether or not to refresh it.
670 * Since the tmo granularity is in msecs, it is not too expensive
671 * to refresh the timer, lets say every '8' msecs.
672 * Either the user can set the 'tmo' or we can derive it based on
673 * a) line-speed and b) block-size.
674 * prb_calc_retire_blk_tmo() calculates the tmo.
677 static void prb_retire_rx_blk_timer_expired(unsigned long data)
679 struct packet_sock *po = (struct packet_sock *)data;
680 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
682 struct tpacket_block_desc *pbd;
684 spin_lock(&po->sk.sk_receive_queue.lock);
686 frozen = prb_queue_frozen(pkc);
687 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
689 if (unlikely(pkc->delete_blk_timer))
692 /* We only need to plug the race when the block is partially filled.
694 * lock(); increment BLOCK_NUM_PKTS; unlock()
695 * copy_bits() is in progress ...
696 * timer fires on other cpu:
697 * we can't retire the current block because copy_bits
701 if (BLOCK_NUM_PKTS(pbd)) {
702 while (atomic_read(&pkc->blk_fill_in_prog)) {
703 /* Waiting for skb_copy_bits to finish... */
708 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
710 if (!BLOCK_NUM_PKTS(pbd)) {
711 /* An empty block. Just refresh the timer. */
714 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
715 if (!prb_dispatch_next_block(pkc, po))
720 /* Case 1. Queue was frozen because user-space was
723 if (prb_curr_blk_in_use(pkc, pbd)) {
725 * Ok, user-space is still behind.
726 * So just refresh the timer.
730 /* Case 2. queue was frozen,user-space caught up,
731 * now the link went idle && the timer fired.
732 * We don't have a block to close.So we open this
733 * block and restart the timer.
734 * opening a block thaws the queue,restarts timer
735 * Thawing/timer-refresh is a side effect.
737 prb_open_block(pkc, pbd);
744 _prb_refresh_rx_retire_blk_timer(pkc);
747 spin_unlock(&po->sk.sk_receive_queue.lock);
750 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
751 struct tpacket_block_desc *pbd1, __u32 status)
753 /* Flush everything minus the block header */
755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
760 /* Skip the block header(we know header WILL fit in 4K) */
763 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
764 for (; start < end; start += PAGE_SIZE)
765 flush_dcache_page(pgv_to_page(start));
770 /* Now update the block status. */
772 BLOCK_STATUS(pbd1) = status;
774 /* Flush the block header */
776 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
778 flush_dcache_page(pgv_to_page(start));
788 * 2) Increment active_blk_num
790 * Note:We DONT refresh the timer on purpose.
791 * Because almost always the next block will be opened.
793 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
794 struct tpacket_block_desc *pbd1,
795 struct packet_sock *po, unsigned int stat)
797 __u32 status = TP_STATUS_USER | stat;
799 struct tpacket3_hdr *last_pkt;
800 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
801 struct sock *sk = &po->sk;
803 if (po->stats.stats3.tp_drops)
804 status |= TP_STATUS_LOSING;
806 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
807 last_pkt->tp_next_offset = 0;
809 /* Get the ts of the last pkt */
810 if (BLOCK_NUM_PKTS(pbd1)) {
811 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
812 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
814 /* Ok, we tmo'd - so get the current time.
816 * It shouldn't really happen as we don't close empty
817 * blocks. See prb_retire_rx_blk_timer_expired().
821 h1->ts_last_pkt.ts_sec = ts.tv_sec;
822 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
827 /* Flush the block */
828 prb_flush_block(pkc1, pbd1, status);
830 sk->sk_data_ready(sk);
832 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
835 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
837 pkc->reset_pending_on_curr_blk = 0;
841 * Side effect of opening a block:
843 * 1) prb_queue is thawed.
844 * 2) retire_blk_timer is refreshed.
847 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
848 struct tpacket_block_desc *pbd1)
851 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
855 /* We could have just memset this but we will lose the
856 * flexibility of making the priv area sticky
859 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
860 BLOCK_NUM_PKTS(pbd1) = 0;
861 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 h1->ts_first_pkt.ts_sec = ts.tv_sec;
866 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
868 pkc1->pkblk_start = (char *)pbd1;
869 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
871 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
874 pbd1->version = pkc1->version;
875 pkc1->prev = pkc1->nxt_offset;
876 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
878 prb_thaw_queue(pkc1);
879 _prb_refresh_rx_retire_blk_timer(pkc1);
885 * Queue freeze logic:
886 * 1) Assume tp_block_nr = 8 blocks.
887 * 2) At time 't0', user opens Rx ring.
888 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
889 * 4) user-space is either sleeping or processing block '0'.
890 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
891 * it will close block-7,loop around and try to fill block '0'.
893 * __packet_lookup_frame_in_block
894 * prb_retire_current_block()
895 * prb_dispatch_next_block()
896 * |->(BLOCK_STATUS == USER) evaluates to true
897 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
898 * 6) Now there are two cases:
899 * 6.1) Link goes idle right after the queue is frozen.
900 * But remember, the last open_block() refreshed the timer.
901 * When this timer expires,it will refresh itself so that we can
902 * re-open block-0 in near future.
903 * 6.2) Link is busy and keeps on receiving packets. This is a simple
904 * case and __packet_lookup_frame_in_block will check if block-0
905 * is free and can now be re-used.
907 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
908 struct packet_sock *po)
910 pkc->reset_pending_on_curr_blk = 1;
911 po->stats.stats3.tp_freeze_q_cnt++;
914 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
917 * If the next block is free then we will dispatch it
918 * and return a good offset.
919 * Else, we will freeze the queue.
920 * So, caller must check the return value.
922 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
923 struct packet_sock *po)
925 struct tpacket_block_desc *pbd;
929 /* 1. Get current block num */
930 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
932 /* 2. If this block is currently in_use then freeze the queue */
933 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
934 prb_freeze_queue(pkc, po);
940 * open this block and return the offset where the first packet
941 * needs to get stored.
943 prb_open_block(pkc, pbd);
944 return (void *)pkc->nxt_offset;
947 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
948 struct packet_sock *po, unsigned int status)
950 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
952 /* retire/close the current block */
953 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
955 * Plug the case where copy_bits() is in progress on
956 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
957 * have space to copy the pkt in the current block and
958 * called prb_retire_current_block()
960 * We don't need to worry about the TMO case because
961 * the timer-handler already handled this case.
963 if (!(status & TP_STATUS_BLK_TMO)) {
964 while (atomic_read(&pkc->blk_fill_in_prog)) {
965 /* Waiting for skb_copy_bits to finish... */
969 prb_close_block(pkc, pbd, po, status);
974 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
975 struct tpacket_block_desc *pbd)
977 return TP_STATUS_USER & BLOCK_STATUS(pbd);
980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
982 return pkc->reset_pending_on_curr_blk;
985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
987 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
988 atomic_dec(&pkc->blk_fill_in_prog);
991 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
992 struct tpacket3_hdr *ppd)
994 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
997 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
998 struct tpacket3_hdr *ppd)
1000 ppd->hv1.tp_rxhash = 0;
1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1004 struct tpacket3_hdr *ppd)
1006 if (skb_vlan_tag_present(pkc->skb)) {
1007 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1008 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1009 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1011 ppd->hv1.tp_vlan_tci = 0;
1012 ppd->hv1.tp_vlan_tpid = 0;
1013 ppd->tp_status = TP_STATUS_AVAILABLE;
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1018 struct tpacket3_hdr *ppd)
1020 ppd->hv1.tp_padding = 0;
1021 prb_fill_vlan_info(pkc, ppd);
1023 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1024 prb_fill_rxhash(pkc, ppd);
1026 prb_clear_rxhash(pkc, ppd);
1029 static void prb_fill_curr_block(char *curr,
1030 struct tpacket_kbdq_core *pkc,
1031 struct tpacket_block_desc *pbd,
1034 struct tpacket3_hdr *ppd;
1036 ppd = (struct tpacket3_hdr *)curr;
1037 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1039 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 BLOCK_NUM_PKTS(pbd) += 1;
1042 atomic_inc(&pkc->blk_fill_in_prog);
1043 prb_run_all_ft_ops(pkc, ppd);
1046 /* Assumes caller has the sk->rx_queue.lock */
1047 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1048 struct sk_buff *skb,
1053 struct tpacket_kbdq_core *pkc;
1054 struct tpacket_block_desc *pbd;
1057 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1058 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1060 /* Queue is frozen when user space is lagging behind */
1061 if (prb_queue_frozen(pkc)) {
1063 * Check if that last block which caused the queue to freeze,
1064 * is still in_use by user-space.
1066 if (prb_curr_blk_in_use(pkc, pbd)) {
1067 /* Can't record this packet */
1071 * Ok, the block was released by user-space.
1072 * Now let's open that block.
1073 * opening a block also thaws the queue.
1074 * Thawing is a side effect.
1076 prb_open_block(pkc, pbd);
1081 curr = pkc->nxt_offset;
1083 end = (char *)pbd + pkc->kblk_size;
1085 /* first try the current block */
1086 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1087 prb_fill_curr_block(curr, pkc, pbd, len);
1088 return (void *)curr;
1091 /* Ok, close the current block */
1092 prb_retire_current_block(pkc, po, 0);
1094 /* Now, try to dispatch the next block */
1095 curr = (char *)prb_dispatch_next_block(pkc, po);
1097 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1098 prb_fill_curr_block(curr, pkc, pbd, len);
1099 return (void *)curr;
1103 * No free blocks are available.user_space hasn't caught up yet.
1104 * Queue was just frozen and now this packet will get dropped.
1109 static void *packet_current_rx_frame(struct packet_sock *po,
1110 struct sk_buff *skb,
1111 int status, unsigned int len)
1114 switch (po->tp_version) {
1117 curr = packet_lookup_frame(po, &po->rx_ring,
1118 po->rx_ring.head, status);
1121 return __packet_lookup_frame_in_block(po, skb, status, len);
1123 WARN(1, "TPACKET version not supported\n");
1129 static void *prb_lookup_block(struct packet_sock *po,
1130 struct packet_ring_buffer *rb,
1134 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1135 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1137 if (status != BLOCK_STATUS(pbd))
1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1145 if (rb->prb_bdqc.kactive_blk_num)
1146 prev = rb->prb_bdqc.kactive_blk_num-1;
1148 prev = rb->prb_bdqc.knum_blocks-1;
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock *po,
1154 struct packet_ring_buffer *rb,
1157 unsigned int previous = prb_previous_blk_num(rb);
1158 return prb_lookup_block(po, rb, previous, status);
1161 static void *packet_previous_rx_frame(struct packet_sock *po,
1162 struct packet_ring_buffer *rb,
1165 if (po->tp_version <= TPACKET_V2)
1166 return packet_previous_frame(po, rb, status);
1168 return __prb_previous_block(po, rb, status);
1171 static void packet_increment_rx_head(struct packet_sock *po,
1172 struct packet_ring_buffer *rb)
1174 switch (po->tp_version) {
1177 return packet_increment_head(rb);
1180 WARN(1, "TPACKET version not supported.\n");
1186 static void *packet_previous_frame(struct packet_sock *po,
1187 struct packet_ring_buffer *rb,
1190 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1191 return packet_lookup_frame(po, rb, previous, status);
1194 static void packet_increment_head(struct packet_ring_buffer *buff)
1196 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1199 static void packet_inc_pending(struct packet_ring_buffer *rb)
1201 this_cpu_inc(*rb->pending_refcnt);
1204 static void packet_dec_pending(struct packet_ring_buffer *rb)
1206 this_cpu_dec(*rb->pending_refcnt);
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1211 unsigned int refcnt = 0;
1214 /* We don't use pending refcount in rx_ring. */
1215 if (rb->pending_refcnt == NULL)
1218 for_each_possible_cpu(cpu)
1219 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1224 static int packet_alloc_pending(struct packet_sock *po)
1226 po->rx_ring.pending_refcnt = NULL;
1228 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1229 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1235 static void packet_free_pending(struct packet_sock *po)
1237 free_percpu(po->tx_ring.pending_refcnt);
1240 #define ROOM_POW_OFF 2
1241 #define ROOM_NONE 0x0
1242 #define ROOM_LOW 0x1
1243 #define ROOM_NORMAL 0x2
1245 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1249 len = po->rx_ring.frame_max + 1;
1250 idx = po->rx_ring.head;
1252 idx += len >> pow_off;
1255 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1258 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1262 len = po->rx_ring.prb_bdqc.knum_blocks;
1263 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1265 idx += len >> pow_off;
1268 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1271 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1273 struct sock *sk = &po->sk;
1274 int ret = ROOM_NONE;
1276 if (po->prot_hook.func != tpacket_rcv) {
1277 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1278 - (skb ? skb->truesize : 0);
1279 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1287 if (po->tp_version == TPACKET_V3) {
1288 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1290 else if (__tpacket_v3_has_room(po, 0))
1293 if (__tpacket_has_room(po, ROOM_POW_OFF))
1295 else if (__tpacket_has_room(po, 0))
1302 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1307 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1308 ret = __packet_rcv_has_room(po, skb);
1309 has_room = ret == ROOM_NORMAL;
1310 if (po->pressure == has_room)
1311 po->pressure = !has_room;
1312 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1317 static void packet_sock_destruct(struct sock *sk)
1319 skb_queue_purge(&sk->sk_error_queue);
1321 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1322 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1324 if (!sock_flag(sk, SOCK_DEAD)) {
1325 pr_err("Attempt to release alive packet socket: %p\n", sk);
1329 sk_refcnt_debug_dec(sk);
1332 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1337 rxhash = skb_get_hash(skb);
1338 for (i = 0; i < ROLLOVER_HLEN; i++)
1339 if (po->rollover->history[i] == rxhash)
1342 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1343 return count > (ROLLOVER_HLEN >> 1);
1346 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1353 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1354 struct sk_buff *skb,
1357 unsigned int val = atomic_inc_return(&f->rr_cur);
1362 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return smp_processor_id() % num;
1369 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1370 struct sk_buff *skb,
1373 return prandom_u32_max(num);
1376 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1377 struct sk_buff *skb,
1378 unsigned int idx, bool try_self,
1381 struct packet_sock *po, *po_next, *po_skip = NULL;
1382 unsigned int i, j, room = ROOM_NONE;
1384 po = pkt_sk(f->arr[idx]);
1387 room = packet_rcv_has_room(po, skb);
1388 if (room == ROOM_NORMAL ||
1389 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1394 i = j = min_t(int, po->rollover->sock, num - 1);
1396 po_next = pkt_sk(f->arr[i]);
1397 if (po_next != po_skip && !po_next->pressure &&
1398 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1400 po->rollover->sock = i;
1401 atomic_long_inc(&po->rollover->num);
1402 if (room == ROOM_LOW)
1403 atomic_long_inc(&po->rollover->num_huge);
1411 atomic_long_inc(&po->rollover->num_failed);
1415 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1416 struct sk_buff *skb,
1419 return skb_get_queue_mapping(skb) % num;
1422 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1423 struct sk_buff *skb,
1426 struct bpf_prog *prog;
1427 unsigned int ret = 0;
1430 prog = rcu_dereference(f->bpf_prog);
1432 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1438 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1440 return f->flags & (flag >> 8);
1443 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1444 struct packet_type *pt, struct net_device *orig_dev)
1446 struct packet_fanout *f = pt->af_packet_priv;
1447 unsigned int num = READ_ONCE(f->num_members);
1448 struct net *net = read_pnet(&f->net);
1449 struct packet_sock *po;
1452 if (!net_eq(dev_net(dev), net) || !num) {
1457 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1458 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1463 case PACKET_FANOUT_HASH:
1465 idx = fanout_demux_hash(f, skb, num);
1467 case PACKET_FANOUT_LB:
1468 idx = fanout_demux_lb(f, skb, num);
1470 case PACKET_FANOUT_CPU:
1471 idx = fanout_demux_cpu(f, skb, num);
1473 case PACKET_FANOUT_RND:
1474 idx = fanout_demux_rnd(f, skb, num);
1476 case PACKET_FANOUT_QM:
1477 idx = fanout_demux_qm(f, skb, num);
1479 case PACKET_FANOUT_ROLLOVER:
1480 idx = fanout_demux_rollover(f, skb, 0, false, num);
1482 case PACKET_FANOUT_CBPF:
1483 case PACKET_FANOUT_EBPF:
1484 idx = fanout_demux_bpf(f, skb, num);
1488 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1489 idx = fanout_demux_rollover(f, skb, idx, true, num);
1491 po = pkt_sk(f->arr[idx]);
1492 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1495 DEFINE_MUTEX(fanout_mutex);
1496 EXPORT_SYMBOL_GPL(fanout_mutex);
1497 static LIST_HEAD(fanout_list);
1499 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1501 struct packet_fanout *f = po->fanout;
1503 spin_lock(&f->lock);
1504 f->arr[f->num_members] = sk;
1507 if (f->num_members == 1)
1508 dev_add_pack(&f->prot_hook);
1509 spin_unlock(&f->lock);
1512 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1514 struct packet_fanout *f = po->fanout;
1517 spin_lock(&f->lock);
1518 for (i = 0; i < f->num_members; i++) {
1519 if (f->arr[i] == sk)
1522 BUG_ON(i >= f->num_members);
1523 f->arr[i] = f->arr[f->num_members - 1];
1525 if (f->num_members == 0)
1526 __dev_remove_pack(&f->prot_hook);
1527 spin_unlock(&f->lock);
1530 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1532 if (sk->sk_family != PF_PACKET)
1535 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1538 static void fanout_init_data(struct packet_fanout *f)
1541 case PACKET_FANOUT_LB:
1542 atomic_set(&f->rr_cur, 0);
1544 case PACKET_FANOUT_CBPF:
1545 case PACKET_FANOUT_EBPF:
1546 RCU_INIT_POINTER(f->bpf_prog, NULL);
1551 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1553 struct bpf_prog *old;
1555 spin_lock(&f->lock);
1556 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1557 rcu_assign_pointer(f->bpf_prog, new);
1558 spin_unlock(&f->lock);
1562 bpf_prog_destroy(old);
1566 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1569 struct bpf_prog *new;
1570 struct sock_fprog fprog;
1573 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1575 if (len != sizeof(fprog))
1577 if (copy_from_user(&fprog, data, len))
1580 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1584 __fanout_set_data_bpf(po->fanout, new);
1588 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1591 struct bpf_prog *new;
1594 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596 if (len != sizeof(fd))
1598 if (copy_from_user(&fd, data, len))
1601 new = bpf_prog_get(fd);
1603 return PTR_ERR(new);
1604 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1609 __fanout_set_data_bpf(po->fanout, new);
1613 static int fanout_set_data(struct packet_sock *po, char __user *data,
1616 switch (po->fanout->type) {
1617 case PACKET_FANOUT_CBPF:
1618 return fanout_set_data_cbpf(po, data, len);
1619 case PACKET_FANOUT_EBPF:
1620 return fanout_set_data_ebpf(po, data, len);
1626 static void fanout_release_data(struct packet_fanout *f)
1629 case PACKET_FANOUT_CBPF:
1630 case PACKET_FANOUT_EBPF:
1631 __fanout_set_data_bpf(f, NULL);
1635 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1637 struct packet_rollover *rollover = NULL;
1638 struct packet_sock *po = pkt_sk(sk);
1639 struct packet_fanout *f, *match;
1640 u8 type = type_flags & 0xff;
1641 u8 flags = type_flags >> 8;
1645 case PACKET_FANOUT_ROLLOVER:
1646 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1648 case PACKET_FANOUT_HASH:
1649 case PACKET_FANOUT_LB:
1650 case PACKET_FANOUT_CPU:
1651 case PACKET_FANOUT_RND:
1652 case PACKET_FANOUT_QM:
1653 case PACKET_FANOUT_CBPF:
1654 case PACKET_FANOUT_EBPF:
1660 mutex_lock(&fanout_mutex);
1666 if (type == PACKET_FANOUT_ROLLOVER ||
1667 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1669 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1672 atomic_long_set(&rollover->num, 0);
1673 atomic_long_set(&rollover->num_huge, 0);
1674 atomic_long_set(&rollover->num_failed, 0);
1678 list_for_each_entry(f, &fanout_list, list) {
1680 read_pnet(&f->net) == sock_net(sk)) {
1686 if (match && match->flags != flags)
1690 match = kzalloc(sizeof(*match), GFP_KERNEL);
1693 write_pnet(&match->net, sock_net(sk));
1696 match->flags = flags;
1697 INIT_LIST_HEAD(&match->list);
1698 spin_lock_init(&match->lock);
1699 atomic_set(&match->sk_ref, 0);
1700 fanout_init_data(match);
1701 match->prot_hook.type = po->prot_hook.type;
1702 match->prot_hook.dev = po->prot_hook.dev;
1703 match->prot_hook.func = packet_rcv_fanout;
1704 match->prot_hook.af_packet_priv = match;
1705 match->prot_hook.id_match = match_fanout_group;
1706 list_add(&match->list, &fanout_list);
1710 spin_lock(&po->bind_lock);
1712 match->type == type &&
1713 match->prot_hook.type == po->prot_hook.type &&
1714 match->prot_hook.dev == po->prot_hook.dev) {
1716 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1717 __dev_remove_pack(&po->prot_hook);
1719 po->rollover = rollover;
1721 atomic_inc(&match->sk_ref);
1722 __fanout_link(sk, po);
1726 spin_unlock(&po->bind_lock);
1728 if (err && !atomic_read(&match->sk_ref)) {
1729 list_del(&match->list);
1735 mutex_unlock(&fanout_mutex);
1739 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1740 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1741 * It is the responsibility of the caller to call fanout_release_data() and
1742 * free the returned packet_fanout (after synchronize_net())
1744 static struct packet_fanout *fanout_release(struct sock *sk)
1746 struct packet_sock *po = pkt_sk(sk);
1747 struct packet_fanout *f;
1749 mutex_lock(&fanout_mutex);
1754 if (atomic_dec_and_test(&f->sk_ref))
1759 mutex_unlock(&fanout_mutex);
1764 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1765 struct sk_buff *skb)
1767 /* Earlier code assumed this would be a VLAN pkt, double-check
1768 * this now that we have the actual packet in hand. We can only
1769 * do this check on Ethernet devices.
1771 if (unlikely(dev->type != ARPHRD_ETHER))
1774 skb_reset_mac_header(skb);
1775 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1778 static const struct proto_ops packet_ops;
1780 static const struct proto_ops packet_ops_spkt;
1782 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1783 struct packet_type *pt, struct net_device *orig_dev)
1786 struct sockaddr_pkt *spkt;
1789 * When we registered the protocol we saved the socket in the data
1790 * field for just this event.
1793 sk = pt->af_packet_priv;
1796 * Yank back the headers [hope the device set this
1797 * right or kerboom...]
1799 * Incoming packets have ll header pulled,
1802 * For outgoing ones skb->data == skb_mac_header(skb)
1803 * so that this procedure is noop.
1806 if (skb->pkt_type == PACKET_LOOPBACK)
1809 if (!net_eq(dev_net(dev), sock_net(sk)))
1812 skb = skb_share_check(skb, GFP_ATOMIC);
1816 /* drop any routing info */
1819 /* drop conntrack reference */
1822 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1824 skb_push(skb, skb->data - skb_mac_header(skb));
1827 * The SOCK_PACKET socket receives _all_ frames.
1830 spkt->spkt_family = dev->type;
1831 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1832 spkt->spkt_protocol = skb->protocol;
1835 * Charge the memory to the socket. This is done specifically
1836 * to prevent sockets using all the memory up.
1839 if (sock_queue_rcv_skb(sk, skb) == 0)
1850 * Output a raw packet to a device layer. This bypasses all the other
1851 * protocol layers and you must therefore supply it with a complete frame
1854 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1857 struct sock *sk = sock->sk;
1858 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1859 struct sk_buff *skb = NULL;
1860 struct net_device *dev;
1866 * Get and verify the address.
1870 if (msg->msg_namelen < sizeof(struct sockaddr))
1872 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1873 proto = saddr->spkt_protocol;
1875 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1878 * Find the device first to size check it
1881 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1884 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1890 if (!(dev->flags & IFF_UP))
1894 * You may not queue a frame bigger than the mtu. This is the lowest level
1895 * raw protocol and you must do your own fragmentation at this level.
1898 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1899 if (!netif_supports_nofcs(dev)) {
1900 err = -EPROTONOSUPPORT;
1903 extra_len = 4; /* We're doing our own CRC */
1907 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1911 size_t reserved = LL_RESERVED_SPACE(dev);
1912 int tlen = dev->needed_tailroom;
1913 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1916 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1919 /* FIXME: Save some space for broken drivers that write a hard
1920 * header at transmission time by themselves. PPP is the notable
1921 * one here. This should really be fixed at the driver level.
1923 skb_reserve(skb, reserved);
1924 skb_reset_network_header(skb);
1926 /* Try to align data part correctly */
1931 skb_reset_network_header(skb);
1933 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1939 if (!dev_validate_header(dev, skb->data, len)) {
1943 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1944 !packet_extra_vlan_len_allowed(dev, skb)) {
1949 skb->protocol = proto;
1951 skb->priority = sk->sk_priority;
1952 skb->mark = sk->sk_mark;
1954 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1956 if (unlikely(extra_len == 4))
1959 skb_probe_transport_header(skb, 0);
1961 dev_queue_xmit(skb);
1972 static unsigned int run_filter(struct sk_buff *skb,
1973 const struct sock *sk,
1976 struct sk_filter *filter;
1979 filter = rcu_dereference(sk->sk_filter);
1981 res = bpf_prog_run_clear_cb(filter->prog, skb);
1988 * This function makes lazy skb cloning in hope that most of packets
1989 * are discarded by BPF.
1991 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1992 * and skb->cb are mangled. It works because (and until) packets
1993 * falling here are owned by current CPU. Output packets are cloned
1994 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1995 * sequencially, so that if we return skb to original state on exit,
1996 * we will not harm anyone.
1999 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2000 struct packet_type *pt, struct net_device *orig_dev)
2003 struct sockaddr_ll *sll;
2004 struct packet_sock *po;
2005 u8 *skb_head = skb->data;
2006 int skb_len = skb->len;
2007 unsigned int snaplen, res;
2009 if (skb->pkt_type == PACKET_LOOPBACK)
2012 sk = pt->af_packet_priv;
2015 if (!net_eq(dev_net(dev), sock_net(sk)))
2020 if (dev->header_ops) {
2021 /* The device has an explicit notion of ll header,
2022 * exported to higher levels.
2024 * Otherwise, the device hides details of its frame
2025 * structure, so that corresponding packet head is
2026 * never delivered to user.
2028 if (sk->sk_type != SOCK_DGRAM)
2029 skb_push(skb, skb->data - skb_mac_header(skb));
2030 else if (skb->pkt_type == PACKET_OUTGOING) {
2031 /* Special case: outgoing packets have ll header at head */
2032 skb_pull(skb, skb_network_offset(skb));
2038 res = run_filter(skb, sk, snaplen);
2040 goto drop_n_restore;
2044 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2047 if (skb_shared(skb)) {
2048 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2052 if (skb_head != skb->data) {
2053 skb->data = skb_head;
2060 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2062 sll = &PACKET_SKB_CB(skb)->sa.ll;
2063 sll->sll_hatype = dev->type;
2064 sll->sll_pkttype = skb->pkt_type;
2065 if (unlikely(po->origdev))
2066 sll->sll_ifindex = orig_dev->ifindex;
2068 sll->sll_ifindex = dev->ifindex;
2070 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2072 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2073 * Use their space for storing the original skb length.
2075 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2077 if (pskb_trim(skb, snaplen))
2080 skb_set_owner_r(skb, sk);
2084 /* drop conntrack reference */
2087 spin_lock(&sk->sk_receive_queue.lock);
2088 po->stats.stats1.tp_packets++;
2089 sock_skb_set_dropcount(sk, skb);
2090 __skb_queue_tail(&sk->sk_receive_queue, skb);
2091 spin_unlock(&sk->sk_receive_queue.lock);
2092 sk->sk_data_ready(sk);
2096 spin_lock(&sk->sk_receive_queue.lock);
2097 po->stats.stats1.tp_drops++;
2098 atomic_inc(&sk->sk_drops);
2099 spin_unlock(&sk->sk_receive_queue.lock);
2102 if (skb_head != skb->data && skb_shared(skb)) {
2103 skb->data = skb_head;
2111 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2112 struct packet_type *pt, struct net_device *orig_dev)
2115 struct packet_sock *po;
2116 struct sockaddr_ll *sll;
2117 union tpacket_uhdr h;
2118 u8 *skb_head = skb->data;
2119 int skb_len = skb->len;
2120 unsigned int snaplen, res;
2121 unsigned long status = TP_STATUS_USER;
2122 unsigned short macoff, netoff, hdrlen;
2123 struct sk_buff *copy_skb = NULL;
2127 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2128 * We may add members to them until current aligned size without forcing
2129 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2131 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2132 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2134 if (skb->pkt_type == PACKET_LOOPBACK)
2137 sk = pt->af_packet_priv;
2140 if (!net_eq(dev_net(dev), sock_net(sk)))
2143 if (dev->header_ops) {
2144 if (sk->sk_type != SOCK_DGRAM)
2145 skb_push(skb, skb->data - skb_mac_header(skb));
2146 else if (skb->pkt_type == PACKET_OUTGOING) {
2147 /* Special case: outgoing packets have ll header at head */
2148 skb_pull(skb, skb_network_offset(skb));
2154 res = run_filter(skb, sk, snaplen);
2156 goto drop_n_restore;
2158 if (skb->ip_summed == CHECKSUM_PARTIAL)
2159 status |= TP_STATUS_CSUMNOTREADY;
2160 else if (skb->pkt_type != PACKET_OUTGOING &&
2161 (skb->ip_summed == CHECKSUM_COMPLETE ||
2162 skb_csum_unnecessary(skb)))
2163 status |= TP_STATUS_CSUM_VALID;
2168 if (sk->sk_type == SOCK_DGRAM) {
2169 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2172 unsigned int maclen = skb_network_offset(skb);
2173 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2174 (maclen < 16 ? 16 : maclen)) +
2176 macoff = netoff - maclen;
2178 if (po->tp_version <= TPACKET_V2) {
2179 if (macoff + snaplen > po->rx_ring.frame_size) {
2180 if (po->copy_thresh &&
2181 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2182 if (skb_shared(skb)) {
2183 copy_skb = skb_clone(skb, GFP_ATOMIC);
2185 copy_skb = skb_get(skb);
2186 skb_head = skb->data;
2189 skb_set_owner_r(copy_skb, sk);
2191 snaplen = po->rx_ring.frame_size - macoff;
2192 if ((int)snaplen < 0)
2195 } else if (unlikely(macoff + snaplen >
2196 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2199 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2200 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2201 snaplen, nval, macoff);
2203 if (unlikely((int)snaplen < 0)) {
2205 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2208 spin_lock(&sk->sk_receive_queue.lock);
2209 h.raw = packet_current_rx_frame(po, skb,
2210 TP_STATUS_KERNEL, (macoff+snaplen));
2213 if (po->tp_version <= TPACKET_V2) {
2214 packet_increment_rx_head(po, &po->rx_ring);
2216 * LOSING will be reported till you read the stats,
2217 * because it's COR - Clear On Read.
2218 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2221 if (po->stats.stats1.tp_drops)
2222 status |= TP_STATUS_LOSING;
2224 po->stats.stats1.tp_packets++;
2226 status |= TP_STATUS_COPY;
2227 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2229 spin_unlock(&sk->sk_receive_queue.lock);
2231 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2233 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2234 getnstimeofday(&ts);
2236 status |= ts_status;
2238 switch (po->tp_version) {
2240 h.h1->tp_len = skb->len;
2241 h.h1->tp_snaplen = snaplen;
2242 h.h1->tp_mac = macoff;
2243 h.h1->tp_net = netoff;
2244 h.h1->tp_sec = ts.tv_sec;
2245 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2246 hdrlen = sizeof(*h.h1);
2249 h.h2->tp_len = skb->len;
2250 h.h2->tp_snaplen = snaplen;
2251 h.h2->tp_mac = macoff;
2252 h.h2->tp_net = netoff;
2253 h.h2->tp_sec = ts.tv_sec;
2254 h.h2->tp_nsec = ts.tv_nsec;
2255 if (skb_vlan_tag_present(skb)) {
2256 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2257 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2258 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2260 h.h2->tp_vlan_tci = 0;
2261 h.h2->tp_vlan_tpid = 0;
2263 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2264 hdrlen = sizeof(*h.h2);
2267 /* tp_nxt_offset,vlan are already populated above.
2268 * So DONT clear those fields here
2270 h.h3->tp_status |= status;
2271 h.h3->tp_len = skb->len;
2272 h.h3->tp_snaplen = snaplen;
2273 h.h3->tp_mac = macoff;
2274 h.h3->tp_net = netoff;
2275 h.h3->tp_sec = ts.tv_sec;
2276 h.h3->tp_nsec = ts.tv_nsec;
2277 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2278 hdrlen = sizeof(*h.h3);
2284 sll = h.raw + TPACKET_ALIGN(hdrlen);
2285 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2286 sll->sll_family = AF_PACKET;
2287 sll->sll_hatype = dev->type;
2288 sll->sll_protocol = skb->protocol;
2289 sll->sll_pkttype = skb->pkt_type;
2290 if (unlikely(po->origdev))
2291 sll->sll_ifindex = orig_dev->ifindex;
2293 sll->sll_ifindex = dev->ifindex;
2297 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2298 if (po->tp_version <= TPACKET_V2) {
2301 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2304 for (start = h.raw; start < end; start += PAGE_SIZE)
2305 flush_dcache_page(pgv_to_page(start));
2310 if (po->tp_version <= TPACKET_V2) {
2311 __packet_set_status(po, h.raw, status);
2312 sk->sk_data_ready(sk);
2314 prb_clear_blk_fill_status(&po->rx_ring);
2318 if (skb_head != skb->data && skb_shared(skb)) {
2319 skb->data = skb_head;
2327 po->stats.stats1.tp_drops++;
2328 spin_unlock(&sk->sk_receive_queue.lock);
2330 sk->sk_data_ready(sk);
2331 kfree_skb(copy_skb);
2332 goto drop_n_restore;
2335 static void tpacket_destruct_skb(struct sk_buff *skb)
2337 struct packet_sock *po = pkt_sk(skb->sk);
2339 if (likely(po->tx_ring.pg_vec)) {
2343 ph = skb_shinfo(skb)->destructor_arg;
2344 packet_dec_pending(&po->tx_ring);
2346 ts = __packet_set_timestamp(po, ph, skb);
2347 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2353 static void tpacket_set_protocol(const struct net_device *dev,
2354 struct sk_buff *skb)
2356 if (dev->type == ARPHRD_ETHER) {
2357 skb_reset_mac_header(skb);
2358 skb->protocol = eth_hdr(skb)->h_proto;
2362 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2363 void *frame, struct net_device *dev, int size_max,
2364 __be16 proto, unsigned char *addr, int hlen)
2366 union tpacket_uhdr ph;
2367 int to_write, offset, len, tp_len, nr_frags, len_max;
2368 struct socket *sock = po->sk.sk_socket;
2375 skb->protocol = proto;
2377 skb->priority = po->sk.sk_priority;
2378 skb->mark = po->sk.sk_mark;
2379 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2380 skb_shinfo(skb)->destructor_arg = ph.raw;
2382 switch (po->tp_version) {
2384 tp_len = ph.h2->tp_len;
2387 tp_len = ph.h1->tp_len;
2390 if (unlikely(tp_len > size_max)) {
2391 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2395 skb_reserve(skb, hlen);
2396 skb_reset_network_header(skb);
2398 if (unlikely(po->tp_tx_has_off)) {
2399 int off_min, off_max, off;
2400 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2401 off_max = po->tx_ring.frame_size - tp_len;
2402 if (sock->type == SOCK_DGRAM) {
2403 switch (po->tp_version) {
2405 off = ph.h2->tp_net;
2408 off = ph.h1->tp_net;
2412 switch (po->tp_version) {
2414 off = ph.h2->tp_mac;
2417 off = ph.h1->tp_mac;
2421 if (unlikely((off < off_min) || (off_max < off)))
2423 data = ph.raw + off;
2425 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2429 if (sock->type == SOCK_DGRAM) {
2430 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2432 if (unlikely(err < 0))
2434 } else if (dev->hard_header_len) {
2435 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2437 skb_push(skb, dev->hard_header_len);
2438 err = skb_store_bits(skb, 0, data, hdrlen);
2441 if (!dev_validate_header(dev, skb->data, hdrlen))
2444 tpacket_set_protocol(dev, skb);
2450 offset = offset_in_page(data);
2451 len_max = PAGE_SIZE - offset;
2452 len = ((to_write > len_max) ? len_max : to_write);
2454 skb->data_len = to_write;
2455 skb->len += to_write;
2456 skb->truesize += to_write;
2457 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2459 while (likely(to_write)) {
2460 nr_frags = skb_shinfo(skb)->nr_frags;
2462 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2463 pr_err("Packet exceed the number of skb frags(%lu)\n",
2468 page = pgv_to_page(data);
2470 flush_dcache_page(page);
2472 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2475 len_max = PAGE_SIZE;
2476 len = ((to_write > len_max) ? len_max : to_write);
2479 skb_probe_transport_header(skb, 0);
2484 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2486 struct sk_buff *skb;
2487 struct net_device *dev;
2489 int err, reserve = 0;
2491 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2492 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2493 unsigned char *addr = NULL;
2494 int tp_len, size_max;
2496 int status = TP_STATUS_AVAILABLE;
2499 mutex_lock(&po->pg_vec_lock);
2501 if (likely(saddr == NULL)) {
2502 dev = packet_cached_dev_get(po);
2507 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2509 if (msg->msg_namelen < (saddr->sll_halen
2510 + offsetof(struct sockaddr_ll,
2513 proto = saddr->sll_protocol;
2514 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2515 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2516 if (dev && msg->msg_namelen < dev->addr_len +
2517 offsetof(struct sockaddr_ll, sll_addr))
2519 addr = saddr->sll_addr;
2524 if (unlikely(dev == NULL))
2527 if (unlikely(!(dev->flags & IFF_UP)))
2530 if (po->sk.sk_socket->type == SOCK_RAW)
2531 reserve = dev->hard_header_len;
2532 size_max = po->tx_ring.frame_size
2533 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2535 if (size_max > dev->mtu + reserve + VLAN_HLEN)
2536 size_max = dev->mtu + reserve + VLAN_HLEN;
2539 ph = packet_current_frame(po, &po->tx_ring,
2540 TP_STATUS_SEND_REQUEST);
2541 if (unlikely(ph == NULL)) {
2542 if (need_wait && need_resched())
2547 status = TP_STATUS_SEND_REQUEST;
2548 hlen = LL_RESERVED_SPACE(dev);
2549 tlen = dev->needed_tailroom;
2550 skb = sock_alloc_send_skb(&po->sk,
2551 hlen + tlen + sizeof(struct sockaddr_ll),
2554 if (unlikely(skb == NULL)) {
2555 /* we assume the socket was initially writeable ... */
2556 if (likely(len_sum > 0))
2560 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2562 if (likely(tp_len >= 0) &&
2563 tp_len > dev->mtu + reserve &&
2564 !packet_extra_vlan_len_allowed(dev, skb))
2567 if (unlikely(tp_len < 0)) {
2569 __packet_set_status(po, ph,
2570 TP_STATUS_AVAILABLE);
2571 packet_increment_head(&po->tx_ring);
2575 status = TP_STATUS_WRONG_FORMAT;
2581 packet_pick_tx_queue(dev, skb);
2583 skb->destructor = tpacket_destruct_skb;
2584 __packet_set_status(po, ph, TP_STATUS_SENDING);
2585 packet_inc_pending(&po->tx_ring);
2587 status = TP_STATUS_SEND_REQUEST;
2588 err = po->xmit(skb);
2589 if (unlikely(err > 0)) {
2590 err = net_xmit_errno(err);
2591 if (err && __packet_get_status(po, ph) ==
2592 TP_STATUS_AVAILABLE) {
2593 /* skb was destructed already */
2598 * skb was dropped but not destructed yet;
2599 * let's treat it like congestion or err < 0
2603 packet_increment_head(&po->tx_ring);
2605 } while (likely((ph != NULL) ||
2606 /* Note: packet_read_pending() might be slow if we have
2607 * to call it as it's per_cpu variable, but in fast-path
2608 * we already short-circuit the loop with the first
2609 * condition, and luckily don't have to go that path
2612 (need_wait && packet_read_pending(&po->tx_ring))));
2618 __packet_set_status(po, ph, status);
2623 mutex_unlock(&po->pg_vec_lock);
2627 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2628 size_t reserve, size_t len,
2629 size_t linear, int noblock,
2632 struct sk_buff *skb;
2634 /* Under a page? Don't bother with paged skb. */
2635 if (prepad + len < PAGE_SIZE || !linear)
2638 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2643 skb_reserve(skb, reserve);
2644 skb_put(skb, linear);
2645 skb->data_len = len - linear;
2646 skb->len += len - linear;
2651 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2653 struct sock *sk = sock->sk;
2654 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2655 struct sk_buff *skb;
2656 struct net_device *dev;
2658 unsigned char *addr = NULL;
2659 int err, reserve = 0;
2660 struct sockcm_cookie sockc;
2661 struct virtio_net_hdr vnet_hdr = { 0 };
2664 struct packet_sock *po = pkt_sk(sk);
2665 unsigned short gso_type = 0;
2666 bool has_vnet_hdr = false;
2667 int hlen, tlen, linear;
2672 * Get and verify the address.
2675 if (likely(saddr == NULL)) {
2676 dev = packet_cached_dev_get(po);
2680 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2682 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2684 proto = saddr->sll_protocol;
2685 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2686 if (sock->type == SOCK_DGRAM) {
2687 if (dev && msg->msg_namelen < dev->addr_len +
2688 offsetof(struct sockaddr_ll, sll_addr))
2690 addr = saddr->sll_addr;
2695 if (unlikely(dev == NULL))
2698 if (unlikely(!(dev->flags & IFF_UP)))
2701 sockc.mark = sk->sk_mark;
2702 if (msg->msg_controllen) {
2703 err = sock_cmsg_send(sk, msg, &sockc);
2708 if (sock->type == SOCK_RAW)
2709 reserve = dev->hard_header_len;
2710 if (po->has_vnet_hdr) {
2711 vnet_hdr_len = sizeof(vnet_hdr);
2714 if (len < vnet_hdr_len)
2717 len -= vnet_hdr_len;
2720 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2721 if (n != vnet_hdr_len)
2724 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2725 (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2726 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2727 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2728 vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2729 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2730 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2733 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2736 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2737 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2738 case VIRTIO_NET_HDR_GSO_TCPV4:
2739 gso_type = SKB_GSO_TCPV4;
2741 case VIRTIO_NET_HDR_GSO_TCPV6:
2742 gso_type = SKB_GSO_TCPV6;
2744 case VIRTIO_NET_HDR_GSO_UDP:
2745 gso_type = SKB_GSO_UDP;
2751 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2752 gso_type |= SKB_GSO_TCP_ECN;
2754 if (vnet_hdr.gso_size == 0)
2758 has_vnet_hdr = true;
2761 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2762 if (!netif_supports_nofcs(dev)) {
2763 err = -EPROTONOSUPPORT;
2766 extra_len = 4; /* We're doing our own CRC */
2770 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2774 hlen = LL_RESERVED_SPACE(dev);
2775 tlen = dev->needed_tailroom;
2776 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2777 linear = max(linear, min_t(int, len, dev->hard_header_len));
2778 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2779 msg->msg_flags & MSG_DONTWAIT, &err);
2783 skb_reset_network_header(skb);
2786 if (sock->type == SOCK_DGRAM) {
2787 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2788 if (unlikely(offset < 0))
2790 } else if (reserve) {
2791 skb_reserve(skb, -reserve);
2793 skb_reset_network_header(skb);
2796 /* Returns -EFAULT on error */
2797 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2801 if (sock->type == SOCK_RAW &&
2802 !dev_validate_header(dev, skb->data, len)) {
2807 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2809 if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2810 !packet_extra_vlan_len_allowed(dev, skb)) {
2815 skb->protocol = proto;
2817 skb->priority = sk->sk_priority;
2818 skb->mark = sockc.mark;
2820 packet_pick_tx_queue(dev, skb);
2823 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2824 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2825 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2826 if (!skb_partial_csum_set(skb, s, o)) {
2832 skb_shinfo(skb)->gso_size =
2833 __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2834 skb_shinfo(skb)->gso_type = gso_type;
2836 /* Header must be checked, and gso_segs computed. */
2837 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2838 skb_shinfo(skb)->gso_segs = 0;
2840 len += vnet_hdr_len;
2843 skb_probe_transport_header(skb, reserve);
2845 if (unlikely(extra_len == 4))
2848 err = po->xmit(skb);
2849 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2865 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2867 struct sock *sk = sock->sk;
2868 struct packet_sock *po = pkt_sk(sk);
2870 if (po->tx_ring.pg_vec)
2871 return tpacket_snd(po, msg);
2873 return packet_snd(sock, msg, len);
2877 * Close a PACKET socket. This is fairly simple. We immediately go
2878 * to 'closed' state and remove our protocol entry in the device list.
2881 static int packet_release(struct socket *sock)
2883 struct sock *sk = sock->sk;
2884 struct packet_sock *po;
2885 struct packet_fanout *f;
2887 union tpacket_req_u req_u;
2895 mutex_lock(&net->packet.sklist_lock);
2896 sk_del_node_init_rcu(sk);
2897 mutex_unlock(&net->packet.sklist_lock);
2900 sock_prot_inuse_add(net, sk->sk_prot, -1);
2903 spin_lock(&po->bind_lock);
2904 unregister_prot_hook(sk, false);
2905 packet_cached_dev_reset(po);
2907 if (po->prot_hook.dev) {
2908 dev_put(po->prot_hook.dev);
2909 po->prot_hook.dev = NULL;
2911 spin_unlock(&po->bind_lock);
2913 packet_flush_mclist(sk);
2916 if (po->rx_ring.pg_vec) {
2917 memset(&req_u, 0, sizeof(req_u));
2918 packet_set_ring(sk, &req_u, 1, 0);
2921 if (po->tx_ring.pg_vec) {
2922 memset(&req_u, 0, sizeof(req_u));
2923 packet_set_ring(sk, &req_u, 1, 1);
2927 f = fanout_release(sk);
2932 kfree(po->rollover);
2933 fanout_release_data(f);
2937 * Now the socket is dead. No more input will appear.
2944 skb_queue_purge(&sk->sk_receive_queue);
2945 packet_free_pending(po);
2946 sk_refcnt_debug_release(sk);
2953 * Attach a packet hook.
2956 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2959 struct packet_sock *po = pkt_sk(sk);
2960 struct net_device *dev_curr;
2963 struct net_device *dev = NULL;
2965 bool unlisted = false;
2968 spin_lock(&po->bind_lock);
2977 dev = dev_get_by_name_rcu(sock_net(sk), name);
2982 } else if (ifindex) {
2983 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2993 proto_curr = po->prot_hook.type;
2994 dev_curr = po->prot_hook.dev;
2996 need_rehook = proto_curr != proto || dev_curr != dev;
3001 /* prevents packet_notifier() from calling
3002 * register_prot_hook()
3005 __unregister_prot_hook(sk, true);
3007 dev_curr = po->prot_hook.dev;
3009 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3013 BUG_ON(po->running);
3015 po->prot_hook.type = proto;
3017 if (unlikely(unlisted)) {
3019 po->prot_hook.dev = NULL;
3021 packet_cached_dev_reset(po);
3023 po->prot_hook.dev = dev;
3024 po->ifindex = dev ? dev->ifindex : 0;
3025 packet_cached_dev_assign(po, dev);
3031 if (proto == 0 || !need_rehook)
3034 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3035 register_prot_hook(sk);
3037 sk->sk_err = ENETDOWN;
3038 if (!sock_flag(sk, SOCK_DEAD))
3039 sk->sk_error_report(sk);
3044 spin_unlock(&po->bind_lock);
3050 * Bind a packet socket to a device
3053 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3056 struct sock *sk = sock->sk;
3057 char name[sizeof(uaddr->sa_data) + 1];
3063 if (addr_len != sizeof(struct sockaddr))
3065 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3068 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3069 name[sizeof(uaddr->sa_data)] = 0;
3071 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3074 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3076 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3077 struct sock *sk = sock->sk;
3083 if (addr_len < sizeof(struct sockaddr_ll))
3085 if (sll->sll_family != AF_PACKET)
3088 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3089 sll->sll_protocol ? : pkt_sk(sk)->num);
3092 static struct proto packet_proto = {
3094 .owner = THIS_MODULE,
3095 .obj_size = sizeof(struct packet_sock),
3099 * Create a packet of type SOCK_PACKET.
3102 static int packet_create(struct net *net, struct socket *sock, int protocol,
3106 struct packet_sock *po;
3107 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3110 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3112 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3113 sock->type != SOCK_PACKET)
3114 return -ESOCKTNOSUPPORT;
3116 sock->state = SS_UNCONNECTED;
3119 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3123 sock->ops = &packet_ops;
3124 if (sock->type == SOCK_PACKET)
3125 sock->ops = &packet_ops_spkt;
3127 sock_init_data(sock, sk);
3130 sk->sk_family = PF_PACKET;
3132 po->xmit = dev_queue_xmit;
3134 err = packet_alloc_pending(po);
3138 packet_cached_dev_reset(po);
3140 sk->sk_destruct = packet_sock_destruct;
3141 sk_refcnt_debug_inc(sk);
3144 * Attach a protocol block
3147 spin_lock_init(&po->bind_lock);
3148 mutex_init(&po->pg_vec_lock);
3149 po->rollover = NULL;
3150 po->prot_hook.func = packet_rcv;
3152 if (sock->type == SOCK_PACKET)
3153 po->prot_hook.func = packet_rcv_spkt;
3155 po->prot_hook.af_packet_priv = sk;
3158 po->prot_hook.type = proto;
3159 __register_prot_hook(sk);
3162 mutex_lock(&net->packet.sklist_lock);
3163 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3164 mutex_unlock(&net->packet.sklist_lock);
3167 sock_prot_inuse_add(net, &packet_proto, 1);
3178 * Pull a packet from our receive queue and hand it to the user.
3179 * If necessary we block.
3182 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3185 struct sock *sk = sock->sk;
3186 struct sk_buff *skb;
3188 int vnet_hdr_len = 0;
3189 unsigned int origlen = 0;
3192 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3196 /* What error should we return now? EUNATTACH? */
3197 if (pkt_sk(sk)->ifindex < 0)
3201 if (flags & MSG_ERRQUEUE) {
3202 err = sock_recv_errqueue(sk, msg, len,
3203 SOL_PACKET, PACKET_TX_TIMESTAMP);
3208 * Call the generic datagram receiver. This handles all sorts
3209 * of horrible races and re-entrancy so we can forget about it
3210 * in the protocol layers.
3212 * Now it will return ENETDOWN, if device have just gone down,
3213 * but then it will block.
3216 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3219 * An error occurred so return it. Because skb_recv_datagram()
3220 * handles the blocking we don't see and worry about blocking
3227 if (pkt_sk(sk)->pressure)
3228 packet_rcv_has_room(pkt_sk(sk), NULL);
3230 if (pkt_sk(sk)->has_vnet_hdr) {
3231 struct virtio_net_hdr vnet_hdr = { 0 };
3234 vnet_hdr_len = sizeof(vnet_hdr);
3235 if (len < vnet_hdr_len)
3238 len -= vnet_hdr_len;
3240 if (skb_is_gso(skb)) {
3241 struct skb_shared_info *sinfo = skb_shinfo(skb);
3243 /* This is a hint as to how much should be linear. */
3245 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3247 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3248 if (sinfo->gso_type & SKB_GSO_TCPV4)
3249 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3250 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3251 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3252 else if (sinfo->gso_type & SKB_GSO_UDP)
3253 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3254 else if (sinfo->gso_type & SKB_GSO_FCOE)
3258 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3259 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3261 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3263 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3264 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3265 vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3266 skb_checksum_start_offset(skb));
3267 vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3269 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3270 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3271 } /* else everything is zero */
3273 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3278 /* You lose any data beyond the buffer you gave. If it worries
3279 * a user program they can ask the device for its MTU
3285 msg->msg_flags |= MSG_TRUNC;
3288 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3292 if (sock->type != SOCK_PACKET) {
3293 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3295 /* Original length was stored in sockaddr_ll fields */
3296 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3297 sll->sll_family = AF_PACKET;
3298 sll->sll_protocol = skb->protocol;
3301 sock_recv_ts_and_drops(msg, sk, skb);
3303 if (msg->msg_name) {
3304 /* If the address length field is there to be filled
3305 * in, we fill it in now.
3307 if (sock->type == SOCK_PACKET) {
3308 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3309 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3311 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3313 msg->msg_namelen = sll->sll_halen +
3314 offsetof(struct sockaddr_ll, sll_addr);
3316 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3320 if (pkt_sk(sk)->auxdata) {
3321 struct tpacket_auxdata aux;
3323 aux.tp_status = TP_STATUS_USER;
3324 if (skb->ip_summed == CHECKSUM_PARTIAL)
3325 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3326 else if (skb->pkt_type != PACKET_OUTGOING &&
3327 (skb->ip_summed == CHECKSUM_COMPLETE ||
3328 skb_csum_unnecessary(skb)))
3329 aux.tp_status |= TP_STATUS_CSUM_VALID;
3331 aux.tp_len = origlen;
3332 aux.tp_snaplen = skb->len;
3334 aux.tp_net = skb_network_offset(skb);
3335 if (skb_vlan_tag_present(skb)) {
3336 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3337 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3338 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3340 aux.tp_vlan_tci = 0;
3341 aux.tp_vlan_tpid = 0;
3343 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3347 * Free or return the buffer as appropriate. Again this
3348 * hides all the races and re-entrancy issues from us.
3350 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3353 skb_free_datagram(sk, skb);
3358 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3359 int *uaddr_len, int peer)
3361 struct net_device *dev;
3362 struct sock *sk = sock->sk;
3367 uaddr->sa_family = AF_PACKET;
3368 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3370 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3372 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3374 *uaddr_len = sizeof(*uaddr);
3379 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3380 int *uaddr_len, int peer)
3382 struct net_device *dev;
3383 struct sock *sk = sock->sk;
3384 struct packet_sock *po = pkt_sk(sk);
3385 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3390 sll->sll_family = AF_PACKET;
3391 sll->sll_ifindex = po->ifindex;
3392 sll->sll_protocol = po->num;
3393 sll->sll_pkttype = 0;
3395 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3397 sll->sll_hatype = dev->type;
3398 sll->sll_halen = dev->addr_len;
3399 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3401 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3405 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3410 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3414 case PACKET_MR_MULTICAST:
3415 if (i->alen != dev->addr_len)
3418 return dev_mc_add(dev, i->addr);
3420 return dev_mc_del(dev, i->addr);
3422 case PACKET_MR_PROMISC:
3423 return dev_set_promiscuity(dev, what);
3424 case PACKET_MR_ALLMULTI:
3425 return dev_set_allmulti(dev, what);
3426 case PACKET_MR_UNICAST:
3427 if (i->alen != dev->addr_len)
3430 return dev_uc_add(dev, i->addr);
3432 return dev_uc_del(dev, i->addr);
3440 static void packet_dev_mclist_delete(struct net_device *dev,
3441 struct packet_mclist **mlp)
3443 struct packet_mclist *ml;
3445 while ((ml = *mlp) != NULL) {
3446 if (ml->ifindex == dev->ifindex) {
3447 packet_dev_mc(dev, ml, -1);
3455 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3457 struct packet_sock *po = pkt_sk(sk);
3458 struct packet_mclist *ml, *i;
3459 struct net_device *dev;
3465 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3470 if (mreq->mr_alen > dev->addr_len)
3474 i = kmalloc(sizeof(*i), GFP_KERNEL);
3479 for (ml = po->mclist; ml; ml = ml->next) {
3480 if (ml->ifindex == mreq->mr_ifindex &&
3481 ml->type == mreq->mr_type &&
3482 ml->alen == mreq->mr_alen &&
3483 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3485 /* Free the new element ... */
3491 i->type = mreq->mr_type;
3492 i->ifindex = mreq->mr_ifindex;
3493 i->alen = mreq->mr_alen;
3494 memcpy(i->addr, mreq->mr_address, i->alen);
3495 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3497 i->next = po->mclist;
3499 err = packet_dev_mc(dev, i, 1);
3501 po->mclist = i->next;
3510 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3512 struct packet_mclist *ml, **mlp;
3516 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3517 if (ml->ifindex == mreq->mr_ifindex &&
3518 ml->type == mreq->mr_type &&
3519 ml->alen == mreq->mr_alen &&
3520 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3521 if (--ml->count == 0) {
3522 struct net_device *dev;
3524 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3526 packet_dev_mc(dev, ml, -1);
3536 static void packet_flush_mclist(struct sock *sk)
3538 struct packet_sock *po = pkt_sk(sk);
3539 struct packet_mclist *ml;
3545 while ((ml = po->mclist) != NULL) {
3546 struct net_device *dev;
3548 po->mclist = ml->next;
3549 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3551 packet_dev_mc(dev, ml, -1);
3558 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3560 struct sock *sk = sock->sk;
3561 struct packet_sock *po = pkt_sk(sk);
3564 if (level != SOL_PACKET)
3565 return -ENOPROTOOPT;
3568 case PACKET_ADD_MEMBERSHIP:
3569 case PACKET_DROP_MEMBERSHIP:
3571 struct packet_mreq_max mreq;
3573 memset(&mreq, 0, sizeof(mreq));
3574 if (len < sizeof(struct packet_mreq))
3576 if (len > sizeof(mreq))
3578 if (copy_from_user(&mreq, optval, len))
3580 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3582 if (optname == PACKET_ADD_MEMBERSHIP)
3583 ret = packet_mc_add(sk, &mreq);
3585 ret = packet_mc_drop(sk, &mreq);
3589 case PACKET_RX_RING:
3590 case PACKET_TX_RING:
3592 union tpacket_req_u req_u;
3596 switch (po->tp_version) {
3599 len = sizeof(req_u.req);
3603 len = sizeof(req_u.req3);
3609 if (pkt_sk(sk)->has_vnet_hdr) {
3612 if (copy_from_user(&req_u.req, optval, len))
3615 ret = packet_set_ring(sk, &req_u, 0,
3616 optname == PACKET_TX_RING);
3622 case PACKET_COPY_THRESH:
3626 if (optlen != sizeof(val))
3628 if (copy_from_user(&val, optval, sizeof(val)))
3631 pkt_sk(sk)->copy_thresh = val;
3634 case PACKET_VERSION:
3638 if (optlen != sizeof(val))
3640 if (copy_from_user(&val, optval, sizeof(val)))
3651 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3654 po->tp_version = val;
3660 case PACKET_RESERVE:
3664 if (optlen != sizeof(val))
3666 if (copy_from_user(&val, optval, sizeof(val)))
3671 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3674 po->tp_reserve = val;
3684 if (optlen != sizeof(val))
3686 if (copy_from_user(&val, optval, sizeof(val)))
3690 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3693 po->tp_loss = !!val;
3699 case PACKET_AUXDATA:
3703 if (optlen < sizeof(val))
3705 if (copy_from_user(&val, optval, sizeof(val)))
3709 po->auxdata = !!val;
3713 case PACKET_ORIGDEV:
3717 if (optlen < sizeof(val))
3719 if (copy_from_user(&val, optval, sizeof(val)))
3723 po->origdev = !!val;
3727 case PACKET_VNET_HDR:
3731 if (sock->type != SOCK_RAW)
3733 if (optlen < sizeof(val))
3735 if (copy_from_user(&val, optval, sizeof(val)))
3739 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3742 po->has_vnet_hdr = !!val;
3748 case PACKET_TIMESTAMP:
3752 if (optlen != sizeof(val))
3754 if (copy_from_user(&val, optval, sizeof(val)))
3757 po->tp_tstamp = val;
3764 if (optlen != sizeof(val))
3766 if (copy_from_user(&val, optval, sizeof(val)))
3769 return fanout_add(sk, val & 0xffff, val >> 16);
3771 case PACKET_FANOUT_DATA:
3776 return fanout_set_data(po, optval, optlen);
3778 case PACKET_TX_HAS_OFF:
3782 if (optlen != sizeof(val))
3784 if (copy_from_user(&val, optval, sizeof(val)))
3788 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3791 po->tp_tx_has_off = !!val;
3797 case PACKET_QDISC_BYPASS:
3801 if (optlen != sizeof(val))
3803 if (copy_from_user(&val, optval, sizeof(val)))
3806 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3810 return -ENOPROTOOPT;
3814 static int packet_getsockopt(struct socket *sock, int level, int optname,
3815 char __user *optval, int __user *optlen)
3818 int val, lv = sizeof(val);
3819 struct sock *sk = sock->sk;
3820 struct packet_sock *po = pkt_sk(sk);
3822 union tpacket_stats_u st;
3823 struct tpacket_rollover_stats rstats;
3825 if (level != SOL_PACKET)
3826 return -ENOPROTOOPT;
3828 if (get_user(len, optlen))
3835 case PACKET_STATISTICS:
3836 spin_lock_bh(&sk->sk_receive_queue.lock);
3837 memcpy(&st, &po->stats, sizeof(st));
3838 memset(&po->stats, 0, sizeof(po->stats));
3839 spin_unlock_bh(&sk->sk_receive_queue.lock);
3841 if (po->tp_version == TPACKET_V3) {
3842 lv = sizeof(struct tpacket_stats_v3);
3843 st.stats3.tp_packets += st.stats3.tp_drops;
3846 lv = sizeof(struct tpacket_stats);
3847 st.stats1.tp_packets += st.stats1.tp_drops;
3852 case PACKET_AUXDATA:
3855 case PACKET_ORIGDEV:
3858 case PACKET_VNET_HDR:
3859 val = po->has_vnet_hdr;
3861 case PACKET_VERSION:
3862 val = po->tp_version;
3865 if (len > sizeof(int))
3867 if (len < sizeof(int))
3869 if (copy_from_user(&val, optval, len))
3873 val = sizeof(struct tpacket_hdr);
3876 val = sizeof(struct tpacket2_hdr);
3879 val = sizeof(struct tpacket3_hdr);
3885 case PACKET_RESERVE:
3886 val = po->tp_reserve;
3891 case PACKET_TIMESTAMP:
3892 val = po->tp_tstamp;
3896 ((u32)po->fanout->id |
3897 ((u32)po->fanout->type << 16) |
3898 ((u32)po->fanout->flags << 24)) :
3901 case PACKET_ROLLOVER_STATS:
3904 rstats.tp_all = atomic_long_read(&po->rollover->num);
3905 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3906 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3908 lv = sizeof(rstats);
3910 case PACKET_TX_HAS_OFF:
3911 val = po->tp_tx_has_off;
3913 case PACKET_QDISC_BYPASS:
3914 val = packet_use_direct_xmit(po);
3917 return -ENOPROTOOPT;
3922 if (put_user(len, optlen))
3924 if (copy_to_user(optval, data, len))
3930 static int packet_notifier(struct notifier_block *this,
3931 unsigned long msg, void *ptr)
3934 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3935 struct net *net = dev_net(dev);
3938 sk_for_each_rcu(sk, &net->packet.sklist) {
3939 struct packet_sock *po = pkt_sk(sk);
3942 case NETDEV_UNREGISTER:
3944 packet_dev_mclist_delete(dev, &po->mclist);
3948 if (dev->ifindex == po->ifindex) {
3949 spin_lock(&po->bind_lock);
3951 __unregister_prot_hook(sk, false);
3952 sk->sk_err = ENETDOWN;
3953 if (!sock_flag(sk, SOCK_DEAD))
3954 sk->sk_error_report(sk);
3956 if (msg == NETDEV_UNREGISTER) {
3957 packet_cached_dev_reset(po);
3959 if (po->prot_hook.dev)
3960 dev_put(po->prot_hook.dev);
3961 po->prot_hook.dev = NULL;
3963 spin_unlock(&po->bind_lock);
3967 if (dev->ifindex == po->ifindex) {
3968 spin_lock(&po->bind_lock);
3970 register_prot_hook(sk);
3971 spin_unlock(&po->bind_lock);
3981 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3984 struct sock *sk = sock->sk;
3989 int amount = sk_wmem_alloc_get(sk);
3991 return put_user(amount, (int __user *)arg);
3995 struct sk_buff *skb;
3998 spin_lock_bh(&sk->sk_receive_queue.lock);
3999 skb = skb_peek(&sk->sk_receive_queue);
4002 spin_unlock_bh(&sk->sk_receive_queue.lock);
4003 return put_user(amount, (int __user *)arg);
4006 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4008 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4018 case SIOCGIFBRDADDR:
4019 case SIOCSIFBRDADDR:
4020 case SIOCGIFNETMASK:
4021 case SIOCSIFNETMASK:
4022 case SIOCGIFDSTADDR:
4023 case SIOCSIFDSTADDR:
4025 return inet_dgram_ops.ioctl(sock, cmd, arg);
4029 return -ENOIOCTLCMD;
4034 static unsigned int packet_poll(struct file *file, struct socket *sock,
4037 struct sock *sk = sock->sk;
4038 struct packet_sock *po = pkt_sk(sk);
4039 unsigned int mask = datagram_poll(file, sock, wait);
4041 spin_lock_bh(&sk->sk_receive_queue.lock);
4042 if (po->rx_ring.pg_vec) {
4043 if (!packet_previous_rx_frame(po, &po->rx_ring,
4045 mask |= POLLIN | POLLRDNORM;
4047 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4049 spin_unlock_bh(&sk->sk_receive_queue.lock);
4050 spin_lock_bh(&sk->sk_write_queue.lock);
4051 if (po->tx_ring.pg_vec) {
4052 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4053 mask |= POLLOUT | POLLWRNORM;
4055 spin_unlock_bh(&sk->sk_write_queue.lock);
4060 /* Dirty? Well, I still did not learn better way to account
4064 static void packet_mm_open(struct vm_area_struct *vma)
4066 struct file *file = vma->vm_file;
4067 struct socket *sock = file->private_data;
4068 struct sock *sk = sock->sk;
4071 atomic_inc(&pkt_sk(sk)->mapped);
4074 static void packet_mm_close(struct vm_area_struct *vma)
4076 struct file *file = vma->vm_file;
4077 struct socket *sock = file->private_data;
4078 struct sock *sk = sock->sk;
4081 atomic_dec(&pkt_sk(sk)->mapped);
4084 static const struct vm_operations_struct packet_mmap_ops = {
4085 .open = packet_mm_open,
4086 .close = packet_mm_close,
4089 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4094 for (i = 0; i < len; i++) {
4095 if (likely(pg_vec[i].buffer)) {
4096 if (is_vmalloc_addr(pg_vec[i].buffer))
4097 vfree(pg_vec[i].buffer);
4099 free_pages((unsigned long)pg_vec[i].buffer,
4101 pg_vec[i].buffer = NULL;
4107 static char *alloc_one_pg_vec_page(unsigned long order)
4110 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4111 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4113 buffer = (char *) __get_free_pages(gfp_flags, order);
4117 /* __get_free_pages failed, fall back to vmalloc */
4118 buffer = vzalloc((1 << order) * PAGE_SIZE);
4122 /* vmalloc failed, lets dig into swap here */
4123 gfp_flags &= ~__GFP_NORETRY;
4124 buffer = (char *) __get_free_pages(gfp_flags, order);
4128 /* complete and utter failure */
4132 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4134 unsigned int block_nr = req->tp_block_nr;
4138 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4139 if (unlikely(!pg_vec))
4142 for (i = 0; i < block_nr; i++) {
4143 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4144 if (unlikely(!pg_vec[i].buffer))
4145 goto out_free_pgvec;
4152 free_pg_vec(pg_vec, order, block_nr);
4157 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4158 int closing, int tx_ring)
4160 struct pgv *pg_vec = NULL;
4161 struct packet_sock *po = pkt_sk(sk);
4162 int was_running, order = 0;
4163 struct packet_ring_buffer *rb;
4164 struct sk_buff_head *rb_queue;
4167 /* Added to avoid minimal code churn */
4168 struct tpacket_req *req = &req_u->req;
4170 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4171 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4172 WARN(1, "Tx-ring is not supported.\n");
4176 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4177 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4181 if (atomic_read(&po->mapped))
4183 if (packet_read_pending(rb))
4187 if (req->tp_block_nr) {
4188 unsigned int min_frame_size;
4190 /* Sanity tests and some calculations */
4192 if (unlikely(rb->pg_vec))
4195 switch (po->tp_version) {
4197 po->tp_hdrlen = TPACKET_HDRLEN;
4200 po->tp_hdrlen = TPACKET2_HDRLEN;
4203 po->tp_hdrlen = TPACKET3_HDRLEN;
4208 if (unlikely((int)req->tp_block_size <= 0))
4210 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4212 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4213 if (po->tp_version >= TPACKET_V3 &&
4214 req->tp_block_size <
4215 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4217 if (unlikely(req->tp_frame_size < min_frame_size))
4219 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4222 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4223 if (unlikely(rb->frames_per_block == 0))
4225 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4227 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4232 order = get_order(req->tp_block_size);
4233 pg_vec = alloc_pg_vec(req, order);
4234 if (unlikely(!pg_vec))
4236 switch (po->tp_version) {
4238 /* Transmit path is not supported. We checked
4239 * it above but just being paranoid
4242 init_prb_bdqc(po, rb, pg_vec, req_u);
4251 if (unlikely(req->tp_frame_nr))
4256 /* Detach socket from network */
4257 spin_lock(&po->bind_lock);
4258 was_running = po->running;
4262 __unregister_prot_hook(sk, false);
4264 spin_unlock(&po->bind_lock);
4269 mutex_lock(&po->pg_vec_lock);
4270 if (closing || atomic_read(&po->mapped) == 0) {
4272 spin_lock_bh(&rb_queue->lock);
4273 swap(rb->pg_vec, pg_vec);
4274 rb->frame_max = (req->tp_frame_nr - 1);
4276 rb->frame_size = req->tp_frame_size;
4277 spin_unlock_bh(&rb_queue->lock);
4279 swap(rb->pg_vec_order, order);
4280 swap(rb->pg_vec_len, req->tp_block_nr);
4282 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4283 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4284 tpacket_rcv : packet_rcv;
4285 skb_queue_purge(rb_queue);
4286 if (atomic_read(&po->mapped))
4287 pr_err("packet_mmap: vma is busy: %d\n",
4288 atomic_read(&po->mapped));
4290 mutex_unlock(&po->pg_vec_lock);
4292 spin_lock(&po->bind_lock);
4295 register_prot_hook(sk);
4297 spin_unlock(&po->bind_lock);
4298 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4299 /* Because we don't support block-based V3 on tx-ring */
4301 prb_shutdown_retire_blk_timer(po, rb_queue);
4305 free_pg_vec(pg_vec, order, req->tp_block_nr);
4310 static int packet_mmap(struct file *file, struct socket *sock,
4311 struct vm_area_struct *vma)
4313 struct sock *sk = sock->sk;
4314 struct packet_sock *po = pkt_sk(sk);
4315 unsigned long size, expected_size;
4316 struct packet_ring_buffer *rb;
4317 unsigned long start;
4324 mutex_lock(&po->pg_vec_lock);
4327 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4329 expected_size += rb->pg_vec_len
4335 if (expected_size == 0)
4338 size = vma->vm_end - vma->vm_start;
4339 if (size != expected_size)
4342 start = vma->vm_start;
4343 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4344 if (rb->pg_vec == NULL)
4347 for (i = 0; i < rb->pg_vec_len; i++) {
4349 void *kaddr = rb->pg_vec[i].buffer;
4352 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4353 page = pgv_to_page(kaddr);
4354 err = vm_insert_page(vma, start, page);
4363 atomic_inc(&po->mapped);
4364 vma->vm_ops = &packet_mmap_ops;
4368 mutex_unlock(&po->pg_vec_lock);
4372 static const struct proto_ops packet_ops_spkt = {
4373 .family = PF_PACKET,
4374 .owner = THIS_MODULE,
4375 .release = packet_release,
4376 .bind = packet_bind_spkt,
4377 .connect = sock_no_connect,
4378 .socketpair = sock_no_socketpair,
4379 .accept = sock_no_accept,
4380 .getname = packet_getname_spkt,
4381 .poll = datagram_poll,
4382 .ioctl = packet_ioctl,
4383 .listen = sock_no_listen,
4384 .shutdown = sock_no_shutdown,
4385 .setsockopt = sock_no_setsockopt,
4386 .getsockopt = sock_no_getsockopt,
4387 .sendmsg = packet_sendmsg_spkt,
4388 .recvmsg = packet_recvmsg,
4389 .mmap = sock_no_mmap,
4390 .sendpage = sock_no_sendpage,
4393 static const struct proto_ops packet_ops = {
4394 .family = PF_PACKET,
4395 .owner = THIS_MODULE,
4396 .release = packet_release,
4397 .bind = packet_bind,
4398 .connect = sock_no_connect,
4399 .socketpair = sock_no_socketpair,
4400 .accept = sock_no_accept,
4401 .getname = packet_getname,
4402 .poll = packet_poll,
4403 .ioctl = packet_ioctl,
4404 .listen = sock_no_listen,
4405 .shutdown = sock_no_shutdown,
4406 .setsockopt = packet_setsockopt,
4407 .getsockopt = packet_getsockopt,
4408 .sendmsg = packet_sendmsg,
4409 .recvmsg = packet_recvmsg,
4410 .mmap = packet_mmap,
4411 .sendpage = sock_no_sendpage,
4414 static const struct net_proto_family packet_family_ops = {
4415 .family = PF_PACKET,
4416 .create = packet_create,
4417 .owner = THIS_MODULE,
4420 static struct notifier_block packet_netdev_notifier = {
4421 .notifier_call = packet_notifier,
4424 #ifdef CONFIG_PROC_FS
4426 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4429 struct net *net = seq_file_net(seq);
4432 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4435 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4437 struct net *net = seq_file_net(seq);
4438 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4441 static void packet_seq_stop(struct seq_file *seq, void *v)
4447 static int packet_seq_show(struct seq_file *seq, void *v)
4449 if (v == SEQ_START_TOKEN)
4450 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4452 struct sock *s = sk_entry(v);
4453 const struct packet_sock *po = pkt_sk(s);
4456 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4458 atomic_read(&s->sk_refcnt),
4463 atomic_read(&s->sk_rmem_alloc),
4464 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4471 static const struct seq_operations packet_seq_ops = {
4472 .start = packet_seq_start,
4473 .next = packet_seq_next,
4474 .stop = packet_seq_stop,
4475 .show = packet_seq_show,
4478 static int packet_seq_open(struct inode *inode, struct file *file)
4480 return seq_open_net(inode, file, &packet_seq_ops,
4481 sizeof(struct seq_net_private));
4484 static const struct file_operations packet_seq_fops = {
4485 .owner = THIS_MODULE,
4486 .open = packet_seq_open,
4488 .llseek = seq_lseek,
4489 .release = seq_release_net,
4494 static int __net_init packet_net_init(struct net *net)
4496 mutex_init(&net->packet.sklist_lock);
4497 INIT_HLIST_HEAD(&net->packet.sklist);
4499 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4505 static void __net_exit packet_net_exit(struct net *net)
4507 remove_proc_entry("packet", net->proc_net);
4510 static struct pernet_operations packet_net_ops = {
4511 .init = packet_net_init,
4512 .exit = packet_net_exit,
4516 static void __exit packet_exit(void)
4518 unregister_netdevice_notifier(&packet_netdev_notifier);
4519 unregister_pernet_subsys(&packet_net_ops);
4520 sock_unregister(PF_PACKET);
4521 proto_unregister(&packet_proto);
4524 static int __init packet_init(void)
4528 rc = proto_register(&packet_proto, 0);
4531 rc = sock_register(&packet_family_ops);
4534 rc = register_pernet_subsys(&packet_net_ops);
4537 rc = register_netdevice_notifier(&packet_netdev_notifier);
4544 unregister_pernet_subsys(&packet_net_ops);
4546 sock_unregister(PF_PACKET);
4548 proto_unregister(&packet_proto);
4553 module_init(packet_init);
4554 module_exit(packet_exit);
4555 MODULE_LICENSE("GPL");
4556 MODULE_ALIAS_NETPROTO(PF_PACKET);