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/ethtool.h>
50 #include <linux/types.h>
52 #include <linux/capability.h>
53 #include <linux/fcntl.h>
54 #include <linux/socket.h>
56 #include <linux/inet.h>
57 #include <linux/netdevice.h>
58 #include <linux/if_packet.h>
59 #include <linux/wireless.h>
60 #include <linux/kernel.h>
61 #include <linux/kmod.h>
62 #include <linux/slab.h>
63 #include <linux/vmalloc.h>
64 #include <net/net_namespace.h>
66 #include <net/protocol.h>
67 #include <linux/skbuff.h>
69 #include <linux/errno.h>
70 #include <linux/timer.h>
71 #include <linux/uaccess.h>
72 #include <asm/ioctls.h>
74 #include <asm/cacheflush.h>
76 #include <linux/proc_fs.h>
77 #include <linux/seq_file.h>
78 #include <linux/poll.h>
79 #include <linux/module.h>
80 #include <linux/init.h>
81 #include <linux/mutex.h>
82 #include <linux/if_vlan.h>
83 #include <linux/virtio_net.h>
84 #include <linux/errqueue.h>
85 #include <linux/net_tstamp.h>
86 #include <linux/percpu.h>
88 #include <net/inet_common.h>
90 #include <linux/bpf.h>
91 #include <net/compat.h>
97 - If the device has no dev->header_ops->create, there is no LL header
98 visible above the device. In this case, its hard_header_len should be 0.
99 The device may prepend its own header internally. In this case, its
100 needed_headroom should be set to the space needed for it to add its
102 For example, a WiFi driver pretending to be an Ethernet driver should
103 set its hard_header_len to be the Ethernet header length, and set its
104 needed_headroom to be (the real WiFi header length - the fake Ethernet
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev_has_header(dev) == true
113 mac_header -> ll header
116 Outgoing, dev_has_header(dev) == true
117 mac_header -> ll header
120 Incoming, dev_has_header(dev) == false
122 However drivers often make it point to the ll header.
123 This is incorrect because the ll header should be invisible to us.
126 Outgoing, dev_has_header(dev) == false
127 mac_header -> data. ll header is invisible to us.
131 If dev_has_header(dev) == false we are unable to restore the ll header,
132 because it is invisible to us.
138 dev_has_header(dev) == true
139 mac_header -> ll header
142 dev_has_header(dev) == false (ll header is invisible to us)
146 We should set network_header on output to the correct position,
147 packet classifier depends on it.
150 /* Private packet socket structures. */
152 /* identical to struct packet_mreq except it has
153 * a longer address field.
155 struct packet_mreq_max {
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
180 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
181 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
182 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
183 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
184 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
188 struct packet_type *pt, struct net_device *orig_dev);
190 static void *packet_previous_frame(struct packet_sock *po,
191 struct packet_ring_buffer *rb,
193 static void packet_increment_head(struct packet_ring_buffer *buff);
194 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
195 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
196 struct packet_sock *);
197 static void prb_retire_current_block(struct tpacket_kbdq_core *,
198 struct packet_sock *, unsigned int status);
199 static int prb_queue_frozen(struct tpacket_kbdq_core *);
200 static void prb_open_block(struct tpacket_kbdq_core *,
201 struct tpacket_block_desc *);
202 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
203 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
204 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
205 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
206 struct tpacket3_hdr *);
207 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
208 struct tpacket3_hdr *);
209 static void packet_flush_mclist(struct sock *sk);
210 static u16 packet_pick_tx_queue(struct sk_buff *skb);
212 struct packet_skb_cb {
214 struct sockaddr_pkt pkt;
216 /* Trick: alias skb original length with
217 * ll.sll_family and ll.protocol in order
220 unsigned int origlen;
221 struct sockaddr_ll ll;
226 #define vio_le() virtio_legacy_is_little_endian()
228 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
230 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
231 #define GET_PBLOCK_DESC(x, bid) \
232 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
233 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
234 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
235 #define GET_NEXT_PRB_BLK_NUM(x) \
236 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
237 ((x)->kactive_blk_num+1) : 0)
239 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
240 static void __fanout_link(struct sock *sk, struct packet_sock *po);
242 static int packet_direct_xmit(struct sk_buff *skb)
244 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
247 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
249 struct net_device *dev;
252 dev = rcu_dereference(po->cached_dev);
260 static void packet_cached_dev_assign(struct packet_sock *po,
261 struct net_device *dev)
263 rcu_assign_pointer(po->cached_dev, dev);
266 static void packet_cached_dev_reset(struct packet_sock *po)
268 RCU_INIT_POINTER(po->cached_dev, NULL);
271 static bool packet_use_direct_xmit(const struct packet_sock *po)
273 return po->xmit == packet_direct_xmit;
276 static u16 packet_pick_tx_queue(struct sk_buff *skb)
278 struct net_device *dev = skb->dev;
279 const struct net_device_ops *ops = dev->netdev_ops;
280 int cpu = raw_smp_processor_id();
284 skb->sender_cpu = cpu + 1;
286 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
287 if (ops->ndo_select_queue) {
288 queue_index = ops->ndo_select_queue(dev, skb, NULL);
289 queue_index = netdev_cap_txqueue(dev, queue_index);
291 queue_index = netdev_pick_tx(dev, skb, NULL);
297 /* __register_prot_hook must be invoked through register_prot_hook
298 * or from a context in which asynchronous accesses to the packet
299 * socket is not possible (packet_create()).
301 static void __register_prot_hook(struct sock *sk)
303 struct packet_sock *po = pkt_sk(sk);
307 __fanout_link(sk, po);
309 dev_add_pack(&po->prot_hook);
316 static void register_prot_hook(struct sock *sk)
318 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
319 __register_prot_hook(sk);
322 /* If the sync parameter is true, we will temporarily drop
323 * the po->bind_lock and do a synchronize_net to make sure no
324 * asynchronous packet processing paths still refer to the elements
325 * of po->prot_hook. If the sync parameter is false, it is the
326 * callers responsibility to take care of this.
328 static void __unregister_prot_hook(struct sock *sk, bool sync)
330 struct packet_sock *po = pkt_sk(sk);
332 lockdep_assert_held_once(&po->bind_lock);
337 __fanout_unlink(sk, po);
339 __dev_remove_pack(&po->prot_hook);
344 spin_unlock(&po->bind_lock);
346 spin_lock(&po->bind_lock);
350 static void unregister_prot_hook(struct sock *sk, bool sync)
352 struct packet_sock *po = pkt_sk(sk);
355 __unregister_prot_hook(sk, sync);
358 static inline struct page * __pure pgv_to_page(void *addr)
360 if (is_vmalloc_addr(addr))
361 return vmalloc_to_page(addr);
362 return virt_to_page(addr);
365 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
367 union tpacket_uhdr h;
370 switch (po->tp_version) {
372 h.h1->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
376 h.h2->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
380 h.h3->tp_status = status;
381 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
384 WARN(1, "TPACKET version not supported.\n");
391 static int __packet_get_status(const struct packet_sock *po, void *frame)
393 union tpacket_uhdr h;
398 switch (po->tp_version) {
400 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
401 return h.h1->tp_status;
403 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
404 return h.h2->tp_status;
406 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
407 return h.h3->tp_status;
409 WARN(1, "TPACKET version not supported.\n");
415 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
418 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
421 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
422 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
423 return TP_STATUS_TS_RAW_HARDWARE;
425 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
426 ktime_to_timespec64_cond(skb->tstamp, ts))
427 return TP_STATUS_TS_SOFTWARE;
432 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
435 union tpacket_uhdr h;
436 struct timespec64 ts;
439 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
444 * versions 1 through 3 overflow the timestamps in y2106, since they
445 * all store the seconds in a 32-bit unsigned integer.
446 * If we create a version 4, that should have a 64-bit timestamp,
447 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
450 switch (po->tp_version) {
452 h.h1->tp_sec = ts.tv_sec;
453 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
456 h.h2->tp_sec = ts.tv_sec;
457 h.h2->tp_nsec = ts.tv_nsec;
460 h.h3->tp_sec = ts.tv_sec;
461 h.h3->tp_nsec = ts.tv_nsec;
464 WARN(1, "TPACKET version not supported.\n");
468 /* one flush is safe, as both fields always lie on the same cacheline */
469 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
475 static void *packet_lookup_frame(const struct packet_sock *po,
476 const struct packet_ring_buffer *rb,
477 unsigned int position,
480 unsigned int pg_vec_pos, frame_offset;
481 union tpacket_uhdr h;
483 pg_vec_pos = position / rb->frames_per_block;
484 frame_offset = position % rb->frames_per_block;
486 h.raw = rb->pg_vec[pg_vec_pos].buffer +
487 (frame_offset * rb->frame_size);
489 if (status != __packet_get_status(po, h.raw))
495 static void *packet_current_frame(struct packet_sock *po,
496 struct packet_ring_buffer *rb,
499 return packet_lookup_frame(po, rb, rb->head, status);
502 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
504 del_timer_sync(&pkc->retire_blk_timer);
507 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
508 struct sk_buff_head *rb_queue)
510 struct tpacket_kbdq_core *pkc;
512 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
514 spin_lock_bh(&rb_queue->lock);
515 pkc->delete_blk_timer = 1;
516 spin_unlock_bh(&rb_queue->lock);
518 prb_del_retire_blk_timer(pkc);
521 static void prb_setup_retire_blk_timer(struct packet_sock *po)
523 struct tpacket_kbdq_core *pkc;
525 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
526 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
528 pkc->retire_blk_timer.expires = jiffies;
531 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
532 int blk_size_in_bytes)
534 struct net_device *dev;
535 unsigned int mbits, div;
536 struct ethtool_link_ksettings ecmd;
540 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
541 if (unlikely(!dev)) {
543 return DEFAULT_PRB_RETIRE_TOV;
545 err = __ethtool_get_link_ksettings(dev, &ecmd);
548 return DEFAULT_PRB_RETIRE_TOV;
550 /* If the link speed is so slow you don't really
551 * need to worry about perf anyways
553 if (ecmd.base.speed < SPEED_1000 ||
554 ecmd.base.speed == SPEED_UNKNOWN)
555 return DEFAULT_PRB_RETIRE_TOV;
557 div = ecmd.base.speed / 1000;
558 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
568 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
569 union tpacket_req_u *req_u)
571 p1->feature_req_word = req_u->req3.tp_feature_req_word;
574 static void init_prb_bdqc(struct packet_sock *po,
575 struct packet_ring_buffer *rb,
577 union tpacket_req_u *req_u)
579 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
580 struct tpacket_block_desc *pbd;
582 memset(p1, 0x0, sizeof(*p1));
584 p1->knxt_seq_num = 1;
586 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
587 p1->pkblk_start = pg_vec[0].buffer;
588 p1->kblk_size = req_u->req3.tp_block_size;
589 p1->knum_blocks = req_u->req3.tp_block_nr;
590 p1->hdrlen = po->tp_hdrlen;
591 p1->version = po->tp_version;
592 p1->last_kactive_blk_num = 0;
593 po->stats.stats3.tp_freeze_q_cnt = 0;
594 if (req_u->req3.tp_retire_blk_tov)
595 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
597 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
598 req_u->req3.tp_block_size);
599 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
600 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
601 rwlock_init(&p1->blk_fill_in_prog_lock);
603 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
604 prb_init_ft_ops(p1, req_u);
605 prb_setup_retire_blk_timer(po);
606 prb_open_block(p1, pbd);
609 /* Do NOT update the last_blk_num first.
610 * Assumes sk_buff_head lock is held.
612 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
614 mod_timer(&pkc->retire_blk_timer,
615 jiffies + pkc->tov_in_jiffies);
616 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
621 * 1) We refresh the timer only when we open a block.
622 * By doing this we don't waste cycles refreshing the timer
623 * on packet-by-packet basis.
625 * With a 1MB block-size, on a 1Gbps line, it will take
626 * i) ~8 ms to fill a block + ii) memcpy etc.
627 * In this cut we are not accounting for the memcpy time.
629 * So, if the user sets the 'tmo' to 10ms then the timer
630 * will never fire while the block is still getting filled
631 * (which is what we want). However, the user could choose
632 * to close a block early and that's fine.
634 * But when the timer does fire, we check whether or not to refresh it.
635 * Since the tmo granularity is in msecs, it is not too expensive
636 * to refresh the timer, lets say every '8' msecs.
637 * Either the user can set the 'tmo' or we can derive it based on
638 * a) line-speed and b) block-size.
639 * prb_calc_retire_blk_tmo() calculates the tmo.
642 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
644 struct packet_sock *po =
645 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
646 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
648 struct tpacket_block_desc *pbd;
650 spin_lock(&po->sk.sk_receive_queue.lock);
652 frozen = prb_queue_frozen(pkc);
653 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
655 if (unlikely(pkc->delete_blk_timer))
658 /* We only need to plug the race when the block is partially filled.
660 * lock(); increment BLOCK_NUM_PKTS; unlock()
661 * copy_bits() is in progress ...
662 * timer fires on other cpu:
663 * we can't retire the current block because copy_bits
667 if (BLOCK_NUM_PKTS(pbd)) {
668 /* Waiting for skb_copy_bits to finish... */
669 write_lock(&pkc->blk_fill_in_prog_lock);
670 write_unlock(&pkc->blk_fill_in_prog_lock);
673 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
675 if (!BLOCK_NUM_PKTS(pbd)) {
676 /* An empty block. Just refresh the timer. */
679 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
680 if (!prb_dispatch_next_block(pkc, po))
685 /* Case 1. Queue was frozen because user-space was
688 if (prb_curr_blk_in_use(pbd)) {
690 * Ok, user-space is still behind.
691 * So just refresh the timer.
695 /* Case 2. queue was frozen,user-space caught up,
696 * now the link went idle && the timer fired.
697 * We don't have a block to close.So we open this
698 * block and restart the timer.
699 * opening a block thaws the queue,restarts timer
700 * Thawing/timer-refresh is a side effect.
702 prb_open_block(pkc, pbd);
709 _prb_refresh_rx_retire_blk_timer(pkc);
712 spin_unlock(&po->sk.sk_receive_queue.lock);
715 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
716 struct tpacket_block_desc *pbd1, __u32 status)
718 /* Flush everything minus the block header */
720 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
725 /* Skip the block header(we know header WILL fit in 4K) */
728 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
729 for (; start < end; start += PAGE_SIZE)
730 flush_dcache_page(pgv_to_page(start));
735 /* Now update the block status. */
737 BLOCK_STATUS(pbd1) = status;
739 /* Flush the block header */
741 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 flush_dcache_page(pgv_to_page(start));
753 * 2) Increment active_blk_num
755 * Note:We DONT refresh the timer on purpose.
756 * Because almost always the next block will be opened.
758 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
759 struct tpacket_block_desc *pbd1,
760 struct packet_sock *po, unsigned int stat)
762 __u32 status = TP_STATUS_USER | stat;
764 struct tpacket3_hdr *last_pkt;
765 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
766 struct sock *sk = &po->sk;
768 if (atomic_read(&po->tp_drops))
769 status |= TP_STATUS_LOSING;
771 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
772 last_pkt->tp_next_offset = 0;
774 /* Get the ts of the last pkt */
775 if (BLOCK_NUM_PKTS(pbd1)) {
776 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
777 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
779 /* Ok, we tmo'd - so get the current time.
781 * It shouldn't really happen as we don't close empty
782 * blocks. See prb_retire_rx_blk_timer_expired().
784 struct timespec64 ts;
785 ktime_get_real_ts64(&ts);
786 h1->ts_last_pkt.ts_sec = ts.tv_sec;
787 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
792 /* Flush the block */
793 prb_flush_block(pkc1, pbd1, status);
795 sk->sk_data_ready(sk);
797 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
800 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
802 pkc->reset_pending_on_curr_blk = 0;
806 * Side effect of opening a block:
808 * 1) prb_queue is thawed.
809 * 2) retire_blk_timer is refreshed.
812 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
813 struct tpacket_block_desc *pbd1)
815 struct timespec64 ts;
816 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
820 /* We could have just memset this but we will lose the
821 * flexibility of making the priv area sticky
824 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
825 BLOCK_NUM_PKTS(pbd1) = 0;
826 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
828 ktime_get_real_ts64(&ts);
830 h1->ts_first_pkt.ts_sec = ts.tv_sec;
831 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
833 pkc1->pkblk_start = (char *)pbd1;
834 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
836 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
839 pbd1->version = pkc1->version;
840 pkc1->prev = pkc1->nxt_offset;
841 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
843 prb_thaw_queue(pkc1);
844 _prb_refresh_rx_retire_blk_timer(pkc1);
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats.stats3.tp_freeze_q_cnt++;
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
890 struct tpacket_block_desc *pbd;
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
905 * open this block and return the offset where the first packet
906 * needs to get stored.
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 /* Waiting for skb_copy_bits to finish... */
930 write_lock(&pkc->blk_fill_in_prog_lock);
931 write_unlock(&pkc->blk_fill_in_prog_lock);
933 prb_close_block(pkc, pbd, po, status);
938 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
940 return TP_STATUS_USER & BLOCK_STATUS(pbd);
943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
945 return pkc->reset_pending_on_curr_blk;
948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
949 __releases(&pkc->blk_fill_in_prog_lock)
951 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
953 read_unlock(&pkc->blk_fill_in_prog_lock);
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957 struct tpacket3_hdr *ppd)
959 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963 struct tpacket3_hdr *ppd)
965 ppd->hv1.tp_rxhash = 0;
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969 struct tpacket3_hdr *ppd)
971 if (skb_vlan_tag_present(pkc->skb)) {
972 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
973 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
974 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
976 ppd->hv1.tp_vlan_tci = 0;
977 ppd->hv1.tp_vlan_tpid = 0;
978 ppd->tp_status = TP_STATUS_AVAILABLE;
982 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
983 struct tpacket3_hdr *ppd)
985 ppd->hv1.tp_padding = 0;
986 prb_fill_vlan_info(pkc, ppd);
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
991 prb_clear_rxhash(pkc, ppd);
994 static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
998 __acquires(&pkc->blk_fill_in_prog_lock)
1000 struct tpacket3_hdr *ppd;
1002 ppd = (struct tpacket3_hdr *)curr;
1003 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1007 BLOCK_NUM_PKTS(pbd) += 1;
1008 read_lock(&pkc->blk_fill_in_prog_lock);
1009 prb_run_all_ft_ops(pkc, ppd);
1012 /* Assumes caller has the sk->rx_queue.lock */
1013 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1014 struct sk_buff *skb,
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1022 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1031 if (prb_curr_blk_in_use(pbd)) {
1032 /* Can't record this packet */
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1041 prb_open_block(pkc, pbd);
1046 curr = pkc->nxt_offset;
1048 end = (char *)pbd + pkc->kblk_size;
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1079 switch (po->tp_version) {
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1086 return __packet_lookup_frame_in_block(po, skb, len);
1088 WARN(1, "TPACKET version not supported\n");
1094 static void *prb_lookup_block(const struct packet_sock *po,
1095 const struct packet_ring_buffer *rb,
1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1102 if (status != BLOCK_STATUS(pbd))
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1113 prev = rb->prb_bdqc.knum_blocks-1;
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1133 return __prb_previous_block(po, rb, status);
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137 struct packet_ring_buffer *rb)
1139 switch (po->tp_version) {
1142 return packet_increment_head(rb);
1145 WARN(1, "TPACKET version not supported.\n");
1151 static void *packet_previous_frame(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1164 static void packet_inc_pending(struct packet_ring_buffer *rb)
1166 this_cpu_inc(*rb->pending_refcnt);
1169 static void packet_dec_pending(struct packet_ring_buffer *rb)
1171 this_cpu_dec(*rb->pending_refcnt);
1174 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1176 unsigned int refcnt = 0;
1179 /* We don't use pending refcount in rx_ring. */
1180 if (rb->pending_refcnt == NULL)
1183 for_each_possible_cpu(cpu)
1184 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1189 static int packet_alloc_pending(struct packet_sock *po)
1191 po->rx_ring.pending_refcnt = NULL;
1193 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1194 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1200 static void packet_free_pending(struct packet_sock *po)
1202 free_percpu(po->tx_ring.pending_refcnt);
1205 #define ROOM_POW_OFF 2
1206 #define ROOM_NONE 0x0
1207 #define ROOM_LOW 0x1
1208 #define ROOM_NORMAL 0x2
1210 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1214 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1215 idx = READ_ONCE(po->rx_ring.head);
1217 idx += len >> pow_off;
1220 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1223 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1227 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1228 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1230 idx += len >> pow_off;
1233 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1236 static int __packet_rcv_has_room(const struct packet_sock *po,
1237 const struct sk_buff *skb)
1239 const struct sock *sk = &po->sk;
1240 int ret = ROOM_NONE;
1242 if (po->prot_hook.func != tpacket_rcv) {
1243 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1244 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1245 - (skb ? skb->truesize : 0);
1247 if (avail > (rcvbuf >> ROOM_POW_OFF))
1255 if (po->tp_version == TPACKET_V3) {
1256 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1258 else if (__tpacket_v3_has_room(po, 0))
1261 if (__tpacket_has_room(po, ROOM_POW_OFF))
1263 else if (__tpacket_has_room(po, 0))
1270 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1274 ret = __packet_rcv_has_room(po, skb);
1275 pressure = ret != ROOM_NORMAL;
1277 if (READ_ONCE(po->pressure) != pressure)
1278 WRITE_ONCE(po->pressure, pressure);
1283 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1285 if (READ_ONCE(po->pressure) &&
1286 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1287 WRITE_ONCE(po->pressure, 0);
1290 static void packet_sock_destruct(struct sock *sk)
1292 skb_queue_purge(&sk->sk_error_queue);
1294 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1295 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1297 if (!sock_flag(sk, SOCK_DEAD)) {
1298 pr_err("Attempt to release alive packet socket: %p\n", sk);
1302 sk_refcnt_debug_dec(sk);
1305 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1307 u32 *history = po->rollover->history;
1311 rxhash = skb_get_hash(skb);
1312 for (i = 0; i < ROLLOVER_HLEN; i++)
1313 if (READ_ONCE(history[i]) == rxhash)
1316 victim = prandom_u32() % ROLLOVER_HLEN;
1318 /* Avoid dirtying the cache line if possible */
1319 if (READ_ONCE(history[victim]) != rxhash)
1320 WRITE_ONCE(history[victim], rxhash);
1322 return count > (ROLLOVER_HLEN >> 1);
1325 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1326 struct sk_buff *skb,
1329 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1332 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1333 struct sk_buff *skb,
1336 unsigned int val = atomic_inc_return(&f->rr_cur);
1341 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1342 struct sk_buff *skb,
1345 return smp_processor_id() % num;
1348 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1349 struct sk_buff *skb,
1352 return prandom_u32_max(num);
1355 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1356 struct sk_buff *skb,
1357 unsigned int idx, bool try_self,
1360 struct packet_sock *po, *po_next, *po_skip = NULL;
1361 unsigned int i, j, room = ROOM_NONE;
1363 po = pkt_sk(rcu_dereference(f->arr[idx]));
1366 room = packet_rcv_has_room(po, skb);
1367 if (room == ROOM_NORMAL ||
1368 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1373 i = j = min_t(int, po->rollover->sock, num - 1);
1375 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1376 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1377 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1379 po->rollover->sock = i;
1380 atomic_long_inc(&po->rollover->num);
1381 if (room == ROOM_LOW)
1382 atomic_long_inc(&po->rollover->num_huge);
1390 atomic_long_inc(&po->rollover->num_failed);
1394 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1395 struct sk_buff *skb,
1398 return skb_get_queue_mapping(skb) % num;
1401 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1402 struct sk_buff *skb,
1405 struct bpf_prog *prog;
1406 unsigned int ret = 0;
1409 prog = rcu_dereference(f->bpf_prog);
1411 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1417 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1419 return f->flags & (flag >> 8);
1422 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1423 struct packet_type *pt, struct net_device *orig_dev)
1425 struct packet_fanout *f = pt->af_packet_priv;
1426 unsigned int num = READ_ONCE(f->num_members);
1427 struct net *net = read_pnet(&f->net);
1428 struct packet_sock *po;
1431 if (!net_eq(dev_net(dev), net) || !num) {
1436 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1437 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1442 case PACKET_FANOUT_HASH:
1444 idx = fanout_demux_hash(f, skb, num);
1446 case PACKET_FANOUT_LB:
1447 idx = fanout_demux_lb(f, skb, num);
1449 case PACKET_FANOUT_CPU:
1450 idx = fanout_demux_cpu(f, skb, num);
1452 case PACKET_FANOUT_RND:
1453 idx = fanout_demux_rnd(f, skb, num);
1455 case PACKET_FANOUT_QM:
1456 idx = fanout_demux_qm(f, skb, num);
1458 case PACKET_FANOUT_ROLLOVER:
1459 idx = fanout_demux_rollover(f, skb, 0, false, num);
1461 case PACKET_FANOUT_CBPF:
1462 case PACKET_FANOUT_EBPF:
1463 idx = fanout_demux_bpf(f, skb, num);
1467 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1468 idx = fanout_demux_rollover(f, skb, idx, true, num);
1470 po = pkt_sk(rcu_dereference(f->arr[idx]));
1471 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1474 DEFINE_MUTEX(fanout_mutex);
1475 EXPORT_SYMBOL_GPL(fanout_mutex);
1476 static LIST_HEAD(fanout_list);
1477 static u16 fanout_next_id;
1479 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1481 struct packet_fanout *f = po->fanout;
1483 spin_lock(&f->lock);
1484 rcu_assign_pointer(f->arr[f->num_members], sk);
1487 if (f->num_members == 1)
1488 dev_add_pack(&f->prot_hook);
1489 spin_unlock(&f->lock);
1492 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1494 struct packet_fanout *f = po->fanout;
1497 spin_lock(&f->lock);
1498 for (i = 0; i < f->num_members; i++) {
1499 if (rcu_dereference_protected(f->arr[i],
1500 lockdep_is_held(&f->lock)) == sk)
1503 BUG_ON(i >= f->num_members);
1504 rcu_assign_pointer(f->arr[i],
1505 rcu_dereference_protected(f->arr[f->num_members - 1],
1506 lockdep_is_held(&f->lock)));
1508 if (f->num_members == 0)
1509 __dev_remove_pack(&f->prot_hook);
1510 spin_unlock(&f->lock);
1513 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1515 if (sk->sk_family != PF_PACKET)
1518 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1521 static void fanout_init_data(struct packet_fanout *f)
1524 case PACKET_FANOUT_LB:
1525 atomic_set(&f->rr_cur, 0);
1527 case PACKET_FANOUT_CBPF:
1528 case PACKET_FANOUT_EBPF:
1529 RCU_INIT_POINTER(f->bpf_prog, NULL);
1534 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1536 struct bpf_prog *old;
1538 spin_lock(&f->lock);
1539 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1540 rcu_assign_pointer(f->bpf_prog, new);
1541 spin_unlock(&f->lock);
1545 bpf_prog_destroy(old);
1549 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1552 struct bpf_prog *new;
1553 struct sock_fprog fprog;
1556 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1559 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1563 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1567 __fanout_set_data_bpf(po->fanout, new);
1571 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1574 struct bpf_prog *new;
1577 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1579 if (len != sizeof(fd))
1581 if (copy_from_sockptr(&fd, data, len))
1584 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1586 return PTR_ERR(new);
1588 __fanout_set_data_bpf(po->fanout, new);
1592 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1595 switch (po->fanout->type) {
1596 case PACKET_FANOUT_CBPF:
1597 return fanout_set_data_cbpf(po, data, len);
1598 case PACKET_FANOUT_EBPF:
1599 return fanout_set_data_ebpf(po, data, len);
1605 static void fanout_release_data(struct packet_fanout *f)
1608 case PACKET_FANOUT_CBPF:
1609 case PACKET_FANOUT_EBPF:
1610 __fanout_set_data_bpf(f, NULL);
1614 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1616 struct packet_fanout *f;
1618 list_for_each_entry(f, &fanout_list, list) {
1619 if (f->id == candidate_id &&
1620 read_pnet(&f->net) == sock_net(sk)) {
1627 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1629 u16 id = fanout_next_id;
1632 if (__fanout_id_is_free(sk, id)) {
1634 fanout_next_id = id + 1;
1639 } while (id != fanout_next_id);
1644 static int fanout_add(struct sock *sk, struct fanout_args *args)
1646 struct packet_rollover *rollover = NULL;
1647 struct packet_sock *po = pkt_sk(sk);
1648 u16 type_flags = args->type_flags;
1649 struct packet_fanout *f, *match;
1650 u8 type = type_flags & 0xff;
1651 u8 flags = type_flags >> 8;
1656 case PACKET_FANOUT_ROLLOVER:
1657 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1659 case PACKET_FANOUT_HASH:
1660 case PACKET_FANOUT_LB:
1661 case PACKET_FANOUT_CPU:
1662 case PACKET_FANOUT_RND:
1663 case PACKET_FANOUT_QM:
1664 case PACKET_FANOUT_CBPF:
1665 case PACKET_FANOUT_EBPF:
1671 mutex_lock(&fanout_mutex);
1677 if (type == PACKET_FANOUT_ROLLOVER ||
1678 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1680 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1683 atomic_long_set(&rollover->num, 0);
1684 atomic_long_set(&rollover->num_huge, 0);
1685 atomic_long_set(&rollover->num_failed, 0);
1688 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1693 if (!fanout_find_new_id(sk, &id)) {
1697 /* ephemeral flag for the first socket in the group: drop it */
1698 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1702 list_for_each_entry(f, &fanout_list, list) {
1704 read_pnet(&f->net) == sock_net(sk)) {
1711 if (match->flags != flags)
1713 if (args->max_num_members &&
1714 args->max_num_members != match->max_num_members)
1717 if (args->max_num_members > PACKET_FANOUT_MAX)
1719 if (!args->max_num_members)
1720 /* legacy PACKET_FANOUT_MAX */
1721 args->max_num_members = 256;
1723 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1727 write_pnet(&match->net, sock_net(sk));
1730 match->flags = flags;
1731 INIT_LIST_HEAD(&match->list);
1732 spin_lock_init(&match->lock);
1733 refcount_set(&match->sk_ref, 0);
1734 fanout_init_data(match);
1735 match->prot_hook.type = po->prot_hook.type;
1736 match->prot_hook.dev = po->prot_hook.dev;
1737 match->prot_hook.func = packet_rcv_fanout;
1738 match->prot_hook.af_packet_priv = match;
1739 match->prot_hook.id_match = match_fanout_group;
1740 match->max_num_members = args->max_num_members;
1741 list_add(&match->list, &fanout_list);
1745 spin_lock(&po->bind_lock);
1747 match->type == type &&
1748 match->prot_hook.type == po->prot_hook.type &&
1749 match->prot_hook.dev == po->prot_hook.dev) {
1751 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1752 __dev_remove_pack(&po->prot_hook);
1754 po->rollover = rollover;
1756 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1757 __fanout_link(sk, po);
1761 spin_unlock(&po->bind_lock);
1763 if (err && !refcount_read(&match->sk_ref)) {
1764 list_del(&match->list);
1770 mutex_unlock(&fanout_mutex);
1774 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1775 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1776 * It is the responsibility of the caller to call fanout_release_data() and
1777 * free the returned packet_fanout (after synchronize_net())
1779 static struct packet_fanout *fanout_release(struct sock *sk)
1781 struct packet_sock *po = pkt_sk(sk);
1782 struct packet_fanout *f;
1784 mutex_lock(&fanout_mutex);
1789 if (refcount_dec_and_test(&f->sk_ref))
1794 mutex_unlock(&fanout_mutex);
1799 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1800 struct sk_buff *skb)
1802 /* Earlier code assumed this would be a VLAN pkt, double-check
1803 * this now that we have the actual packet in hand. We can only
1804 * do this check on Ethernet devices.
1806 if (unlikely(dev->type != ARPHRD_ETHER))
1809 skb_reset_mac_header(skb);
1810 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1813 static const struct proto_ops packet_ops;
1815 static const struct proto_ops packet_ops_spkt;
1817 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1818 struct packet_type *pt, struct net_device *orig_dev)
1821 struct sockaddr_pkt *spkt;
1824 * When we registered the protocol we saved the socket in the data
1825 * field for just this event.
1828 sk = pt->af_packet_priv;
1831 * Yank back the headers [hope the device set this
1832 * right or kerboom...]
1834 * Incoming packets have ll header pulled,
1837 * For outgoing ones skb->data == skb_mac_header(skb)
1838 * so that this procedure is noop.
1841 if (skb->pkt_type == PACKET_LOOPBACK)
1844 if (!net_eq(dev_net(dev), sock_net(sk)))
1847 skb = skb_share_check(skb, GFP_ATOMIC);
1851 /* drop any routing info */
1854 /* drop conntrack reference */
1857 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1859 skb_push(skb, skb->data - skb_mac_header(skb));
1862 * The SOCK_PACKET socket receives _all_ frames.
1865 spkt->spkt_family = dev->type;
1866 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1867 spkt->spkt_protocol = skb->protocol;
1870 * Charge the memory to the socket. This is done specifically
1871 * to prevent sockets using all the memory up.
1874 if (sock_queue_rcv_skb(sk, skb) == 0)
1883 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1885 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1886 sock->type == SOCK_RAW) {
1887 skb_reset_mac_header(skb);
1888 skb->protocol = dev_parse_header_protocol(skb);
1891 skb_probe_transport_header(skb);
1895 * Output a raw packet to a device layer. This bypasses all the other
1896 * protocol layers and you must therefore supply it with a complete frame
1899 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1902 struct sock *sk = sock->sk;
1903 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1904 struct sk_buff *skb = NULL;
1905 struct net_device *dev;
1906 struct sockcm_cookie sockc;
1912 * Get and verify the address.
1916 if (msg->msg_namelen < sizeof(struct sockaddr))
1918 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1919 proto = saddr->spkt_protocol;
1921 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1924 * Find the device first to size check it
1927 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1930 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1936 if (!(dev->flags & IFF_UP))
1940 * You may not queue a frame bigger than the mtu. This is the lowest level
1941 * raw protocol and you must do your own fragmentation at this level.
1944 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1945 if (!netif_supports_nofcs(dev)) {
1946 err = -EPROTONOSUPPORT;
1949 extra_len = 4; /* We're doing our own CRC */
1953 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1957 size_t reserved = LL_RESERVED_SPACE(dev);
1958 int tlen = dev->needed_tailroom;
1959 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1962 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1965 /* FIXME: Save some space for broken drivers that write a hard
1966 * header at transmission time by themselves. PPP is the notable
1967 * one here. This should really be fixed at the driver level.
1969 skb_reserve(skb, reserved);
1970 skb_reset_network_header(skb);
1972 /* Try to align data part correctly */
1977 skb_reset_network_header(skb);
1979 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1985 if (!dev_validate_header(dev, skb->data, len)) {
1989 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1990 !packet_extra_vlan_len_allowed(dev, skb)) {
1995 sockcm_init(&sockc, sk);
1996 if (msg->msg_controllen) {
1997 err = sock_cmsg_send(sk, msg, &sockc);
2002 skb->protocol = proto;
2004 skb->priority = sk->sk_priority;
2005 skb->mark = sk->sk_mark;
2006 skb->tstamp = sockc.transmit_time;
2008 skb_setup_tx_timestamp(skb, sockc.tsflags);
2010 if (unlikely(extra_len == 4))
2013 packet_parse_headers(skb, sock);
2015 dev_queue_xmit(skb);
2026 static unsigned int run_filter(struct sk_buff *skb,
2027 const struct sock *sk,
2030 struct sk_filter *filter;
2033 filter = rcu_dereference(sk->sk_filter);
2035 res = bpf_prog_run_clear_cb(filter->prog, skb);
2041 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2044 struct virtio_net_hdr vnet_hdr;
2046 if (*len < sizeof(vnet_hdr))
2048 *len -= sizeof(vnet_hdr);
2050 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2053 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2057 * This function makes lazy skb cloning in hope that most of packets
2058 * are discarded by BPF.
2060 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2061 * and skb->cb are mangled. It works because (and until) packets
2062 * falling here are owned by current CPU. Output packets are cloned
2063 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2064 * sequentially, so that if we return skb to original state on exit,
2065 * we will not harm anyone.
2068 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2069 struct packet_type *pt, struct net_device *orig_dev)
2072 struct sockaddr_ll *sll;
2073 struct packet_sock *po;
2074 u8 *skb_head = skb->data;
2075 int skb_len = skb->len;
2076 unsigned int snaplen, res;
2077 bool is_drop_n_account = false;
2079 if (skb->pkt_type == PACKET_LOOPBACK)
2082 sk = pt->af_packet_priv;
2085 if (!net_eq(dev_net(dev), sock_net(sk)))
2090 if (dev_has_header(dev)) {
2091 /* The device has an explicit notion of ll header,
2092 * exported to higher levels.
2094 * Otherwise, the device hides details of its frame
2095 * structure, so that corresponding packet head is
2096 * never delivered to user.
2098 if (sk->sk_type != SOCK_DGRAM)
2099 skb_push(skb, skb->data - skb_mac_header(skb));
2100 else if (skb->pkt_type == PACKET_OUTGOING) {
2101 /* Special case: outgoing packets have ll header at head */
2102 skb_pull(skb, skb_network_offset(skb));
2108 res = run_filter(skb, sk, snaplen);
2110 goto drop_n_restore;
2114 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2117 if (skb_shared(skb)) {
2118 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2122 if (skb_head != skb->data) {
2123 skb->data = skb_head;
2130 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2132 sll = &PACKET_SKB_CB(skb)->sa.ll;
2133 sll->sll_hatype = dev->type;
2134 sll->sll_pkttype = skb->pkt_type;
2135 if (unlikely(po->origdev))
2136 sll->sll_ifindex = orig_dev->ifindex;
2138 sll->sll_ifindex = dev->ifindex;
2140 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2142 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2143 * Use their space for storing the original skb length.
2145 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2147 if (pskb_trim(skb, snaplen))
2150 skb_set_owner_r(skb, sk);
2154 /* drop conntrack reference */
2157 spin_lock(&sk->sk_receive_queue.lock);
2158 po->stats.stats1.tp_packets++;
2159 sock_skb_set_dropcount(sk, skb);
2160 __skb_queue_tail(&sk->sk_receive_queue, skb);
2161 spin_unlock(&sk->sk_receive_queue.lock);
2162 sk->sk_data_ready(sk);
2166 is_drop_n_account = true;
2167 atomic_inc(&po->tp_drops);
2168 atomic_inc(&sk->sk_drops);
2171 if (skb_head != skb->data && skb_shared(skb)) {
2172 skb->data = skb_head;
2176 if (!is_drop_n_account)
2183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2184 struct packet_type *pt, struct net_device *orig_dev)
2187 struct packet_sock *po;
2188 struct sockaddr_ll *sll;
2189 union tpacket_uhdr h;
2190 u8 *skb_head = skb->data;
2191 int skb_len = skb->len;
2192 unsigned int snaplen, res;
2193 unsigned long status = TP_STATUS_USER;
2194 unsigned short macoff, hdrlen;
2195 unsigned int netoff;
2196 struct sk_buff *copy_skb = NULL;
2197 struct timespec64 ts;
2199 bool is_drop_n_account = false;
2200 unsigned int slot_id = 0;
2201 bool do_vnet = false;
2203 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2204 * We may add members to them until current aligned size without forcing
2205 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2207 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2208 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2210 if (skb->pkt_type == PACKET_LOOPBACK)
2213 sk = pt->af_packet_priv;
2216 if (!net_eq(dev_net(dev), sock_net(sk)))
2219 if (dev_has_header(dev)) {
2220 if (sk->sk_type != SOCK_DGRAM)
2221 skb_push(skb, skb->data - skb_mac_header(skb));
2222 else if (skb->pkt_type == PACKET_OUTGOING) {
2223 /* Special case: outgoing packets have ll header at head */
2224 skb_pull(skb, skb_network_offset(skb));
2230 res = run_filter(skb, sk, snaplen);
2232 goto drop_n_restore;
2234 /* If we are flooded, just give up */
2235 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2236 atomic_inc(&po->tp_drops);
2237 goto drop_n_restore;
2240 if (skb->ip_summed == CHECKSUM_PARTIAL)
2241 status |= TP_STATUS_CSUMNOTREADY;
2242 else if (skb->pkt_type != PACKET_OUTGOING &&
2243 (skb->ip_summed == CHECKSUM_COMPLETE ||
2244 skb_csum_unnecessary(skb)))
2245 status |= TP_STATUS_CSUM_VALID;
2250 if (sk->sk_type == SOCK_DGRAM) {
2251 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2254 unsigned int maclen = skb_network_offset(skb);
2255 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2256 (maclen < 16 ? 16 : maclen)) +
2258 if (po->has_vnet_hdr) {
2259 netoff += sizeof(struct virtio_net_hdr);
2262 macoff = netoff - maclen;
2264 if (netoff > USHRT_MAX) {
2265 atomic_inc(&po->tp_drops);
2266 goto drop_n_restore;
2268 if (po->tp_version <= TPACKET_V2) {
2269 if (macoff + snaplen > po->rx_ring.frame_size) {
2270 if (po->copy_thresh &&
2271 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2272 if (skb_shared(skb)) {
2273 copy_skb = skb_clone(skb, GFP_ATOMIC);
2275 copy_skb = skb_get(skb);
2276 skb_head = skb->data;
2279 skb_set_owner_r(copy_skb, sk);
2281 snaplen = po->rx_ring.frame_size - macoff;
2282 if ((int)snaplen < 0) {
2287 } else if (unlikely(macoff + snaplen >
2288 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2291 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2292 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2293 snaplen, nval, macoff);
2295 if (unlikely((int)snaplen < 0)) {
2297 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2301 spin_lock(&sk->sk_receive_queue.lock);
2302 h.raw = packet_current_rx_frame(po, skb,
2303 TP_STATUS_KERNEL, (macoff+snaplen));
2305 goto drop_n_account;
2307 if (po->tp_version <= TPACKET_V2) {
2308 slot_id = po->rx_ring.head;
2309 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2310 goto drop_n_account;
2311 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2315 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2316 sizeof(struct virtio_net_hdr),
2317 vio_le(), true, 0)) {
2318 if (po->tp_version == TPACKET_V3)
2319 prb_clear_blk_fill_status(&po->rx_ring);
2320 goto drop_n_account;
2323 if (po->tp_version <= TPACKET_V2) {
2324 packet_increment_rx_head(po, &po->rx_ring);
2326 * LOSING will be reported till you read the stats,
2327 * because it's COR - Clear On Read.
2328 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2331 if (atomic_read(&po->tp_drops))
2332 status |= TP_STATUS_LOSING;
2335 po->stats.stats1.tp_packets++;
2337 status |= TP_STATUS_COPY;
2338 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2340 spin_unlock(&sk->sk_receive_queue.lock);
2342 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2344 /* Always timestamp; prefer an existing software timestamp taken
2345 * closer to the time of capture.
2347 ts_status = tpacket_get_timestamp(skb, &ts,
2348 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2350 ktime_get_real_ts64(&ts);
2352 status |= ts_status;
2354 switch (po->tp_version) {
2356 h.h1->tp_len = skb->len;
2357 h.h1->tp_snaplen = snaplen;
2358 h.h1->tp_mac = macoff;
2359 h.h1->tp_net = netoff;
2360 h.h1->tp_sec = ts.tv_sec;
2361 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2362 hdrlen = sizeof(*h.h1);
2365 h.h2->tp_len = skb->len;
2366 h.h2->tp_snaplen = snaplen;
2367 h.h2->tp_mac = macoff;
2368 h.h2->tp_net = netoff;
2369 h.h2->tp_sec = ts.tv_sec;
2370 h.h2->tp_nsec = ts.tv_nsec;
2371 if (skb_vlan_tag_present(skb)) {
2372 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2373 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2374 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2376 h.h2->tp_vlan_tci = 0;
2377 h.h2->tp_vlan_tpid = 0;
2379 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2380 hdrlen = sizeof(*h.h2);
2383 /* tp_nxt_offset,vlan are already populated above.
2384 * So DONT clear those fields here
2386 h.h3->tp_status |= status;
2387 h.h3->tp_len = skb->len;
2388 h.h3->tp_snaplen = snaplen;
2389 h.h3->tp_mac = macoff;
2390 h.h3->tp_net = netoff;
2391 h.h3->tp_sec = ts.tv_sec;
2392 h.h3->tp_nsec = ts.tv_nsec;
2393 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2394 hdrlen = sizeof(*h.h3);
2400 sll = h.raw + TPACKET_ALIGN(hdrlen);
2401 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2402 sll->sll_family = AF_PACKET;
2403 sll->sll_hatype = dev->type;
2404 sll->sll_protocol = skb->protocol;
2405 sll->sll_pkttype = skb->pkt_type;
2406 if (unlikely(po->origdev))
2407 sll->sll_ifindex = orig_dev->ifindex;
2409 sll->sll_ifindex = dev->ifindex;
2413 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2414 if (po->tp_version <= TPACKET_V2) {
2417 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2420 for (start = h.raw; start < end; start += PAGE_SIZE)
2421 flush_dcache_page(pgv_to_page(start));
2426 if (po->tp_version <= TPACKET_V2) {
2427 spin_lock(&sk->sk_receive_queue.lock);
2428 __packet_set_status(po, h.raw, status);
2429 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2430 spin_unlock(&sk->sk_receive_queue.lock);
2431 sk->sk_data_ready(sk);
2432 } else if (po->tp_version == TPACKET_V3) {
2433 prb_clear_blk_fill_status(&po->rx_ring);
2437 if (skb_head != skb->data && skb_shared(skb)) {
2438 skb->data = skb_head;
2442 if (!is_drop_n_account)
2449 spin_unlock(&sk->sk_receive_queue.lock);
2450 atomic_inc(&po->tp_drops);
2451 is_drop_n_account = true;
2453 sk->sk_data_ready(sk);
2454 kfree_skb(copy_skb);
2455 goto drop_n_restore;
2458 static void tpacket_destruct_skb(struct sk_buff *skb)
2460 struct packet_sock *po = pkt_sk(skb->sk);
2462 if (likely(po->tx_ring.pg_vec)) {
2466 ph = skb_zcopy_get_nouarg(skb);
2467 packet_dec_pending(&po->tx_ring);
2469 ts = __packet_set_timestamp(po, ph, skb);
2470 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2472 if (!packet_read_pending(&po->tx_ring))
2473 complete(&po->skb_completion);
2479 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2481 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2482 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2483 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2484 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2485 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2486 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2487 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2489 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2495 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2496 struct virtio_net_hdr *vnet_hdr)
2498 if (*len < sizeof(*vnet_hdr))
2500 *len -= sizeof(*vnet_hdr);
2502 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2505 return __packet_snd_vnet_parse(vnet_hdr, *len);
2508 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2509 void *frame, struct net_device *dev, void *data, int tp_len,
2510 __be16 proto, unsigned char *addr, int hlen, int copylen,
2511 const struct sockcm_cookie *sockc)
2513 union tpacket_uhdr ph;
2514 int to_write, offset, len, nr_frags, len_max;
2515 struct socket *sock = po->sk.sk_socket;
2521 skb->protocol = proto;
2523 skb->priority = po->sk.sk_priority;
2524 skb->mark = po->sk.sk_mark;
2525 skb->tstamp = sockc->transmit_time;
2526 skb_setup_tx_timestamp(skb, sockc->tsflags);
2527 skb_zcopy_set_nouarg(skb, ph.raw);
2529 skb_reserve(skb, hlen);
2530 skb_reset_network_header(skb);
2534 if (sock->type == SOCK_DGRAM) {
2535 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2537 if (unlikely(err < 0))
2539 } else if (copylen) {
2540 int hdrlen = min_t(int, copylen, tp_len);
2542 skb_push(skb, dev->hard_header_len);
2543 skb_put(skb, copylen - dev->hard_header_len);
2544 err = skb_store_bits(skb, 0, data, hdrlen);
2547 if (!dev_validate_header(dev, skb->data, hdrlen))
2554 offset = offset_in_page(data);
2555 len_max = PAGE_SIZE - offset;
2556 len = ((to_write > len_max) ? len_max : to_write);
2558 skb->data_len = to_write;
2559 skb->len += to_write;
2560 skb->truesize += to_write;
2561 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2563 while (likely(to_write)) {
2564 nr_frags = skb_shinfo(skb)->nr_frags;
2566 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2567 pr_err("Packet exceed the number of skb frags(%lu)\n",
2572 page = pgv_to_page(data);
2574 flush_dcache_page(page);
2576 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2579 len_max = PAGE_SIZE;
2580 len = ((to_write > len_max) ? len_max : to_write);
2583 packet_parse_headers(skb, sock);
2588 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2589 int size_max, void **data)
2591 union tpacket_uhdr ph;
2596 switch (po->tp_version) {
2598 if (ph.h3->tp_next_offset != 0) {
2599 pr_warn_once("variable sized slot not supported");
2602 tp_len = ph.h3->tp_len;
2605 tp_len = ph.h2->tp_len;
2608 tp_len = ph.h1->tp_len;
2611 if (unlikely(tp_len > size_max)) {
2612 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2616 if (unlikely(po->tp_tx_has_off)) {
2617 int off_min, off_max;
2619 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2620 off_max = po->tx_ring.frame_size - tp_len;
2621 if (po->sk.sk_type == SOCK_DGRAM) {
2622 switch (po->tp_version) {
2624 off = ph.h3->tp_net;
2627 off = ph.h2->tp_net;
2630 off = ph.h1->tp_net;
2634 switch (po->tp_version) {
2636 off = ph.h3->tp_mac;
2639 off = ph.h2->tp_mac;
2642 off = ph.h1->tp_mac;
2646 if (unlikely((off < off_min) || (off_max < off)))
2649 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2652 *data = frame + off;
2656 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2658 struct sk_buff *skb = NULL;
2659 struct net_device *dev;
2660 struct virtio_net_hdr *vnet_hdr = NULL;
2661 struct sockcm_cookie sockc;
2663 int err, reserve = 0;
2665 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2666 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2667 unsigned char *addr = NULL;
2668 int tp_len, size_max;
2671 int status = TP_STATUS_AVAILABLE;
2672 int hlen, tlen, copylen = 0;
2675 mutex_lock(&po->pg_vec_lock);
2677 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2678 * we need to confirm it under protection of pg_vec_lock.
2680 if (unlikely(!po->tx_ring.pg_vec)) {
2684 if (likely(saddr == NULL)) {
2685 dev = packet_cached_dev_get(po);
2686 proto = READ_ONCE(po->num);
2689 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2691 if (msg->msg_namelen < (saddr->sll_halen
2692 + offsetof(struct sockaddr_ll,
2695 proto = saddr->sll_protocol;
2696 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2697 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2698 if (dev && msg->msg_namelen < dev->addr_len +
2699 offsetof(struct sockaddr_ll, sll_addr))
2701 addr = saddr->sll_addr;
2706 if (unlikely(dev == NULL))
2709 if (unlikely(!(dev->flags & IFF_UP)))
2712 sockcm_init(&sockc, &po->sk);
2713 if (msg->msg_controllen) {
2714 err = sock_cmsg_send(&po->sk, msg, &sockc);
2719 if (po->sk.sk_socket->type == SOCK_RAW)
2720 reserve = dev->hard_header_len;
2721 size_max = po->tx_ring.frame_size
2722 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2724 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2725 size_max = dev->mtu + reserve + VLAN_HLEN;
2727 reinit_completion(&po->skb_completion);
2730 ph = packet_current_frame(po, &po->tx_ring,
2731 TP_STATUS_SEND_REQUEST);
2732 if (unlikely(ph == NULL)) {
2733 if (need_wait && skb) {
2734 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2735 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2737 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2741 /* check for additional frames */
2746 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2750 status = TP_STATUS_SEND_REQUEST;
2751 hlen = LL_RESERVED_SPACE(dev);
2752 tlen = dev->needed_tailroom;
2753 if (po->has_vnet_hdr) {
2755 data += sizeof(*vnet_hdr);
2756 tp_len -= sizeof(*vnet_hdr);
2758 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2762 copylen = __virtio16_to_cpu(vio_le(),
2765 copylen = max_t(int, copylen, dev->hard_header_len);
2766 skb = sock_alloc_send_skb(&po->sk,
2767 hlen + tlen + sizeof(struct sockaddr_ll) +
2768 (copylen - dev->hard_header_len),
2771 if (unlikely(skb == NULL)) {
2772 /* we assume the socket was initially writeable ... */
2773 if (likely(len_sum > 0))
2777 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2778 addr, hlen, copylen, &sockc);
2779 if (likely(tp_len >= 0) &&
2780 tp_len > dev->mtu + reserve &&
2781 !po->has_vnet_hdr &&
2782 !packet_extra_vlan_len_allowed(dev, skb))
2785 if (unlikely(tp_len < 0)) {
2788 __packet_set_status(po, ph,
2789 TP_STATUS_AVAILABLE);
2790 packet_increment_head(&po->tx_ring);
2794 status = TP_STATUS_WRONG_FORMAT;
2800 if (po->has_vnet_hdr) {
2801 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2805 virtio_net_hdr_set_proto(skb, vnet_hdr);
2808 skb->destructor = tpacket_destruct_skb;
2809 __packet_set_status(po, ph, TP_STATUS_SENDING);
2810 packet_inc_pending(&po->tx_ring);
2812 status = TP_STATUS_SEND_REQUEST;
2813 err = po->xmit(skb);
2814 if (unlikely(err > 0)) {
2815 err = net_xmit_errno(err);
2816 if (err && __packet_get_status(po, ph) ==
2817 TP_STATUS_AVAILABLE) {
2818 /* skb was destructed already */
2823 * skb was dropped but not destructed yet;
2824 * let's treat it like congestion or err < 0
2828 packet_increment_head(&po->tx_ring);
2830 } while (likely((ph != NULL) ||
2831 /* Note: packet_read_pending() might be slow if we have
2832 * to call it as it's per_cpu variable, but in fast-path
2833 * we already short-circuit the loop with the first
2834 * condition, and luckily don't have to go that path
2837 (need_wait && packet_read_pending(&po->tx_ring))));
2843 __packet_set_status(po, ph, status);
2848 mutex_unlock(&po->pg_vec_lock);
2852 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2853 size_t reserve, size_t len,
2854 size_t linear, int noblock,
2857 struct sk_buff *skb;
2859 /* Under a page? Don't bother with paged skb. */
2860 if (prepad + len < PAGE_SIZE || !linear)
2863 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2868 skb_reserve(skb, reserve);
2869 skb_put(skb, linear);
2870 skb->data_len = len - linear;
2871 skb->len += len - linear;
2876 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2878 struct sock *sk = sock->sk;
2879 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2880 struct sk_buff *skb;
2881 struct net_device *dev;
2883 unsigned char *addr = NULL;
2884 int err, reserve = 0;
2885 struct sockcm_cookie sockc;
2886 struct virtio_net_hdr vnet_hdr = { 0 };
2888 struct packet_sock *po = pkt_sk(sk);
2889 bool has_vnet_hdr = false;
2890 int hlen, tlen, linear;
2894 * Get and verify the address.
2897 if (likely(saddr == NULL)) {
2898 dev = packet_cached_dev_get(po);
2899 proto = READ_ONCE(po->num);
2902 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2904 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2906 proto = saddr->sll_protocol;
2907 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2908 if (sock->type == SOCK_DGRAM) {
2909 if (dev && msg->msg_namelen < dev->addr_len +
2910 offsetof(struct sockaddr_ll, sll_addr))
2912 addr = saddr->sll_addr;
2917 if (unlikely(dev == NULL))
2920 if (unlikely(!(dev->flags & IFF_UP)))
2923 sockcm_init(&sockc, sk);
2924 sockc.mark = sk->sk_mark;
2925 if (msg->msg_controllen) {
2926 err = sock_cmsg_send(sk, msg, &sockc);
2931 if (sock->type == SOCK_RAW)
2932 reserve = dev->hard_header_len;
2933 if (po->has_vnet_hdr) {
2934 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2937 has_vnet_hdr = true;
2940 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2941 if (!netif_supports_nofcs(dev)) {
2942 err = -EPROTONOSUPPORT;
2945 extra_len = 4; /* We're doing our own CRC */
2949 if (!vnet_hdr.gso_type &&
2950 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2954 hlen = LL_RESERVED_SPACE(dev);
2955 tlen = dev->needed_tailroom;
2956 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2957 linear = max(linear, min_t(int, len, dev->hard_header_len));
2958 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2959 msg->msg_flags & MSG_DONTWAIT, &err);
2963 skb_reset_network_header(skb);
2966 if (sock->type == SOCK_DGRAM) {
2967 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2968 if (unlikely(offset < 0))
2970 } else if (reserve) {
2971 skb_reserve(skb, -reserve);
2972 if (len < reserve + sizeof(struct ipv6hdr) &&
2973 dev->min_header_len != dev->hard_header_len)
2974 skb_reset_network_header(skb);
2977 /* Returns -EFAULT on error */
2978 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2982 if (sock->type == SOCK_RAW &&
2983 !dev_validate_header(dev, skb->data, len)) {
2988 skb_setup_tx_timestamp(skb, sockc.tsflags);
2990 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2991 !packet_extra_vlan_len_allowed(dev, skb)) {
2996 skb->protocol = proto;
2998 skb->priority = sk->sk_priority;
2999 skb->mark = sockc.mark;
3000 skb->tstamp = sockc.transmit_time;
3003 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3006 len += sizeof(vnet_hdr);
3007 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3010 packet_parse_headers(skb, sock);
3012 if (unlikely(extra_len == 4))
3015 err = po->xmit(skb);
3016 if (err > 0 && (err = net_xmit_errno(err)) != 0)
3032 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3034 struct sock *sk = sock->sk;
3035 struct packet_sock *po = pkt_sk(sk);
3037 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3038 * tpacket_snd() will redo the check safely.
3040 if (data_race(po->tx_ring.pg_vec))
3041 return tpacket_snd(po, msg);
3043 return packet_snd(sock, msg, len);
3047 * Close a PACKET socket. This is fairly simple. We immediately go
3048 * to 'closed' state and remove our protocol entry in the device list.
3051 static int packet_release(struct socket *sock)
3053 struct sock *sk = sock->sk;
3054 struct packet_sock *po;
3055 struct packet_fanout *f;
3057 union tpacket_req_u req_u;
3065 mutex_lock(&net->packet.sklist_lock);
3066 sk_del_node_init_rcu(sk);
3067 mutex_unlock(&net->packet.sklist_lock);
3070 sock_prot_inuse_add(net, sk->sk_prot, -1);
3073 spin_lock(&po->bind_lock);
3074 unregister_prot_hook(sk, false);
3075 packet_cached_dev_reset(po);
3077 if (po->prot_hook.dev) {
3078 dev_put(po->prot_hook.dev);
3079 po->prot_hook.dev = NULL;
3081 spin_unlock(&po->bind_lock);
3083 packet_flush_mclist(sk);
3086 if (po->rx_ring.pg_vec) {
3087 memset(&req_u, 0, sizeof(req_u));
3088 packet_set_ring(sk, &req_u, 1, 0);
3091 if (po->tx_ring.pg_vec) {
3092 memset(&req_u, 0, sizeof(req_u));
3093 packet_set_ring(sk, &req_u, 1, 1);
3097 f = fanout_release(sk);
3101 kfree(po->rollover);
3103 fanout_release_data(f);
3107 * Now the socket is dead. No more input will appear.
3114 skb_queue_purge(&sk->sk_receive_queue);
3115 packet_free_pending(po);
3116 sk_refcnt_debug_release(sk);
3123 * Attach a packet hook.
3126 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3129 struct packet_sock *po = pkt_sk(sk);
3130 struct net_device *dev_curr;
3133 struct net_device *dev = NULL;
3135 bool unlisted = false;
3138 spin_lock(&po->bind_lock);
3147 dev = dev_get_by_name_rcu(sock_net(sk), name);
3152 } else if (ifindex) {
3153 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3163 proto_curr = po->prot_hook.type;
3164 dev_curr = po->prot_hook.dev;
3166 need_rehook = proto_curr != proto || dev_curr != dev;
3171 /* prevents packet_notifier() from calling
3172 * register_prot_hook()
3174 WRITE_ONCE(po->num, 0);
3175 __unregister_prot_hook(sk, true);
3177 dev_curr = po->prot_hook.dev;
3179 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3183 BUG_ON(po->running);
3184 WRITE_ONCE(po->num, proto);
3185 po->prot_hook.type = proto;
3187 if (unlikely(unlisted)) {
3189 po->prot_hook.dev = NULL;
3190 WRITE_ONCE(po->ifindex, -1);
3191 packet_cached_dev_reset(po);
3193 po->prot_hook.dev = dev;
3194 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3195 packet_cached_dev_assign(po, dev);
3201 if (proto == 0 || !need_rehook)
3204 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3205 register_prot_hook(sk);
3207 sk->sk_err = ENETDOWN;
3208 if (!sock_flag(sk, SOCK_DEAD))
3209 sk->sk_error_report(sk);
3214 spin_unlock(&po->bind_lock);
3220 * Bind a packet socket to a device
3223 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3226 struct sock *sk = sock->sk;
3227 char name[sizeof(uaddr->sa_data) + 1];
3233 if (addr_len != sizeof(struct sockaddr))
3235 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3238 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3239 name[sizeof(uaddr->sa_data)] = 0;
3241 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3244 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3246 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3247 struct sock *sk = sock->sk;
3253 if (addr_len < sizeof(struct sockaddr_ll))
3255 if (sll->sll_family != AF_PACKET)
3258 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3259 sll->sll_protocol ? : pkt_sk(sk)->num);
3262 static struct proto packet_proto = {
3264 .owner = THIS_MODULE,
3265 .obj_size = sizeof(struct packet_sock),
3269 * Create a packet of type SOCK_PACKET.
3272 static int packet_create(struct net *net, struct socket *sock, int protocol,
3276 struct packet_sock *po;
3277 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3280 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3282 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3283 sock->type != SOCK_PACKET)
3284 return -ESOCKTNOSUPPORT;
3286 sock->state = SS_UNCONNECTED;
3289 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3293 sock->ops = &packet_ops;
3294 if (sock->type == SOCK_PACKET)
3295 sock->ops = &packet_ops_spkt;
3297 sock_init_data(sock, sk);
3300 init_completion(&po->skb_completion);
3301 sk->sk_family = PF_PACKET;
3303 po->xmit = dev_queue_xmit;
3305 err = packet_alloc_pending(po);
3309 packet_cached_dev_reset(po);
3311 sk->sk_destruct = packet_sock_destruct;
3312 sk_refcnt_debug_inc(sk);
3315 * Attach a protocol block
3318 spin_lock_init(&po->bind_lock);
3319 mutex_init(&po->pg_vec_lock);
3320 po->rollover = NULL;
3321 po->prot_hook.func = packet_rcv;
3323 if (sock->type == SOCK_PACKET)
3324 po->prot_hook.func = packet_rcv_spkt;
3326 po->prot_hook.af_packet_priv = sk;
3329 po->prot_hook.type = proto;
3330 __register_prot_hook(sk);
3333 mutex_lock(&net->packet.sklist_lock);
3334 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3335 mutex_unlock(&net->packet.sklist_lock);
3338 sock_prot_inuse_add(net, &packet_proto, 1);
3349 * Pull a packet from our receive queue and hand it to the user.
3350 * If necessary we block.
3353 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3356 struct sock *sk = sock->sk;
3357 struct sk_buff *skb;
3359 int vnet_hdr_len = 0;
3360 unsigned int origlen = 0;
3363 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3367 /* What error should we return now? EUNATTACH? */
3368 if (pkt_sk(sk)->ifindex < 0)
3372 if (flags & MSG_ERRQUEUE) {
3373 err = sock_recv_errqueue(sk, msg, len,
3374 SOL_PACKET, PACKET_TX_TIMESTAMP);
3379 * Call the generic datagram receiver. This handles all sorts
3380 * of horrible races and re-entrancy so we can forget about it
3381 * in the protocol layers.
3383 * Now it will return ENETDOWN, if device have just gone down,
3384 * but then it will block.
3387 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3390 * An error occurred so return it. Because skb_recv_datagram()
3391 * handles the blocking we don't see and worry about blocking
3398 packet_rcv_try_clear_pressure(pkt_sk(sk));
3400 if (pkt_sk(sk)->has_vnet_hdr) {
3401 err = packet_rcv_vnet(msg, skb, &len);
3404 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3407 /* You lose any data beyond the buffer you gave. If it worries
3408 * a user program they can ask the device for its MTU
3414 msg->msg_flags |= MSG_TRUNC;
3417 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3421 if (sock->type != SOCK_PACKET) {
3422 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3424 /* Original length was stored in sockaddr_ll fields */
3425 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3426 sll->sll_family = AF_PACKET;
3427 sll->sll_protocol = skb->protocol;
3430 sock_recv_ts_and_drops(msg, sk, skb);
3432 if (msg->msg_name) {
3435 /* If the address length field is there to be filled
3436 * in, we fill it in now.
3438 if (sock->type == SOCK_PACKET) {
3439 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3440 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3441 copy_len = msg->msg_namelen;
3443 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3445 msg->msg_namelen = sll->sll_halen +
3446 offsetof(struct sockaddr_ll, sll_addr);
3447 copy_len = msg->msg_namelen;
3448 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3449 memset(msg->msg_name +
3450 offsetof(struct sockaddr_ll, sll_addr),
3451 0, sizeof(sll->sll_addr));
3452 msg->msg_namelen = sizeof(struct sockaddr_ll);
3455 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3458 if (pkt_sk(sk)->auxdata) {
3459 struct tpacket_auxdata aux;
3461 aux.tp_status = TP_STATUS_USER;
3462 if (skb->ip_summed == CHECKSUM_PARTIAL)
3463 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3464 else if (skb->pkt_type != PACKET_OUTGOING &&
3465 (skb->ip_summed == CHECKSUM_COMPLETE ||
3466 skb_csum_unnecessary(skb)))
3467 aux.tp_status |= TP_STATUS_CSUM_VALID;
3469 aux.tp_len = origlen;
3470 aux.tp_snaplen = skb->len;
3472 aux.tp_net = skb_network_offset(skb);
3473 if (skb_vlan_tag_present(skb)) {
3474 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3475 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3476 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3478 aux.tp_vlan_tci = 0;
3479 aux.tp_vlan_tpid = 0;
3481 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3485 * Free or return the buffer as appropriate. Again this
3486 * hides all the races and re-entrancy issues from us.
3488 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3491 skb_free_datagram(sk, skb);
3496 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3499 struct net_device *dev;
3500 struct sock *sk = sock->sk;
3505 uaddr->sa_family = AF_PACKET;
3506 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3508 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3510 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3513 return sizeof(*uaddr);
3516 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3519 struct net_device *dev;
3520 struct sock *sk = sock->sk;
3521 struct packet_sock *po = pkt_sk(sk);
3522 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3528 ifindex = READ_ONCE(po->ifindex);
3529 sll->sll_family = AF_PACKET;
3530 sll->sll_ifindex = ifindex;
3531 sll->sll_protocol = READ_ONCE(po->num);
3532 sll->sll_pkttype = 0;
3534 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3536 sll->sll_hatype = dev->type;
3537 sll->sll_halen = dev->addr_len;
3538 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3540 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3545 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3548 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3552 case PACKET_MR_MULTICAST:
3553 if (i->alen != dev->addr_len)
3556 return dev_mc_add(dev, i->addr);
3558 return dev_mc_del(dev, i->addr);
3560 case PACKET_MR_PROMISC:
3561 return dev_set_promiscuity(dev, what);
3562 case PACKET_MR_ALLMULTI:
3563 return dev_set_allmulti(dev, what);
3564 case PACKET_MR_UNICAST:
3565 if (i->alen != dev->addr_len)
3568 return dev_uc_add(dev, i->addr);
3570 return dev_uc_del(dev, i->addr);
3578 static void packet_dev_mclist_delete(struct net_device *dev,
3579 struct packet_mclist **mlp)
3581 struct packet_mclist *ml;
3583 while ((ml = *mlp) != NULL) {
3584 if (ml->ifindex == dev->ifindex) {
3585 packet_dev_mc(dev, ml, -1);
3593 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3595 struct packet_sock *po = pkt_sk(sk);
3596 struct packet_mclist *ml, *i;
3597 struct net_device *dev;
3603 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3608 if (mreq->mr_alen > dev->addr_len)
3612 i = kmalloc(sizeof(*i), GFP_KERNEL);
3617 for (ml = po->mclist; ml; ml = ml->next) {
3618 if (ml->ifindex == mreq->mr_ifindex &&
3619 ml->type == mreq->mr_type &&
3620 ml->alen == mreq->mr_alen &&
3621 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3623 /* Free the new element ... */
3629 i->type = mreq->mr_type;
3630 i->ifindex = mreq->mr_ifindex;
3631 i->alen = mreq->mr_alen;
3632 memcpy(i->addr, mreq->mr_address, i->alen);
3633 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3635 i->next = po->mclist;
3637 err = packet_dev_mc(dev, i, 1);
3639 po->mclist = i->next;
3648 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3650 struct packet_mclist *ml, **mlp;
3654 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3655 if (ml->ifindex == mreq->mr_ifindex &&
3656 ml->type == mreq->mr_type &&
3657 ml->alen == mreq->mr_alen &&
3658 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3659 if (--ml->count == 0) {
3660 struct net_device *dev;
3662 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3664 packet_dev_mc(dev, ml, -1);
3674 static void packet_flush_mclist(struct sock *sk)
3676 struct packet_sock *po = pkt_sk(sk);
3677 struct packet_mclist *ml;
3683 while ((ml = po->mclist) != NULL) {
3684 struct net_device *dev;
3686 po->mclist = ml->next;
3687 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3689 packet_dev_mc(dev, ml, -1);
3696 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3697 unsigned int optlen)
3699 struct sock *sk = sock->sk;
3700 struct packet_sock *po = pkt_sk(sk);
3703 if (level != SOL_PACKET)
3704 return -ENOPROTOOPT;
3707 case PACKET_ADD_MEMBERSHIP:
3708 case PACKET_DROP_MEMBERSHIP:
3710 struct packet_mreq_max mreq;
3712 memset(&mreq, 0, sizeof(mreq));
3713 if (len < sizeof(struct packet_mreq))
3715 if (len > sizeof(mreq))
3717 if (copy_from_sockptr(&mreq, optval, len))
3719 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3721 if (optname == PACKET_ADD_MEMBERSHIP)
3722 ret = packet_mc_add(sk, &mreq);
3724 ret = packet_mc_drop(sk, &mreq);
3728 case PACKET_RX_RING:
3729 case PACKET_TX_RING:
3731 union tpacket_req_u req_u;
3735 switch (po->tp_version) {
3738 len = sizeof(req_u.req);
3742 len = sizeof(req_u.req3);
3748 if (copy_from_sockptr(&req_u.req, optval, len))
3751 ret = packet_set_ring(sk, &req_u, 0,
3752 optname == PACKET_TX_RING);
3757 case PACKET_COPY_THRESH:
3761 if (optlen != sizeof(val))
3763 if (copy_from_sockptr(&val, optval, sizeof(val)))
3766 pkt_sk(sk)->copy_thresh = val;
3769 case PACKET_VERSION:
3773 if (optlen != sizeof(val))
3775 if (copy_from_sockptr(&val, optval, sizeof(val)))
3786 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3789 po->tp_version = val;
3795 case PACKET_RESERVE:
3799 if (optlen != sizeof(val))
3801 if (copy_from_sockptr(&val, optval, sizeof(val)))
3806 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3809 po->tp_reserve = val;
3819 if (optlen != sizeof(val))
3821 if (copy_from_sockptr(&val, optval, sizeof(val)))
3825 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3828 po->tp_loss = !!val;
3834 case PACKET_AUXDATA:
3838 if (optlen < sizeof(val))
3840 if (copy_from_sockptr(&val, optval, sizeof(val)))
3844 po->auxdata = !!val;
3848 case PACKET_ORIGDEV:
3852 if (optlen < sizeof(val))
3854 if (copy_from_sockptr(&val, optval, sizeof(val)))
3858 po->origdev = !!val;
3862 case PACKET_VNET_HDR:
3866 if (sock->type != SOCK_RAW)
3868 if (optlen < sizeof(val))
3870 if (copy_from_sockptr(&val, optval, sizeof(val)))
3874 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3877 po->has_vnet_hdr = !!val;
3883 case PACKET_TIMESTAMP:
3887 if (optlen != sizeof(val))
3889 if (copy_from_sockptr(&val, optval, sizeof(val)))
3892 po->tp_tstamp = val;
3897 struct fanout_args args = { 0 };
3899 if (optlen != sizeof(int) && optlen != sizeof(args))
3901 if (copy_from_sockptr(&args, optval, optlen))
3904 return fanout_add(sk, &args);
3906 case PACKET_FANOUT_DATA:
3911 return fanout_set_data(po, optval, optlen);
3913 case PACKET_IGNORE_OUTGOING:
3917 if (optlen != sizeof(val))
3919 if (copy_from_sockptr(&val, optval, sizeof(val)))
3921 if (val < 0 || val > 1)
3924 po->prot_hook.ignore_outgoing = !!val;
3927 case PACKET_TX_HAS_OFF:
3931 if (optlen != sizeof(val))
3933 if (copy_from_sockptr(&val, optval, sizeof(val)))
3937 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3940 po->tp_tx_has_off = !!val;
3946 case PACKET_QDISC_BYPASS:
3950 if (optlen != sizeof(val))
3952 if (copy_from_sockptr(&val, optval, sizeof(val)))
3955 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3959 return -ENOPROTOOPT;
3963 static int packet_getsockopt(struct socket *sock, int level, int optname,
3964 char __user *optval, int __user *optlen)
3967 int val, lv = sizeof(val);
3968 struct sock *sk = sock->sk;
3969 struct packet_sock *po = pkt_sk(sk);
3971 union tpacket_stats_u st;
3972 struct tpacket_rollover_stats rstats;
3975 if (level != SOL_PACKET)
3976 return -ENOPROTOOPT;
3978 if (get_user(len, optlen))
3985 case PACKET_STATISTICS:
3986 spin_lock_bh(&sk->sk_receive_queue.lock);
3987 memcpy(&st, &po->stats, sizeof(st));
3988 memset(&po->stats, 0, sizeof(po->stats));
3989 spin_unlock_bh(&sk->sk_receive_queue.lock);
3990 drops = atomic_xchg(&po->tp_drops, 0);
3992 if (po->tp_version == TPACKET_V3) {
3993 lv = sizeof(struct tpacket_stats_v3);
3994 st.stats3.tp_drops = drops;
3995 st.stats3.tp_packets += drops;
3998 lv = sizeof(struct tpacket_stats);
3999 st.stats1.tp_drops = drops;
4000 st.stats1.tp_packets += drops;
4005 case PACKET_AUXDATA:
4008 case PACKET_ORIGDEV:
4011 case PACKET_VNET_HDR:
4012 val = po->has_vnet_hdr;
4014 case PACKET_VERSION:
4015 val = po->tp_version;
4018 if (len > sizeof(int))
4020 if (len < sizeof(int))
4022 if (copy_from_user(&val, optval, len))
4026 val = sizeof(struct tpacket_hdr);
4029 val = sizeof(struct tpacket2_hdr);
4032 val = sizeof(struct tpacket3_hdr);
4038 case PACKET_RESERVE:
4039 val = po->tp_reserve;
4044 case PACKET_TIMESTAMP:
4045 val = po->tp_tstamp;
4049 ((u32)po->fanout->id |
4050 ((u32)po->fanout->type << 16) |
4051 ((u32)po->fanout->flags << 24)) :
4054 case PACKET_IGNORE_OUTGOING:
4055 val = po->prot_hook.ignore_outgoing;
4057 case PACKET_ROLLOVER_STATS:
4060 rstats.tp_all = atomic_long_read(&po->rollover->num);
4061 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4062 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4064 lv = sizeof(rstats);
4066 case PACKET_TX_HAS_OFF:
4067 val = po->tp_tx_has_off;
4069 case PACKET_QDISC_BYPASS:
4070 val = packet_use_direct_xmit(po);
4073 return -ENOPROTOOPT;
4078 if (put_user(len, optlen))
4080 if (copy_to_user(optval, data, len))
4085 static int packet_notifier(struct notifier_block *this,
4086 unsigned long msg, void *ptr)
4089 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4090 struct net *net = dev_net(dev);
4093 sk_for_each_rcu(sk, &net->packet.sklist) {
4094 struct packet_sock *po = pkt_sk(sk);
4097 case NETDEV_UNREGISTER:
4099 packet_dev_mclist_delete(dev, &po->mclist);
4103 if (dev->ifindex == po->ifindex) {
4104 spin_lock(&po->bind_lock);
4106 __unregister_prot_hook(sk, false);
4107 sk->sk_err = ENETDOWN;
4108 if (!sock_flag(sk, SOCK_DEAD))
4109 sk->sk_error_report(sk);
4111 if (msg == NETDEV_UNREGISTER) {
4112 packet_cached_dev_reset(po);
4113 WRITE_ONCE(po->ifindex, -1);
4114 if (po->prot_hook.dev)
4115 dev_put(po->prot_hook.dev);
4116 po->prot_hook.dev = NULL;
4118 spin_unlock(&po->bind_lock);
4122 if (dev->ifindex == po->ifindex) {
4123 spin_lock(&po->bind_lock);
4125 register_prot_hook(sk);
4126 spin_unlock(&po->bind_lock);
4136 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4139 struct sock *sk = sock->sk;
4144 int amount = sk_wmem_alloc_get(sk);
4146 return put_user(amount, (int __user *)arg);
4150 struct sk_buff *skb;
4153 spin_lock_bh(&sk->sk_receive_queue.lock);
4154 skb = skb_peek(&sk->sk_receive_queue);
4157 spin_unlock_bh(&sk->sk_receive_queue.lock);
4158 return put_user(amount, (int __user *)arg);
4168 case SIOCGIFBRDADDR:
4169 case SIOCSIFBRDADDR:
4170 case SIOCGIFNETMASK:
4171 case SIOCSIFNETMASK:
4172 case SIOCGIFDSTADDR:
4173 case SIOCSIFDSTADDR:
4175 return inet_dgram_ops.ioctl(sock, cmd, arg);
4179 return -ENOIOCTLCMD;
4184 static __poll_t packet_poll(struct file *file, struct socket *sock,
4187 struct sock *sk = sock->sk;
4188 struct packet_sock *po = pkt_sk(sk);
4189 __poll_t mask = datagram_poll(file, sock, wait);
4191 spin_lock_bh(&sk->sk_receive_queue.lock);
4192 if (po->rx_ring.pg_vec) {
4193 if (!packet_previous_rx_frame(po, &po->rx_ring,
4195 mask |= EPOLLIN | EPOLLRDNORM;
4197 packet_rcv_try_clear_pressure(po);
4198 spin_unlock_bh(&sk->sk_receive_queue.lock);
4199 spin_lock_bh(&sk->sk_write_queue.lock);
4200 if (po->tx_ring.pg_vec) {
4201 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4202 mask |= EPOLLOUT | EPOLLWRNORM;
4204 spin_unlock_bh(&sk->sk_write_queue.lock);
4209 /* Dirty? Well, I still did not learn better way to account
4213 static void packet_mm_open(struct vm_area_struct *vma)
4215 struct file *file = vma->vm_file;
4216 struct socket *sock = file->private_data;
4217 struct sock *sk = sock->sk;
4220 atomic_inc(&pkt_sk(sk)->mapped);
4223 static void packet_mm_close(struct vm_area_struct *vma)
4225 struct file *file = vma->vm_file;
4226 struct socket *sock = file->private_data;
4227 struct sock *sk = sock->sk;
4230 atomic_dec(&pkt_sk(sk)->mapped);
4233 static const struct vm_operations_struct packet_mmap_ops = {
4234 .open = packet_mm_open,
4235 .close = packet_mm_close,
4238 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4243 for (i = 0; i < len; i++) {
4244 if (likely(pg_vec[i].buffer)) {
4245 if (is_vmalloc_addr(pg_vec[i].buffer))
4246 vfree(pg_vec[i].buffer);
4248 free_pages((unsigned long)pg_vec[i].buffer,
4250 pg_vec[i].buffer = NULL;
4256 static char *alloc_one_pg_vec_page(unsigned long order)
4259 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4260 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4262 buffer = (char *) __get_free_pages(gfp_flags, order);
4266 /* __get_free_pages failed, fall back to vmalloc */
4267 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4271 /* vmalloc failed, lets dig into swap here */
4272 gfp_flags &= ~__GFP_NORETRY;
4273 buffer = (char *) __get_free_pages(gfp_flags, order);
4277 /* complete and utter failure */
4281 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4283 unsigned int block_nr = req->tp_block_nr;
4287 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4288 if (unlikely(!pg_vec))
4291 for (i = 0; i < block_nr; i++) {
4292 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4293 if (unlikely(!pg_vec[i].buffer))
4294 goto out_free_pgvec;
4301 free_pg_vec(pg_vec, order, block_nr);
4306 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4307 int closing, int tx_ring)
4309 struct pgv *pg_vec = NULL;
4310 struct packet_sock *po = pkt_sk(sk);
4311 unsigned long *rx_owner_map = NULL;
4312 int was_running, order = 0;
4313 struct packet_ring_buffer *rb;
4314 struct sk_buff_head *rb_queue;
4317 /* Added to avoid minimal code churn */
4318 struct tpacket_req *req = &req_u->req;
4320 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4321 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4325 if (atomic_read(&po->mapped))
4327 if (packet_read_pending(rb))
4331 if (req->tp_block_nr) {
4332 unsigned int min_frame_size;
4334 /* Sanity tests and some calculations */
4336 if (unlikely(rb->pg_vec))
4339 switch (po->tp_version) {
4341 po->tp_hdrlen = TPACKET_HDRLEN;
4344 po->tp_hdrlen = TPACKET2_HDRLEN;
4347 po->tp_hdrlen = TPACKET3_HDRLEN;
4352 if (unlikely((int)req->tp_block_size <= 0))
4354 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4356 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4357 if (po->tp_version >= TPACKET_V3 &&
4358 req->tp_block_size <
4359 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4361 if (unlikely(req->tp_frame_size < min_frame_size))
4363 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4366 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4367 if (unlikely(rb->frames_per_block == 0))
4369 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4371 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4376 order = get_order(req->tp_block_size);
4377 pg_vec = alloc_pg_vec(req, order);
4378 if (unlikely(!pg_vec))
4380 switch (po->tp_version) {
4382 /* Block transmit is not supported yet */
4384 init_prb_bdqc(po, rb, pg_vec, req_u);
4386 struct tpacket_req3 *req3 = &req_u->req3;
4388 if (req3->tp_retire_blk_tov ||
4389 req3->tp_sizeof_priv ||
4390 req3->tp_feature_req_word) {
4392 goto out_free_pg_vec;
4398 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4399 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4401 goto out_free_pg_vec;
4409 if (unlikely(req->tp_frame_nr))
4414 /* Detach socket from network */
4415 spin_lock(&po->bind_lock);
4416 was_running = po->running;
4419 WRITE_ONCE(po->num, 0);
4420 __unregister_prot_hook(sk, false);
4422 spin_unlock(&po->bind_lock);
4427 mutex_lock(&po->pg_vec_lock);
4428 if (closing || atomic_read(&po->mapped) == 0) {
4430 spin_lock_bh(&rb_queue->lock);
4431 swap(rb->pg_vec, pg_vec);
4432 if (po->tp_version <= TPACKET_V2)
4433 swap(rb->rx_owner_map, rx_owner_map);
4434 rb->frame_max = (req->tp_frame_nr - 1);
4436 rb->frame_size = req->tp_frame_size;
4437 spin_unlock_bh(&rb_queue->lock);
4439 swap(rb->pg_vec_order, order);
4440 swap(rb->pg_vec_len, req->tp_block_nr);
4442 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4443 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4444 tpacket_rcv : packet_rcv;
4445 skb_queue_purge(rb_queue);
4446 if (atomic_read(&po->mapped))
4447 pr_err("packet_mmap: vma is busy: %d\n",
4448 atomic_read(&po->mapped));
4450 mutex_unlock(&po->pg_vec_lock);
4452 spin_lock(&po->bind_lock);
4454 WRITE_ONCE(po->num, num);
4455 register_prot_hook(sk);
4457 spin_unlock(&po->bind_lock);
4458 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4459 /* Because we don't support block-based V3 on tx-ring */
4461 prb_shutdown_retire_blk_timer(po, rb_queue);
4465 bitmap_free(rx_owner_map);
4467 free_pg_vec(pg_vec, order, req->tp_block_nr);
4472 static int packet_mmap(struct file *file, struct socket *sock,
4473 struct vm_area_struct *vma)
4475 struct sock *sk = sock->sk;
4476 struct packet_sock *po = pkt_sk(sk);
4477 unsigned long size, expected_size;
4478 struct packet_ring_buffer *rb;
4479 unsigned long start;
4486 mutex_lock(&po->pg_vec_lock);
4489 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4491 expected_size += rb->pg_vec_len
4497 if (expected_size == 0)
4500 size = vma->vm_end - vma->vm_start;
4501 if (size != expected_size)
4504 start = vma->vm_start;
4505 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4506 if (rb->pg_vec == NULL)
4509 for (i = 0; i < rb->pg_vec_len; i++) {
4511 void *kaddr = rb->pg_vec[i].buffer;
4514 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4515 page = pgv_to_page(kaddr);
4516 err = vm_insert_page(vma, start, page);
4525 atomic_inc(&po->mapped);
4526 vma->vm_ops = &packet_mmap_ops;
4530 mutex_unlock(&po->pg_vec_lock);
4534 static const struct proto_ops packet_ops_spkt = {
4535 .family = PF_PACKET,
4536 .owner = THIS_MODULE,
4537 .release = packet_release,
4538 .bind = packet_bind_spkt,
4539 .connect = sock_no_connect,
4540 .socketpair = sock_no_socketpair,
4541 .accept = sock_no_accept,
4542 .getname = packet_getname_spkt,
4543 .poll = datagram_poll,
4544 .ioctl = packet_ioctl,
4545 .gettstamp = sock_gettstamp,
4546 .listen = sock_no_listen,
4547 .shutdown = sock_no_shutdown,
4548 .sendmsg = packet_sendmsg_spkt,
4549 .recvmsg = packet_recvmsg,
4550 .mmap = sock_no_mmap,
4551 .sendpage = sock_no_sendpage,
4554 static const struct proto_ops packet_ops = {
4555 .family = PF_PACKET,
4556 .owner = THIS_MODULE,
4557 .release = packet_release,
4558 .bind = packet_bind,
4559 .connect = sock_no_connect,
4560 .socketpair = sock_no_socketpair,
4561 .accept = sock_no_accept,
4562 .getname = packet_getname,
4563 .poll = packet_poll,
4564 .ioctl = packet_ioctl,
4565 .gettstamp = sock_gettstamp,
4566 .listen = sock_no_listen,
4567 .shutdown = sock_no_shutdown,
4568 .setsockopt = packet_setsockopt,
4569 .getsockopt = packet_getsockopt,
4570 .sendmsg = packet_sendmsg,
4571 .recvmsg = packet_recvmsg,
4572 .mmap = packet_mmap,
4573 .sendpage = sock_no_sendpage,
4576 static const struct net_proto_family packet_family_ops = {
4577 .family = PF_PACKET,
4578 .create = packet_create,
4579 .owner = THIS_MODULE,
4582 static struct notifier_block packet_netdev_notifier = {
4583 .notifier_call = packet_notifier,
4586 #ifdef CONFIG_PROC_FS
4588 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4591 struct net *net = seq_file_net(seq);
4594 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4597 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4599 struct net *net = seq_file_net(seq);
4600 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4603 static void packet_seq_stop(struct seq_file *seq, void *v)
4609 static int packet_seq_show(struct seq_file *seq, void *v)
4611 if (v == SEQ_START_TOKEN)
4613 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4614 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4616 struct sock *s = sk_entry(v);
4617 const struct packet_sock *po = pkt_sk(s);
4620 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4622 refcount_read(&s->sk_refcnt),
4624 ntohs(READ_ONCE(po->num)),
4625 READ_ONCE(po->ifindex),
4627 atomic_read(&s->sk_rmem_alloc),
4628 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4635 static const struct seq_operations packet_seq_ops = {
4636 .start = packet_seq_start,
4637 .next = packet_seq_next,
4638 .stop = packet_seq_stop,
4639 .show = packet_seq_show,
4643 static int __net_init packet_net_init(struct net *net)
4645 mutex_init(&net->packet.sklist_lock);
4646 INIT_HLIST_HEAD(&net->packet.sklist);
4648 #ifdef CONFIG_PROC_FS
4649 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4650 sizeof(struct seq_net_private)))
4652 #endif /* CONFIG_PROC_FS */
4657 static void __net_exit packet_net_exit(struct net *net)
4659 remove_proc_entry("packet", net->proc_net);
4660 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4663 static struct pernet_operations packet_net_ops = {
4664 .init = packet_net_init,
4665 .exit = packet_net_exit,
4669 static void __exit packet_exit(void)
4671 unregister_netdevice_notifier(&packet_netdev_notifier);
4672 unregister_pernet_subsys(&packet_net_ops);
4673 sock_unregister(PF_PACKET);
4674 proto_unregister(&packet_proto);
4677 static int __init packet_init(void)
4681 rc = proto_register(&packet_proto, 0);
4684 rc = sock_register(&packet_family_ops);
4687 rc = register_pernet_subsys(&packet_net_ops);
4690 rc = register_netdevice_notifier(&packet_netdev_notifier);
4697 unregister_pernet_subsys(&packet_net_ops);
4699 sock_unregister(PF_PACKET);
4701 proto_unregister(&packet_proto);
4706 module_init(packet_init);
4707 module_exit(packet_exit);
4708 MODULE_LICENSE("GPL");
4709 MODULE_ALIAS_NETPROTO(PF_PACKET);