1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #include <linux/types.h>
51 #include <linux/capability.h>
52 #include <linux/fcntl.h>
53 #include <linux/socket.h>
55 #include <linux/inet.h>
56 #include <linux/netdevice.h>
57 #include <linux/if_packet.h>
58 #include <linux/wireless.h>
59 #include <linux/kernel.h>
60 #include <linux/kmod.h>
61 #include <linux/slab.h>
62 #include <linux/vmalloc.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <linux/uaccess.h>
71 #include <asm/ioctls.h>
73 #include <asm/cacheflush.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80 #include <linux/mutex.h>
81 #include <linux/if_vlan.h>
82 #include <linux/virtio_net.h>
83 #include <linux/errqueue.h>
84 #include <linux/net_tstamp.h>
85 #include <linux/percpu.h>
87 #include <net/inet_common.h>
89 #include <linux/bpf.h>
90 #include <net/compat.h>
96 - if device has no dev->hard_header routine, it adds and removes ll header
97 inside itself. In this case ll header is invisible outside of device,
98 but higher levels still should reserve dev->hard_header_len.
99 Some devices are enough clever to reallocate skb, when header
100 will not fit to reserved space (tunnel), another ones are silly
102 - packet socket receives packets with pulled ll header,
103 so that SOCK_RAW should push it back.
108 Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
112 Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
116 Incoming, dev->hard_header==NULL
117 mac_header -> UNKNOWN position. It is very likely, that it points to ll
118 header. PPP makes it, that is wrong, because introduce
119 assymetry between rx and tx paths.
122 Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133 dev->hard_header != NULL
134 mac_header -> ll header
137 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141 We should set nh.raw on output to correct posistion,
142 packet classifier depends on it.
145 /* Private packet socket structures. */
147 /* identical to struct packet_mreq except it has
148 * a longer address field.
150 struct packet_mreq_max {
152 unsigned short mr_type;
153 unsigned short mr_alen;
154 unsigned char mr_address[MAX_ADDR_LEN];
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
164 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
165 int closing, int tx_ring);
167 #define V3_ALIGNMENT (8)
169 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
171 #define BLK_PLUS_PRIV(sz_of_priv) \
172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
186 static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
189 static void packet_increment_head(struct packet_ring_buffer *buff);
190 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
196 static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205 static void packet_flush_mclist(struct sock *sk);
206 static u16 packet_pick_tx_queue(struct sk_buff *skb);
208 struct packet_skb_cb {
210 struct sockaddr_pkt pkt;
212 /* Trick: alias skb original length with
213 * ll.sll_family and ll.protocol in order
216 unsigned int origlen;
217 struct sockaddr_ll ll;
222 #define vio_le() virtio_legacy_is_little_endian()
224 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
226 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
227 #define GET_PBLOCK_DESC(x, bid) \
228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
229 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
231 #define GET_NEXT_PRB_BLK_NUM(x) \
232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
233 ((x)->kactive_blk_num+1) : 0)
235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236 static void __fanout_link(struct sock *sk, struct packet_sock *po);
238 static int packet_direct_xmit(struct sk_buff *skb)
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
243 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
245 struct net_device *dev;
248 dev = rcu_dereference(po->cached_dev);
256 static void packet_cached_dev_assign(struct packet_sock *po,
257 struct net_device *dev)
259 rcu_assign_pointer(po->cached_dev, dev);
262 static void packet_cached_dev_reset(struct packet_sock *po)
264 RCU_INIT_POINTER(po->cached_dev, NULL);
267 static bool packet_use_direct_xmit(const struct packet_sock *po)
269 return po->xmit == packet_direct_xmit;
272 static u16 packet_pick_tx_queue(struct sk_buff *skb)
274 struct net_device *dev = skb->dev;
275 const struct net_device_ops *ops = dev->netdev_ops;
276 int cpu = raw_smp_processor_id();
280 skb->sender_cpu = cpu + 1;
282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
283 if (ops->ndo_select_queue) {
284 queue_index = ops->ndo_select_queue(dev, skb, NULL);
285 queue_index = netdev_cap_txqueue(dev, queue_index);
287 queue_index = netdev_pick_tx(dev, skb, NULL);
293 /* __register_prot_hook must be invoked through register_prot_hook
294 * or from a context in which asynchronous accesses to the packet
295 * socket is not possible (packet_create()).
297 static void __register_prot_hook(struct sock *sk)
299 struct packet_sock *po = pkt_sk(sk);
303 __fanout_link(sk, po);
305 dev_add_pack(&po->prot_hook);
312 static void register_prot_hook(struct sock *sk)
314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
315 __register_prot_hook(sk);
318 /* If the sync parameter is true, we will temporarily drop
319 * the po->bind_lock and do a synchronize_net to make sure no
320 * asynchronous packet processing paths still refer to the elements
321 * of po->prot_hook. If the sync parameter is false, it is the
322 * callers responsibility to take care of this.
324 static void __unregister_prot_hook(struct sock *sk, bool sync)
326 struct packet_sock *po = pkt_sk(sk);
328 lockdep_assert_held_once(&po->bind_lock);
333 __fanout_unlink(sk, po);
335 __dev_remove_pack(&po->prot_hook);
340 spin_unlock(&po->bind_lock);
342 spin_lock(&po->bind_lock);
346 static void unregister_prot_hook(struct sock *sk, bool sync)
348 struct packet_sock *po = pkt_sk(sk);
351 __unregister_prot_hook(sk, sync);
354 static inline struct page * __pure pgv_to_page(void *addr)
356 if (is_vmalloc_addr(addr))
357 return vmalloc_to_page(addr);
358 return virt_to_page(addr);
361 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
363 union tpacket_uhdr h;
366 switch (po->tp_version) {
368 h.h1->tp_status = status;
369 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
372 h.h2->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
376 h.h3->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
380 WARN(1, "TPACKET version not supported.\n");
387 static int __packet_get_status(const struct packet_sock *po, void *frame)
389 union tpacket_uhdr h;
394 switch (po->tp_version) {
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 return h.h1->tp_status;
399 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
400 return h.h2->tp_status;
402 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
403 return h.h3->tp_status;
405 WARN(1, "TPACKET version not supported.\n");
411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
418 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
421 if (ktime_to_timespec64_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
430 union tpacket_uhdr h;
431 struct timespec64 ts;
434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
439 * versions 1 through 3 overflow the timestamps in y2106, since they
440 * all store the seconds in a 32-bit unsigned integer.
441 * If we create a version 4, that should have a 64-bit timestamp,
442 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
445 switch (po->tp_version) {
447 h.h1->tp_sec = ts.tv_sec;
448 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
451 h.h2->tp_sec = ts.tv_sec;
452 h.h2->tp_nsec = ts.tv_nsec;
455 h.h3->tp_sec = ts.tv_sec;
456 h.h3->tp_nsec = ts.tv_nsec;
459 WARN(1, "TPACKET version not supported.\n");
463 /* one flush is safe, as both fields always lie on the same cacheline */
464 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
470 static void *packet_lookup_frame(const struct packet_sock *po,
471 const struct packet_ring_buffer *rb,
472 unsigned int position,
475 unsigned int pg_vec_pos, frame_offset;
476 union tpacket_uhdr h;
478 pg_vec_pos = position / rb->frames_per_block;
479 frame_offset = position % rb->frames_per_block;
481 h.raw = rb->pg_vec[pg_vec_pos].buffer +
482 (frame_offset * rb->frame_size);
484 if (status != __packet_get_status(po, h.raw))
490 static void *packet_current_frame(struct packet_sock *po,
491 struct packet_ring_buffer *rb,
494 return packet_lookup_frame(po, rb, rb->head, status);
497 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
499 del_timer_sync(&pkc->retire_blk_timer);
502 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
503 struct sk_buff_head *rb_queue)
505 struct tpacket_kbdq_core *pkc;
507 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
509 spin_lock_bh(&rb_queue->lock);
510 pkc->delete_blk_timer = 1;
511 spin_unlock_bh(&rb_queue->lock);
513 prb_del_retire_blk_timer(pkc);
516 static void prb_setup_retire_blk_timer(struct packet_sock *po)
518 struct tpacket_kbdq_core *pkc;
520 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
521 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
523 pkc->retire_blk_timer.expires = jiffies;
526 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
527 int blk_size_in_bytes)
529 struct net_device *dev;
530 unsigned int mbits, div;
531 struct ethtool_link_ksettings ecmd;
535 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
536 if (unlikely(!dev)) {
538 return DEFAULT_PRB_RETIRE_TOV;
540 err = __ethtool_get_link_ksettings(dev, &ecmd);
543 return DEFAULT_PRB_RETIRE_TOV;
545 /* If the link speed is so slow you don't really
546 * need to worry about perf anyways
548 if (ecmd.base.speed < SPEED_1000 ||
549 ecmd.base.speed == SPEED_UNKNOWN)
550 return DEFAULT_PRB_RETIRE_TOV;
552 div = ecmd.base.speed / 1000;
553 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
563 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
564 union tpacket_req_u *req_u)
566 p1->feature_req_word = req_u->req3.tp_feature_req_word;
569 static void init_prb_bdqc(struct packet_sock *po,
570 struct packet_ring_buffer *rb,
572 union tpacket_req_u *req_u)
574 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
575 struct tpacket_block_desc *pbd;
577 memset(p1, 0x0, sizeof(*p1));
579 p1->knxt_seq_num = 1;
581 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
582 p1->pkblk_start = pg_vec[0].buffer;
583 p1->kblk_size = req_u->req3.tp_block_size;
584 p1->knum_blocks = req_u->req3.tp_block_nr;
585 p1->hdrlen = po->tp_hdrlen;
586 p1->version = po->tp_version;
587 p1->last_kactive_blk_num = 0;
588 po->stats.stats3.tp_freeze_q_cnt = 0;
589 if (req_u->req3.tp_retire_blk_tov)
590 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
592 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
593 req_u->req3.tp_block_size);
594 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
595 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
597 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
598 prb_init_ft_ops(p1, req_u);
599 prb_setup_retire_blk_timer(po);
600 prb_open_block(p1, pbd);
603 /* Do NOT update the last_blk_num first.
604 * Assumes sk_buff_head lock is held.
606 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
608 mod_timer(&pkc->retire_blk_timer,
609 jiffies + pkc->tov_in_jiffies);
610 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
615 * 1) We refresh the timer only when we open a block.
616 * By doing this we don't waste cycles refreshing the timer
617 * on packet-by-packet basis.
619 * With a 1MB block-size, on a 1Gbps line, it will take
620 * i) ~8 ms to fill a block + ii) memcpy etc.
621 * In this cut we are not accounting for the memcpy time.
623 * So, if the user sets the 'tmo' to 10ms then the timer
624 * will never fire while the block is still getting filled
625 * (which is what we want). However, the user could choose
626 * to close a block early and that's fine.
628 * But when the timer does fire, we check whether or not to refresh it.
629 * Since the tmo granularity is in msecs, it is not too expensive
630 * to refresh the timer, lets say every '8' msecs.
631 * Either the user can set the 'tmo' or we can derive it based on
632 * a) line-speed and b) block-size.
633 * prb_calc_retire_blk_tmo() calculates the tmo.
636 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
638 struct packet_sock *po =
639 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
640 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
642 struct tpacket_block_desc *pbd;
644 spin_lock(&po->sk.sk_receive_queue.lock);
646 frozen = prb_queue_frozen(pkc);
647 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
649 if (unlikely(pkc->delete_blk_timer))
652 /* We only need to plug the race when the block is partially filled.
654 * lock(); increment BLOCK_NUM_PKTS; unlock()
655 * copy_bits() is in progress ...
656 * timer fires on other cpu:
657 * we can't retire the current block because copy_bits
661 if (BLOCK_NUM_PKTS(pbd)) {
662 while (atomic_read(&pkc->blk_fill_in_prog)) {
663 /* Waiting for skb_copy_bits to finish... */
668 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
670 if (!BLOCK_NUM_PKTS(pbd)) {
671 /* An empty block. Just refresh the timer. */
674 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
675 if (!prb_dispatch_next_block(pkc, po))
680 /* Case 1. Queue was frozen because user-space was
683 if (prb_curr_blk_in_use(pbd)) {
685 * Ok, user-space is still behind.
686 * So just refresh the timer.
690 /* Case 2. queue was frozen,user-space caught up,
691 * now the link went idle && the timer fired.
692 * We don't have a block to close.So we open this
693 * block and restart the timer.
694 * opening a block thaws the queue,restarts timer
695 * Thawing/timer-refresh is a side effect.
697 prb_open_block(pkc, pbd);
704 _prb_refresh_rx_retire_blk_timer(pkc);
707 spin_unlock(&po->sk.sk_receive_queue.lock);
710 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
711 struct tpacket_block_desc *pbd1, __u32 status)
713 /* Flush everything minus the block header */
715 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
720 /* Skip the block header(we know header WILL fit in 4K) */
723 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
724 for (; start < end; start += PAGE_SIZE)
725 flush_dcache_page(pgv_to_page(start));
730 /* Now update the block status. */
732 BLOCK_STATUS(pbd1) = status;
734 /* Flush the block header */
736 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
738 flush_dcache_page(pgv_to_page(start));
748 * 2) Increment active_blk_num
750 * Note:We DONT refresh the timer on purpose.
751 * Because almost always the next block will be opened.
753 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
754 struct tpacket_block_desc *pbd1,
755 struct packet_sock *po, unsigned int stat)
757 __u32 status = TP_STATUS_USER | stat;
759 struct tpacket3_hdr *last_pkt;
760 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
761 struct sock *sk = &po->sk;
763 if (atomic_read(&po->tp_drops))
764 status |= TP_STATUS_LOSING;
766 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
767 last_pkt->tp_next_offset = 0;
769 /* Get the ts of the last pkt */
770 if (BLOCK_NUM_PKTS(pbd1)) {
771 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
772 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
774 /* Ok, we tmo'd - so get the current time.
776 * It shouldn't really happen as we don't close empty
777 * blocks. See prb_retire_rx_blk_timer_expired().
779 struct timespec64 ts;
780 ktime_get_real_ts64(&ts);
781 h1->ts_last_pkt.ts_sec = ts.tv_sec;
782 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
787 /* Flush the block */
788 prb_flush_block(pkc1, pbd1, status);
790 sk->sk_data_ready(sk);
792 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
795 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
797 pkc->reset_pending_on_curr_blk = 0;
801 * Side effect of opening a block:
803 * 1) prb_queue is thawed.
804 * 2) retire_blk_timer is refreshed.
807 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
808 struct tpacket_block_desc *pbd1)
810 struct timespec64 ts;
811 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
815 /* We could have just memset this but we will lose the
816 * flexibility of making the priv area sticky
819 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
820 BLOCK_NUM_PKTS(pbd1) = 0;
821 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
823 ktime_get_real_ts64(&ts);
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
828 pkc1->pkblk_start = (char *)pbd1;
829 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
832 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
834 pbd1->version = pkc1->version;
835 pkc1->prev = pkc1->nxt_offset;
836 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
838 prb_thaw_queue(pkc1);
839 _prb_refresh_rx_retire_blk_timer(pkc1);
845 * Queue freeze logic:
846 * 1) Assume tp_block_nr = 8 blocks.
847 * 2) At time 't0', user opens Rx ring.
848 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
849 * 4) user-space is either sleeping or processing block '0'.
850 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
851 * it will close block-7,loop around and try to fill block '0'.
853 * __packet_lookup_frame_in_block
854 * prb_retire_current_block()
855 * prb_dispatch_next_block()
856 * |->(BLOCK_STATUS == USER) evaluates to true
857 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
858 * 6) Now there are two cases:
859 * 6.1) Link goes idle right after the queue is frozen.
860 * But remember, the last open_block() refreshed the timer.
861 * When this timer expires,it will refresh itself so that we can
862 * re-open block-0 in near future.
863 * 6.2) Link is busy and keeps on receiving packets. This is a simple
864 * case and __packet_lookup_frame_in_block will check if block-0
865 * is free and can now be re-used.
867 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
868 struct packet_sock *po)
870 pkc->reset_pending_on_curr_blk = 1;
871 po->stats.stats3.tp_freeze_q_cnt++;
874 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
877 * If the next block is free then we will dispatch it
878 * and return a good offset.
879 * Else, we will freeze the queue.
880 * So, caller must check the return value.
882 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
883 struct packet_sock *po)
885 struct tpacket_block_desc *pbd;
889 /* 1. Get current block num */
890 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
892 /* 2. If this block is currently in_use then freeze the queue */
893 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
894 prb_freeze_queue(pkc, po);
900 * open this block and return the offset where the first packet
901 * needs to get stored.
903 prb_open_block(pkc, pbd);
904 return (void *)pkc->nxt_offset;
907 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
908 struct packet_sock *po, unsigned int status)
910 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
912 /* retire/close the current block */
913 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
915 * Plug the case where copy_bits() is in progress on
916 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
917 * have space to copy the pkt in the current block and
918 * called prb_retire_current_block()
920 * We don't need to worry about the TMO case because
921 * the timer-handler already handled this case.
923 if (!(status & TP_STATUS_BLK_TMO)) {
924 while (atomic_read(&pkc->blk_fill_in_prog)) {
925 /* Waiting for skb_copy_bits to finish... */
929 prb_close_block(pkc, pbd, po, status);
934 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
936 return TP_STATUS_USER & BLOCK_STATUS(pbd);
939 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
941 return pkc->reset_pending_on_curr_blk;
944 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
946 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
947 atomic_dec(&pkc->blk_fill_in_prog);
950 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
951 struct tpacket3_hdr *ppd)
953 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
956 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
957 struct tpacket3_hdr *ppd)
959 ppd->hv1.tp_rxhash = 0;
962 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
963 struct tpacket3_hdr *ppd)
965 if (skb_vlan_tag_present(pkc->skb)) {
966 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
967 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
968 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
970 ppd->hv1.tp_vlan_tci = 0;
971 ppd->hv1.tp_vlan_tpid = 0;
972 ppd->tp_status = TP_STATUS_AVAILABLE;
976 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
977 struct tpacket3_hdr *ppd)
979 ppd->hv1.tp_padding = 0;
980 prb_fill_vlan_info(pkc, ppd);
982 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
983 prb_fill_rxhash(pkc, ppd);
985 prb_clear_rxhash(pkc, ppd);
988 static void prb_fill_curr_block(char *curr,
989 struct tpacket_kbdq_core *pkc,
990 struct tpacket_block_desc *pbd,
993 struct tpacket3_hdr *ppd;
995 ppd = (struct tpacket3_hdr *)curr;
996 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
998 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
999 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1000 BLOCK_NUM_PKTS(pbd) += 1;
1001 atomic_inc(&pkc->blk_fill_in_prog);
1002 prb_run_all_ft_ops(pkc, ppd);
1005 /* Assumes caller has the sk->rx_queue.lock */
1006 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1007 struct sk_buff *skb,
1011 struct tpacket_kbdq_core *pkc;
1012 struct tpacket_block_desc *pbd;
1015 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1016 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1018 /* Queue is frozen when user space is lagging behind */
1019 if (prb_queue_frozen(pkc)) {
1021 * Check if that last block which caused the queue to freeze,
1022 * is still in_use by user-space.
1024 if (prb_curr_blk_in_use(pbd)) {
1025 /* Can't record this packet */
1029 * Ok, the block was released by user-space.
1030 * Now let's open that block.
1031 * opening a block also thaws the queue.
1032 * Thawing is a side effect.
1034 prb_open_block(pkc, pbd);
1039 curr = pkc->nxt_offset;
1041 end = (char *)pbd + pkc->kblk_size;
1043 /* first try the current block */
1044 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1045 prb_fill_curr_block(curr, pkc, pbd, len);
1046 return (void *)curr;
1049 /* Ok, close the current block */
1050 prb_retire_current_block(pkc, po, 0);
1052 /* Now, try to dispatch the next block */
1053 curr = (char *)prb_dispatch_next_block(pkc, po);
1055 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1056 prb_fill_curr_block(curr, pkc, pbd, len);
1057 return (void *)curr;
1061 * No free blocks are available.user_space hasn't caught up yet.
1062 * Queue was just frozen and now this packet will get dropped.
1067 static void *packet_current_rx_frame(struct packet_sock *po,
1068 struct sk_buff *skb,
1069 int status, unsigned int len)
1072 switch (po->tp_version) {
1075 curr = packet_lookup_frame(po, &po->rx_ring,
1076 po->rx_ring.head, status);
1079 return __packet_lookup_frame_in_block(po, skb, len);
1081 WARN(1, "TPACKET version not supported\n");
1087 static void *prb_lookup_block(const struct packet_sock *po,
1088 const struct packet_ring_buffer *rb,
1092 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1093 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1095 if (status != BLOCK_STATUS(pbd))
1100 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1103 if (rb->prb_bdqc.kactive_blk_num)
1104 prev = rb->prb_bdqc.kactive_blk_num-1;
1106 prev = rb->prb_bdqc.knum_blocks-1;
1110 /* Assumes caller has held the rx_queue.lock */
1111 static void *__prb_previous_block(struct packet_sock *po,
1112 struct packet_ring_buffer *rb,
1115 unsigned int previous = prb_previous_blk_num(rb);
1116 return prb_lookup_block(po, rb, previous, status);
1119 static void *packet_previous_rx_frame(struct packet_sock *po,
1120 struct packet_ring_buffer *rb,
1123 if (po->tp_version <= TPACKET_V2)
1124 return packet_previous_frame(po, rb, status);
1126 return __prb_previous_block(po, rb, status);
1129 static void packet_increment_rx_head(struct packet_sock *po,
1130 struct packet_ring_buffer *rb)
1132 switch (po->tp_version) {
1135 return packet_increment_head(rb);
1138 WARN(1, "TPACKET version not supported.\n");
1144 static void *packet_previous_frame(struct packet_sock *po,
1145 struct packet_ring_buffer *rb,
1148 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1149 return packet_lookup_frame(po, rb, previous, status);
1152 static void packet_increment_head(struct packet_ring_buffer *buff)
1154 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1157 static void packet_inc_pending(struct packet_ring_buffer *rb)
1159 this_cpu_inc(*rb->pending_refcnt);
1162 static void packet_dec_pending(struct packet_ring_buffer *rb)
1164 this_cpu_dec(*rb->pending_refcnt);
1167 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1169 unsigned int refcnt = 0;
1172 /* We don't use pending refcount in rx_ring. */
1173 if (rb->pending_refcnt == NULL)
1176 for_each_possible_cpu(cpu)
1177 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1182 static int packet_alloc_pending(struct packet_sock *po)
1184 po->rx_ring.pending_refcnt = NULL;
1186 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1187 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1193 static void packet_free_pending(struct packet_sock *po)
1195 free_percpu(po->tx_ring.pending_refcnt);
1198 #define ROOM_POW_OFF 2
1199 #define ROOM_NONE 0x0
1200 #define ROOM_LOW 0x1
1201 #define ROOM_NORMAL 0x2
1203 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1207 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1208 idx = READ_ONCE(po->rx_ring.head);
1210 idx += len >> pow_off;
1213 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1216 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1220 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1221 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1223 idx += len >> pow_off;
1226 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1229 static int __packet_rcv_has_room(const struct packet_sock *po,
1230 const struct sk_buff *skb)
1232 const struct sock *sk = &po->sk;
1233 int ret = ROOM_NONE;
1235 if (po->prot_hook.func != tpacket_rcv) {
1236 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1237 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1238 - (skb ? skb->truesize : 0);
1240 if (avail > (rcvbuf >> ROOM_POW_OFF))
1248 if (po->tp_version == TPACKET_V3) {
1249 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1251 else if (__tpacket_v3_has_room(po, 0))
1254 if (__tpacket_has_room(po, ROOM_POW_OFF))
1256 else if (__tpacket_has_room(po, 0))
1263 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1267 ret = __packet_rcv_has_room(po, skb);
1268 pressure = ret != ROOM_NORMAL;
1270 if (READ_ONCE(po->pressure) != pressure)
1271 WRITE_ONCE(po->pressure, pressure);
1276 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1278 if (READ_ONCE(po->pressure) &&
1279 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1280 WRITE_ONCE(po->pressure, 0);
1283 static void packet_sock_destruct(struct sock *sk)
1285 skb_queue_purge(&sk->sk_error_queue);
1287 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1288 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1290 if (!sock_flag(sk, SOCK_DEAD)) {
1291 pr_err("Attempt to release alive packet socket: %p\n", sk);
1295 sk_refcnt_debug_dec(sk);
1298 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1300 u32 *history = po->rollover->history;
1304 rxhash = skb_get_hash(skb);
1305 for (i = 0; i < ROLLOVER_HLEN; i++)
1306 if (READ_ONCE(history[i]) == rxhash)
1309 victim = prandom_u32() % ROLLOVER_HLEN;
1311 /* Avoid dirtying the cache line if possible */
1312 if (READ_ONCE(history[victim]) != rxhash)
1313 WRITE_ONCE(history[victim], rxhash);
1315 return count > (ROLLOVER_HLEN >> 1);
1318 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1319 struct sk_buff *skb,
1322 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1325 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1326 struct sk_buff *skb,
1329 unsigned int val = atomic_inc_return(&f->rr_cur);
1334 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1335 struct sk_buff *skb,
1338 return smp_processor_id() % num;
1341 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1342 struct sk_buff *skb,
1345 return prandom_u32_max(num);
1348 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1349 struct sk_buff *skb,
1350 unsigned int idx, bool try_self,
1353 struct packet_sock *po, *po_next, *po_skip = NULL;
1354 unsigned int i, j, room = ROOM_NONE;
1356 po = pkt_sk(f->arr[idx]);
1359 room = packet_rcv_has_room(po, skb);
1360 if (room == ROOM_NORMAL ||
1361 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1366 i = j = min_t(int, po->rollover->sock, num - 1);
1368 po_next = pkt_sk(f->arr[i]);
1369 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1370 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1372 po->rollover->sock = i;
1373 atomic_long_inc(&po->rollover->num);
1374 if (room == ROOM_LOW)
1375 atomic_long_inc(&po->rollover->num_huge);
1383 atomic_long_inc(&po->rollover->num_failed);
1387 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1388 struct sk_buff *skb,
1391 return skb_get_queue_mapping(skb) % num;
1394 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1395 struct sk_buff *skb,
1398 struct bpf_prog *prog;
1399 unsigned int ret = 0;
1402 prog = rcu_dereference(f->bpf_prog);
1404 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1410 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1412 return f->flags & (flag >> 8);
1415 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1416 struct packet_type *pt, struct net_device *orig_dev)
1418 struct packet_fanout *f = pt->af_packet_priv;
1419 unsigned int num = READ_ONCE(f->num_members);
1420 struct net *net = read_pnet(&f->net);
1421 struct packet_sock *po;
1424 if (!net_eq(dev_net(dev), net) || !num) {
1429 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1430 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1435 case PACKET_FANOUT_HASH:
1437 idx = fanout_demux_hash(f, skb, num);
1439 case PACKET_FANOUT_LB:
1440 idx = fanout_demux_lb(f, skb, num);
1442 case PACKET_FANOUT_CPU:
1443 idx = fanout_demux_cpu(f, skb, num);
1445 case PACKET_FANOUT_RND:
1446 idx = fanout_demux_rnd(f, skb, num);
1448 case PACKET_FANOUT_QM:
1449 idx = fanout_demux_qm(f, skb, num);
1451 case PACKET_FANOUT_ROLLOVER:
1452 idx = fanout_demux_rollover(f, skb, 0, false, num);
1454 case PACKET_FANOUT_CBPF:
1455 case PACKET_FANOUT_EBPF:
1456 idx = fanout_demux_bpf(f, skb, num);
1460 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1461 idx = fanout_demux_rollover(f, skb, idx, true, num);
1463 po = pkt_sk(f->arr[idx]);
1464 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1467 DEFINE_MUTEX(fanout_mutex);
1468 EXPORT_SYMBOL_GPL(fanout_mutex);
1469 static LIST_HEAD(fanout_list);
1470 static u16 fanout_next_id;
1472 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1474 struct packet_fanout *f = po->fanout;
1476 spin_lock(&f->lock);
1477 f->arr[f->num_members] = sk;
1480 if (f->num_members == 1)
1481 dev_add_pack(&f->prot_hook);
1482 spin_unlock(&f->lock);
1485 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1487 struct packet_fanout *f = po->fanout;
1490 spin_lock(&f->lock);
1491 for (i = 0; i < f->num_members; i++) {
1492 if (f->arr[i] == sk)
1495 BUG_ON(i >= f->num_members);
1496 f->arr[i] = f->arr[f->num_members - 1];
1498 if (f->num_members == 0)
1499 __dev_remove_pack(&f->prot_hook);
1500 spin_unlock(&f->lock);
1503 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1505 if (sk->sk_family != PF_PACKET)
1508 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1511 static void fanout_init_data(struct packet_fanout *f)
1514 case PACKET_FANOUT_LB:
1515 atomic_set(&f->rr_cur, 0);
1517 case PACKET_FANOUT_CBPF:
1518 case PACKET_FANOUT_EBPF:
1519 RCU_INIT_POINTER(f->bpf_prog, NULL);
1524 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1526 struct bpf_prog *old;
1528 spin_lock(&f->lock);
1529 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1530 rcu_assign_pointer(f->bpf_prog, new);
1531 spin_unlock(&f->lock);
1535 bpf_prog_destroy(old);
1539 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1542 struct bpf_prog *new;
1543 struct sock_fprog fprog;
1546 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1548 if (len != sizeof(fprog))
1550 if (copy_from_user(&fprog, data, len))
1553 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1557 __fanout_set_data_bpf(po->fanout, new);
1561 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1564 struct bpf_prog *new;
1567 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1569 if (len != sizeof(fd))
1571 if (copy_from_user(&fd, data, len))
1574 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1576 return PTR_ERR(new);
1578 __fanout_set_data_bpf(po->fanout, new);
1582 static int fanout_set_data(struct packet_sock *po, char __user *data,
1585 switch (po->fanout->type) {
1586 case PACKET_FANOUT_CBPF:
1587 return fanout_set_data_cbpf(po, data, len);
1588 case PACKET_FANOUT_EBPF:
1589 return fanout_set_data_ebpf(po, data, len);
1595 static void fanout_release_data(struct packet_fanout *f)
1598 case PACKET_FANOUT_CBPF:
1599 case PACKET_FANOUT_EBPF:
1600 __fanout_set_data_bpf(f, NULL);
1604 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1606 struct packet_fanout *f;
1608 list_for_each_entry(f, &fanout_list, list) {
1609 if (f->id == candidate_id &&
1610 read_pnet(&f->net) == sock_net(sk)) {
1617 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1619 u16 id = fanout_next_id;
1622 if (__fanout_id_is_free(sk, id)) {
1624 fanout_next_id = id + 1;
1629 } while (id != fanout_next_id);
1634 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1636 struct packet_rollover *rollover = NULL;
1637 struct packet_sock *po = pkt_sk(sk);
1638 struct packet_fanout *f, *match;
1639 u8 type = type_flags & 0xff;
1640 u8 flags = type_flags >> 8;
1644 case PACKET_FANOUT_ROLLOVER:
1645 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1647 case PACKET_FANOUT_HASH:
1648 case PACKET_FANOUT_LB:
1649 case PACKET_FANOUT_CPU:
1650 case PACKET_FANOUT_RND:
1651 case PACKET_FANOUT_QM:
1652 case PACKET_FANOUT_CBPF:
1653 case PACKET_FANOUT_EBPF:
1659 mutex_lock(&fanout_mutex);
1665 if (type == PACKET_FANOUT_ROLLOVER ||
1666 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1668 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1671 atomic_long_set(&rollover->num, 0);
1672 atomic_long_set(&rollover->num_huge, 0);
1673 atomic_long_set(&rollover->num_failed, 0);
1676 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1681 if (!fanout_find_new_id(sk, &id)) {
1685 /* ephemeral flag for the first socket in the group: drop it */
1686 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1690 list_for_each_entry(f, &fanout_list, list) {
1692 read_pnet(&f->net) == sock_net(sk)) {
1698 if (match && match->flags != flags)
1702 match = kzalloc(sizeof(*match), GFP_KERNEL);
1705 write_pnet(&match->net, sock_net(sk));
1708 match->flags = flags;
1709 INIT_LIST_HEAD(&match->list);
1710 spin_lock_init(&match->lock);
1711 refcount_set(&match->sk_ref, 0);
1712 fanout_init_data(match);
1713 match->prot_hook.type = po->prot_hook.type;
1714 match->prot_hook.dev = po->prot_hook.dev;
1715 match->prot_hook.func = packet_rcv_fanout;
1716 match->prot_hook.af_packet_priv = match;
1717 match->prot_hook.id_match = match_fanout_group;
1718 list_add(&match->list, &fanout_list);
1722 spin_lock(&po->bind_lock);
1724 match->type == type &&
1725 match->prot_hook.type == po->prot_hook.type &&
1726 match->prot_hook.dev == po->prot_hook.dev) {
1728 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1729 __dev_remove_pack(&po->prot_hook);
1731 po->rollover = rollover;
1733 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1734 __fanout_link(sk, po);
1738 spin_unlock(&po->bind_lock);
1740 if (err && !refcount_read(&match->sk_ref)) {
1741 list_del(&match->list);
1747 mutex_unlock(&fanout_mutex);
1751 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1752 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1753 * It is the responsibility of the caller to call fanout_release_data() and
1754 * free the returned packet_fanout (after synchronize_net())
1756 static struct packet_fanout *fanout_release(struct sock *sk)
1758 struct packet_sock *po = pkt_sk(sk);
1759 struct packet_fanout *f;
1761 mutex_lock(&fanout_mutex);
1766 if (refcount_dec_and_test(&f->sk_ref))
1771 mutex_unlock(&fanout_mutex);
1776 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1777 struct sk_buff *skb)
1779 /* Earlier code assumed this would be a VLAN pkt, double-check
1780 * this now that we have the actual packet in hand. We can only
1781 * do this check on Ethernet devices.
1783 if (unlikely(dev->type != ARPHRD_ETHER))
1786 skb_reset_mac_header(skb);
1787 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1790 static const struct proto_ops packet_ops;
1792 static const struct proto_ops packet_ops_spkt;
1794 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1795 struct packet_type *pt, struct net_device *orig_dev)
1798 struct sockaddr_pkt *spkt;
1801 * When we registered the protocol we saved the socket in the data
1802 * field for just this event.
1805 sk = pt->af_packet_priv;
1808 * Yank back the headers [hope the device set this
1809 * right or kerboom...]
1811 * Incoming packets have ll header pulled,
1814 * For outgoing ones skb->data == skb_mac_header(skb)
1815 * so that this procedure is noop.
1818 if (skb->pkt_type == PACKET_LOOPBACK)
1821 if (!net_eq(dev_net(dev), sock_net(sk)))
1824 skb = skb_share_check(skb, GFP_ATOMIC);
1828 /* drop any routing info */
1831 /* drop conntrack reference */
1834 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1836 skb_push(skb, skb->data - skb_mac_header(skb));
1839 * The SOCK_PACKET socket receives _all_ frames.
1842 spkt->spkt_family = dev->type;
1843 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1844 spkt->spkt_protocol = skb->protocol;
1847 * Charge the memory to the socket. This is done specifically
1848 * to prevent sockets using all the memory up.
1851 if (sock_queue_rcv_skb(sk, skb) == 0)
1860 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1862 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1863 sock->type == SOCK_RAW) {
1864 skb_reset_mac_header(skb);
1865 skb->protocol = dev_parse_header_protocol(skb);
1868 skb_probe_transport_header(skb);
1872 * Output a raw packet to a device layer. This bypasses all the other
1873 * protocol layers and you must therefore supply it with a complete frame
1876 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1879 struct sock *sk = sock->sk;
1880 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1881 struct sk_buff *skb = NULL;
1882 struct net_device *dev;
1883 struct sockcm_cookie sockc;
1889 * Get and verify the address.
1893 if (msg->msg_namelen < sizeof(struct sockaddr))
1895 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1896 proto = saddr->spkt_protocol;
1898 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1901 * Find the device first to size check it
1904 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1907 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1913 if (!(dev->flags & IFF_UP))
1917 * You may not queue a frame bigger than the mtu. This is the lowest level
1918 * raw protocol and you must do your own fragmentation at this level.
1921 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1922 if (!netif_supports_nofcs(dev)) {
1923 err = -EPROTONOSUPPORT;
1926 extra_len = 4; /* We're doing our own CRC */
1930 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1934 size_t reserved = LL_RESERVED_SPACE(dev);
1935 int tlen = dev->needed_tailroom;
1936 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1939 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1942 /* FIXME: Save some space for broken drivers that write a hard
1943 * header at transmission time by themselves. PPP is the notable
1944 * one here. This should really be fixed at the driver level.
1946 skb_reserve(skb, reserved);
1947 skb_reset_network_header(skb);
1949 /* Try to align data part correctly */
1954 skb_reset_network_header(skb);
1956 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1962 if (!dev_validate_header(dev, skb->data, len)) {
1966 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1967 !packet_extra_vlan_len_allowed(dev, skb)) {
1972 sockcm_init(&sockc, sk);
1973 if (msg->msg_controllen) {
1974 err = sock_cmsg_send(sk, msg, &sockc);
1979 skb->protocol = proto;
1981 skb->priority = sk->sk_priority;
1982 skb->mark = sk->sk_mark;
1983 skb->tstamp = sockc.transmit_time;
1985 skb_setup_tx_timestamp(skb, sockc.tsflags);
1987 if (unlikely(extra_len == 4))
1990 packet_parse_headers(skb, sock);
1992 dev_queue_xmit(skb);
2003 static unsigned int run_filter(struct sk_buff *skb,
2004 const struct sock *sk,
2007 struct sk_filter *filter;
2010 filter = rcu_dereference(sk->sk_filter);
2012 res = bpf_prog_run_clear_cb(filter->prog, skb);
2018 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2021 struct virtio_net_hdr vnet_hdr;
2023 if (*len < sizeof(vnet_hdr))
2025 *len -= sizeof(vnet_hdr);
2027 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2030 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2034 * This function makes lazy skb cloning in hope that most of packets
2035 * are discarded by BPF.
2037 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2038 * and skb->cb are mangled. It works because (and until) packets
2039 * falling here are owned by current CPU. Output packets are cloned
2040 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2041 * sequencially, so that if we return skb to original state on exit,
2042 * we will not harm anyone.
2045 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2046 struct packet_type *pt, struct net_device *orig_dev)
2049 struct sockaddr_ll *sll;
2050 struct packet_sock *po;
2051 u8 *skb_head = skb->data;
2052 int skb_len = skb->len;
2053 unsigned int snaplen, res;
2054 bool is_drop_n_account = false;
2056 if (skb->pkt_type == PACKET_LOOPBACK)
2059 sk = pt->af_packet_priv;
2062 if (!net_eq(dev_net(dev), sock_net(sk)))
2067 if (dev->header_ops) {
2068 /* The device has an explicit notion of ll header,
2069 * exported to higher levels.
2071 * Otherwise, the device hides details of its frame
2072 * structure, so that corresponding packet head is
2073 * never delivered to user.
2075 if (sk->sk_type != SOCK_DGRAM)
2076 skb_push(skb, skb->data - skb_mac_header(skb));
2077 else if (skb->pkt_type == PACKET_OUTGOING) {
2078 /* Special case: outgoing packets have ll header at head */
2079 skb_pull(skb, skb_network_offset(skb));
2085 res = run_filter(skb, sk, snaplen);
2087 goto drop_n_restore;
2091 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2094 if (skb_shared(skb)) {
2095 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2099 if (skb_head != skb->data) {
2100 skb->data = skb_head;
2107 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2109 sll = &PACKET_SKB_CB(skb)->sa.ll;
2110 sll->sll_hatype = dev->type;
2111 sll->sll_pkttype = skb->pkt_type;
2112 if (unlikely(po->origdev))
2113 sll->sll_ifindex = orig_dev->ifindex;
2115 sll->sll_ifindex = dev->ifindex;
2117 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2119 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2120 * Use their space for storing the original skb length.
2122 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2124 if (pskb_trim(skb, snaplen))
2127 skb_set_owner_r(skb, sk);
2131 /* drop conntrack reference */
2134 spin_lock(&sk->sk_receive_queue.lock);
2135 po->stats.stats1.tp_packets++;
2136 sock_skb_set_dropcount(sk, skb);
2137 __skb_queue_tail(&sk->sk_receive_queue, skb);
2138 spin_unlock(&sk->sk_receive_queue.lock);
2139 sk->sk_data_ready(sk);
2143 is_drop_n_account = true;
2144 atomic_inc(&po->tp_drops);
2145 atomic_inc(&sk->sk_drops);
2148 if (skb_head != skb->data && skb_shared(skb)) {
2149 skb->data = skb_head;
2153 if (!is_drop_n_account)
2160 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2161 struct packet_type *pt, struct net_device *orig_dev)
2164 struct packet_sock *po;
2165 struct sockaddr_ll *sll;
2166 union tpacket_uhdr h;
2167 u8 *skb_head = skb->data;
2168 int skb_len = skb->len;
2169 unsigned int snaplen, res;
2170 unsigned long status = TP_STATUS_USER;
2171 unsigned short macoff, netoff, hdrlen;
2172 struct sk_buff *copy_skb = NULL;
2173 struct timespec64 ts;
2175 bool is_drop_n_account = false;
2176 bool do_vnet = false;
2178 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2179 * We may add members to them until current aligned size without forcing
2180 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2182 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2183 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2185 if (skb->pkt_type == PACKET_LOOPBACK)
2188 sk = pt->af_packet_priv;
2191 if (!net_eq(dev_net(dev), sock_net(sk)))
2194 if (dev->header_ops) {
2195 if (sk->sk_type != SOCK_DGRAM)
2196 skb_push(skb, skb->data - skb_mac_header(skb));
2197 else if (skb->pkt_type == PACKET_OUTGOING) {
2198 /* Special case: outgoing packets have ll header at head */
2199 skb_pull(skb, skb_network_offset(skb));
2205 res = run_filter(skb, sk, snaplen);
2207 goto drop_n_restore;
2209 /* If we are flooded, just give up */
2210 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2211 atomic_inc(&po->tp_drops);
2212 goto drop_n_restore;
2215 if (skb->ip_summed == CHECKSUM_PARTIAL)
2216 status |= TP_STATUS_CSUMNOTREADY;
2217 else if (skb->pkt_type != PACKET_OUTGOING &&
2218 (skb->ip_summed == CHECKSUM_COMPLETE ||
2219 skb_csum_unnecessary(skb)))
2220 status |= TP_STATUS_CSUM_VALID;
2225 if (sk->sk_type == SOCK_DGRAM) {
2226 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2229 unsigned int maclen = skb_network_offset(skb);
2230 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2231 (maclen < 16 ? 16 : maclen)) +
2233 if (po->has_vnet_hdr) {
2234 netoff += sizeof(struct virtio_net_hdr);
2237 macoff = netoff - maclen;
2239 if (po->tp_version <= TPACKET_V2) {
2240 if (macoff + snaplen > po->rx_ring.frame_size) {
2241 if (po->copy_thresh &&
2242 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2243 if (skb_shared(skb)) {
2244 copy_skb = skb_clone(skb, GFP_ATOMIC);
2246 copy_skb = skb_get(skb);
2247 skb_head = skb->data;
2250 skb_set_owner_r(copy_skb, sk);
2252 snaplen = po->rx_ring.frame_size - macoff;
2253 if ((int)snaplen < 0) {
2258 } else if (unlikely(macoff + snaplen >
2259 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2262 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2263 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2264 snaplen, nval, macoff);
2266 if (unlikely((int)snaplen < 0)) {
2268 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2272 spin_lock(&sk->sk_receive_queue.lock);
2273 h.raw = packet_current_rx_frame(po, skb,
2274 TP_STATUS_KERNEL, (macoff+snaplen));
2276 goto drop_n_account;
2279 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2280 sizeof(struct virtio_net_hdr),
2282 goto drop_n_account;
2284 if (po->tp_version <= TPACKET_V2) {
2285 packet_increment_rx_head(po, &po->rx_ring);
2287 * LOSING will be reported till you read the stats,
2288 * because it's COR - Clear On Read.
2289 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2292 if (atomic_read(&po->tp_drops))
2293 status |= TP_STATUS_LOSING;
2296 po->stats.stats1.tp_packets++;
2298 status |= TP_STATUS_COPY;
2299 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2301 spin_unlock(&sk->sk_receive_queue.lock);
2303 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2305 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2306 ktime_get_real_ts64(&ts);
2308 status |= ts_status;
2310 switch (po->tp_version) {
2312 h.h1->tp_len = skb->len;
2313 h.h1->tp_snaplen = snaplen;
2314 h.h1->tp_mac = macoff;
2315 h.h1->tp_net = netoff;
2316 h.h1->tp_sec = ts.tv_sec;
2317 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2318 hdrlen = sizeof(*h.h1);
2321 h.h2->tp_len = skb->len;
2322 h.h2->tp_snaplen = snaplen;
2323 h.h2->tp_mac = macoff;
2324 h.h2->tp_net = netoff;
2325 h.h2->tp_sec = ts.tv_sec;
2326 h.h2->tp_nsec = ts.tv_nsec;
2327 if (skb_vlan_tag_present(skb)) {
2328 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2329 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2330 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2332 h.h2->tp_vlan_tci = 0;
2333 h.h2->tp_vlan_tpid = 0;
2335 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2336 hdrlen = sizeof(*h.h2);
2339 /* tp_nxt_offset,vlan are already populated above.
2340 * So DONT clear those fields here
2342 h.h3->tp_status |= status;
2343 h.h3->tp_len = skb->len;
2344 h.h3->tp_snaplen = snaplen;
2345 h.h3->tp_mac = macoff;
2346 h.h3->tp_net = netoff;
2347 h.h3->tp_sec = ts.tv_sec;
2348 h.h3->tp_nsec = ts.tv_nsec;
2349 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2350 hdrlen = sizeof(*h.h3);
2356 sll = h.raw + TPACKET_ALIGN(hdrlen);
2357 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2358 sll->sll_family = AF_PACKET;
2359 sll->sll_hatype = dev->type;
2360 sll->sll_protocol = skb->protocol;
2361 sll->sll_pkttype = skb->pkt_type;
2362 if (unlikely(po->origdev))
2363 sll->sll_ifindex = orig_dev->ifindex;
2365 sll->sll_ifindex = dev->ifindex;
2369 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2370 if (po->tp_version <= TPACKET_V2) {
2373 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2376 for (start = h.raw; start < end; start += PAGE_SIZE)
2377 flush_dcache_page(pgv_to_page(start));
2382 if (po->tp_version <= TPACKET_V2) {
2383 __packet_set_status(po, h.raw, status);
2384 sk->sk_data_ready(sk);
2386 prb_clear_blk_fill_status(&po->rx_ring);
2390 if (skb_head != skb->data && skb_shared(skb)) {
2391 skb->data = skb_head;
2395 if (!is_drop_n_account)
2402 spin_unlock(&sk->sk_receive_queue.lock);
2403 atomic_inc(&po->tp_drops);
2404 is_drop_n_account = true;
2406 sk->sk_data_ready(sk);
2407 kfree_skb(copy_skb);
2408 goto drop_n_restore;
2411 static void tpacket_destruct_skb(struct sk_buff *skb)
2413 struct packet_sock *po = pkt_sk(skb->sk);
2415 if (likely(po->tx_ring.pg_vec)) {
2419 ph = skb_zcopy_get_nouarg(skb);
2420 packet_dec_pending(&po->tx_ring);
2422 ts = __packet_set_timestamp(po, ph, skb);
2423 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2425 if (!packet_read_pending(&po->tx_ring))
2426 complete(&po->skb_completion);
2432 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2434 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2435 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2436 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2437 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2438 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2439 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2440 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2442 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2448 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2449 struct virtio_net_hdr *vnet_hdr)
2451 if (*len < sizeof(*vnet_hdr))
2453 *len -= sizeof(*vnet_hdr);
2455 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2458 return __packet_snd_vnet_parse(vnet_hdr, *len);
2461 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2462 void *frame, struct net_device *dev, void *data, int tp_len,
2463 __be16 proto, unsigned char *addr, int hlen, int copylen,
2464 const struct sockcm_cookie *sockc)
2466 union tpacket_uhdr ph;
2467 int to_write, offset, len, nr_frags, len_max;
2468 struct socket *sock = po->sk.sk_socket;
2474 skb->protocol = proto;
2476 skb->priority = po->sk.sk_priority;
2477 skb->mark = po->sk.sk_mark;
2478 skb->tstamp = sockc->transmit_time;
2479 skb_setup_tx_timestamp(skb, sockc->tsflags);
2480 skb_zcopy_set_nouarg(skb, ph.raw);
2482 skb_reserve(skb, hlen);
2483 skb_reset_network_header(skb);
2487 if (sock->type == SOCK_DGRAM) {
2488 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2490 if (unlikely(err < 0))
2492 } else if (copylen) {
2493 int hdrlen = min_t(int, copylen, tp_len);
2495 skb_push(skb, dev->hard_header_len);
2496 skb_put(skb, copylen - dev->hard_header_len);
2497 err = skb_store_bits(skb, 0, data, hdrlen);
2500 if (!dev_validate_header(dev, skb->data, hdrlen))
2507 offset = offset_in_page(data);
2508 len_max = PAGE_SIZE - offset;
2509 len = ((to_write > len_max) ? len_max : to_write);
2511 skb->data_len = to_write;
2512 skb->len += to_write;
2513 skb->truesize += to_write;
2514 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2516 while (likely(to_write)) {
2517 nr_frags = skb_shinfo(skb)->nr_frags;
2519 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2520 pr_err("Packet exceed the number of skb frags(%lu)\n",
2525 page = pgv_to_page(data);
2527 flush_dcache_page(page);
2529 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2532 len_max = PAGE_SIZE;
2533 len = ((to_write > len_max) ? len_max : to_write);
2536 packet_parse_headers(skb, sock);
2541 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2542 int size_max, void **data)
2544 union tpacket_uhdr ph;
2549 switch (po->tp_version) {
2551 if (ph.h3->tp_next_offset != 0) {
2552 pr_warn_once("variable sized slot not supported");
2555 tp_len = ph.h3->tp_len;
2558 tp_len = ph.h2->tp_len;
2561 tp_len = ph.h1->tp_len;
2564 if (unlikely(tp_len > size_max)) {
2565 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2569 if (unlikely(po->tp_tx_has_off)) {
2570 int off_min, off_max;
2572 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2573 off_max = po->tx_ring.frame_size - tp_len;
2574 if (po->sk.sk_type == SOCK_DGRAM) {
2575 switch (po->tp_version) {
2577 off = ph.h3->tp_net;
2580 off = ph.h2->tp_net;
2583 off = ph.h1->tp_net;
2587 switch (po->tp_version) {
2589 off = ph.h3->tp_mac;
2592 off = ph.h2->tp_mac;
2595 off = ph.h1->tp_mac;
2599 if (unlikely((off < off_min) || (off_max < off)))
2602 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2605 *data = frame + off;
2609 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2611 struct sk_buff *skb = NULL;
2612 struct net_device *dev;
2613 struct virtio_net_hdr *vnet_hdr = NULL;
2614 struct sockcm_cookie sockc;
2616 int err, reserve = 0;
2618 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2619 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2620 unsigned char *addr = NULL;
2621 int tp_len, size_max;
2624 int status = TP_STATUS_AVAILABLE;
2625 int hlen, tlen, copylen = 0;
2628 mutex_lock(&po->pg_vec_lock);
2630 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2631 * we need to confirm it under protection of pg_vec_lock.
2633 if (unlikely(!po->tx_ring.pg_vec)) {
2637 if (likely(saddr == NULL)) {
2638 dev = packet_cached_dev_get(po);
2642 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2644 if (msg->msg_namelen < (saddr->sll_halen
2645 + offsetof(struct sockaddr_ll,
2648 proto = saddr->sll_protocol;
2649 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2650 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2651 if (dev && msg->msg_namelen < dev->addr_len +
2652 offsetof(struct sockaddr_ll, sll_addr))
2654 addr = saddr->sll_addr;
2659 if (unlikely(dev == NULL))
2662 if (unlikely(!(dev->flags & IFF_UP)))
2665 sockcm_init(&sockc, &po->sk);
2666 if (msg->msg_controllen) {
2667 err = sock_cmsg_send(&po->sk, msg, &sockc);
2672 if (po->sk.sk_socket->type == SOCK_RAW)
2673 reserve = dev->hard_header_len;
2674 size_max = po->tx_ring.frame_size
2675 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2677 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2678 size_max = dev->mtu + reserve + VLAN_HLEN;
2680 reinit_completion(&po->skb_completion);
2683 ph = packet_current_frame(po, &po->tx_ring,
2684 TP_STATUS_SEND_REQUEST);
2685 if (unlikely(ph == NULL)) {
2686 if (need_wait && skb) {
2687 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2688 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2690 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2694 /* check for additional frames */
2699 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2703 status = TP_STATUS_SEND_REQUEST;
2704 hlen = LL_RESERVED_SPACE(dev);
2705 tlen = dev->needed_tailroom;
2706 if (po->has_vnet_hdr) {
2708 data += sizeof(*vnet_hdr);
2709 tp_len -= sizeof(*vnet_hdr);
2711 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2715 copylen = __virtio16_to_cpu(vio_le(),
2718 copylen = max_t(int, copylen, dev->hard_header_len);
2719 skb = sock_alloc_send_skb(&po->sk,
2720 hlen + tlen + sizeof(struct sockaddr_ll) +
2721 (copylen - dev->hard_header_len),
2724 if (unlikely(skb == NULL)) {
2725 /* we assume the socket was initially writeable ... */
2726 if (likely(len_sum > 0))
2730 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2731 addr, hlen, copylen, &sockc);
2732 if (likely(tp_len >= 0) &&
2733 tp_len > dev->mtu + reserve &&
2734 !po->has_vnet_hdr &&
2735 !packet_extra_vlan_len_allowed(dev, skb))
2738 if (unlikely(tp_len < 0)) {
2741 __packet_set_status(po, ph,
2742 TP_STATUS_AVAILABLE);
2743 packet_increment_head(&po->tx_ring);
2747 status = TP_STATUS_WRONG_FORMAT;
2753 if (po->has_vnet_hdr) {
2754 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2758 virtio_net_hdr_set_proto(skb, vnet_hdr);
2761 skb->destructor = tpacket_destruct_skb;
2762 __packet_set_status(po, ph, TP_STATUS_SENDING);
2763 packet_inc_pending(&po->tx_ring);
2765 status = TP_STATUS_SEND_REQUEST;
2766 err = po->xmit(skb);
2767 if (unlikely(err > 0)) {
2768 err = net_xmit_errno(err);
2769 if (err && __packet_get_status(po, ph) ==
2770 TP_STATUS_AVAILABLE) {
2771 /* skb was destructed already */
2776 * skb was dropped but not destructed yet;
2777 * let's treat it like congestion or err < 0
2781 packet_increment_head(&po->tx_ring);
2783 } while (likely((ph != NULL) ||
2784 /* Note: packet_read_pending() might be slow if we have
2785 * to call it as it's per_cpu variable, but in fast-path
2786 * we already short-circuit the loop with the first
2787 * condition, and luckily don't have to go that path
2790 (need_wait && packet_read_pending(&po->tx_ring))));
2796 __packet_set_status(po, ph, status);
2801 mutex_unlock(&po->pg_vec_lock);
2805 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2806 size_t reserve, size_t len,
2807 size_t linear, int noblock,
2810 struct sk_buff *skb;
2812 /* Under a page? Don't bother with paged skb. */
2813 if (prepad + len < PAGE_SIZE || !linear)
2816 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2821 skb_reserve(skb, reserve);
2822 skb_put(skb, linear);
2823 skb->data_len = len - linear;
2824 skb->len += len - linear;
2829 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2831 struct sock *sk = sock->sk;
2832 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2833 struct sk_buff *skb;
2834 struct net_device *dev;
2836 unsigned char *addr = NULL;
2837 int err, reserve = 0;
2838 struct sockcm_cookie sockc;
2839 struct virtio_net_hdr vnet_hdr = { 0 };
2841 struct packet_sock *po = pkt_sk(sk);
2842 bool has_vnet_hdr = false;
2843 int hlen, tlen, linear;
2847 * Get and verify the address.
2850 if (likely(saddr == NULL)) {
2851 dev = packet_cached_dev_get(po);
2855 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2857 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2859 proto = saddr->sll_protocol;
2860 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2861 if (sock->type == SOCK_DGRAM) {
2862 if (dev && msg->msg_namelen < dev->addr_len +
2863 offsetof(struct sockaddr_ll, sll_addr))
2865 addr = saddr->sll_addr;
2870 if (unlikely(dev == NULL))
2873 if (unlikely(!(dev->flags & IFF_UP)))
2876 sockcm_init(&sockc, sk);
2877 sockc.mark = sk->sk_mark;
2878 if (msg->msg_controllen) {
2879 err = sock_cmsg_send(sk, msg, &sockc);
2884 if (sock->type == SOCK_RAW)
2885 reserve = dev->hard_header_len;
2886 if (po->has_vnet_hdr) {
2887 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2890 has_vnet_hdr = true;
2893 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2894 if (!netif_supports_nofcs(dev)) {
2895 err = -EPROTONOSUPPORT;
2898 extra_len = 4; /* We're doing our own CRC */
2902 if (!vnet_hdr.gso_type &&
2903 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2907 hlen = LL_RESERVED_SPACE(dev);
2908 tlen = dev->needed_tailroom;
2909 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2910 linear = max(linear, min_t(int, len, dev->hard_header_len));
2911 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2912 msg->msg_flags & MSG_DONTWAIT, &err);
2916 skb_reset_network_header(skb);
2919 if (sock->type == SOCK_DGRAM) {
2920 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2921 if (unlikely(offset < 0))
2923 } else if (reserve) {
2924 skb_reserve(skb, -reserve);
2925 if (len < reserve + sizeof(struct ipv6hdr) &&
2926 dev->min_header_len != dev->hard_header_len)
2927 skb_reset_network_header(skb);
2930 /* Returns -EFAULT on error */
2931 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2935 if (sock->type == SOCK_RAW &&
2936 !dev_validate_header(dev, skb->data, len)) {
2941 skb_setup_tx_timestamp(skb, sockc.tsflags);
2943 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2944 !packet_extra_vlan_len_allowed(dev, skb)) {
2949 skb->protocol = proto;
2951 skb->priority = sk->sk_priority;
2952 skb->mark = sockc.mark;
2953 skb->tstamp = sockc.transmit_time;
2956 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2959 len += sizeof(vnet_hdr);
2960 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2963 packet_parse_headers(skb, sock);
2965 if (unlikely(extra_len == 4))
2968 err = po->xmit(skb);
2969 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2985 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2987 struct sock *sk = sock->sk;
2988 struct packet_sock *po = pkt_sk(sk);
2990 if (po->tx_ring.pg_vec)
2991 return tpacket_snd(po, msg);
2993 return packet_snd(sock, msg, len);
2997 * Close a PACKET socket. This is fairly simple. We immediately go
2998 * to 'closed' state and remove our protocol entry in the device list.
3001 static int packet_release(struct socket *sock)
3003 struct sock *sk = sock->sk;
3004 struct packet_sock *po;
3005 struct packet_fanout *f;
3007 union tpacket_req_u req_u;
3015 mutex_lock(&net->packet.sklist_lock);
3016 sk_del_node_init_rcu(sk);
3017 mutex_unlock(&net->packet.sklist_lock);
3020 sock_prot_inuse_add(net, sk->sk_prot, -1);
3023 spin_lock(&po->bind_lock);
3024 unregister_prot_hook(sk, false);
3025 packet_cached_dev_reset(po);
3027 if (po->prot_hook.dev) {
3028 dev_put(po->prot_hook.dev);
3029 po->prot_hook.dev = NULL;
3031 spin_unlock(&po->bind_lock);
3033 packet_flush_mclist(sk);
3036 if (po->rx_ring.pg_vec) {
3037 memset(&req_u, 0, sizeof(req_u));
3038 packet_set_ring(sk, &req_u, 1, 0);
3041 if (po->tx_ring.pg_vec) {
3042 memset(&req_u, 0, sizeof(req_u));
3043 packet_set_ring(sk, &req_u, 1, 1);
3047 f = fanout_release(sk);
3051 kfree(po->rollover);
3053 fanout_release_data(f);
3057 * Now the socket is dead. No more input will appear.
3064 skb_queue_purge(&sk->sk_receive_queue);
3065 packet_free_pending(po);
3066 sk_refcnt_debug_release(sk);
3073 * Attach a packet hook.
3076 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3079 struct packet_sock *po = pkt_sk(sk);
3080 struct net_device *dev_curr;
3083 struct net_device *dev = NULL;
3085 bool unlisted = false;
3088 spin_lock(&po->bind_lock);
3097 dev = dev_get_by_name_rcu(sock_net(sk), name);
3102 } else if (ifindex) {
3103 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3113 proto_curr = po->prot_hook.type;
3114 dev_curr = po->prot_hook.dev;
3116 need_rehook = proto_curr != proto || dev_curr != dev;
3121 /* prevents packet_notifier() from calling
3122 * register_prot_hook()
3125 __unregister_prot_hook(sk, true);
3127 dev_curr = po->prot_hook.dev;
3129 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3133 BUG_ON(po->running);
3135 po->prot_hook.type = proto;
3137 if (unlikely(unlisted)) {
3139 po->prot_hook.dev = NULL;
3141 packet_cached_dev_reset(po);
3143 po->prot_hook.dev = dev;
3144 po->ifindex = dev ? dev->ifindex : 0;
3145 packet_cached_dev_assign(po, dev);
3151 if (proto == 0 || !need_rehook)
3154 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3155 register_prot_hook(sk);
3157 sk->sk_err = ENETDOWN;
3158 if (!sock_flag(sk, SOCK_DEAD))
3159 sk->sk_error_report(sk);
3164 spin_unlock(&po->bind_lock);
3170 * Bind a packet socket to a device
3173 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3176 struct sock *sk = sock->sk;
3177 char name[sizeof(uaddr->sa_data) + 1];
3183 if (addr_len != sizeof(struct sockaddr))
3185 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3188 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3189 name[sizeof(uaddr->sa_data)] = 0;
3191 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3194 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3196 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3197 struct sock *sk = sock->sk;
3203 if (addr_len < sizeof(struct sockaddr_ll))
3205 if (sll->sll_family != AF_PACKET)
3208 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3209 sll->sll_protocol ? : pkt_sk(sk)->num);
3212 static struct proto packet_proto = {
3214 .owner = THIS_MODULE,
3215 .obj_size = sizeof(struct packet_sock),
3219 * Create a packet of type SOCK_PACKET.
3222 static int packet_create(struct net *net, struct socket *sock, int protocol,
3226 struct packet_sock *po;
3227 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3230 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3232 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3233 sock->type != SOCK_PACKET)
3234 return -ESOCKTNOSUPPORT;
3236 sock->state = SS_UNCONNECTED;
3239 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3243 sock->ops = &packet_ops;
3244 if (sock->type == SOCK_PACKET)
3245 sock->ops = &packet_ops_spkt;
3247 sock_init_data(sock, sk);
3250 init_completion(&po->skb_completion);
3251 sk->sk_family = PF_PACKET;
3253 po->xmit = dev_queue_xmit;
3255 err = packet_alloc_pending(po);
3259 packet_cached_dev_reset(po);
3261 sk->sk_destruct = packet_sock_destruct;
3262 sk_refcnt_debug_inc(sk);
3265 * Attach a protocol block
3268 spin_lock_init(&po->bind_lock);
3269 mutex_init(&po->pg_vec_lock);
3270 po->rollover = NULL;
3271 po->prot_hook.func = packet_rcv;
3273 if (sock->type == SOCK_PACKET)
3274 po->prot_hook.func = packet_rcv_spkt;
3276 po->prot_hook.af_packet_priv = sk;
3279 po->prot_hook.type = proto;
3280 __register_prot_hook(sk);
3283 mutex_lock(&net->packet.sklist_lock);
3284 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3285 mutex_unlock(&net->packet.sklist_lock);
3288 sock_prot_inuse_add(net, &packet_proto, 1);
3299 * Pull a packet from our receive queue and hand it to the user.
3300 * If necessary we block.
3303 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3306 struct sock *sk = sock->sk;
3307 struct sk_buff *skb;
3309 int vnet_hdr_len = 0;
3310 unsigned int origlen = 0;
3313 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3317 /* What error should we return now? EUNATTACH? */
3318 if (pkt_sk(sk)->ifindex < 0)
3322 if (flags & MSG_ERRQUEUE) {
3323 err = sock_recv_errqueue(sk, msg, len,
3324 SOL_PACKET, PACKET_TX_TIMESTAMP);
3329 * Call the generic datagram receiver. This handles all sorts
3330 * of horrible races and re-entrancy so we can forget about it
3331 * in the protocol layers.
3333 * Now it will return ENETDOWN, if device have just gone down,
3334 * but then it will block.
3337 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3340 * An error occurred so return it. Because skb_recv_datagram()
3341 * handles the blocking we don't see and worry about blocking
3348 packet_rcv_try_clear_pressure(pkt_sk(sk));
3350 if (pkt_sk(sk)->has_vnet_hdr) {
3351 err = packet_rcv_vnet(msg, skb, &len);
3354 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3357 /* You lose any data beyond the buffer you gave. If it worries
3358 * a user program they can ask the device for its MTU
3364 msg->msg_flags |= MSG_TRUNC;
3367 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3371 if (sock->type != SOCK_PACKET) {
3372 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3374 /* Original length was stored in sockaddr_ll fields */
3375 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3376 sll->sll_family = AF_PACKET;
3377 sll->sll_protocol = skb->protocol;
3380 sock_recv_ts_and_drops(msg, sk, skb);
3382 if (msg->msg_name) {
3385 /* If the address length field is there to be filled
3386 * in, we fill it in now.
3388 if (sock->type == SOCK_PACKET) {
3389 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3390 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3391 copy_len = msg->msg_namelen;
3393 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3395 msg->msg_namelen = sll->sll_halen +
3396 offsetof(struct sockaddr_ll, sll_addr);
3397 copy_len = msg->msg_namelen;
3398 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3399 memset(msg->msg_name +
3400 offsetof(struct sockaddr_ll, sll_addr),
3401 0, sizeof(sll->sll_addr));
3402 msg->msg_namelen = sizeof(struct sockaddr_ll);
3405 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3408 if (pkt_sk(sk)->auxdata) {
3409 struct tpacket_auxdata aux;
3411 aux.tp_status = TP_STATUS_USER;
3412 if (skb->ip_summed == CHECKSUM_PARTIAL)
3413 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3414 else if (skb->pkt_type != PACKET_OUTGOING &&
3415 (skb->ip_summed == CHECKSUM_COMPLETE ||
3416 skb_csum_unnecessary(skb)))
3417 aux.tp_status |= TP_STATUS_CSUM_VALID;
3419 aux.tp_len = origlen;
3420 aux.tp_snaplen = skb->len;
3422 aux.tp_net = skb_network_offset(skb);
3423 if (skb_vlan_tag_present(skb)) {
3424 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3425 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3426 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3428 aux.tp_vlan_tci = 0;
3429 aux.tp_vlan_tpid = 0;
3431 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3435 * Free or return the buffer as appropriate. Again this
3436 * hides all the races and re-entrancy issues from us.
3438 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3441 skb_free_datagram(sk, skb);
3446 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3449 struct net_device *dev;
3450 struct sock *sk = sock->sk;
3455 uaddr->sa_family = AF_PACKET;
3456 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3458 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3460 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3463 return sizeof(*uaddr);
3466 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3469 struct net_device *dev;
3470 struct sock *sk = sock->sk;
3471 struct packet_sock *po = pkt_sk(sk);
3472 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3477 sll->sll_family = AF_PACKET;
3478 sll->sll_ifindex = po->ifindex;
3479 sll->sll_protocol = po->num;
3480 sll->sll_pkttype = 0;
3482 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3484 sll->sll_hatype = dev->type;
3485 sll->sll_halen = dev->addr_len;
3486 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3488 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3493 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3496 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3500 case PACKET_MR_MULTICAST:
3501 if (i->alen != dev->addr_len)
3504 return dev_mc_add(dev, i->addr);
3506 return dev_mc_del(dev, i->addr);
3508 case PACKET_MR_PROMISC:
3509 return dev_set_promiscuity(dev, what);
3510 case PACKET_MR_ALLMULTI:
3511 return dev_set_allmulti(dev, what);
3512 case PACKET_MR_UNICAST:
3513 if (i->alen != dev->addr_len)
3516 return dev_uc_add(dev, i->addr);
3518 return dev_uc_del(dev, i->addr);
3526 static void packet_dev_mclist_delete(struct net_device *dev,
3527 struct packet_mclist **mlp)
3529 struct packet_mclist *ml;
3531 while ((ml = *mlp) != NULL) {
3532 if (ml->ifindex == dev->ifindex) {
3533 packet_dev_mc(dev, ml, -1);
3541 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3543 struct packet_sock *po = pkt_sk(sk);
3544 struct packet_mclist *ml, *i;
3545 struct net_device *dev;
3551 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3556 if (mreq->mr_alen > dev->addr_len)
3560 i = kmalloc(sizeof(*i), GFP_KERNEL);
3565 for (ml = po->mclist; ml; ml = ml->next) {
3566 if (ml->ifindex == mreq->mr_ifindex &&
3567 ml->type == mreq->mr_type &&
3568 ml->alen == mreq->mr_alen &&
3569 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3571 /* Free the new element ... */
3577 i->type = mreq->mr_type;
3578 i->ifindex = mreq->mr_ifindex;
3579 i->alen = mreq->mr_alen;
3580 memcpy(i->addr, mreq->mr_address, i->alen);
3581 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3583 i->next = po->mclist;
3585 err = packet_dev_mc(dev, i, 1);
3587 po->mclist = i->next;
3596 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3598 struct packet_mclist *ml, **mlp;
3602 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3603 if (ml->ifindex == mreq->mr_ifindex &&
3604 ml->type == mreq->mr_type &&
3605 ml->alen == mreq->mr_alen &&
3606 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3607 if (--ml->count == 0) {
3608 struct net_device *dev;
3610 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3612 packet_dev_mc(dev, ml, -1);
3622 static void packet_flush_mclist(struct sock *sk)
3624 struct packet_sock *po = pkt_sk(sk);
3625 struct packet_mclist *ml;
3631 while ((ml = po->mclist) != NULL) {
3632 struct net_device *dev;
3634 po->mclist = ml->next;
3635 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3637 packet_dev_mc(dev, ml, -1);
3644 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3646 struct sock *sk = sock->sk;
3647 struct packet_sock *po = pkt_sk(sk);
3650 if (level != SOL_PACKET)
3651 return -ENOPROTOOPT;
3654 case PACKET_ADD_MEMBERSHIP:
3655 case PACKET_DROP_MEMBERSHIP:
3657 struct packet_mreq_max mreq;
3659 memset(&mreq, 0, sizeof(mreq));
3660 if (len < sizeof(struct packet_mreq))
3662 if (len > sizeof(mreq))
3664 if (copy_from_user(&mreq, optval, len))
3666 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3668 if (optname == PACKET_ADD_MEMBERSHIP)
3669 ret = packet_mc_add(sk, &mreq);
3671 ret = packet_mc_drop(sk, &mreq);
3675 case PACKET_RX_RING:
3676 case PACKET_TX_RING:
3678 union tpacket_req_u req_u;
3682 switch (po->tp_version) {
3685 len = sizeof(req_u.req);
3689 len = sizeof(req_u.req3);
3695 if (copy_from_user(&req_u.req, optval, len))
3698 ret = packet_set_ring(sk, &req_u, 0,
3699 optname == PACKET_TX_RING);
3704 case PACKET_COPY_THRESH:
3708 if (optlen != sizeof(val))
3710 if (copy_from_user(&val, optval, sizeof(val)))
3713 pkt_sk(sk)->copy_thresh = val;
3716 case PACKET_VERSION:
3720 if (optlen != sizeof(val))
3722 if (copy_from_user(&val, optval, sizeof(val)))
3733 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3736 po->tp_version = val;
3742 case PACKET_RESERVE:
3746 if (optlen != sizeof(val))
3748 if (copy_from_user(&val, optval, sizeof(val)))
3753 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3756 po->tp_reserve = val;
3766 if (optlen != sizeof(val))
3768 if (copy_from_user(&val, optval, sizeof(val)))
3772 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3775 po->tp_loss = !!val;
3781 case PACKET_AUXDATA:
3785 if (optlen < sizeof(val))
3787 if (copy_from_user(&val, optval, sizeof(val)))
3791 po->auxdata = !!val;
3795 case PACKET_ORIGDEV:
3799 if (optlen < sizeof(val))
3801 if (copy_from_user(&val, optval, sizeof(val)))
3805 po->origdev = !!val;
3809 case PACKET_VNET_HDR:
3813 if (sock->type != SOCK_RAW)
3815 if (optlen < sizeof(val))
3817 if (copy_from_user(&val, optval, sizeof(val)))
3821 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3824 po->has_vnet_hdr = !!val;
3830 case PACKET_TIMESTAMP:
3834 if (optlen != sizeof(val))
3836 if (copy_from_user(&val, optval, sizeof(val)))
3839 po->tp_tstamp = val;
3846 if (optlen != sizeof(val))
3848 if (copy_from_user(&val, optval, sizeof(val)))
3851 return fanout_add(sk, val & 0xffff, val >> 16);
3853 case PACKET_FANOUT_DATA:
3858 return fanout_set_data(po, optval, optlen);
3860 case PACKET_IGNORE_OUTGOING:
3864 if (optlen != sizeof(val))
3866 if (copy_from_user(&val, optval, sizeof(val)))
3868 if (val < 0 || val > 1)
3871 po->prot_hook.ignore_outgoing = !!val;
3874 case PACKET_TX_HAS_OFF:
3878 if (optlen != sizeof(val))
3880 if (copy_from_user(&val, optval, sizeof(val)))
3884 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3887 po->tp_tx_has_off = !!val;
3893 case PACKET_QDISC_BYPASS:
3897 if (optlen != sizeof(val))
3899 if (copy_from_user(&val, optval, sizeof(val)))
3902 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3906 return -ENOPROTOOPT;
3910 static int packet_getsockopt(struct socket *sock, int level, int optname,
3911 char __user *optval, int __user *optlen)
3914 int val, lv = sizeof(val);
3915 struct sock *sk = sock->sk;
3916 struct packet_sock *po = pkt_sk(sk);
3918 union tpacket_stats_u st;
3919 struct tpacket_rollover_stats rstats;
3922 if (level != SOL_PACKET)
3923 return -ENOPROTOOPT;
3925 if (get_user(len, optlen))
3932 case PACKET_STATISTICS:
3933 spin_lock_bh(&sk->sk_receive_queue.lock);
3934 memcpy(&st, &po->stats, sizeof(st));
3935 memset(&po->stats, 0, sizeof(po->stats));
3936 spin_unlock_bh(&sk->sk_receive_queue.lock);
3937 drops = atomic_xchg(&po->tp_drops, 0);
3939 if (po->tp_version == TPACKET_V3) {
3940 lv = sizeof(struct tpacket_stats_v3);
3941 st.stats3.tp_drops = drops;
3942 st.stats3.tp_packets += drops;
3945 lv = sizeof(struct tpacket_stats);
3946 st.stats1.tp_drops = drops;
3947 st.stats1.tp_packets += drops;
3952 case PACKET_AUXDATA:
3955 case PACKET_ORIGDEV:
3958 case PACKET_VNET_HDR:
3959 val = po->has_vnet_hdr;
3961 case PACKET_VERSION:
3962 val = po->tp_version;
3965 if (len > sizeof(int))
3967 if (len < sizeof(int))
3969 if (copy_from_user(&val, optval, len))
3973 val = sizeof(struct tpacket_hdr);
3976 val = sizeof(struct tpacket2_hdr);
3979 val = sizeof(struct tpacket3_hdr);
3985 case PACKET_RESERVE:
3986 val = po->tp_reserve;
3991 case PACKET_TIMESTAMP:
3992 val = po->tp_tstamp;
3996 ((u32)po->fanout->id |
3997 ((u32)po->fanout->type << 16) |
3998 ((u32)po->fanout->flags << 24)) :
4001 case PACKET_IGNORE_OUTGOING:
4002 val = po->prot_hook.ignore_outgoing;
4004 case PACKET_ROLLOVER_STATS:
4007 rstats.tp_all = atomic_long_read(&po->rollover->num);
4008 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4009 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4011 lv = sizeof(rstats);
4013 case PACKET_TX_HAS_OFF:
4014 val = po->tp_tx_has_off;
4016 case PACKET_QDISC_BYPASS:
4017 val = packet_use_direct_xmit(po);
4020 return -ENOPROTOOPT;
4025 if (put_user(len, optlen))
4027 if (copy_to_user(optval, data, len))
4033 #ifdef CONFIG_COMPAT
4034 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4035 char __user *optval, unsigned int optlen)
4037 struct packet_sock *po = pkt_sk(sock->sk);
4039 if (level != SOL_PACKET)
4040 return -ENOPROTOOPT;
4042 if (optname == PACKET_FANOUT_DATA &&
4043 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4044 optval = (char __user *)get_compat_bpf_fprog(optval);
4047 optlen = sizeof(struct sock_fprog);
4050 return packet_setsockopt(sock, level, optname, optval, optlen);
4054 static int packet_notifier(struct notifier_block *this,
4055 unsigned long msg, void *ptr)
4058 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4059 struct net *net = dev_net(dev);
4062 sk_for_each_rcu(sk, &net->packet.sklist) {
4063 struct packet_sock *po = pkt_sk(sk);
4066 case NETDEV_UNREGISTER:
4068 packet_dev_mclist_delete(dev, &po->mclist);
4072 if (dev->ifindex == po->ifindex) {
4073 spin_lock(&po->bind_lock);
4075 __unregister_prot_hook(sk, false);
4076 sk->sk_err = ENETDOWN;
4077 if (!sock_flag(sk, SOCK_DEAD))
4078 sk->sk_error_report(sk);
4080 if (msg == NETDEV_UNREGISTER) {
4081 packet_cached_dev_reset(po);
4083 if (po->prot_hook.dev)
4084 dev_put(po->prot_hook.dev);
4085 po->prot_hook.dev = NULL;
4087 spin_unlock(&po->bind_lock);
4091 if (dev->ifindex == po->ifindex) {
4092 spin_lock(&po->bind_lock);
4094 register_prot_hook(sk);
4095 spin_unlock(&po->bind_lock);
4105 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4108 struct sock *sk = sock->sk;
4113 int amount = sk_wmem_alloc_get(sk);
4115 return put_user(amount, (int __user *)arg);
4119 struct sk_buff *skb;
4122 spin_lock_bh(&sk->sk_receive_queue.lock);
4123 skb = skb_peek(&sk->sk_receive_queue);
4126 spin_unlock_bh(&sk->sk_receive_queue.lock);
4127 return put_user(amount, (int __user *)arg);
4137 case SIOCGIFBRDADDR:
4138 case SIOCSIFBRDADDR:
4139 case SIOCGIFNETMASK:
4140 case SIOCSIFNETMASK:
4141 case SIOCGIFDSTADDR:
4142 case SIOCSIFDSTADDR:
4144 return inet_dgram_ops.ioctl(sock, cmd, arg);
4148 return -ENOIOCTLCMD;
4153 static __poll_t packet_poll(struct file *file, struct socket *sock,
4156 struct sock *sk = sock->sk;
4157 struct packet_sock *po = pkt_sk(sk);
4158 __poll_t mask = datagram_poll(file, sock, wait);
4160 spin_lock_bh(&sk->sk_receive_queue.lock);
4161 if (po->rx_ring.pg_vec) {
4162 if (!packet_previous_rx_frame(po, &po->rx_ring,
4164 mask |= EPOLLIN | EPOLLRDNORM;
4166 packet_rcv_try_clear_pressure(po);
4167 spin_unlock_bh(&sk->sk_receive_queue.lock);
4168 spin_lock_bh(&sk->sk_write_queue.lock);
4169 if (po->tx_ring.pg_vec) {
4170 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4171 mask |= EPOLLOUT | EPOLLWRNORM;
4173 spin_unlock_bh(&sk->sk_write_queue.lock);
4178 /* Dirty? Well, I still did not learn better way to account
4182 static void packet_mm_open(struct vm_area_struct *vma)
4184 struct file *file = vma->vm_file;
4185 struct socket *sock = file->private_data;
4186 struct sock *sk = sock->sk;
4189 atomic_inc(&pkt_sk(sk)->mapped);
4192 static void packet_mm_close(struct vm_area_struct *vma)
4194 struct file *file = vma->vm_file;
4195 struct socket *sock = file->private_data;
4196 struct sock *sk = sock->sk;
4199 atomic_dec(&pkt_sk(sk)->mapped);
4202 static const struct vm_operations_struct packet_mmap_ops = {
4203 .open = packet_mm_open,
4204 .close = packet_mm_close,
4207 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4212 for (i = 0; i < len; i++) {
4213 if (likely(pg_vec[i].buffer)) {
4214 if (is_vmalloc_addr(pg_vec[i].buffer))
4215 vfree(pg_vec[i].buffer);
4217 free_pages((unsigned long)pg_vec[i].buffer,
4219 pg_vec[i].buffer = NULL;
4225 static char *alloc_one_pg_vec_page(unsigned long order)
4228 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4229 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4231 buffer = (char *) __get_free_pages(gfp_flags, order);
4235 /* __get_free_pages failed, fall back to vmalloc */
4236 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4240 /* vmalloc failed, lets dig into swap here */
4241 gfp_flags &= ~__GFP_NORETRY;
4242 buffer = (char *) __get_free_pages(gfp_flags, order);
4246 /* complete and utter failure */
4250 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4252 unsigned int block_nr = req->tp_block_nr;
4256 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4257 if (unlikely(!pg_vec))
4260 for (i = 0; i < block_nr; i++) {
4261 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4262 if (unlikely(!pg_vec[i].buffer))
4263 goto out_free_pgvec;
4270 free_pg_vec(pg_vec, order, block_nr);
4275 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4276 int closing, int tx_ring)
4278 struct pgv *pg_vec = NULL;
4279 struct packet_sock *po = pkt_sk(sk);
4280 int was_running, order = 0;
4281 struct packet_ring_buffer *rb;
4282 struct sk_buff_head *rb_queue;
4285 /* Added to avoid minimal code churn */
4286 struct tpacket_req *req = &req_u->req;
4288 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4289 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4293 if (atomic_read(&po->mapped))
4295 if (packet_read_pending(rb))
4299 if (req->tp_block_nr) {
4300 unsigned int min_frame_size;
4302 /* Sanity tests and some calculations */
4304 if (unlikely(rb->pg_vec))
4307 switch (po->tp_version) {
4309 po->tp_hdrlen = TPACKET_HDRLEN;
4312 po->tp_hdrlen = TPACKET2_HDRLEN;
4315 po->tp_hdrlen = TPACKET3_HDRLEN;
4320 if (unlikely((int)req->tp_block_size <= 0))
4322 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4324 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4325 if (po->tp_version >= TPACKET_V3 &&
4326 req->tp_block_size <
4327 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4329 if (unlikely(req->tp_frame_size < min_frame_size))
4331 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4334 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4335 if (unlikely(rb->frames_per_block == 0))
4337 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4339 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4344 order = get_order(req->tp_block_size);
4345 pg_vec = alloc_pg_vec(req, order);
4346 if (unlikely(!pg_vec))
4348 switch (po->tp_version) {
4350 /* Block transmit is not supported yet */
4352 init_prb_bdqc(po, rb, pg_vec, req_u);
4354 struct tpacket_req3 *req3 = &req_u->req3;
4356 if (req3->tp_retire_blk_tov ||
4357 req3->tp_sizeof_priv ||
4358 req3->tp_feature_req_word) {
4360 goto out_free_pg_vec;
4371 if (unlikely(req->tp_frame_nr))
4376 /* Detach socket from network */
4377 spin_lock(&po->bind_lock);
4378 was_running = po->running;
4382 __unregister_prot_hook(sk, false);
4384 spin_unlock(&po->bind_lock);
4389 mutex_lock(&po->pg_vec_lock);
4390 if (closing || atomic_read(&po->mapped) == 0) {
4392 spin_lock_bh(&rb_queue->lock);
4393 swap(rb->pg_vec, pg_vec);
4394 rb->frame_max = (req->tp_frame_nr - 1);
4396 rb->frame_size = req->tp_frame_size;
4397 spin_unlock_bh(&rb_queue->lock);
4399 swap(rb->pg_vec_order, order);
4400 swap(rb->pg_vec_len, req->tp_block_nr);
4402 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4403 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4404 tpacket_rcv : packet_rcv;
4405 skb_queue_purge(rb_queue);
4406 if (atomic_read(&po->mapped))
4407 pr_err("packet_mmap: vma is busy: %d\n",
4408 atomic_read(&po->mapped));
4410 mutex_unlock(&po->pg_vec_lock);
4412 spin_lock(&po->bind_lock);
4415 register_prot_hook(sk);
4417 spin_unlock(&po->bind_lock);
4418 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4419 /* Because we don't support block-based V3 on tx-ring */
4421 prb_shutdown_retire_blk_timer(po, rb_queue);
4426 free_pg_vec(pg_vec, order, req->tp_block_nr);
4431 static int packet_mmap(struct file *file, struct socket *sock,
4432 struct vm_area_struct *vma)
4434 struct sock *sk = sock->sk;
4435 struct packet_sock *po = pkt_sk(sk);
4436 unsigned long size, expected_size;
4437 struct packet_ring_buffer *rb;
4438 unsigned long start;
4445 mutex_lock(&po->pg_vec_lock);
4448 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4450 expected_size += rb->pg_vec_len
4456 if (expected_size == 0)
4459 size = vma->vm_end - vma->vm_start;
4460 if (size != expected_size)
4463 start = vma->vm_start;
4464 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4465 if (rb->pg_vec == NULL)
4468 for (i = 0; i < rb->pg_vec_len; i++) {
4470 void *kaddr = rb->pg_vec[i].buffer;
4473 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4474 page = pgv_to_page(kaddr);
4475 err = vm_insert_page(vma, start, page);
4484 atomic_inc(&po->mapped);
4485 vma->vm_ops = &packet_mmap_ops;
4489 mutex_unlock(&po->pg_vec_lock);
4493 static const struct proto_ops packet_ops_spkt = {
4494 .family = PF_PACKET,
4495 .owner = THIS_MODULE,
4496 .release = packet_release,
4497 .bind = packet_bind_spkt,
4498 .connect = sock_no_connect,
4499 .socketpair = sock_no_socketpair,
4500 .accept = sock_no_accept,
4501 .getname = packet_getname_spkt,
4502 .poll = datagram_poll,
4503 .ioctl = packet_ioctl,
4504 .gettstamp = sock_gettstamp,
4505 .listen = sock_no_listen,
4506 .shutdown = sock_no_shutdown,
4507 .setsockopt = sock_no_setsockopt,
4508 .getsockopt = sock_no_getsockopt,
4509 .sendmsg = packet_sendmsg_spkt,
4510 .recvmsg = packet_recvmsg,
4511 .mmap = sock_no_mmap,
4512 .sendpage = sock_no_sendpage,
4515 static const struct proto_ops packet_ops = {
4516 .family = PF_PACKET,
4517 .owner = THIS_MODULE,
4518 .release = packet_release,
4519 .bind = packet_bind,
4520 .connect = sock_no_connect,
4521 .socketpair = sock_no_socketpair,
4522 .accept = sock_no_accept,
4523 .getname = packet_getname,
4524 .poll = packet_poll,
4525 .ioctl = packet_ioctl,
4526 .gettstamp = sock_gettstamp,
4527 .listen = sock_no_listen,
4528 .shutdown = sock_no_shutdown,
4529 .setsockopt = packet_setsockopt,
4530 .getsockopt = packet_getsockopt,
4531 #ifdef CONFIG_COMPAT
4532 .compat_setsockopt = compat_packet_setsockopt,
4534 .sendmsg = packet_sendmsg,
4535 .recvmsg = packet_recvmsg,
4536 .mmap = packet_mmap,
4537 .sendpage = sock_no_sendpage,
4540 static const struct net_proto_family packet_family_ops = {
4541 .family = PF_PACKET,
4542 .create = packet_create,
4543 .owner = THIS_MODULE,
4546 static struct notifier_block packet_netdev_notifier = {
4547 .notifier_call = packet_notifier,
4550 #ifdef CONFIG_PROC_FS
4552 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4555 struct net *net = seq_file_net(seq);
4558 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4561 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4563 struct net *net = seq_file_net(seq);
4564 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4567 static void packet_seq_stop(struct seq_file *seq, void *v)
4573 static int packet_seq_show(struct seq_file *seq, void *v)
4575 if (v == SEQ_START_TOKEN)
4576 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4578 struct sock *s = sk_entry(v);
4579 const struct packet_sock *po = pkt_sk(s);
4582 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4584 refcount_read(&s->sk_refcnt),
4589 atomic_read(&s->sk_rmem_alloc),
4590 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4597 static const struct seq_operations packet_seq_ops = {
4598 .start = packet_seq_start,
4599 .next = packet_seq_next,
4600 .stop = packet_seq_stop,
4601 .show = packet_seq_show,
4605 static int __net_init packet_net_init(struct net *net)
4607 mutex_init(&net->packet.sklist_lock);
4608 INIT_HLIST_HEAD(&net->packet.sklist);
4610 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4611 sizeof(struct seq_net_private)))
4617 static void __net_exit packet_net_exit(struct net *net)
4619 remove_proc_entry("packet", net->proc_net);
4620 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4623 static struct pernet_operations packet_net_ops = {
4624 .init = packet_net_init,
4625 .exit = packet_net_exit,
4629 static void __exit packet_exit(void)
4631 unregister_netdevice_notifier(&packet_netdev_notifier);
4632 unregister_pernet_subsys(&packet_net_ops);
4633 sock_unregister(PF_PACKET);
4634 proto_unregister(&packet_proto);
4637 static int __init packet_init(void)
4641 rc = proto_register(&packet_proto, 0);
4644 rc = sock_register(&packet_family_ops);
4647 rc = register_pernet_subsys(&packet_net_ops);
4650 rc = register_netdevice_notifier(&packet_netdev_notifier);
4657 unregister_pernet_subsys(&packet_net_ops);
4659 sock_unregister(PF_PACKET);
4661 proto_unregister(&packet_proto);
4666 module_init(packet_init);
4667 module_exit(packet_exit);
4668 MODULE_LICENSE("GPL");
4669 MODULE_ALIAS_NETPROTO(PF_PACKET);