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;
596 rwlock_init(&p1->blk_fill_in_prog_lock);
598 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
599 prb_init_ft_ops(p1, req_u);
600 prb_setup_retire_blk_timer(po);
601 prb_open_block(p1, pbd);
604 /* Do NOT update the last_blk_num first.
605 * Assumes sk_buff_head lock is held.
607 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
609 mod_timer(&pkc->retire_blk_timer,
610 jiffies + pkc->tov_in_jiffies);
611 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
616 * 1) We refresh the timer only when we open a block.
617 * By doing this we don't waste cycles refreshing the timer
618 * on packet-by-packet basis.
620 * With a 1MB block-size, on a 1Gbps line, it will take
621 * i) ~8 ms to fill a block + ii) memcpy etc.
622 * In this cut we are not accounting for the memcpy time.
624 * So, if the user sets the 'tmo' to 10ms then the timer
625 * will never fire while the block is still getting filled
626 * (which is what we want). However, the user could choose
627 * to close a block early and that's fine.
629 * But when the timer does fire, we check whether or not to refresh it.
630 * Since the tmo granularity is in msecs, it is not too expensive
631 * to refresh the timer, lets say every '8' msecs.
632 * Either the user can set the 'tmo' or we can derive it based on
633 * a) line-speed and b) block-size.
634 * prb_calc_retire_blk_tmo() calculates the tmo.
637 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
639 struct packet_sock *po =
640 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
641 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
643 struct tpacket_block_desc *pbd;
645 spin_lock(&po->sk.sk_receive_queue.lock);
647 frozen = prb_queue_frozen(pkc);
648 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
650 if (unlikely(pkc->delete_blk_timer))
653 /* We only need to plug the race when the block is partially filled.
655 * lock(); increment BLOCK_NUM_PKTS; unlock()
656 * copy_bits() is in progress ...
657 * timer fires on other cpu:
658 * we can't retire the current block because copy_bits
662 if (BLOCK_NUM_PKTS(pbd)) {
663 /* Waiting for skb_copy_bits to finish... */
664 write_lock(&pkc->blk_fill_in_prog_lock);
665 write_unlock(&pkc->blk_fill_in_prog_lock);
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 /* Waiting for skb_copy_bits to finish... */
925 write_lock(&pkc->blk_fill_in_prog_lock);
926 write_unlock(&pkc->blk_fill_in_prog_lock);
928 prb_close_block(pkc, pbd, po, status);
933 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
935 return TP_STATUS_USER & BLOCK_STATUS(pbd);
938 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
940 return pkc->reset_pending_on_curr_blk;
943 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
944 __releases(&pkc->blk_fill_in_prog_lock)
946 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
948 read_unlock(&pkc->blk_fill_in_prog_lock);
951 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
952 struct tpacket3_hdr *ppd)
954 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
957 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
958 struct tpacket3_hdr *ppd)
960 ppd->hv1.tp_rxhash = 0;
963 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
964 struct tpacket3_hdr *ppd)
966 if (skb_vlan_tag_present(pkc->skb)) {
967 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
968 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
969 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
971 ppd->hv1.tp_vlan_tci = 0;
972 ppd->hv1.tp_vlan_tpid = 0;
973 ppd->tp_status = TP_STATUS_AVAILABLE;
977 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
978 struct tpacket3_hdr *ppd)
980 ppd->hv1.tp_padding = 0;
981 prb_fill_vlan_info(pkc, ppd);
983 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
984 prb_fill_rxhash(pkc, ppd);
986 prb_clear_rxhash(pkc, ppd);
989 static void prb_fill_curr_block(char *curr,
990 struct tpacket_kbdq_core *pkc,
991 struct tpacket_block_desc *pbd,
993 __acquires(&pkc->blk_fill_in_prog_lock)
995 struct tpacket3_hdr *ppd;
997 ppd = (struct tpacket3_hdr *)curr;
998 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1000 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002 BLOCK_NUM_PKTS(pbd) += 1;
1003 read_lock(&pkc->blk_fill_in_prog_lock);
1004 prb_run_all_ft_ops(pkc, ppd);
1007 /* Assumes caller has the sk->rx_queue.lock */
1008 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009 struct sk_buff *skb,
1013 struct tpacket_kbdq_core *pkc;
1014 struct tpacket_block_desc *pbd;
1017 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1018 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1020 /* Queue is frozen when user space is lagging behind */
1021 if (prb_queue_frozen(pkc)) {
1023 * Check if that last block which caused the queue to freeze,
1024 * is still in_use by user-space.
1026 if (prb_curr_blk_in_use(pbd)) {
1027 /* Can't record this packet */
1031 * Ok, the block was released by user-space.
1032 * Now let's open that block.
1033 * opening a block also thaws the queue.
1034 * Thawing is a side effect.
1036 prb_open_block(pkc, pbd);
1041 curr = pkc->nxt_offset;
1043 end = (char *)pbd + pkc->kblk_size;
1045 /* first try the current block */
1046 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1047 prb_fill_curr_block(curr, pkc, pbd, len);
1048 return (void *)curr;
1051 /* Ok, close the current block */
1052 prb_retire_current_block(pkc, po, 0);
1054 /* Now, try to dispatch the next block */
1055 curr = (char *)prb_dispatch_next_block(pkc, po);
1057 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058 prb_fill_curr_block(curr, pkc, pbd, len);
1059 return (void *)curr;
1063 * No free blocks are available.user_space hasn't caught up yet.
1064 * Queue was just frozen and now this packet will get dropped.
1069 static void *packet_current_rx_frame(struct packet_sock *po,
1070 struct sk_buff *skb,
1071 int status, unsigned int len)
1074 switch (po->tp_version) {
1077 curr = packet_lookup_frame(po, &po->rx_ring,
1078 po->rx_ring.head, status);
1081 return __packet_lookup_frame_in_block(po, skb, len);
1083 WARN(1, "TPACKET version not supported\n");
1089 static void *prb_lookup_block(const struct packet_sock *po,
1090 const struct packet_ring_buffer *rb,
1094 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1095 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1097 if (status != BLOCK_STATUS(pbd))
1102 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1105 if (rb->prb_bdqc.kactive_blk_num)
1106 prev = rb->prb_bdqc.kactive_blk_num-1;
1108 prev = rb->prb_bdqc.knum_blocks-1;
1112 /* Assumes caller has held the rx_queue.lock */
1113 static void *__prb_previous_block(struct packet_sock *po,
1114 struct packet_ring_buffer *rb,
1117 unsigned int previous = prb_previous_blk_num(rb);
1118 return prb_lookup_block(po, rb, previous, status);
1121 static void *packet_previous_rx_frame(struct packet_sock *po,
1122 struct packet_ring_buffer *rb,
1125 if (po->tp_version <= TPACKET_V2)
1126 return packet_previous_frame(po, rb, status);
1128 return __prb_previous_block(po, rb, status);
1131 static void packet_increment_rx_head(struct packet_sock *po,
1132 struct packet_ring_buffer *rb)
1134 switch (po->tp_version) {
1137 return packet_increment_head(rb);
1140 WARN(1, "TPACKET version not supported.\n");
1146 static void *packet_previous_frame(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1150 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1151 return packet_lookup_frame(po, rb, previous, status);
1154 static void packet_increment_head(struct packet_ring_buffer *buff)
1156 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1159 static void packet_inc_pending(struct packet_ring_buffer *rb)
1161 this_cpu_inc(*rb->pending_refcnt);
1164 static void packet_dec_pending(struct packet_ring_buffer *rb)
1166 this_cpu_dec(*rb->pending_refcnt);
1169 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1171 unsigned int refcnt = 0;
1174 /* We don't use pending refcount in rx_ring. */
1175 if (rb->pending_refcnt == NULL)
1178 for_each_possible_cpu(cpu)
1179 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1184 static int packet_alloc_pending(struct packet_sock *po)
1186 po->rx_ring.pending_refcnt = NULL;
1188 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1189 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1195 static void packet_free_pending(struct packet_sock *po)
1197 free_percpu(po->tx_ring.pending_refcnt);
1200 #define ROOM_POW_OFF 2
1201 #define ROOM_NONE 0x0
1202 #define ROOM_LOW 0x1
1203 #define ROOM_NORMAL 0x2
1205 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1209 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1210 idx = READ_ONCE(po->rx_ring.head);
1212 idx += len >> pow_off;
1215 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1218 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1222 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1223 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1225 idx += len >> pow_off;
1228 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1231 static int __packet_rcv_has_room(const struct packet_sock *po,
1232 const struct sk_buff *skb)
1234 const struct sock *sk = &po->sk;
1235 int ret = ROOM_NONE;
1237 if (po->prot_hook.func != tpacket_rcv) {
1238 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1239 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1240 - (skb ? skb->truesize : 0);
1242 if (avail > (rcvbuf >> ROOM_POW_OFF))
1250 if (po->tp_version == TPACKET_V3) {
1251 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1253 else if (__tpacket_v3_has_room(po, 0))
1256 if (__tpacket_has_room(po, ROOM_POW_OFF))
1258 else if (__tpacket_has_room(po, 0))
1265 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1269 ret = __packet_rcv_has_room(po, skb);
1270 pressure = ret != ROOM_NORMAL;
1272 if (READ_ONCE(po->pressure) != pressure)
1273 WRITE_ONCE(po->pressure, pressure);
1278 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1280 if (READ_ONCE(po->pressure) &&
1281 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1282 WRITE_ONCE(po->pressure, 0);
1285 static void packet_sock_destruct(struct sock *sk)
1287 skb_queue_purge(&sk->sk_error_queue);
1289 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1290 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1292 if (!sock_flag(sk, SOCK_DEAD)) {
1293 pr_err("Attempt to release alive packet socket: %p\n", sk);
1297 sk_refcnt_debug_dec(sk);
1300 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1302 u32 *history = po->rollover->history;
1306 rxhash = skb_get_hash(skb);
1307 for (i = 0; i < ROLLOVER_HLEN; i++)
1308 if (READ_ONCE(history[i]) == rxhash)
1311 victim = prandom_u32() % ROLLOVER_HLEN;
1313 /* Avoid dirtying the cache line if possible */
1314 if (READ_ONCE(history[victim]) != rxhash)
1315 WRITE_ONCE(history[victim], rxhash);
1317 return count > (ROLLOVER_HLEN >> 1);
1320 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1321 struct sk_buff *skb,
1324 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1327 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1328 struct sk_buff *skb,
1331 unsigned int val = atomic_inc_return(&f->rr_cur);
1336 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1337 struct sk_buff *skb,
1340 return smp_processor_id() % num;
1343 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1344 struct sk_buff *skb,
1347 return prandom_u32_max(num);
1350 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1351 struct sk_buff *skb,
1352 unsigned int idx, bool try_self,
1355 struct packet_sock *po, *po_next, *po_skip = NULL;
1356 unsigned int i, j, room = ROOM_NONE;
1358 po = pkt_sk(f->arr[idx]);
1361 room = packet_rcv_has_room(po, skb);
1362 if (room == ROOM_NORMAL ||
1363 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1368 i = j = min_t(int, po->rollover->sock, num - 1);
1370 po_next = pkt_sk(f->arr[i]);
1371 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1372 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1374 po->rollover->sock = i;
1375 atomic_long_inc(&po->rollover->num);
1376 if (room == ROOM_LOW)
1377 atomic_long_inc(&po->rollover->num_huge);
1385 atomic_long_inc(&po->rollover->num_failed);
1389 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1390 struct sk_buff *skb,
1393 return skb_get_queue_mapping(skb) % num;
1396 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1397 struct sk_buff *skb,
1400 struct bpf_prog *prog;
1401 unsigned int ret = 0;
1404 prog = rcu_dereference(f->bpf_prog);
1406 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1412 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1414 return f->flags & (flag >> 8);
1417 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1418 struct packet_type *pt, struct net_device *orig_dev)
1420 struct packet_fanout *f = pt->af_packet_priv;
1421 unsigned int num = READ_ONCE(f->num_members);
1422 struct net *net = read_pnet(&f->net);
1423 struct packet_sock *po;
1426 if (!net_eq(dev_net(dev), net) || !num) {
1431 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1432 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1437 case PACKET_FANOUT_HASH:
1439 idx = fanout_demux_hash(f, skb, num);
1441 case PACKET_FANOUT_LB:
1442 idx = fanout_demux_lb(f, skb, num);
1444 case PACKET_FANOUT_CPU:
1445 idx = fanout_demux_cpu(f, skb, num);
1447 case PACKET_FANOUT_RND:
1448 idx = fanout_demux_rnd(f, skb, num);
1450 case PACKET_FANOUT_QM:
1451 idx = fanout_demux_qm(f, skb, num);
1453 case PACKET_FANOUT_ROLLOVER:
1454 idx = fanout_demux_rollover(f, skb, 0, false, num);
1456 case PACKET_FANOUT_CBPF:
1457 case PACKET_FANOUT_EBPF:
1458 idx = fanout_demux_bpf(f, skb, num);
1462 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1463 idx = fanout_demux_rollover(f, skb, idx, true, num);
1465 po = pkt_sk(f->arr[idx]);
1466 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1469 DEFINE_MUTEX(fanout_mutex);
1470 EXPORT_SYMBOL_GPL(fanout_mutex);
1471 static LIST_HEAD(fanout_list);
1472 static u16 fanout_next_id;
1474 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1476 struct packet_fanout *f = po->fanout;
1478 spin_lock(&f->lock);
1479 f->arr[f->num_members] = sk;
1482 if (f->num_members == 1)
1483 dev_add_pack(&f->prot_hook);
1484 spin_unlock(&f->lock);
1487 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1489 struct packet_fanout *f = po->fanout;
1492 spin_lock(&f->lock);
1493 for (i = 0; i < f->num_members; i++) {
1494 if (f->arr[i] == sk)
1497 BUG_ON(i >= f->num_members);
1498 f->arr[i] = f->arr[f->num_members - 1];
1500 if (f->num_members == 0)
1501 __dev_remove_pack(&f->prot_hook);
1502 spin_unlock(&f->lock);
1505 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1507 if (sk->sk_family != PF_PACKET)
1510 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1513 static void fanout_init_data(struct packet_fanout *f)
1516 case PACKET_FANOUT_LB:
1517 atomic_set(&f->rr_cur, 0);
1519 case PACKET_FANOUT_CBPF:
1520 case PACKET_FANOUT_EBPF:
1521 RCU_INIT_POINTER(f->bpf_prog, NULL);
1526 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1528 struct bpf_prog *old;
1530 spin_lock(&f->lock);
1531 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1532 rcu_assign_pointer(f->bpf_prog, new);
1533 spin_unlock(&f->lock);
1537 bpf_prog_destroy(old);
1541 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1544 struct bpf_prog *new;
1545 struct sock_fprog fprog;
1548 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1551 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1555 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1559 __fanout_set_data_bpf(po->fanout, new);
1563 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1566 struct bpf_prog *new;
1569 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1571 if (len != sizeof(fd))
1573 if (copy_from_sockptr(&fd, data, len))
1576 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1578 return PTR_ERR(new);
1580 __fanout_set_data_bpf(po->fanout, new);
1584 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1587 switch (po->fanout->type) {
1588 case PACKET_FANOUT_CBPF:
1589 return fanout_set_data_cbpf(po, data, len);
1590 case PACKET_FANOUT_EBPF:
1591 return fanout_set_data_ebpf(po, data, len);
1597 static void fanout_release_data(struct packet_fanout *f)
1600 case PACKET_FANOUT_CBPF:
1601 case PACKET_FANOUT_EBPF:
1602 __fanout_set_data_bpf(f, NULL);
1606 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1608 struct packet_fanout *f;
1610 list_for_each_entry(f, &fanout_list, list) {
1611 if (f->id == candidate_id &&
1612 read_pnet(&f->net) == sock_net(sk)) {
1619 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1621 u16 id = fanout_next_id;
1624 if (__fanout_id_is_free(sk, id)) {
1626 fanout_next_id = id + 1;
1631 } while (id != fanout_next_id);
1636 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1638 struct packet_rollover *rollover = NULL;
1639 struct packet_sock *po = pkt_sk(sk);
1640 struct packet_fanout *f, *match;
1641 u8 type = type_flags & 0xff;
1642 u8 flags = type_flags >> 8;
1646 case PACKET_FANOUT_ROLLOVER:
1647 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1649 case PACKET_FANOUT_HASH:
1650 case PACKET_FANOUT_LB:
1651 case PACKET_FANOUT_CPU:
1652 case PACKET_FANOUT_RND:
1653 case PACKET_FANOUT_QM:
1654 case PACKET_FANOUT_CBPF:
1655 case PACKET_FANOUT_EBPF:
1661 mutex_lock(&fanout_mutex);
1667 if (type == PACKET_FANOUT_ROLLOVER ||
1668 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1670 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1673 atomic_long_set(&rollover->num, 0);
1674 atomic_long_set(&rollover->num_huge, 0);
1675 atomic_long_set(&rollover->num_failed, 0);
1678 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1683 if (!fanout_find_new_id(sk, &id)) {
1687 /* ephemeral flag for the first socket in the group: drop it */
1688 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1692 list_for_each_entry(f, &fanout_list, list) {
1694 read_pnet(&f->net) == sock_net(sk)) {
1700 if (match && match->flags != flags)
1704 match = kzalloc(sizeof(*match), GFP_KERNEL);
1707 write_pnet(&match->net, sock_net(sk));
1710 match->flags = flags;
1711 INIT_LIST_HEAD(&match->list);
1712 spin_lock_init(&match->lock);
1713 refcount_set(&match->sk_ref, 0);
1714 fanout_init_data(match);
1715 match->prot_hook.type = po->prot_hook.type;
1716 match->prot_hook.dev = po->prot_hook.dev;
1717 match->prot_hook.func = packet_rcv_fanout;
1718 match->prot_hook.af_packet_priv = match;
1719 match->prot_hook.id_match = match_fanout_group;
1720 list_add(&match->list, &fanout_list);
1724 spin_lock(&po->bind_lock);
1726 match->type == type &&
1727 match->prot_hook.type == po->prot_hook.type &&
1728 match->prot_hook.dev == po->prot_hook.dev) {
1730 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1731 __dev_remove_pack(&po->prot_hook);
1733 po->rollover = rollover;
1735 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1736 __fanout_link(sk, po);
1740 spin_unlock(&po->bind_lock);
1742 if (err && !refcount_read(&match->sk_ref)) {
1743 list_del(&match->list);
1749 mutex_unlock(&fanout_mutex);
1753 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1754 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1755 * It is the responsibility of the caller to call fanout_release_data() and
1756 * free the returned packet_fanout (after synchronize_net())
1758 static struct packet_fanout *fanout_release(struct sock *sk)
1760 struct packet_sock *po = pkt_sk(sk);
1761 struct packet_fanout *f;
1763 mutex_lock(&fanout_mutex);
1768 if (refcount_dec_and_test(&f->sk_ref))
1773 mutex_unlock(&fanout_mutex);
1778 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1779 struct sk_buff *skb)
1781 /* Earlier code assumed this would be a VLAN pkt, double-check
1782 * this now that we have the actual packet in hand. We can only
1783 * do this check on Ethernet devices.
1785 if (unlikely(dev->type != ARPHRD_ETHER))
1788 skb_reset_mac_header(skb);
1789 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1792 static const struct proto_ops packet_ops;
1794 static const struct proto_ops packet_ops_spkt;
1796 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1797 struct packet_type *pt, struct net_device *orig_dev)
1800 struct sockaddr_pkt *spkt;
1803 * When we registered the protocol we saved the socket in the data
1804 * field for just this event.
1807 sk = pt->af_packet_priv;
1810 * Yank back the headers [hope the device set this
1811 * right or kerboom...]
1813 * Incoming packets have ll header pulled,
1816 * For outgoing ones skb->data == skb_mac_header(skb)
1817 * so that this procedure is noop.
1820 if (skb->pkt_type == PACKET_LOOPBACK)
1823 if (!net_eq(dev_net(dev), sock_net(sk)))
1826 skb = skb_share_check(skb, GFP_ATOMIC);
1830 /* drop any routing info */
1833 /* drop conntrack reference */
1836 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1838 skb_push(skb, skb->data - skb_mac_header(skb));
1841 * The SOCK_PACKET socket receives _all_ frames.
1844 spkt->spkt_family = dev->type;
1845 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1846 spkt->spkt_protocol = skb->protocol;
1849 * Charge the memory to the socket. This is done specifically
1850 * to prevent sockets using all the memory up.
1853 if (sock_queue_rcv_skb(sk, skb) == 0)
1862 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1864 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1865 sock->type == SOCK_RAW) {
1866 skb_reset_mac_header(skb);
1867 skb->protocol = dev_parse_header_protocol(skb);
1870 skb_probe_transport_header(skb);
1874 * Output a raw packet to a device layer. This bypasses all the other
1875 * protocol layers and you must therefore supply it with a complete frame
1878 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1881 struct sock *sk = sock->sk;
1882 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1883 struct sk_buff *skb = NULL;
1884 struct net_device *dev;
1885 struct sockcm_cookie sockc;
1891 * Get and verify the address.
1895 if (msg->msg_namelen < sizeof(struct sockaddr))
1897 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1898 proto = saddr->spkt_protocol;
1900 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1903 * Find the device first to size check it
1906 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1909 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1915 if (!(dev->flags & IFF_UP))
1919 * You may not queue a frame bigger than the mtu. This is the lowest level
1920 * raw protocol and you must do your own fragmentation at this level.
1923 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1924 if (!netif_supports_nofcs(dev)) {
1925 err = -EPROTONOSUPPORT;
1928 extra_len = 4; /* We're doing our own CRC */
1932 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1936 size_t reserved = LL_RESERVED_SPACE(dev);
1937 int tlen = dev->needed_tailroom;
1938 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1941 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1944 /* FIXME: Save some space for broken drivers that write a hard
1945 * header at transmission time by themselves. PPP is the notable
1946 * one here. This should really be fixed at the driver level.
1948 skb_reserve(skb, reserved);
1949 skb_reset_network_header(skb);
1951 /* Try to align data part correctly */
1956 skb_reset_network_header(skb);
1958 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1964 if (!dev_validate_header(dev, skb->data, len)) {
1968 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1969 !packet_extra_vlan_len_allowed(dev, skb)) {
1974 sockcm_init(&sockc, sk);
1975 if (msg->msg_controllen) {
1976 err = sock_cmsg_send(sk, msg, &sockc);
1981 skb->protocol = proto;
1983 skb->priority = sk->sk_priority;
1984 skb->mark = sk->sk_mark;
1985 skb->tstamp = sockc.transmit_time;
1987 skb_setup_tx_timestamp(skb, sockc.tsflags);
1989 if (unlikely(extra_len == 4))
1992 packet_parse_headers(skb, sock);
1994 dev_queue_xmit(skb);
2005 static unsigned int run_filter(struct sk_buff *skb,
2006 const struct sock *sk,
2009 struct sk_filter *filter;
2012 filter = rcu_dereference(sk->sk_filter);
2014 res = bpf_prog_run_clear_cb(filter->prog, skb);
2020 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2023 struct virtio_net_hdr vnet_hdr;
2025 if (*len < sizeof(vnet_hdr))
2027 *len -= sizeof(vnet_hdr);
2029 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2032 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2036 * This function makes lazy skb cloning in hope that most of packets
2037 * are discarded by BPF.
2039 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2040 * and skb->cb are mangled. It works because (and until) packets
2041 * falling here are owned by current CPU. Output packets are cloned
2042 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2043 * sequencially, so that if we return skb to original state on exit,
2044 * we will not harm anyone.
2047 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2048 struct packet_type *pt, struct net_device *orig_dev)
2051 struct sockaddr_ll *sll;
2052 struct packet_sock *po;
2053 u8 *skb_head = skb->data;
2054 int skb_len = skb->len;
2055 unsigned int snaplen, res;
2056 bool is_drop_n_account = false;
2058 if (skb->pkt_type == PACKET_LOOPBACK)
2061 sk = pt->af_packet_priv;
2064 if (!net_eq(dev_net(dev), sock_net(sk)))
2069 if (dev->header_ops) {
2070 /* The device has an explicit notion of ll header,
2071 * exported to higher levels.
2073 * Otherwise, the device hides details of its frame
2074 * structure, so that corresponding packet head is
2075 * never delivered to user.
2077 if (sk->sk_type != SOCK_DGRAM)
2078 skb_push(skb, skb->data - skb_mac_header(skb));
2079 else if (skb->pkt_type == PACKET_OUTGOING) {
2080 /* Special case: outgoing packets have ll header at head */
2081 skb_pull(skb, skb_network_offset(skb));
2087 res = run_filter(skb, sk, snaplen);
2089 goto drop_n_restore;
2093 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2096 if (skb_shared(skb)) {
2097 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2101 if (skb_head != skb->data) {
2102 skb->data = skb_head;
2109 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2111 sll = &PACKET_SKB_CB(skb)->sa.ll;
2112 sll->sll_hatype = dev->type;
2113 sll->sll_pkttype = skb->pkt_type;
2114 if (unlikely(po->origdev))
2115 sll->sll_ifindex = orig_dev->ifindex;
2117 sll->sll_ifindex = dev->ifindex;
2119 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2121 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2122 * Use their space for storing the original skb length.
2124 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2126 if (pskb_trim(skb, snaplen))
2129 skb_set_owner_r(skb, sk);
2133 /* drop conntrack reference */
2136 spin_lock(&sk->sk_receive_queue.lock);
2137 po->stats.stats1.tp_packets++;
2138 sock_skb_set_dropcount(sk, skb);
2139 __skb_queue_tail(&sk->sk_receive_queue, skb);
2140 spin_unlock(&sk->sk_receive_queue.lock);
2141 sk->sk_data_ready(sk);
2145 is_drop_n_account = true;
2146 atomic_inc(&po->tp_drops);
2147 atomic_inc(&sk->sk_drops);
2150 if (skb_head != skb->data && skb_shared(skb)) {
2151 skb->data = skb_head;
2155 if (!is_drop_n_account)
2162 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2163 struct packet_type *pt, struct net_device *orig_dev)
2166 struct packet_sock *po;
2167 struct sockaddr_ll *sll;
2168 union tpacket_uhdr h;
2169 u8 *skb_head = skb->data;
2170 int skb_len = skb->len;
2171 unsigned int snaplen, res;
2172 unsigned long status = TP_STATUS_USER;
2173 unsigned short macoff, hdrlen;
2174 unsigned int netoff;
2175 struct sk_buff *copy_skb = NULL;
2176 struct timespec64 ts;
2178 bool is_drop_n_account = false;
2179 unsigned int slot_id = 0;
2180 bool do_vnet = false;
2182 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2183 * We may add members to them until current aligned size without forcing
2184 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2186 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2187 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2189 if (skb->pkt_type == PACKET_LOOPBACK)
2192 sk = pt->af_packet_priv;
2195 if (!net_eq(dev_net(dev), sock_net(sk)))
2198 if (dev->header_ops) {
2199 if (sk->sk_type != SOCK_DGRAM)
2200 skb_push(skb, skb->data - skb_mac_header(skb));
2201 else if (skb->pkt_type == PACKET_OUTGOING) {
2202 /* Special case: outgoing packets have ll header at head */
2203 skb_pull(skb, skb_network_offset(skb));
2209 res = run_filter(skb, sk, snaplen);
2211 goto drop_n_restore;
2213 /* If we are flooded, just give up */
2214 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2215 atomic_inc(&po->tp_drops);
2216 goto drop_n_restore;
2219 if (skb->ip_summed == CHECKSUM_PARTIAL)
2220 status |= TP_STATUS_CSUMNOTREADY;
2221 else if (skb->pkt_type != PACKET_OUTGOING &&
2222 (skb->ip_summed == CHECKSUM_COMPLETE ||
2223 skb_csum_unnecessary(skb)))
2224 status |= TP_STATUS_CSUM_VALID;
2229 if (sk->sk_type == SOCK_DGRAM) {
2230 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2233 unsigned int maclen = skb_network_offset(skb);
2234 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2235 (maclen < 16 ? 16 : maclen)) +
2237 if (po->has_vnet_hdr) {
2238 netoff += sizeof(struct virtio_net_hdr);
2241 macoff = netoff - maclen;
2243 if (netoff > USHRT_MAX) {
2244 atomic_inc(&po->tp_drops);
2245 goto drop_n_restore;
2247 if (po->tp_version <= TPACKET_V2) {
2248 if (macoff + snaplen > po->rx_ring.frame_size) {
2249 if (po->copy_thresh &&
2250 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2251 if (skb_shared(skb)) {
2252 copy_skb = skb_clone(skb, GFP_ATOMIC);
2254 copy_skb = skb_get(skb);
2255 skb_head = skb->data;
2258 skb_set_owner_r(copy_skb, sk);
2260 snaplen = po->rx_ring.frame_size - macoff;
2261 if ((int)snaplen < 0) {
2266 } else if (unlikely(macoff + snaplen >
2267 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2270 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2271 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2272 snaplen, nval, macoff);
2274 if (unlikely((int)snaplen < 0)) {
2276 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2280 spin_lock(&sk->sk_receive_queue.lock);
2281 h.raw = packet_current_rx_frame(po, skb,
2282 TP_STATUS_KERNEL, (macoff+snaplen));
2284 goto drop_n_account;
2286 if (po->tp_version <= TPACKET_V2) {
2287 slot_id = po->rx_ring.head;
2288 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2289 goto drop_n_account;
2290 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2294 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2295 sizeof(struct virtio_net_hdr),
2296 vio_le(), true, 0)) {
2297 if (po->tp_version == TPACKET_V3)
2298 prb_clear_blk_fill_status(&po->rx_ring);
2299 goto drop_n_account;
2302 if (po->tp_version <= TPACKET_V2) {
2303 packet_increment_rx_head(po, &po->rx_ring);
2305 * LOSING will be reported till you read the stats,
2306 * because it's COR - Clear On Read.
2307 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2310 if (atomic_read(&po->tp_drops))
2311 status |= TP_STATUS_LOSING;
2314 po->stats.stats1.tp_packets++;
2316 status |= TP_STATUS_COPY;
2317 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2319 spin_unlock(&sk->sk_receive_queue.lock);
2321 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2323 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2324 ktime_get_real_ts64(&ts);
2326 status |= ts_status;
2328 switch (po->tp_version) {
2330 h.h1->tp_len = skb->len;
2331 h.h1->tp_snaplen = snaplen;
2332 h.h1->tp_mac = macoff;
2333 h.h1->tp_net = netoff;
2334 h.h1->tp_sec = ts.tv_sec;
2335 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2336 hdrlen = sizeof(*h.h1);
2339 h.h2->tp_len = skb->len;
2340 h.h2->tp_snaplen = snaplen;
2341 h.h2->tp_mac = macoff;
2342 h.h2->tp_net = netoff;
2343 h.h2->tp_sec = ts.tv_sec;
2344 h.h2->tp_nsec = ts.tv_nsec;
2345 if (skb_vlan_tag_present(skb)) {
2346 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2347 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2348 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2350 h.h2->tp_vlan_tci = 0;
2351 h.h2->tp_vlan_tpid = 0;
2353 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2354 hdrlen = sizeof(*h.h2);
2357 /* tp_nxt_offset,vlan are already populated above.
2358 * So DONT clear those fields here
2360 h.h3->tp_status |= status;
2361 h.h3->tp_len = skb->len;
2362 h.h3->tp_snaplen = snaplen;
2363 h.h3->tp_mac = macoff;
2364 h.h3->tp_net = netoff;
2365 h.h3->tp_sec = ts.tv_sec;
2366 h.h3->tp_nsec = ts.tv_nsec;
2367 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2368 hdrlen = sizeof(*h.h3);
2374 sll = h.raw + TPACKET_ALIGN(hdrlen);
2375 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2376 sll->sll_family = AF_PACKET;
2377 sll->sll_hatype = dev->type;
2378 sll->sll_protocol = skb->protocol;
2379 sll->sll_pkttype = skb->pkt_type;
2380 if (unlikely(po->origdev))
2381 sll->sll_ifindex = orig_dev->ifindex;
2383 sll->sll_ifindex = dev->ifindex;
2387 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2388 if (po->tp_version <= TPACKET_V2) {
2391 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2394 for (start = h.raw; start < end; start += PAGE_SIZE)
2395 flush_dcache_page(pgv_to_page(start));
2400 if (po->tp_version <= TPACKET_V2) {
2401 spin_lock(&sk->sk_receive_queue.lock);
2402 __packet_set_status(po, h.raw, status);
2403 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2404 spin_unlock(&sk->sk_receive_queue.lock);
2405 sk->sk_data_ready(sk);
2406 } else if (po->tp_version == TPACKET_V3) {
2407 prb_clear_blk_fill_status(&po->rx_ring);
2411 if (skb_head != skb->data && skb_shared(skb)) {
2412 skb->data = skb_head;
2416 if (!is_drop_n_account)
2423 spin_unlock(&sk->sk_receive_queue.lock);
2424 atomic_inc(&po->tp_drops);
2425 is_drop_n_account = true;
2427 sk->sk_data_ready(sk);
2428 kfree_skb(copy_skb);
2429 goto drop_n_restore;
2432 static void tpacket_destruct_skb(struct sk_buff *skb)
2434 struct packet_sock *po = pkt_sk(skb->sk);
2436 if (likely(po->tx_ring.pg_vec)) {
2440 ph = skb_zcopy_get_nouarg(skb);
2441 packet_dec_pending(&po->tx_ring);
2443 ts = __packet_set_timestamp(po, ph, skb);
2444 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2446 if (!packet_read_pending(&po->tx_ring))
2447 complete(&po->skb_completion);
2453 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2455 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2456 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2457 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2458 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2459 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2460 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2461 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2463 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2469 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2470 struct virtio_net_hdr *vnet_hdr)
2472 if (*len < sizeof(*vnet_hdr))
2474 *len -= sizeof(*vnet_hdr);
2476 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2479 return __packet_snd_vnet_parse(vnet_hdr, *len);
2482 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2483 void *frame, struct net_device *dev, void *data, int tp_len,
2484 __be16 proto, unsigned char *addr, int hlen, int copylen,
2485 const struct sockcm_cookie *sockc)
2487 union tpacket_uhdr ph;
2488 int to_write, offset, len, nr_frags, len_max;
2489 struct socket *sock = po->sk.sk_socket;
2495 skb->protocol = proto;
2497 skb->priority = po->sk.sk_priority;
2498 skb->mark = po->sk.sk_mark;
2499 skb->tstamp = sockc->transmit_time;
2500 skb_setup_tx_timestamp(skb, sockc->tsflags);
2501 skb_zcopy_set_nouarg(skb, ph.raw);
2503 skb_reserve(skb, hlen);
2504 skb_reset_network_header(skb);
2508 if (sock->type == SOCK_DGRAM) {
2509 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2511 if (unlikely(err < 0))
2513 } else if (copylen) {
2514 int hdrlen = min_t(int, copylen, tp_len);
2516 skb_push(skb, dev->hard_header_len);
2517 skb_put(skb, copylen - dev->hard_header_len);
2518 err = skb_store_bits(skb, 0, data, hdrlen);
2521 if (!dev_validate_header(dev, skb->data, hdrlen))
2528 offset = offset_in_page(data);
2529 len_max = PAGE_SIZE - offset;
2530 len = ((to_write > len_max) ? len_max : to_write);
2532 skb->data_len = to_write;
2533 skb->len += to_write;
2534 skb->truesize += to_write;
2535 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2537 while (likely(to_write)) {
2538 nr_frags = skb_shinfo(skb)->nr_frags;
2540 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2541 pr_err("Packet exceed the number of skb frags(%lu)\n",
2546 page = pgv_to_page(data);
2548 flush_dcache_page(page);
2550 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2553 len_max = PAGE_SIZE;
2554 len = ((to_write > len_max) ? len_max : to_write);
2557 packet_parse_headers(skb, sock);
2562 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2563 int size_max, void **data)
2565 union tpacket_uhdr ph;
2570 switch (po->tp_version) {
2572 if (ph.h3->tp_next_offset != 0) {
2573 pr_warn_once("variable sized slot not supported");
2576 tp_len = ph.h3->tp_len;
2579 tp_len = ph.h2->tp_len;
2582 tp_len = ph.h1->tp_len;
2585 if (unlikely(tp_len > size_max)) {
2586 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2590 if (unlikely(po->tp_tx_has_off)) {
2591 int off_min, off_max;
2593 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2594 off_max = po->tx_ring.frame_size - tp_len;
2595 if (po->sk.sk_type == SOCK_DGRAM) {
2596 switch (po->tp_version) {
2598 off = ph.h3->tp_net;
2601 off = ph.h2->tp_net;
2604 off = ph.h1->tp_net;
2608 switch (po->tp_version) {
2610 off = ph.h3->tp_mac;
2613 off = ph.h2->tp_mac;
2616 off = ph.h1->tp_mac;
2620 if (unlikely((off < off_min) || (off_max < off)))
2623 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2626 *data = frame + off;
2630 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2632 struct sk_buff *skb = NULL;
2633 struct net_device *dev;
2634 struct virtio_net_hdr *vnet_hdr = NULL;
2635 struct sockcm_cookie sockc;
2637 int err, reserve = 0;
2639 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2640 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2641 unsigned char *addr = NULL;
2642 int tp_len, size_max;
2645 int status = TP_STATUS_AVAILABLE;
2646 int hlen, tlen, copylen = 0;
2649 mutex_lock(&po->pg_vec_lock);
2651 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2652 * we need to confirm it under protection of pg_vec_lock.
2654 if (unlikely(!po->tx_ring.pg_vec)) {
2658 if (likely(saddr == NULL)) {
2659 dev = packet_cached_dev_get(po);
2663 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2665 if (msg->msg_namelen < (saddr->sll_halen
2666 + offsetof(struct sockaddr_ll,
2669 proto = saddr->sll_protocol;
2670 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2671 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2672 if (dev && msg->msg_namelen < dev->addr_len +
2673 offsetof(struct sockaddr_ll, sll_addr))
2675 addr = saddr->sll_addr;
2680 if (unlikely(dev == NULL))
2683 if (unlikely(!(dev->flags & IFF_UP)))
2686 sockcm_init(&sockc, &po->sk);
2687 if (msg->msg_controllen) {
2688 err = sock_cmsg_send(&po->sk, msg, &sockc);
2693 if (po->sk.sk_socket->type == SOCK_RAW)
2694 reserve = dev->hard_header_len;
2695 size_max = po->tx_ring.frame_size
2696 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2698 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2699 size_max = dev->mtu + reserve + VLAN_HLEN;
2701 reinit_completion(&po->skb_completion);
2704 ph = packet_current_frame(po, &po->tx_ring,
2705 TP_STATUS_SEND_REQUEST);
2706 if (unlikely(ph == NULL)) {
2707 if (need_wait && skb) {
2708 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2709 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2711 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2715 /* check for additional frames */
2720 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2724 status = TP_STATUS_SEND_REQUEST;
2725 hlen = LL_RESERVED_SPACE(dev);
2726 tlen = dev->needed_tailroom;
2727 if (po->has_vnet_hdr) {
2729 data += sizeof(*vnet_hdr);
2730 tp_len -= sizeof(*vnet_hdr);
2732 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2736 copylen = __virtio16_to_cpu(vio_le(),
2739 copylen = max_t(int, copylen, dev->hard_header_len);
2740 skb = sock_alloc_send_skb(&po->sk,
2741 hlen + tlen + sizeof(struct sockaddr_ll) +
2742 (copylen - dev->hard_header_len),
2745 if (unlikely(skb == NULL)) {
2746 /* we assume the socket was initially writeable ... */
2747 if (likely(len_sum > 0))
2751 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2752 addr, hlen, copylen, &sockc);
2753 if (likely(tp_len >= 0) &&
2754 tp_len > dev->mtu + reserve &&
2755 !po->has_vnet_hdr &&
2756 !packet_extra_vlan_len_allowed(dev, skb))
2759 if (unlikely(tp_len < 0)) {
2762 __packet_set_status(po, ph,
2763 TP_STATUS_AVAILABLE);
2764 packet_increment_head(&po->tx_ring);
2768 status = TP_STATUS_WRONG_FORMAT;
2774 if (po->has_vnet_hdr) {
2775 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2779 virtio_net_hdr_set_proto(skb, vnet_hdr);
2782 skb->destructor = tpacket_destruct_skb;
2783 __packet_set_status(po, ph, TP_STATUS_SENDING);
2784 packet_inc_pending(&po->tx_ring);
2786 status = TP_STATUS_SEND_REQUEST;
2787 err = po->xmit(skb);
2788 if (unlikely(err > 0)) {
2789 err = net_xmit_errno(err);
2790 if (err && __packet_get_status(po, ph) ==
2791 TP_STATUS_AVAILABLE) {
2792 /* skb was destructed already */
2797 * skb was dropped but not destructed yet;
2798 * let's treat it like congestion or err < 0
2802 packet_increment_head(&po->tx_ring);
2804 } while (likely((ph != NULL) ||
2805 /* Note: packet_read_pending() might be slow if we have
2806 * to call it as it's per_cpu variable, but in fast-path
2807 * we already short-circuit the loop with the first
2808 * condition, and luckily don't have to go that path
2811 (need_wait && packet_read_pending(&po->tx_ring))));
2817 __packet_set_status(po, ph, status);
2822 mutex_unlock(&po->pg_vec_lock);
2826 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2827 size_t reserve, size_t len,
2828 size_t linear, int noblock,
2831 struct sk_buff *skb;
2833 /* Under a page? Don't bother with paged skb. */
2834 if (prepad + len < PAGE_SIZE || !linear)
2837 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2842 skb_reserve(skb, reserve);
2843 skb_put(skb, linear);
2844 skb->data_len = len - linear;
2845 skb->len += len - linear;
2850 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2852 struct sock *sk = sock->sk;
2853 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2854 struct sk_buff *skb;
2855 struct net_device *dev;
2857 unsigned char *addr = NULL;
2858 int err, reserve = 0;
2859 struct sockcm_cookie sockc;
2860 struct virtio_net_hdr vnet_hdr = { 0 };
2862 struct packet_sock *po = pkt_sk(sk);
2863 bool has_vnet_hdr = false;
2864 int hlen, tlen, linear;
2868 * Get and verify the address.
2871 if (likely(saddr == NULL)) {
2872 dev = packet_cached_dev_get(po);
2876 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2878 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2880 proto = saddr->sll_protocol;
2881 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2882 if (sock->type == SOCK_DGRAM) {
2883 if (dev && msg->msg_namelen < dev->addr_len +
2884 offsetof(struct sockaddr_ll, sll_addr))
2886 addr = saddr->sll_addr;
2891 if (unlikely(dev == NULL))
2894 if (unlikely(!(dev->flags & IFF_UP)))
2897 sockcm_init(&sockc, sk);
2898 sockc.mark = sk->sk_mark;
2899 if (msg->msg_controllen) {
2900 err = sock_cmsg_send(sk, msg, &sockc);
2905 if (sock->type == SOCK_RAW)
2906 reserve = dev->hard_header_len;
2907 if (po->has_vnet_hdr) {
2908 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2911 has_vnet_hdr = true;
2914 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2915 if (!netif_supports_nofcs(dev)) {
2916 err = -EPROTONOSUPPORT;
2919 extra_len = 4; /* We're doing our own CRC */
2923 if (!vnet_hdr.gso_type &&
2924 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2928 hlen = LL_RESERVED_SPACE(dev);
2929 tlen = dev->needed_tailroom;
2930 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2931 linear = max(linear, min_t(int, len, dev->hard_header_len));
2932 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2933 msg->msg_flags & MSG_DONTWAIT, &err);
2937 skb_reset_network_header(skb);
2940 if (sock->type == SOCK_DGRAM) {
2941 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2942 if (unlikely(offset < 0))
2944 } else if (reserve) {
2945 skb_reserve(skb, -reserve);
2946 if (len < reserve + sizeof(struct ipv6hdr) &&
2947 dev->min_header_len != dev->hard_header_len)
2948 skb_reset_network_header(skb);
2951 /* Returns -EFAULT on error */
2952 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2956 if (sock->type == SOCK_RAW &&
2957 !dev_validate_header(dev, skb->data, len)) {
2962 skb_setup_tx_timestamp(skb, sockc.tsflags);
2964 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2965 !packet_extra_vlan_len_allowed(dev, skb)) {
2970 skb->protocol = proto;
2972 skb->priority = sk->sk_priority;
2973 skb->mark = sockc.mark;
2974 skb->tstamp = sockc.transmit_time;
2977 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2980 len += sizeof(vnet_hdr);
2981 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2984 packet_parse_headers(skb, sock);
2986 if (unlikely(extra_len == 4))
2989 err = po->xmit(skb);
2990 if (err > 0 && (err = net_xmit_errno(err)) != 0)
3006 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3008 struct sock *sk = sock->sk;
3009 struct packet_sock *po = pkt_sk(sk);
3011 if (po->tx_ring.pg_vec)
3012 return tpacket_snd(po, msg);
3014 return packet_snd(sock, msg, len);
3018 * Close a PACKET socket. This is fairly simple. We immediately go
3019 * to 'closed' state and remove our protocol entry in the device list.
3022 static int packet_release(struct socket *sock)
3024 struct sock *sk = sock->sk;
3025 struct packet_sock *po;
3026 struct packet_fanout *f;
3028 union tpacket_req_u req_u;
3036 mutex_lock(&net->packet.sklist_lock);
3037 sk_del_node_init_rcu(sk);
3038 mutex_unlock(&net->packet.sklist_lock);
3041 sock_prot_inuse_add(net, sk->sk_prot, -1);
3044 spin_lock(&po->bind_lock);
3045 unregister_prot_hook(sk, false);
3046 packet_cached_dev_reset(po);
3048 if (po->prot_hook.dev) {
3049 dev_put(po->prot_hook.dev);
3050 po->prot_hook.dev = NULL;
3052 spin_unlock(&po->bind_lock);
3054 packet_flush_mclist(sk);
3057 if (po->rx_ring.pg_vec) {
3058 memset(&req_u, 0, sizeof(req_u));
3059 packet_set_ring(sk, &req_u, 1, 0);
3062 if (po->tx_ring.pg_vec) {
3063 memset(&req_u, 0, sizeof(req_u));
3064 packet_set_ring(sk, &req_u, 1, 1);
3068 f = fanout_release(sk);
3072 kfree(po->rollover);
3074 fanout_release_data(f);
3078 * Now the socket is dead. No more input will appear.
3085 skb_queue_purge(&sk->sk_receive_queue);
3086 packet_free_pending(po);
3087 sk_refcnt_debug_release(sk);
3094 * Attach a packet hook.
3097 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3100 struct packet_sock *po = pkt_sk(sk);
3101 struct net_device *dev_curr;
3104 struct net_device *dev = NULL;
3106 bool unlisted = false;
3109 spin_lock(&po->bind_lock);
3118 dev = dev_get_by_name_rcu(sock_net(sk), name);
3123 } else if (ifindex) {
3124 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3134 proto_curr = po->prot_hook.type;
3135 dev_curr = po->prot_hook.dev;
3137 need_rehook = proto_curr != proto || dev_curr != dev;
3142 /* prevents packet_notifier() from calling
3143 * register_prot_hook()
3146 __unregister_prot_hook(sk, true);
3148 dev_curr = po->prot_hook.dev;
3150 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3154 BUG_ON(po->running);
3156 po->prot_hook.type = proto;
3158 if (unlikely(unlisted)) {
3160 po->prot_hook.dev = NULL;
3162 packet_cached_dev_reset(po);
3164 po->prot_hook.dev = dev;
3165 po->ifindex = dev ? dev->ifindex : 0;
3166 packet_cached_dev_assign(po, dev);
3172 if (proto == 0 || !need_rehook)
3175 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3176 register_prot_hook(sk);
3178 sk->sk_err = ENETDOWN;
3179 if (!sock_flag(sk, SOCK_DEAD))
3180 sk->sk_error_report(sk);
3185 spin_unlock(&po->bind_lock);
3191 * Bind a packet socket to a device
3194 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3197 struct sock *sk = sock->sk;
3198 char name[sizeof(uaddr->sa_data) + 1];
3204 if (addr_len != sizeof(struct sockaddr))
3206 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3209 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3210 name[sizeof(uaddr->sa_data)] = 0;
3212 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3215 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3217 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3218 struct sock *sk = sock->sk;
3224 if (addr_len < sizeof(struct sockaddr_ll))
3226 if (sll->sll_family != AF_PACKET)
3229 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3230 sll->sll_protocol ? : pkt_sk(sk)->num);
3233 static struct proto packet_proto = {
3235 .owner = THIS_MODULE,
3236 .obj_size = sizeof(struct packet_sock),
3240 * Create a packet of type SOCK_PACKET.
3243 static int packet_create(struct net *net, struct socket *sock, int protocol,
3247 struct packet_sock *po;
3248 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3251 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3253 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3254 sock->type != SOCK_PACKET)
3255 return -ESOCKTNOSUPPORT;
3257 sock->state = SS_UNCONNECTED;
3260 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3264 sock->ops = &packet_ops;
3265 if (sock->type == SOCK_PACKET)
3266 sock->ops = &packet_ops_spkt;
3268 sock_init_data(sock, sk);
3271 init_completion(&po->skb_completion);
3272 sk->sk_family = PF_PACKET;
3274 po->xmit = dev_queue_xmit;
3276 err = packet_alloc_pending(po);
3280 packet_cached_dev_reset(po);
3282 sk->sk_destruct = packet_sock_destruct;
3283 sk_refcnt_debug_inc(sk);
3286 * Attach a protocol block
3289 spin_lock_init(&po->bind_lock);
3290 mutex_init(&po->pg_vec_lock);
3291 po->rollover = NULL;
3292 po->prot_hook.func = packet_rcv;
3294 if (sock->type == SOCK_PACKET)
3295 po->prot_hook.func = packet_rcv_spkt;
3297 po->prot_hook.af_packet_priv = sk;
3300 po->prot_hook.type = proto;
3301 __register_prot_hook(sk);
3304 mutex_lock(&net->packet.sklist_lock);
3305 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3306 mutex_unlock(&net->packet.sklist_lock);
3309 sock_prot_inuse_add(net, &packet_proto, 1);
3320 * Pull a packet from our receive queue and hand it to the user.
3321 * If necessary we block.
3324 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3327 struct sock *sk = sock->sk;
3328 struct sk_buff *skb;
3330 int vnet_hdr_len = 0;
3331 unsigned int origlen = 0;
3334 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3338 /* What error should we return now? EUNATTACH? */
3339 if (pkt_sk(sk)->ifindex < 0)
3343 if (flags & MSG_ERRQUEUE) {
3344 err = sock_recv_errqueue(sk, msg, len,
3345 SOL_PACKET, PACKET_TX_TIMESTAMP);
3350 * Call the generic datagram receiver. This handles all sorts
3351 * of horrible races and re-entrancy so we can forget about it
3352 * in the protocol layers.
3354 * Now it will return ENETDOWN, if device have just gone down,
3355 * but then it will block.
3358 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3361 * An error occurred so return it. Because skb_recv_datagram()
3362 * handles the blocking we don't see and worry about blocking
3369 packet_rcv_try_clear_pressure(pkt_sk(sk));
3371 if (pkt_sk(sk)->has_vnet_hdr) {
3372 err = packet_rcv_vnet(msg, skb, &len);
3375 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3378 /* You lose any data beyond the buffer you gave. If it worries
3379 * a user program they can ask the device for its MTU
3385 msg->msg_flags |= MSG_TRUNC;
3388 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3392 if (sock->type != SOCK_PACKET) {
3393 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3395 /* Original length was stored in sockaddr_ll fields */
3396 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3397 sll->sll_family = AF_PACKET;
3398 sll->sll_protocol = skb->protocol;
3401 sock_recv_ts_and_drops(msg, sk, skb);
3403 if (msg->msg_name) {
3406 /* If the address length field is there to be filled
3407 * in, we fill it in now.
3409 if (sock->type == SOCK_PACKET) {
3410 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3411 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3412 copy_len = msg->msg_namelen;
3414 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3416 msg->msg_namelen = sll->sll_halen +
3417 offsetof(struct sockaddr_ll, sll_addr);
3418 copy_len = msg->msg_namelen;
3419 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3420 memset(msg->msg_name +
3421 offsetof(struct sockaddr_ll, sll_addr),
3422 0, sizeof(sll->sll_addr));
3423 msg->msg_namelen = sizeof(struct sockaddr_ll);
3426 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3429 if (pkt_sk(sk)->auxdata) {
3430 struct tpacket_auxdata aux;
3432 aux.tp_status = TP_STATUS_USER;
3433 if (skb->ip_summed == CHECKSUM_PARTIAL)
3434 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3435 else if (skb->pkt_type != PACKET_OUTGOING &&
3436 (skb->ip_summed == CHECKSUM_COMPLETE ||
3437 skb_csum_unnecessary(skb)))
3438 aux.tp_status |= TP_STATUS_CSUM_VALID;
3440 aux.tp_len = origlen;
3441 aux.tp_snaplen = skb->len;
3443 aux.tp_net = skb_network_offset(skb);
3444 if (skb_vlan_tag_present(skb)) {
3445 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3446 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3447 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3449 aux.tp_vlan_tci = 0;
3450 aux.tp_vlan_tpid = 0;
3452 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3456 * Free or return the buffer as appropriate. Again this
3457 * hides all the races and re-entrancy issues from us.
3459 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3462 skb_free_datagram(sk, skb);
3467 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3470 struct net_device *dev;
3471 struct sock *sk = sock->sk;
3476 uaddr->sa_family = AF_PACKET;
3477 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3479 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3481 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3484 return sizeof(*uaddr);
3487 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3490 struct net_device *dev;
3491 struct sock *sk = sock->sk;
3492 struct packet_sock *po = pkt_sk(sk);
3493 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3498 sll->sll_family = AF_PACKET;
3499 sll->sll_ifindex = po->ifindex;
3500 sll->sll_protocol = po->num;
3501 sll->sll_pkttype = 0;
3503 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3505 sll->sll_hatype = dev->type;
3506 sll->sll_halen = dev->addr_len;
3507 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3509 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3514 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3517 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3521 case PACKET_MR_MULTICAST:
3522 if (i->alen != dev->addr_len)
3525 return dev_mc_add(dev, i->addr);
3527 return dev_mc_del(dev, i->addr);
3529 case PACKET_MR_PROMISC:
3530 return dev_set_promiscuity(dev, what);
3531 case PACKET_MR_ALLMULTI:
3532 return dev_set_allmulti(dev, what);
3533 case PACKET_MR_UNICAST:
3534 if (i->alen != dev->addr_len)
3537 return dev_uc_add(dev, i->addr);
3539 return dev_uc_del(dev, i->addr);
3547 static void packet_dev_mclist_delete(struct net_device *dev,
3548 struct packet_mclist **mlp)
3550 struct packet_mclist *ml;
3552 while ((ml = *mlp) != NULL) {
3553 if (ml->ifindex == dev->ifindex) {
3554 packet_dev_mc(dev, ml, -1);
3562 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3564 struct packet_sock *po = pkt_sk(sk);
3565 struct packet_mclist *ml, *i;
3566 struct net_device *dev;
3572 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3577 if (mreq->mr_alen > dev->addr_len)
3581 i = kmalloc(sizeof(*i), GFP_KERNEL);
3586 for (ml = po->mclist; ml; ml = ml->next) {
3587 if (ml->ifindex == mreq->mr_ifindex &&
3588 ml->type == mreq->mr_type &&
3589 ml->alen == mreq->mr_alen &&
3590 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3592 /* Free the new element ... */
3598 i->type = mreq->mr_type;
3599 i->ifindex = mreq->mr_ifindex;
3600 i->alen = mreq->mr_alen;
3601 memcpy(i->addr, mreq->mr_address, i->alen);
3602 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3604 i->next = po->mclist;
3606 err = packet_dev_mc(dev, i, 1);
3608 po->mclist = i->next;
3617 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3619 struct packet_mclist *ml, **mlp;
3623 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3624 if (ml->ifindex == mreq->mr_ifindex &&
3625 ml->type == mreq->mr_type &&
3626 ml->alen == mreq->mr_alen &&
3627 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3628 if (--ml->count == 0) {
3629 struct net_device *dev;
3631 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3633 packet_dev_mc(dev, ml, -1);
3643 static void packet_flush_mclist(struct sock *sk)
3645 struct packet_sock *po = pkt_sk(sk);
3646 struct packet_mclist *ml;
3652 while ((ml = po->mclist) != NULL) {
3653 struct net_device *dev;
3655 po->mclist = ml->next;
3656 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3658 packet_dev_mc(dev, ml, -1);
3665 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3666 unsigned int optlen)
3668 struct sock *sk = sock->sk;
3669 struct packet_sock *po = pkt_sk(sk);
3672 if (level != SOL_PACKET)
3673 return -ENOPROTOOPT;
3676 case PACKET_ADD_MEMBERSHIP:
3677 case PACKET_DROP_MEMBERSHIP:
3679 struct packet_mreq_max mreq;
3681 memset(&mreq, 0, sizeof(mreq));
3682 if (len < sizeof(struct packet_mreq))
3684 if (len > sizeof(mreq))
3686 if (copy_from_sockptr(&mreq, optval, len))
3688 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3690 if (optname == PACKET_ADD_MEMBERSHIP)
3691 ret = packet_mc_add(sk, &mreq);
3693 ret = packet_mc_drop(sk, &mreq);
3697 case PACKET_RX_RING:
3698 case PACKET_TX_RING:
3700 union tpacket_req_u req_u;
3704 switch (po->tp_version) {
3707 len = sizeof(req_u.req);
3711 len = sizeof(req_u.req3);
3717 if (copy_from_sockptr(&req_u.req, optval, len))
3720 ret = packet_set_ring(sk, &req_u, 0,
3721 optname == PACKET_TX_RING);
3726 case PACKET_COPY_THRESH:
3730 if (optlen != sizeof(val))
3732 if (copy_from_sockptr(&val, optval, sizeof(val)))
3735 pkt_sk(sk)->copy_thresh = val;
3738 case PACKET_VERSION:
3742 if (optlen != sizeof(val))
3744 if (copy_from_sockptr(&val, optval, sizeof(val)))
3755 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3758 po->tp_version = val;
3764 case PACKET_RESERVE:
3768 if (optlen != sizeof(val))
3770 if (copy_from_sockptr(&val, optval, sizeof(val)))
3775 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3778 po->tp_reserve = val;
3788 if (optlen != sizeof(val))
3790 if (copy_from_sockptr(&val, optval, sizeof(val)))
3794 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3797 po->tp_loss = !!val;
3803 case PACKET_AUXDATA:
3807 if (optlen < sizeof(val))
3809 if (copy_from_sockptr(&val, optval, sizeof(val)))
3813 po->auxdata = !!val;
3817 case PACKET_ORIGDEV:
3821 if (optlen < sizeof(val))
3823 if (copy_from_sockptr(&val, optval, sizeof(val)))
3827 po->origdev = !!val;
3831 case PACKET_VNET_HDR:
3835 if (sock->type != SOCK_RAW)
3837 if (optlen < sizeof(val))
3839 if (copy_from_sockptr(&val, optval, sizeof(val)))
3843 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3846 po->has_vnet_hdr = !!val;
3852 case PACKET_TIMESTAMP:
3856 if (optlen != sizeof(val))
3858 if (copy_from_sockptr(&val, optval, sizeof(val)))
3861 po->tp_tstamp = val;
3868 if (optlen != sizeof(val))
3870 if (copy_from_sockptr(&val, optval, sizeof(val)))
3873 return fanout_add(sk, val & 0xffff, val >> 16);
3875 case PACKET_FANOUT_DATA:
3880 return fanout_set_data(po, optval, optlen);
3882 case PACKET_IGNORE_OUTGOING:
3886 if (optlen != sizeof(val))
3888 if (copy_from_sockptr(&val, optval, sizeof(val)))
3890 if (val < 0 || val > 1)
3893 po->prot_hook.ignore_outgoing = !!val;
3896 case PACKET_TX_HAS_OFF:
3900 if (optlen != sizeof(val))
3902 if (copy_from_sockptr(&val, optval, sizeof(val)))
3906 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3909 po->tp_tx_has_off = !!val;
3915 case PACKET_QDISC_BYPASS:
3919 if (optlen != sizeof(val))
3921 if (copy_from_sockptr(&val, optval, sizeof(val)))
3924 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3928 return -ENOPROTOOPT;
3932 static int packet_getsockopt(struct socket *sock, int level, int optname,
3933 char __user *optval, int __user *optlen)
3936 int val, lv = sizeof(val);
3937 struct sock *sk = sock->sk;
3938 struct packet_sock *po = pkt_sk(sk);
3940 union tpacket_stats_u st;
3941 struct tpacket_rollover_stats rstats;
3944 if (level != SOL_PACKET)
3945 return -ENOPROTOOPT;
3947 if (get_user(len, optlen))
3954 case PACKET_STATISTICS:
3955 spin_lock_bh(&sk->sk_receive_queue.lock);
3956 memcpy(&st, &po->stats, sizeof(st));
3957 memset(&po->stats, 0, sizeof(po->stats));
3958 spin_unlock_bh(&sk->sk_receive_queue.lock);
3959 drops = atomic_xchg(&po->tp_drops, 0);
3961 if (po->tp_version == TPACKET_V3) {
3962 lv = sizeof(struct tpacket_stats_v3);
3963 st.stats3.tp_drops = drops;
3964 st.stats3.tp_packets += drops;
3967 lv = sizeof(struct tpacket_stats);
3968 st.stats1.tp_drops = drops;
3969 st.stats1.tp_packets += drops;
3974 case PACKET_AUXDATA:
3977 case PACKET_ORIGDEV:
3980 case PACKET_VNET_HDR:
3981 val = po->has_vnet_hdr;
3983 case PACKET_VERSION:
3984 val = po->tp_version;
3987 if (len > sizeof(int))
3989 if (len < sizeof(int))
3991 if (copy_from_user(&val, optval, len))
3995 val = sizeof(struct tpacket_hdr);
3998 val = sizeof(struct tpacket2_hdr);
4001 val = sizeof(struct tpacket3_hdr);
4007 case PACKET_RESERVE:
4008 val = po->tp_reserve;
4013 case PACKET_TIMESTAMP:
4014 val = po->tp_tstamp;
4018 ((u32)po->fanout->id |
4019 ((u32)po->fanout->type << 16) |
4020 ((u32)po->fanout->flags << 24)) :
4023 case PACKET_IGNORE_OUTGOING:
4024 val = po->prot_hook.ignore_outgoing;
4026 case PACKET_ROLLOVER_STATS:
4029 rstats.tp_all = atomic_long_read(&po->rollover->num);
4030 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4031 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4033 lv = sizeof(rstats);
4035 case PACKET_TX_HAS_OFF:
4036 val = po->tp_tx_has_off;
4038 case PACKET_QDISC_BYPASS:
4039 val = packet_use_direct_xmit(po);
4042 return -ENOPROTOOPT;
4047 if (put_user(len, optlen))
4049 if (copy_to_user(optval, data, len))
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 unsigned long *rx_owner_map = NULL;
4281 int was_running, order = 0;
4282 struct packet_ring_buffer *rb;
4283 struct sk_buff_head *rb_queue;
4286 /* Added to avoid minimal code churn */
4287 struct tpacket_req *req = &req_u->req;
4289 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4290 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4294 if (atomic_read(&po->mapped))
4296 if (packet_read_pending(rb))
4300 if (req->tp_block_nr) {
4301 unsigned int min_frame_size;
4303 /* Sanity tests and some calculations */
4305 if (unlikely(rb->pg_vec))
4308 switch (po->tp_version) {
4310 po->tp_hdrlen = TPACKET_HDRLEN;
4313 po->tp_hdrlen = TPACKET2_HDRLEN;
4316 po->tp_hdrlen = TPACKET3_HDRLEN;
4321 if (unlikely((int)req->tp_block_size <= 0))
4323 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4325 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4326 if (po->tp_version >= TPACKET_V3 &&
4327 req->tp_block_size <
4328 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4330 if (unlikely(req->tp_frame_size < min_frame_size))
4332 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4335 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4336 if (unlikely(rb->frames_per_block == 0))
4338 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4340 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4345 order = get_order(req->tp_block_size);
4346 pg_vec = alloc_pg_vec(req, order);
4347 if (unlikely(!pg_vec))
4349 switch (po->tp_version) {
4351 /* Block transmit is not supported yet */
4353 init_prb_bdqc(po, rb, pg_vec, req_u);
4355 struct tpacket_req3 *req3 = &req_u->req3;
4357 if (req3->tp_retire_blk_tov ||
4358 req3->tp_sizeof_priv ||
4359 req3->tp_feature_req_word) {
4361 goto out_free_pg_vec;
4367 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4368 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4370 goto out_free_pg_vec;
4378 if (unlikely(req->tp_frame_nr))
4383 /* Detach socket from network */
4384 spin_lock(&po->bind_lock);
4385 was_running = po->running;
4389 __unregister_prot_hook(sk, false);
4391 spin_unlock(&po->bind_lock);
4396 mutex_lock(&po->pg_vec_lock);
4397 if (closing || atomic_read(&po->mapped) == 0) {
4399 spin_lock_bh(&rb_queue->lock);
4400 swap(rb->pg_vec, pg_vec);
4401 if (po->tp_version <= TPACKET_V2)
4402 swap(rb->rx_owner_map, rx_owner_map);
4403 rb->frame_max = (req->tp_frame_nr - 1);
4405 rb->frame_size = req->tp_frame_size;
4406 spin_unlock_bh(&rb_queue->lock);
4408 swap(rb->pg_vec_order, order);
4409 swap(rb->pg_vec_len, req->tp_block_nr);
4411 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4412 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4413 tpacket_rcv : packet_rcv;
4414 skb_queue_purge(rb_queue);
4415 if (atomic_read(&po->mapped))
4416 pr_err("packet_mmap: vma is busy: %d\n",
4417 atomic_read(&po->mapped));
4419 mutex_unlock(&po->pg_vec_lock);
4421 spin_lock(&po->bind_lock);
4424 register_prot_hook(sk);
4426 spin_unlock(&po->bind_lock);
4427 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4428 /* Because we don't support block-based V3 on tx-ring */
4430 prb_shutdown_retire_blk_timer(po, rb_queue);
4434 bitmap_free(rx_owner_map);
4436 free_pg_vec(pg_vec, order, req->tp_block_nr);
4441 static int packet_mmap(struct file *file, struct socket *sock,
4442 struct vm_area_struct *vma)
4444 struct sock *sk = sock->sk;
4445 struct packet_sock *po = pkt_sk(sk);
4446 unsigned long size, expected_size;
4447 struct packet_ring_buffer *rb;
4448 unsigned long start;
4455 mutex_lock(&po->pg_vec_lock);
4458 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4460 expected_size += rb->pg_vec_len
4466 if (expected_size == 0)
4469 size = vma->vm_end - vma->vm_start;
4470 if (size != expected_size)
4473 start = vma->vm_start;
4474 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4475 if (rb->pg_vec == NULL)
4478 for (i = 0; i < rb->pg_vec_len; i++) {
4480 void *kaddr = rb->pg_vec[i].buffer;
4483 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4484 page = pgv_to_page(kaddr);
4485 err = vm_insert_page(vma, start, page);
4494 atomic_inc(&po->mapped);
4495 vma->vm_ops = &packet_mmap_ops;
4499 mutex_unlock(&po->pg_vec_lock);
4503 static const struct proto_ops packet_ops_spkt = {
4504 .family = PF_PACKET,
4505 .owner = THIS_MODULE,
4506 .release = packet_release,
4507 .bind = packet_bind_spkt,
4508 .connect = sock_no_connect,
4509 .socketpair = sock_no_socketpair,
4510 .accept = sock_no_accept,
4511 .getname = packet_getname_spkt,
4512 .poll = datagram_poll,
4513 .ioctl = packet_ioctl,
4514 .gettstamp = sock_gettstamp,
4515 .listen = sock_no_listen,
4516 .shutdown = sock_no_shutdown,
4517 .sendmsg = packet_sendmsg_spkt,
4518 .recvmsg = packet_recvmsg,
4519 .mmap = sock_no_mmap,
4520 .sendpage = sock_no_sendpage,
4523 static const struct proto_ops packet_ops = {
4524 .family = PF_PACKET,
4525 .owner = THIS_MODULE,
4526 .release = packet_release,
4527 .bind = packet_bind,
4528 .connect = sock_no_connect,
4529 .socketpair = sock_no_socketpair,
4530 .accept = sock_no_accept,
4531 .getname = packet_getname,
4532 .poll = packet_poll,
4533 .ioctl = packet_ioctl,
4534 .gettstamp = sock_gettstamp,
4535 .listen = sock_no_listen,
4536 .shutdown = sock_no_shutdown,
4537 .setsockopt = packet_setsockopt,
4538 .getsockopt = packet_getsockopt,
4539 .sendmsg = packet_sendmsg,
4540 .recvmsg = packet_recvmsg,
4541 .mmap = packet_mmap,
4542 .sendpage = sock_no_sendpage,
4545 static const struct net_proto_family packet_family_ops = {
4546 .family = PF_PACKET,
4547 .create = packet_create,
4548 .owner = THIS_MODULE,
4551 static struct notifier_block packet_netdev_notifier = {
4552 .notifier_call = packet_notifier,
4555 #ifdef CONFIG_PROC_FS
4557 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4560 struct net *net = seq_file_net(seq);
4563 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4566 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4568 struct net *net = seq_file_net(seq);
4569 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4572 static void packet_seq_stop(struct seq_file *seq, void *v)
4578 static int packet_seq_show(struct seq_file *seq, void *v)
4580 if (v == SEQ_START_TOKEN)
4581 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4583 struct sock *s = sk_entry(v);
4584 const struct packet_sock *po = pkt_sk(s);
4587 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4589 refcount_read(&s->sk_refcnt),
4594 atomic_read(&s->sk_rmem_alloc),
4595 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4602 static const struct seq_operations packet_seq_ops = {
4603 .start = packet_seq_start,
4604 .next = packet_seq_next,
4605 .stop = packet_seq_stop,
4606 .show = packet_seq_show,
4610 static int __net_init packet_net_init(struct net *net)
4612 mutex_init(&net->packet.sklist_lock);
4613 INIT_HLIST_HEAD(&net->packet.sklist);
4615 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4616 sizeof(struct seq_net_private)))
4622 static void __net_exit packet_net_exit(struct net *net)
4624 remove_proc_entry("packet", net->proc_net);
4625 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4628 static struct pernet_operations packet_net_ops = {
4629 .init = packet_net_init,
4630 .exit = packet_net_exit,
4634 static void __exit packet_exit(void)
4636 unregister_netdevice_notifier(&packet_netdev_notifier);
4637 unregister_pernet_subsys(&packet_net_ops);
4638 sock_unregister(PF_PACKET);
4639 proto_unregister(&packet_proto);
4642 static int __init packet_init(void)
4646 rc = proto_register(&packet_proto, 0);
4649 rc = sock_register(&packet_family_ops);
4652 rc = register_pernet_subsys(&packet_net_ops);
4655 rc = register_netdevice_notifier(&packet_netdev_notifier);
4662 unregister_pernet_subsys(&packet_net_ops);
4664 sock_unregister(PF_PACKET);
4666 proto_unregister(&packet_proto);
4671 module_init(packet_init);
4672 module_exit(packet_exit);
4673 MODULE_LICENSE("GPL");
4674 MODULE_ALIAS_NETPROTO(PF_PACKET);