1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #include <linux/types.h>
51 #include <linux/capability.h>
52 #include <linux/fcntl.h>
53 #include <linux/socket.h>
55 #include <linux/inet.h>
56 #include <linux/netdevice.h>
57 #include <linux/if_packet.h>
58 #include <linux/wireless.h>
59 #include <linux/kernel.h>
60 #include <linux/kmod.h>
61 #include <linux/slab.h>
62 #include <linux/vmalloc.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <linux/uaccess.h>
71 #include <asm/ioctls.h>
73 #include <asm/cacheflush.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80 #include <linux/mutex.h>
81 #include <linux/if_vlan.h>
82 #include <linux/virtio_net.h>
83 #include <linux/errqueue.h>
84 #include <linux/net_tstamp.h>
85 #include <linux/percpu.h>
87 #include <net/inet_common.h>
89 #include <linux/bpf.h>
90 #include <net/compat.h>
96 - If the device has no dev->header_ops->create, there is no LL header
97 visible above the device. In this case, its hard_header_len should be 0.
98 The device may prepend its own header internally. In this case, its
99 needed_headroom should be set to the space needed for it to add its
101 For example, a WiFi driver pretending to be an Ethernet driver should
102 set its hard_header_len to be the Ethernet header length, and set its
103 needed_headroom to be (the real WiFi header length - the fake Ethernet
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
111 Incoming, dev_has_header(dev) == true
112 mac_header -> ll header
115 Outgoing, dev_has_header(dev) == true
116 mac_header -> ll header
119 Incoming, dev_has_header(dev) == false
121 However drivers often make it point to the ll header.
122 This is incorrect because the ll header should be invisible to us.
125 Outgoing, dev_has_header(dev) == false
126 mac_header -> data. ll header is invisible to us.
130 If dev_has_header(dev) == false we are unable to restore the ll header,
131 because it is invisible to us.
137 dev->header_ops != NULL
138 mac_header -> ll header
141 dev->header_ops == NULL (ll header is invisible to us)
145 We should set network_header on output to the correct position,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
179 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
180 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
181 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
182 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
183 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
187 struct packet_type *pt, struct net_device *orig_dev);
189 static void *packet_previous_frame(struct packet_sock *po,
190 struct packet_ring_buffer *rb,
192 static void packet_increment_head(struct packet_ring_buffer *buff);
193 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
194 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
195 struct packet_sock *);
196 static void prb_retire_current_block(struct tpacket_kbdq_core *,
197 struct packet_sock *, unsigned int status);
198 static int prb_queue_frozen(struct tpacket_kbdq_core *);
199 static void prb_open_block(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
202 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
203 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
204 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
205 struct tpacket3_hdr *);
206 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
207 struct tpacket3_hdr *);
208 static void packet_flush_mclist(struct sock *sk);
209 static u16 packet_pick_tx_queue(struct sk_buff *skb);
211 struct packet_skb_cb {
213 struct sockaddr_pkt pkt;
215 /* Trick: alias skb original length with
216 * ll.sll_family and ll.protocol in order
219 unsigned int origlen;
220 struct sockaddr_ll ll;
225 #define vio_le() virtio_legacy_is_little_endian()
227 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
229 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
230 #define GET_PBLOCK_DESC(x, bid) \
231 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
232 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
233 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
234 #define GET_NEXT_PRB_BLK_NUM(x) \
235 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
236 ((x)->kactive_blk_num+1) : 0)
238 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
239 static void __fanout_link(struct sock *sk, struct packet_sock *po);
241 static int packet_direct_xmit(struct sk_buff *skb)
243 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
246 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
248 struct net_device *dev;
251 dev = rcu_dereference(po->cached_dev);
259 static void packet_cached_dev_assign(struct packet_sock *po,
260 struct net_device *dev)
262 rcu_assign_pointer(po->cached_dev, dev);
265 static void packet_cached_dev_reset(struct packet_sock *po)
267 RCU_INIT_POINTER(po->cached_dev, NULL);
270 static bool packet_use_direct_xmit(const struct packet_sock *po)
272 /* Paired with WRITE_ONCE() in packet_setsockopt() */
273 return READ_ONCE(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;
369 /* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
372 switch (po->tp_version) {
374 WRITE_ONCE(h.h1->tp_status, status);
375 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
378 WRITE_ONCE(h.h2->tp_status, status);
379 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
382 WRITE_ONCE(h.h3->tp_status, status);
383 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
386 WARN(1, "TPACKET version not supported.\n");
393 static int __packet_get_status(const struct packet_sock *po, void *frame)
395 union tpacket_uhdr h;
399 /* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
402 switch (po->tp_version) {
404 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 return READ_ONCE(h.h1->tp_status);
407 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
408 return READ_ONCE(h.h2->tp_status);
410 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
411 return READ_ONCE(h.h3->tp_status);
413 WARN(1, "TPACKET version not supported.\n");
419 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
422 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
425 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
426 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
427 return TP_STATUS_TS_RAW_HARDWARE;
429 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
430 ktime_to_timespec64_cond(skb->tstamp, ts))
431 return TP_STATUS_TS_SOFTWARE;
436 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
439 union tpacket_uhdr h;
440 struct timespec64 ts;
443 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
448 * versions 1 through 3 overflow the timestamps in y2106, since they
449 * all store the seconds in a 32-bit unsigned integer.
450 * If we create a version 4, that should have a 64-bit timestamp,
451 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
454 switch (po->tp_version) {
456 h.h1->tp_sec = ts.tv_sec;
457 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
460 h.h2->tp_sec = ts.tv_sec;
461 h.h2->tp_nsec = ts.tv_nsec;
464 h.h3->tp_sec = ts.tv_sec;
465 h.h3->tp_nsec = ts.tv_nsec;
468 WARN(1, "TPACKET version not supported.\n");
472 /* one flush is safe, as both fields always lie on the same cacheline */
473 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
479 static void *packet_lookup_frame(const struct packet_sock *po,
480 const struct packet_ring_buffer *rb,
481 unsigned int position,
484 unsigned int pg_vec_pos, frame_offset;
485 union tpacket_uhdr h;
487 pg_vec_pos = position / rb->frames_per_block;
488 frame_offset = position % rb->frames_per_block;
490 h.raw = rb->pg_vec[pg_vec_pos].buffer +
491 (frame_offset * rb->frame_size);
493 if (status != __packet_get_status(po, h.raw))
499 static void *packet_current_frame(struct packet_sock *po,
500 struct packet_ring_buffer *rb,
503 return packet_lookup_frame(po, rb, rb->head, status);
506 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
508 del_timer_sync(&pkc->retire_blk_timer);
511 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
512 struct sk_buff_head *rb_queue)
514 struct tpacket_kbdq_core *pkc;
516 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
518 spin_lock_bh(&rb_queue->lock);
519 pkc->delete_blk_timer = 1;
520 spin_unlock_bh(&rb_queue->lock);
522 prb_del_retire_blk_timer(pkc);
525 static void prb_setup_retire_blk_timer(struct packet_sock *po)
527 struct tpacket_kbdq_core *pkc;
529 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
530 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
532 pkc->retire_blk_timer.expires = jiffies;
535 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
536 int blk_size_in_bytes)
538 struct net_device *dev;
539 unsigned int mbits, div;
540 struct ethtool_link_ksettings ecmd;
544 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
545 if (unlikely(!dev)) {
547 return DEFAULT_PRB_RETIRE_TOV;
549 err = __ethtool_get_link_ksettings(dev, &ecmd);
552 return DEFAULT_PRB_RETIRE_TOV;
554 /* If the link speed is so slow you don't really
555 * need to worry about perf anyways
557 if (ecmd.base.speed < SPEED_1000 ||
558 ecmd.base.speed == SPEED_UNKNOWN)
559 return DEFAULT_PRB_RETIRE_TOV;
561 div = ecmd.base.speed / 1000;
562 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
572 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
573 union tpacket_req_u *req_u)
575 p1->feature_req_word = req_u->req3.tp_feature_req_word;
578 static void init_prb_bdqc(struct packet_sock *po,
579 struct packet_ring_buffer *rb,
581 union tpacket_req_u *req_u)
583 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
584 struct tpacket_block_desc *pbd;
586 memset(p1, 0x0, sizeof(*p1));
588 p1->knxt_seq_num = 1;
590 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
591 p1->pkblk_start = pg_vec[0].buffer;
592 p1->kblk_size = req_u->req3.tp_block_size;
593 p1->knum_blocks = req_u->req3.tp_block_nr;
594 p1->hdrlen = po->tp_hdrlen;
595 p1->version = po->tp_version;
596 p1->last_kactive_blk_num = 0;
597 po->stats.stats3.tp_freeze_q_cnt = 0;
598 if (req_u->req3.tp_retire_blk_tov)
599 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
601 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
602 req_u->req3.tp_block_size);
603 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
604 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
605 rwlock_init(&p1->blk_fill_in_prog_lock);
607 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
608 prb_init_ft_ops(p1, req_u);
609 prb_setup_retire_blk_timer(po);
610 prb_open_block(p1, pbd);
613 /* Do NOT update the last_blk_num first.
614 * Assumes sk_buff_head lock is held.
616 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
618 mod_timer(&pkc->retire_blk_timer,
619 jiffies + pkc->tov_in_jiffies);
620 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
625 * 1) We refresh the timer only when we open a block.
626 * By doing this we don't waste cycles refreshing the timer
627 * on packet-by-packet basis.
629 * With a 1MB block-size, on a 1Gbps line, it will take
630 * i) ~8 ms to fill a block + ii) memcpy etc.
631 * In this cut we are not accounting for the memcpy time.
633 * So, if the user sets the 'tmo' to 10ms then the timer
634 * will never fire while the block is still getting filled
635 * (which is what we want). However, the user could choose
636 * to close a block early and that's fine.
638 * But when the timer does fire, we check whether or not to refresh it.
639 * Since the tmo granularity is in msecs, it is not too expensive
640 * to refresh the timer, lets say every '8' msecs.
641 * Either the user can set the 'tmo' or we can derive it based on
642 * a) line-speed and b) block-size.
643 * prb_calc_retire_blk_tmo() calculates the tmo.
646 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
648 struct packet_sock *po =
649 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
650 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
652 struct tpacket_block_desc *pbd;
654 spin_lock(&po->sk.sk_receive_queue.lock);
656 frozen = prb_queue_frozen(pkc);
657 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
659 if (unlikely(pkc->delete_blk_timer))
662 /* We only need to plug the race when the block is partially filled.
664 * lock(); increment BLOCK_NUM_PKTS; unlock()
665 * copy_bits() is in progress ...
666 * timer fires on other cpu:
667 * we can't retire the current block because copy_bits
671 if (BLOCK_NUM_PKTS(pbd)) {
672 /* Waiting for skb_copy_bits to finish... */
673 write_lock(&pkc->blk_fill_in_prog_lock);
674 write_unlock(&pkc->blk_fill_in_prog_lock);
677 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
679 if (!BLOCK_NUM_PKTS(pbd)) {
680 /* An empty block. Just refresh the timer. */
683 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
684 if (!prb_dispatch_next_block(pkc, po))
689 /* Case 1. Queue was frozen because user-space was
692 if (prb_curr_blk_in_use(pbd)) {
694 * Ok, user-space is still behind.
695 * So just refresh the timer.
699 /* Case 2. queue was frozen,user-space caught up,
700 * now the link went idle && the timer fired.
701 * We don't have a block to close.So we open this
702 * block and restart the timer.
703 * opening a block thaws the queue,restarts timer
704 * Thawing/timer-refresh is a side effect.
706 prb_open_block(pkc, pbd);
713 _prb_refresh_rx_retire_blk_timer(pkc);
716 spin_unlock(&po->sk.sk_receive_queue.lock);
719 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
720 struct tpacket_block_desc *pbd1, __u32 status)
722 /* Flush everything minus the block header */
724 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
729 /* Skip the block header(we know header WILL fit in 4K) */
732 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
733 for (; start < end; start += PAGE_SIZE)
734 flush_dcache_page(pgv_to_page(start));
739 /* Now update the block status. */
741 BLOCK_STATUS(pbd1) = status;
743 /* Flush the block header */
745 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
747 flush_dcache_page(pgv_to_page(start));
757 * 2) Increment active_blk_num
759 * Note:We DONT refresh the timer on purpose.
760 * Because almost always the next block will be opened.
762 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
763 struct tpacket_block_desc *pbd1,
764 struct packet_sock *po, unsigned int stat)
766 __u32 status = TP_STATUS_USER | stat;
768 struct tpacket3_hdr *last_pkt;
769 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
770 struct sock *sk = &po->sk;
772 if (atomic_read(&po->tp_drops))
773 status |= TP_STATUS_LOSING;
775 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
776 last_pkt->tp_next_offset = 0;
778 /* Get the ts of the last pkt */
779 if (BLOCK_NUM_PKTS(pbd1)) {
780 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
781 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
783 /* Ok, we tmo'd - so get the current time.
785 * It shouldn't really happen as we don't close empty
786 * blocks. See prb_retire_rx_blk_timer_expired().
788 struct timespec64 ts;
789 ktime_get_real_ts64(&ts);
790 h1->ts_last_pkt.ts_sec = ts.tv_sec;
791 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
796 /* Flush the block */
797 prb_flush_block(pkc1, pbd1, status);
799 sk->sk_data_ready(sk);
801 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
804 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
806 pkc->reset_pending_on_curr_blk = 0;
810 * Side effect of opening a block:
812 * 1) prb_queue is thawed.
813 * 2) retire_blk_timer is refreshed.
816 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
817 struct tpacket_block_desc *pbd1)
819 struct timespec64 ts;
820 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
824 /* We could have just memset this but we will lose the
825 * flexibility of making the priv area sticky
828 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
829 BLOCK_NUM_PKTS(pbd1) = 0;
830 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
832 ktime_get_real_ts64(&ts);
834 h1->ts_first_pkt.ts_sec = ts.tv_sec;
835 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
837 pkc1->pkblk_start = (char *)pbd1;
838 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
840 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
841 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
843 pbd1->version = pkc1->version;
844 pkc1->prev = pkc1->nxt_offset;
845 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
847 prb_thaw_queue(pkc1);
848 _prb_refresh_rx_retire_blk_timer(pkc1);
854 * Queue freeze logic:
855 * 1) Assume tp_block_nr = 8 blocks.
856 * 2) At time 't0', user opens Rx ring.
857 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
858 * 4) user-space is either sleeping or processing block '0'.
859 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
860 * it will close block-7,loop around and try to fill block '0'.
862 * __packet_lookup_frame_in_block
863 * prb_retire_current_block()
864 * prb_dispatch_next_block()
865 * |->(BLOCK_STATUS == USER) evaluates to true
866 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
867 * 6) Now there are two cases:
868 * 6.1) Link goes idle right after the queue is frozen.
869 * But remember, the last open_block() refreshed the timer.
870 * When this timer expires,it will refresh itself so that we can
871 * re-open block-0 in near future.
872 * 6.2) Link is busy and keeps on receiving packets. This is a simple
873 * case and __packet_lookup_frame_in_block will check if block-0
874 * is free and can now be re-used.
876 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
877 struct packet_sock *po)
879 pkc->reset_pending_on_curr_blk = 1;
880 po->stats.stats3.tp_freeze_q_cnt++;
883 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
886 * If the next block is free then we will dispatch it
887 * and return a good offset.
888 * Else, we will freeze the queue.
889 * So, caller must check the return value.
891 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
892 struct packet_sock *po)
894 struct tpacket_block_desc *pbd;
898 /* 1. Get current block num */
899 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
901 /* 2. If this block is currently in_use then freeze the queue */
902 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
903 prb_freeze_queue(pkc, po);
909 * open this block and return the offset where the first packet
910 * needs to get stored.
912 prb_open_block(pkc, pbd);
913 return (void *)pkc->nxt_offset;
916 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
917 struct packet_sock *po, unsigned int status)
919 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
921 /* retire/close the current block */
922 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
924 * Plug the case where copy_bits() is in progress on
925 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
926 * have space to copy the pkt in the current block and
927 * called prb_retire_current_block()
929 * We don't need to worry about the TMO case because
930 * the timer-handler already handled this case.
932 if (!(status & TP_STATUS_BLK_TMO)) {
933 /* Waiting for skb_copy_bits to finish... */
934 write_lock(&pkc->blk_fill_in_prog_lock);
935 write_unlock(&pkc->blk_fill_in_prog_lock);
937 prb_close_block(pkc, pbd, po, status);
942 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
944 return TP_STATUS_USER & BLOCK_STATUS(pbd);
947 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
949 return pkc->reset_pending_on_curr_blk;
952 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
953 __releases(&pkc->blk_fill_in_prog_lock)
955 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
957 read_unlock(&pkc->blk_fill_in_prog_lock);
960 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
963 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
966 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
967 struct tpacket3_hdr *ppd)
969 ppd->hv1.tp_rxhash = 0;
972 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
973 struct tpacket3_hdr *ppd)
975 if (skb_vlan_tag_present(pkc->skb)) {
976 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
977 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
978 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
980 ppd->hv1.tp_vlan_tci = 0;
981 ppd->hv1.tp_vlan_tpid = 0;
982 ppd->tp_status = TP_STATUS_AVAILABLE;
986 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
987 struct tpacket3_hdr *ppd)
989 ppd->hv1.tp_padding = 0;
990 prb_fill_vlan_info(pkc, ppd);
992 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
993 prb_fill_rxhash(pkc, ppd);
995 prb_clear_rxhash(pkc, ppd);
998 static void prb_fill_curr_block(char *curr,
999 struct tpacket_kbdq_core *pkc,
1000 struct tpacket_block_desc *pbd,
1002 __acquires(&pkc->blk_fill_in_prog_lock)
1004 struct tpacket3_hdr *ppd;
1006 ppd = (struct tpacket3_hdr *)curr;
1007 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1009 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1010 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1011 BLOCK_NUM_PKTS(pbd) += 1;
1012 read_lock(&pkc->blk_fill_in_prog_lock);
1013 prb_run_all_ft_ops(pkc, ppd);
1016 /* Assumes caller has the sk->rx_queue.lock */
1017 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1018 struct sk_buff *skb,
1022 struct tpacket_kbdq_core *pkc;
1023 struct tpacket_block_desc *pbd;
1026 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1027 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1029 /* Queue is frozen when user space is lagging behind */
1030 if (prb_queue_frozen(pkc)) {
1032 * Check if that last block which caused the queue to freeze,
1033 * is still in_use by user-space.
1035 if (prb_curr_blk_in_use(pbd)) {
1036 /* Can't record this packet */
1040 * Ok, the block was released by user-space.
1041 * Now let's open that block.
1042 * opening a block also thaws the queue.
1043 * Thawing is a side effect.
1045 prb_open_block(pkc, pbd);
1050 curr = pkc->nxt_offset;
1052 end = (char *)pbd + pkc->kblk_size;
1054 /* first try the current block */
1055 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1056 prb_fill_curr_block(curr, pkc, pbd, len);
1057 return (void *)curr;
1060 /* Ok, close the current block */
1061 prb_retire_current_block(pkc, po, 0);
1063 /* Now, try to dispatch the next block */
1064 curr = (char *)prb_dispatch_next_block(pkc, po);
1066 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1067 prb_fill_curr_block(curr, pkc, pbd, len);
1068 return (void *)curr;
1072 * No free blocks are available.user_space hasn't caught up yet.
1073 * Queue was just frozen and now this packet will get dropped.
1078 static void *packet_current_rx_frame(struct packet_sock *po,
1079 struct sk_buff *skb,
1080 int status, unsigned int len)
1083 switch (po->tp_version) {
1086 curr = packet_lookup_frame(po, &po->rx_ring,
1087 po->rx_ring.head, status);
1090 return __packet_lookup_frame_in_block(po, skb, len);
1092 WARN(1, "TPACKET version not supported\n");
1098 static void *prb_lookup_block(const struct packet_sock *po,
1099 const struct packet_ring_buffer *rb,
1103 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1104 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1106 if (status != BLOCK_STATUS(pbd))
1111 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1114 if (rb->prb_bdqc.kactive_blk_num)
1115 prev = rb->prb_bdqc.kactive_blk_num-1;
1117 prev = rb->prb_bdqc.knum_blocks-1;
1121 /* Assumes caller has held the rx_queue.lock */
1122 static void *__prb_previous_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1126 unsigned int previous = prb_previous_blk_num(rb);
1127 return prb_lookup_block(po, rb, previous, status);
1130 static void *packet_previous_rx_frame(struct packet_sock *po,
1131 struct packet_ring_buffer *rb,
1134 if (po->tp_version <= TPACKET_V2)
1135 return packet_previous_frame(po, rb, status);
1137 return __prb_previous_block(po, rb, status);
1140 static void packet_increment_rx_head(struct packet_sock *po,
1141 struct packet_ring_buffer *rb)
1143 switch (po->tp_version) {
1146 return packet_increment_head(rb);
1149 WARN(1, "TPACKET version not supported.\n");
1155 static void *packet_previous_frame(struct packet_sock *po,
1156 struct packet_ring_buffer *rb,
1159 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1160 return packet_lookup_frame(po, rb, previous, status);
1163 static void packet_increment_head(struct packet_ring_buffer *buff)
1165 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1168 static void packet_inc_pending(struct packet_ring_buffer *rb)
1170 this_cpu_inc(*rb->pending_refcnt);
1173 static void packet_dec_pending(struct packet_ring_buffer *rb)
1175 this_cpu_dec(*rb->pending_refcnt);
1178 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1180 unsigned int refcnt = 0;
1183 /* We don't use pending refcount in rx_ring. */
1184 if (rb->pending_refcnt == NULL)
1187 for_each_possible_cpu(cpu)
1188 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1193 static int packet_alloc_pending(struct packet_sock *po)
1195 po->rx_ring.pending_refcnt = NULL;
1197 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1198 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1204 static void packet_free_pending(struct packet_sock *po)
1206 free_percpu(po->tx_ring.pending_refcnt);
1209 #define ROOM_POW_OFF 2
1210 #define ROOM_NONE 0x0
1211 #define ROOM_LOW 0x1
1212 #define ROOM_NORMAL 0x2
1214 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1218 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1219 idx = READ_ONCE(po->rx_ring.head);
1221 idx += len >> pow_off;
1224 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1227 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1231 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1232 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1234 idx += len >> pow_off;
1237 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1240 static int __packet_rcv_has_room(const struct packet_sock *po,
1241 const struct sk_buff *skb)
1243 const struct sock *sk = &po->sk;
1244 int ret = ROOM_NONE;
1246 if (po->prot_hook.func != tpacket_rcv) {
1247 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1248 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1249 - (skb ? skb->truesize : 0);
1251 if (avail > (rcvbuf >> ROOM_POW_OFF))
1259 if (po->tp_version == TPACKET_V3) {
1260 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1262 else if (__tpacket_v3_has_room(po, 0))
1265 if (__tpacket_has_room(po, ROOM_POW_OFF))
1267 else if (__tpacket_has_room(po, 0))
1274 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1278 ret = __packet_rcv_has_room(po, skb);
1279 pressure = ret != ROOM_NORMAL;
1281 if (READ_ONCE(po->pressure) != pressure)
1282 WRITE_ONCE(po->pressure, pressure);
1287 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1289 if (READ_ONCE(po->pressure) &&
1290 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1291 WRITE_ONCE(po->pressure, 0);
1294 static void packet_sock_destruct(struct sock *sk)
1296 skb_queue_purge(&sk->sk_error_queue);
1298 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1299 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1301 if (!sock_flag(sk, SOCK_DEAD)) {
1302 pr_err("Attempt to release alive packet socket: %p\n", sk);
1306 sk_refcnt_debug_dec(sk);
1309 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1311 u32 *history = po->rollover->history;
1315 rxhash = skb_get_hash(skb);
1316 for (i = 0; i < ROLLOVER_HLEN; i++)
1317 if (READ_ONCE(history[i]) == rxhash)
1320 victim = prandom_u32() % ROLLOVER_HLEN;
1322 /* Avoid dirtying the cache line if possible */
1323 if (READ_ONCE(history[victim]) != rxhash)
1324 WRITE_ONCE(history[victim], rxhash);
1326 return count > (ROLLOVER_HLEN >> 1);
1329 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1330 struct sk_buff *skb,
1333 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1336 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1337 struct sk_buff *skb,
1340 unsigned int val = atomic_inc_return(&f->rr_cur);
1345 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1346 struct sk_buff *skb,
1349 return smp_processor_id() % num;
1352 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1353 struct sk_buff *skb,
1356 return prandom_u32_max(num);
1359 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1360 struct sk_buff *skb,
1361 unsigned int idx, bool try_self,
1364 struct packet_sock *po, *po_next, *po_skip = NULL;
1365 unsigned int i, j, room = ROOM_NONE;
1367 po = pkt_sk(rcu_dereference(f->arr[idx]));
1370 room = packet_rcv_has_room(po, skb);
1371 if (room == ROOM_NORMAL ||
1372 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1377 i = j = min_t(int, po->rollover->sock, num - 1);
1379 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1380 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1381 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1383 po->rollover->sock = i;
1384 atomic_long_inc(&po->rollover->num);
1385 if (room == ROOM_LOW)
1386 atomic_long_inc(&po->rollover->num_huge);
1394 atomic_long_inc(&po->rollover->num_failed);
1398 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1399 struct sk_buff *skb,
1402 return skb_get_queue_mapping(skb) % num;
1405 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1406 struct sk_buff *skb,
1409 struct bpf_prog *prog;
1410 unsigned int ret = 0;
1413 prog = rcu_dereference(f->bpf_prog);
1415 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1421 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1423 return f->flags & (flag >> 8);
1426 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1427 struct packet_type *pt, struct net_device *orig_dev)
1429 struct packet_fanout *f = pt->af_packet_priv;
1430 unsigned int num = READ_ONCE(f->num_members);
1431 struct net *net = read_pnet(&f->net);
1432 struct packet_sock *po;
1435 if (!net_eq(dev_net(dev), net) || !num) {
1440 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1441 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1446 case PACKET_FANOUT_HASH:
1448 idx = fanout_demux_hash(f, skb, num);
1450 case PACKET_FANOUT_LB:
1451 idx = fanout_demux_lb(f, skb, num);
1453 case PACKET_FANOUT_CPU:
1454 idx = fanout_demux_cpu(f, skb, num);
1456 case PACKET_FANOUT_RND:
1457 idx = fanout_demux_rnd(f, skb, num);
1459 case PACKET_FANOUT_QM:
1460 idx = fanout_demux_qm(f, skb, num);
1462 case PACKET_FANOUT_ROLLOVER:
1463 idx = fanout_demux_rollover(f, skb, 0, false, num);
1465 case PACKET_FANOUT_CBPF:
1466 case PACKET_FANOUT_EBPF:
1467 idx = fanout_demux_bpf(f, skb, num);
1471 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1472 idx = fanout_demux_rollover(f, skb, idx, true, num);
1474 po = pkt_sk(rcu_dereference(f->arr[idx]));
1475 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1478 DEFINE_MUTEX(fanout_mutex);
1479 EXPORT_SYMBOL_GPL(fanout_mutex);
1480 static LIST_HEAD(fanout_list);
1481 static u16 fanout_next_id;
1483 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1485 struct packet_fanout *f = po->fanout;
1487 spin_lock(&f->lock);
1488 rcu_assign_pointer(f->arr[f->num_members], sk);
1491 if (f->num_members == 1)
1492 dev_add_pack(&f->prot_hook);
1493 spin_unlock(&f->lock);
1496 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1498 struct packet_fanout *f = po->fanout;
1501 spin_lock(&f->lock);
1502 for (i = 0; i < f->num_members; i++) {
1503 if (rcu_dereference_protected(f->arr[i],
1504 lockdep_is_held(&f->lock)) == sk)
1507 BUG_ON(i >= f->num_members);
1508 rcu_assign_pointer(f->arr[i],
1509 rcu_dereference_protected(f->arr[f->num_members - 1],
1510 lockdep_is_held(&f->lock)));
1512 if (f->num_members == 0)
1513 __dev_remove_pack(&f->prot_hook);
1514 spin_unlock(&f->lock);
1517 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1519 if (sk->sk_family != PF_PACKET)
1522 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1525 static void fanout_init_data(struct packet_fanout *f)
1528 case PACKET_FANOUT_LB:
1529 atomic_set(&f->rr_cur, 0);
1531 case PACKET_FANOUT_CBPF:
1532 case PACKET_FANOUT_EBPF:
1533 RCU_INIT_POINTER(f->bpf_prog, NULL);
1538 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1540 struct bpf_prog *old;
1542 spin_lock(&f->lock);
1543 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1544 rcu_assign_pointer(f->bpf_prog, new);
1545 spin_unlock(&f->lock);
1549 bpf_prog_destroy(old);
1553 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1556 struct bpf_prog *new;
1557 struct sock_fprog fprog;
1560 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1563 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1567 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1571 __fanout_set_data_bpf(po->fanout, new);
1575 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1578 struct bpf_prog *new;
1581 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1583 if (len != sizeof(fd))
1585 if (copy_from_sockptr(&fd, data, len))
1588 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1590 return PTR_ERR(new);
1592 __fanout_set_data_bpf(po->fanout, new);
1596 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1599 switch (po->fanout->type) {
1600 case PACKET_FANOUT_CBPF:
1601 return fanout_set_data_cbpf(po, data, len);
1602 case PACKET_FANOUT_EBPF:
1603 return fanout_set_data_ebpf(po, data, len);
1609 static void fanout_release_data(struct packet_fanout *f)
1612 case PACKET_FANOUT_CBPF:
1613 case PACKET_FANOUT_EBPF:
1614 __fanout_set_data_bpf(f, NULL);
1618 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1620 struct packet_fanout *f;
1622 list_for_each_entry(f, &fanout_list, list) {
1623 if (f->id == candidate_id &&
1624 read_pnet(&f->net) == sock_net(sk)) {
1631 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1633 u16 id = fanout_next_id;
1636 if (__fanout_id_is_free(sk, id)) {
1638 fanout_next_id = id + 1;
1643 } while (id != fanout_next_id);
1648 static int fanout_add(struct sock *sk, struct fanout_args *args)
1650 struct packet_rollover *rollover = NULL;
1651 struct packet_sock *po = pkt_sk(sk);
1652 u16 type_flags = args->type_flags;
1653 struct packet_fanout *f, *match;
1654 u8 type = type_flags & 0xff;
1655 u8 flags = type_flags >> 8;
1660 case PACKET_FANOUT_ROLLOVER:
1661 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1663 case PACKET_FANOUT_HASH:
1664 case PACKET_FANOUT_LB:
1665 case PACKET_FANOUT_CPU:
1666 case PACKET_FANOUT_RND:
1667 case PACKET_FANOUT_QM:
1668 case PACKET_FANOUT_CBPF:
1669 case PACKET_FANOUT_EBPF:
1675 mutex_lock(&fanout_mutex);
1681 if (type == PACKET_FANOUT_ROLLOVER ||
1682 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1684 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1687 atomic_long_set(&rollover->num, 0);
1688 atomic_long_set(&rollover->num_huge, 0);
1689 atomic_long_set(&rollover->num_failed, 0);
1692 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1697 if (!fanout_find_new_id(sk, &id)) {
1701 /* ephemeral flag for the first socket in the group: drop it */
1702 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1706 list_for_each_entry(f, &fanout_list, list) {
1708 read_pnet(&f->net) == sock_net(sk)) {
1715 if (match->flags != flags)
1717 if (args->max_num_members &&
1718 args->max_num_members != match->max_num_members)
1721 if (args->max_num_members > PACKET_FANOUT_MAX)
1723 if (!args->max_num_members)
1724 /* legacy PACKET_FANOUT_MAX */
1725 args->max_num_members = 256;
1727 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1731 write_pnet(&match->net, sock_net(sk));
1734 match->flags = flags;
1735 INIT_LIST_HEAD(&match->list);
1736 spin_lock_init(&match->lock);
1737 refcount_set(&match->sk_ref, 0);
1738 fanout_init_data(match);
1739 match->prot_hook.type = po->prot_hook.type;
1740 match->prot_hook.dev = po->prot_hook.dev;
1741 match->prot_hook.func = packet_rcv_fanout;
1742 match->prot_hook.af_packet_priv = match;
1743 match->prot_hook.af_packet_net = read_pnet(&match->net);
1744 match->prot_hook.id_match = match_fanout_group;
1745 match->max_num_members = args->max_num_members;
1746 list_add(&match->list, &fanout_list);
1750 spin_lock(&po->bind_lock);
1752 match->type == type &&
1753 match->prot_hook.type == po->prot_hook.type &&
1754 match->prot_hook.dev == po->prot_hook.dev) {
1756 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1757 __dev_remove_pack(&po->prot_hook);
1759 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1760 WRITE_ONCE(po->fanout, match);
1762 po->rollover = rollover;
1764 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1765 __fanout_link(sk, po);
1769 spin_unlock(&po->bind_lock);
1771 if (err && !refcount_read(&match->sk_ref)) {
1772 list_del(&match->list);
1778 mutex_unlock(&fanout_mutex);
1782 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1783 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1784 * It is the responsibility of the caller to call fanout_release_data() and
1785 * free the returned packet_fanout (after synchronize_net())
1787 static struct packet_fanout *fanout_release(struct sock *sk)
1789 struct packet_sock *po = pkt_sk(sk);
1790 struct packet_fanout *f;
1792 mutex_lock(&fanout_mutex);
1797 if (refcount_dec_and_test(&f->sk_ref))
1802 mutex_unlock(&fanout_mutex);
1807 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1808 struct sk_buff *skb)
1810 /* Earlier code assumed this would be a VLAN pkt, double-check
1811 * this now that we have the actual packet in hand. We can only
1812 * do this check on Ethernet devices.
1814 if (unlikely(dev->type != ARPHRD_ETHER))
1817 skb_reset_mac_header(skb);
1818 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1821 static const struct proto_ops packet_ops;
1823 static const struct proto_ops packet_ops_spkt;
1825 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1826 struct packet_type *pt, struct net_device *orig_dev)
1829 struct sockaddr_pkt *spkt;
1832 * When we registered the protocol we saved the socket in the data
1833 * field for just this event.
1836 sk = pt->af_packet_priv;
1839 * Yank back the headers [hope the device set this
1840 * right or kerboom...]
1842 * Incoming packets have ll header pulled,
1845 * For outgoing ones skb->data == skb_mac_header(skb)
1846 * so that this procedure is noop.
1849 if (skb->pkt_type == PACKET_LOOPBACK)
1852 if (!net_eq(dev_net(dev), sock_net(sk)))
1855 skb = skb_share_check(skb, GFP_ATOMIC);
1859 /* drop any routing info */
1862 /* drop conntrack reference */
1865 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1867 skb_push(skb, skb->data - skb_mac_header(skb));
1870 * The SOCK_PACKET socket receives _all_ frames.
1873 spkt->spkt_family = dev->type;
1874 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1875 spkt->spkt_protocol = skb->protocol;
1878 * Charge the memory to the socket. This is done specifically
1879 * to prevent sockets using all the memory up.
1882 if (sock_queue_rcv_skb(sk, skb) == 0)
1891 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1895 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1896 sock->type == SOCK_RAW) {
1897 skb_reset_mac_header(skb);
1898 skb->protocol = dev_parse_header_protocol(skb);
1901 /* Move network header to the right position for VLAN tagged packets */
1902 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1903 eth_type_vlan(skb->protocol) &&
1904 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1905 skb_set_network_header(skb, depth);
1907 skb_probe_transport_header(skb);
1911 * Output a raw packet to a device layer. This bypasses all the other
1912 * protocol layers and you must therefore supply it with a complete frame
1915 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1918 struct sock *sk = sock->sk;
1919 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1920 struct sk_buff *skb = NULL;
1921 struct net_device *dev;
1922 struct sockcm_cookie sockc;
1928 * Get and verify the address.
1932 if (msg->msg_namelen < sizeof(struct sockaddr))
1934 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1935 proto = saddr->spkt_protocol;
1937 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1940 * Find the device first to size check it
1943 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1946 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1952 if (!(dev->flags & IFF_UP))
1956 * You may not queue a frame bigger than the mtu. This is the lowest level
1957 * raw protocol and you must do your own fragmentation at this level.
1960 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1961 if (!netif_supports_nofcs(dev)) {
1962 err = -EPROTONOSUPPORT;
1965 extra_len = 4; /* We're doing our own CRC */
1969 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1973 size_t reserved = LL_RESERVED_SPACE(dev);
1974 int tlen = dev->needed_tailroom;
1975 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1978 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1981 /* FIXME: Save some space for broken drivers that write a hard
1982 * header at transmission time by themselves. PPP is the notable
1983 * one here. This should really be fixed at the driver level.
1985 skb_reserve(skb, reserved);
1986 skb_reset_network_header(skb);
1988 /* Try to align data part correctly */
1993 skb_reset_network_header(skb);
1995 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2001 if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2005 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2006 !packet_extra_vlan_len_allowed(dev, skb)) {
2011 sockcm_init(&sockc, sk);
2012 if (msg->msg_controllen) {
2013 err = sock_cmsg_send(sk, msg, &sockc);
2018 skb->protocol = proto;
2020 skb->priority = sk->sk_priority;
2021 skb->mark = sk->sk_mark;
2022 skb->tstamp = sockc.transmit_time;
2024 skb_setup_tx_timestamp(skb, sockc.tsflags);
2026 if (unlikely(extra_len == 4))
2029 packet_parse_headers(skb, sock);
2031 dev_queue_xmit(skb);
2042 static unsigned int run_filter(struct sk_buff *skb,
2043 const struct sock *sk,
2046 struct sk_filter *filter;
2049 filter = rcu_dereference(sk->sk_filter);
2051 res = bpf_prog_run_clear_cb(filter->prog, skb);
2057 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2060 struct virtio_net_hdr vnet_hdr;
2062 if (*len < sizeof(vnet_hdr))
2064 *len -= sizeof(vnet_hdr);
2066 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2069 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2073 * This function makes lazy skb cloning in hope that most of packets
2074 * are discarded by BPF.
2076 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2077 * and skb->cb are mangled. It works because (and until) packets
2078 * falling here are owned by current CPU. Output packets are cloned
2079 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2080 * sequencially, so that if we return skb to original state on exit,
2081 * we will not harm anyone.
2084 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2085 struct packet_type *pt, struct net_device *orig_dev)
2088 struct sockaddr_ll *sll;
2089 struct packet_sock *po;
2090 u8 *skb_head = skb->data;
2091 int skb_len = skb->len;
2092 unsigned int snaplen, res;
2093 bool is_drop_n_account = false;
2095 if (skb->pkt_type == PACKET_LOOPBACK)
2098 sk = pt->af_packet_priv;
2101 if (!net_eq(dev_net(dev), sock_net(sk)))
2106 if (dev_has_header(dev)) {
2107 /* The device has an explicit notion of ll header,
2108 * exported to higher levels.
2110 * Otherwise, the device hides details of its frame
2111 * structure, so that corresponding packet head is
2112 * never delivered to user.
2114 if (sk->sk_type != SOCK_DGRAM)
2115 skb_push(skb, skb->data - skb_mac_header(skb));
2116 else if (skb->pkt_type == PACKET_OUTGOING) {
2117 /* Special case: outgoing packets have ll header at head */
2118 skb_pull(skb, skb_network_offset(skb));
2124 res = run_filter(skb, sk, snaplen);
2126 goto drop_n_restore;
2130 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2133 if (skb_shared(skb)) {
2134 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2138 if (skb_head != skb->data) {
2139 skb->data = skb_head;
2146 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2148 sll = &PACKET_SKB_CB(skb)->sa.ll;
2149 sll->sll_hatype = dev->type;
2150 sll->sll_pkttype = skb->pkt_type;
2151 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2152 sll->sll_ifindex = orig_dev->ifindex;
2154 sll->sll_ifindex = dev->ifindex;
2156 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2158 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2159 * Use their space for storing the original skb length.
2161 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2163 if (pskb_trim(skb, snaplen))
2166 skb_set_owner_r(skb, sk);
2170 /* drop conntrack reference */
2173 spin_lock(&sk->sk_receive_queue.lock);
2174 po->stats.stats1.tp_packets++;
2175 sock_skb_set_dropcount(sk, skb);
2176 __skb_queue_tail(&sk->sk_receive_queue, skb);
2177 spin_unlock(&sk->sk_receive_queue.lock);
2178 sk->sk_data_ready(sk);
2182 is_drop_n_account = true;
2183 atomic_inc(&po->tp_drops);
2184 atomic_inc(&sk->sk_drops);
2187 if (skb_head != skb->data && skb_shared(skb)) {
2188 skb->data = skb_head;
2192 if (!is_drop_n_account)
2199 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2200 struct packet_type *pt, struct net_device *orig_dev)
2203 struct packet_sock *po;
2204 struct sockaddr_ll *sll;
2205 union tpacket_uhdr h;
2206 u8 *skb_head = skb->data;
2207 int skb_len = skb->len;
2208 unsigned int snaplen, res;
2209 unsigned long status = TP_STATUS_USER;
2210 unsigned short macoff, hdrlen;
2211 unsigned int netoff;
2212 struct sk_buff *copy_skb = NULL;
2213 struct timespec64 ts;
2215 bool is_drop_n_account = false;
2216 unsigned int slot_id = 0;
2217 bool do_vnet = false;
2219 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2220 * We may add members to them until current aligned size without forcing
2221 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2223 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2224 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2226 if (skb->pkt_type == PACKET_LOOPBACK)
2229 sk = pt->af_packet_priv;
2232 if (!net_eq(dev_net(dev), sock_net(sk)))
2235 if (dev_has_header(dev)) {
2236 if (sk->sk_type != SOCK_DGRAM)
2237 skb_push(skb, skb->data - skb_mac_header(skb));
2238 else if (skb->pkt_type == PACKET_OUTGOING) {
2239 /* Special case: outgoing packets have ll header at head */
2240 skb_pull(skb, skb_network_offset(skb));
2246 res = run_filter(skb, sk, snaplen);
2248 goto drop_n_restore;
2250 /* If we are flooded, just give up */
2251 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2252 atomic_inc(&po->tp_drops);
2253 goto drop_n_restore;
2256 if (skb->ip_summed == CHECKSUM_PARTIAL)
2257 status |= TP_STATUS_CSUMNOTREADY;
2258 else if (skb->pkt_type != PACKET_OUTGOING &&
2259 skb_csum_unnecessary(skb))
2260 status |= TP_STATUS_CSUM_VALID;
2265 if (sk->sk_type == SOCK_DGRAM) {
2266 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2269 unsigned int maclen = skb_network_offset(skb);
2270 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2271 (maclen < 16 ? 16 : maclen)) +
2273 if (po->has_vnet_hdr) {
2274 netoff += sizeof(struct virtio_net_hdr);
2277 macoff = netoff - maclen;
2279 if (netoff > USHRT_MAX) {
2280 atomic_inc(&po->tp_drops);
2281 goto drop_n_restore;
2283 if (po->tp_version <= TPACKET_V2) {
2284 if (macoff + snaplen > po->rx_ring.frame_size) {
2285 if (po->copy_thresh &&
2286 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2287 if (skb_shared(skb)) {
2288 copy_skb = skb_clone(skb, GFP_ATOMIC);
2290 copy_skb = skb_get(skb);
2291 skb_head = skb->data;
2294 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2295 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2296 skb_set_owner_r(copy_skb, sk);
2299 snaplen = po->rx_ring.frame_size - macoff;
2300 if ((int)snaplen < 0) {
2305 } else if (unlikely(macoff + snaplen >
2306 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2309 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2310 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2311 snaplen, nval, macoff);
2313 if (unlikely((int)snaplen < 0)) {
2315 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2319 spin_lock(&sk->sk_receive_queue.lock);
2320 h.raw = packet_current_rx_frame(po, skb,
2321 TP_STATUS_KERNEL, (macoff+snaplen));
2323 goto drop_n_account;
2325 if (po->tp_version <= TPACKET_V2) {
2326 slot_id = po->rx_ring.head;
2327 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2328 goto drop_n_account;
2329 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2333 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2334 sizeof(struct virtio_net_hdr),
2335 vio_le(), true, 0)) {
2336 if (po->tp_version == TPACKET_V3)
2337 prb_clear_blk_fill_status(&po->rx_ring);
2338 goto drop_n_account;
2341 if (po->tp_version <= TPACKET_V2) {
2342 packet_increment_rx_head(po, &po->rx_ring);
2344 * LOSING will be reported till you read the stats,
2345 * because it's COR - Clear On Read.
2346 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2349 if (atomic_read(&po->tp_drops))
2350 status |= TP_STATUS_LOSING;
2353 po->stats.stats1.tp_packets++;
2355 status |= TP_STATUS_COPY;
2356 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2358 spin_unlock(&sk->sk_receive_queue.lock);
2360 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2362 /* Always timestamp; prefer an existing software timestamp taken
2363 * closer to the time of capture.
2365 ts_status = tpacket_get_timestamp(skb, &ts,
2366 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2368 ktime_get_real_ts64(&ts);
2370 status |= ts_status;
2372 switch (po->tp_version) {
2374 h.h1->tp_len = skb->len;
2375 h.h1->tp_snaplen = snaplen;
2376 h.h1->tp_mac = macoff;
2377 h.h1->tp_net = netoff;
2378 h.h1->tp_sec = ts.tv_sec;
2379 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2380 hdrlen = sizeof(*h.h1);
2383 h.h2->tp_len = skb->len;
2384 h.h2->tp_snaplen = snaplen;
2385 h.h2->tp_mac = macoff;
2386 h.h2->tp_net = netoff;
2387 h.h2->tp_sec = ts.tv_sec;
2388 h.h2->tp_nsec = ts.tv_nsec;
2389 if (skb_vlan_tag_present(skb)) {
2390 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2391 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2392 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2394 h.h2->tp_vlan_tci = 0;
2395 h.h2->tp_vlan_tpid = 0;
2397 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2398 hdrlen = sizeof(*h.h2);
2401 /* tp_nxt_offset,vlan are already populated above.
2402 * So DONT clear those fields here
2404 h.h3->tp_status |= status;
2405 h.h3->tp_len = skb->len;
2406 h.h3->tp_snaplen = snaplen;
2407 h.h3->tp_mac = macoff;
2408 h.h3->tp_net = netoff;
2409 h.h3->tp_sec = ts.tv_sec;
2410 h.h3->tp_nsec = ts.tv_nsec;
2411 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2412 hdrlen = sizeof(*h.h3);
2418 sll = h.raw + TPACKET_ALIGN(hdrlen);
2419 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2420 sll->sll_family = AF_PACKET;
2421 sll->sll_hatype = dev->type;
2422 sll->sll_protocol = skb->protocol;
2423 sll->sll_pkttype = skb->pkt_type;
2424 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2425 sll->sll_ifindex = orig_dev->ifindex;
2427 sll->sll_ifindex = dev->ifindex;
2431 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2432 if (po->tp_version <= TPACKET_V2) {
2435 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2438 for (start = h.raw; start < end; start += PAGE_SIZE)
2439 flush_dcache_page(pgv_to_page(start));
2444 if (po->tp_version <= TPACKET_V2) {
2445 spin_lock(&sk->sk_receive_queue.lock);
2446 __packet_set_status(po, h.raw, status);
2447 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2448 spin_unlock(&sk->sk_receive_queue.lock);
2449 sk->sk_data_ready(sk);
2450 } else if (po->tp_version == TPACKET_V3) {
2451 prb_clear_blk_fill_status(&po->rx_ring);
2455 if (skb_head != skb->data && skb_shared(skb)) {
2456 skb->data = skb_head;
2460 if (!is_drop_n_account)
2467 spin_unlock(&sk->sk_receive_queue.lock);
2468 atomic_inc(&po->tp_drops);
2469 is_drop_n_account = true;
2471 sk->sk_data_ready(sk);
2472 kfree_skb(copy_skb);
2473 goto drop_n_restore;
2476 static void tpacket_destruct_skb(struct sk_buff *skb)
2478 struct packet_sock *po = pkt_sk(skb->sk);
2480 if (likely(po->tx_ring.pg_vec)) {
2484 ph = skb_zcopy_get_nouarg(skb);
2485 packet_dec_pending(&po->tx_ring);
2487 ts = __packet_set_timestamp(po, ph, skb);
2488 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2490 complete(&po->skb_completion);
2496 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2498 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2499 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2500 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2501 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2502 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2503 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2504 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2506 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2512 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2513 struct virtio_net_hdr *vnet_hdr)
2515 if (*len < sizeof(*vnet_hdr))
2517 *len -= sizeof(*vnet_hdr);
2519 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2522 return __packet_snd_vnet_parse(vnet_hdr, *len);
2525 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2526 void *frame, struct net_device *dev, void *data, int tp_len,
2527 __be16 proto, unsigned char *addr, int hlen, int copylen,
2528 const struct sockcm_cookie *sockc)
2530 union tpacket_uhdr ph;
2531 int to_write, offset, len, nr_frags, len_max;
2532 struct socket *sock = po->sk.sk_socket;
2538 skb->protocol = proto;
2540 skb->priority = po->sk.sk_priority;
2541 skb->mark = po->sk.sk_mark;
2542 skb->tstamp = sockc->transmit_time;
2543 skb_setup_tx_timestamp(skb, sockc->tsflags);
2544 skb_zcopy_set_nouarg(skb, ph.raw);
2546 skb_reserve(skb, hlen);
2547 skb_reset_network_header(skb);
2551 if (sock->type == SOCK_DGRAM) {
2552 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2554 if (unlikely(err < 0))
2556 } else if (copylen) {
2557 int hdrlen = min_t(int, copylen, tp_len);
2559 skb_push(skb, dev->hard_header_len);
2560 skb_put(skb, copylen - dev->hard_header_len);
2561 err = skb_store_bits(skb, 0, data, hdrlen);
2564 if (!dev_validate_header(dev, skb->data, hdrlen))
2571 offset = offset_in_page(data);
2572 len_max = PAGE_SIZE - offset;
2573 len = ((to_write > len_max) ? len_max : to_write);
2575 skb->data_len = to_write;
2576 skb->len += to_write;
2577 skb->truesize += to_write;
2578 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2580 while (likely(to_write)) {
2581 nr_frags = skb_shinfo(skb)->nr_frags;
2583 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2584 pr_err("Packet exceed the number of skb frags(%lu)\n",
2589 page = pgv_to_page(data);
2591 flush_dcache_page(page);
2593 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2596 len_max = PAGE_SIZE;
2597 len = ((to_write > len_max) ? len_max : to_write);
2600 packet_parse_headers(skb, sock);
2605 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2606 int size_max, void **data)
2608 union tpacket_uhdr ph;
2613 switch (po->tp_version) {
2615 if (ph.h3->tp_next_offset != 0) {
2616 pr_warn_once("variable sized slot not supported");
2619 tp_len = ph.h3->tp_len;
2622 tp_len = ph.h2->tp_len;
2625 tp_len = ph.h1->tp_len;
2628 if (unlikely(tp_len > size_max)) {
2629 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2633 if (unlikely(po->tp_tx_has_off)) {
2634 int off_min, off_max;
2636 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2637 off_max = po->tx_ring.frame_size - tp_len;
2638 if (po->sk.sk_type == SOCK_DGRAM) {
2639 switch (po->tp_version) {
2641 off = ph.h3->tp_net;
2644 off = ph.h2->tp_net;
2647 off = ph.h1->tp_net;
2651 switch (po->tp_version) {
2653 off = ph.h3->tp_mac;
2656 off = ph.h2->tp_mac;
2659 off = ph.h1->tp_mac;
2663 if (unlikely((off < off_min) || (off_max < off)))
2666 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2669 *data = frame + off;
2673 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2675 struct sk_buff *skb = NULL;
2676 struct net_device *dev;
2677 struct virtio_net_hdr *vnet_hdr = NULL;
2678 struct sockcm_cookie sockc;
2680 int err, reserve = 0;
2682 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2683 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2684 unsigned char *addr = NULL;
2685 int tp_len, size_max;
2688 int status = TP_STATUS_AVAILABLE;
2689 int hlen, tlen, copylen = 0;
2692 mutex_lock(&po->pg_vec_lock);
2694 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2695 * we need to confirm it under protection of pg_vec_lock.
2697 if (unlikely(!po->tx_ring.pg_vec)) {
2701 if (likely(saddr == NULL)) {
2702 dev = packet_cached_dev_get(po);
2703 proto = READ_ONCE(po->num);
2706 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2708 if (msg->msg_namelen < (saddr->sll_halen
2709 + offsetof(struct sockaddr_ll,
2712 proto = saddr->sll_protocol;
2713 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2714 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2715 if (dev && msg->msg_namelen < dev->addr_len +
2716 offsetof(struct sockaddr_ll, sll_addr))
2718 addr = saddr->sll_addr;
2723 if (unlikely(dev == NULL))
2726 if (unlikely(!(dev->flags & IFF_UP)))
2729 sockcm_init(&sockc, &po->sk);
2730 if (msg->msg_controllen) {
2731 err = sock_cmsg_send(&po->sk, msg, &sockc);
2736 if (po->sk.sk_socket->type == SOCK_RAW)
2737 reserve = dev->hard_header_len;
2738 size_max = po->tx_ring.frame_size
2739 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2741 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2742 size_max = dev->mtu + reserve + VLAN_HLEN;
2744 reinit_completion(&po->skb_completion);
2747 ph = packet_current_frame(po, &po->tx_ring,
2748 TP_STATUS_SEND_REQUEST);
2749 if (unlikely(ph == NULL)) {
2750 if (need_wait && skb) {
2751 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2752 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2754 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2758 /* check for additional frames */
2763 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2767 status = TP_STATUS_SEND_REQUEST;
2768 hlen = LL_RESERVED_SPACE(dev);
2769 tlen = dev->needed_tailroom;
2770 if (po->has_vnet_hdr) {
2772 data += sizeof(*vnet_hdr);
2773 tp_len -= sizeof(*vnet_hdr);
2775 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2779 copylen = __virtio16_to_cpu(vio_le(),
2782 copylen = max_t(int, copylen, dev->hard_header_len);
2783 skb = sock_alloc_send_skb(&po->sk,
2784 hlen + tlen + sizeof(struct sockaddr_ll) +
2785 (copylen - dev->hard_header_len),
2788 if (unlikely(skb == NULL)) {
2789 /* we assume the socket was initially writeable ... */
2790 if (likely(len_sum > 0))
2794 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2795 addr, hlen, copylen, &sockc);
2796 if (likely(tp_len >= 0) &&
2797 tp_len > dev->mtu + reserve &&
2798 !po->has_vnet_hdr &&
2799 !packet_extra_vlan_len_allowed(dev, skb))
2802 if (unlikely(tp_len < 0)) {
2805 __packet_set_status(po, ph,
2806 TP_STATUS_AVAILABLE);
2807 packet_increment_head(&po->tx_ring);
2811 status = TP_STATUS_WRONG_FORMAT;
2817 if (po->has_vnet_hdr) {
2818 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2822 virtio_net_hdr_set_proto(skb, vnet_hdr);
2825 skb->destructor = tpacket_destruct_skb;
2826 __packet_set_status(po, ph, TP_STATUS_SENDING);
2827 packet_inc_pending(&po->tx_ring);
2829 status = TP_STATUS_SEND_REQUEST;
2830 /* Paired with WRITE_ONCE() in packet_setsockopt() */
2831 err = READ_ONCE(po->xmit)(skb);
2832 if (unlikely(err != 0)) {
2834 err = net_xmit_errno(err);
2835 if (err && __packet_get_status(po, ph) ==
2836 TP_STATUS_AVAILABLE) {
2837 /* skb was destructed already */
2842 * skb was dropped but not destructed yet;
2843 * let's treat it like congestion or err < 0
2847 packet_increment_head(&po->tx_ring);
2849 } while (likely((ph != NULL) ||
2850 /* Note: packet_read_pending() might be slow if we have
2851 * to call it as it's per_cpu variable, but in fast-path
2852 * we already short-circuit the loop with the first
2853 * condition, and luckily don't have to go that path
2856 (need_wait && packet_read_pending(&po->tx_ring))));
2862 __packet_set_status(po, ph, status);
2867 mutex_unlock(&po->pg_vec_lock);
2871 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2872 size_t reserve, size_t len,
2873 size_t linear, int noblock,
2876 struct sk_buff *skb;
2878 /* Under a page? Don't bother with paged skb. */
2879 if (prepad + len < PAGE_SIZE || !linear)
2882 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2887 skb_reserve(skb, reserve);
2888 skb_put(skb, linear);
2889 skb->data_len = len - linear;
2890 skb->len += len - linear;
2895 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2897 struct sock *sk = sock->sk;
2898 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2899 struct sk_buff *skb;
2900 struct net_device *dev;
2902 unsigned char *addr = NULL;
2903 int err, reserve = 0;
2904 struct sockcm_cookie sockc;
2905 struct virtio_net_hdr vnet_hdr = { 0 };
2907 struct packet_sock *po = pkt_sk(sk);
2908 bool has_vnet_hdr = false;
2909 int hlen, tlen, linear;
2913 * Get and verify the address.
2916 if (likely(saddr == NULL)) {
2917 dev = packet_cached_dev_get(po);
2918 proto = READ_ONCE(po->num);
2921 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2923 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2925 proto = saddr->sll_protocol;
2926 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2927 if (sock->type == SOCK_DGRAM) {
2928 if (dev && msg->msg_namelen < dev->addr_len +
2929 offsetof(struct sockaddr_ll, sll_addr))
2931 addr = saddr->sll_addr;
2936 if (unlikely(dev == NULL))
2939 if (unlikely(!(dev->flags & IFF_UP)))
2942 sockcm_init(&sockc, sk);
2943 sockc.mark = sk->sk_mark;
2944 if (msg->msg_controllen) {
2945 err = sock_cmsg_send(sk, msg, &sockc);
2950 if (sock->type == SOCK_RAW)
2951 reserve = dev->hard_header_len;
2952 if (po->has_vnet_hdr) {
2953 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2956 has_vnet_hdr = true;
2959 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2960 if (!netif_supports_nofcs(dev)) {
2961 err = -EPROTONOSUPPORT;
2964 extra_len = 4; /* We're doing our own CRC */
2968 if (!vnet_hdr.gso_type &&
2969 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2973 hlen = LL_RESERVED_SPACE(dev);
2974 tlen = dev->needed_tailroom;
2975 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2976 linear = max(linear, min_t(int, len, dev->hard_header_len));
2977 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2978 msg->msg_flags & MSG_DONTWAIT, &err);
2982 skb_reset_network_header(skb);
2985 if (sock->type == SOCK_DGRAM) {
2986 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2987 if (unlikely(offset < 0))
2989 } else if (reserve) {
2990 skb_reserve(skb, -reserve);
2991 if (len < reserve + sizeof(struct ipv6hdr) &&
2992 dev->min_header_len != dev->hard_header_len)
2993 skb_reset_network_header(skb);
2996 /* Returns -EFAULT on error */
2997 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3001 if ((sock->type == SOCK_RAW &&
3002 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3007 skb_setup_tx_timestamp(skb, sockc.tsflags);
3009 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3010 !packet_extra_vlan_len_allowed(dev, skb)) {
3015 skb->protocol = proto;
3017 skb->priority = sk->sk_priority;
3018 skb->mark = sockc.mark;
3019 skb->tstamp = sockc.transmit_time;
3021 if (unlikely(extra_len == 4))
3024 packet_parse_headers(skb, sock);
3027 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3030 len += sizeof(vnet_hdr);
3031 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3034 /* Paired with WRITE_ONCE() in packet_setsockopt() */
3035 err = READ_ONCE(po->xmit)(skb);
3036 if (unlikely(err != 0)) {
3038 err = net_xmit_errno(err);
3056 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3058 struct sock *sk = sock->sk;
3059 struct packet_sock *po = pkt_sk(sk);
3061 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3062 * tpacket_snd() will redo the check safely.
3064 if (data_race(po->tx_ring.pg_vec))
3065 return tpacket_snd(po, msg);
3067 return packet_snd(sock, msg, len);
3071 * Close a PACKET socket. This is fairly simple. We immediately go
3072 * to 'closed' state and remove our protocol entry in the device list.
3075 static int packet_release(struct socket *sock)
3077 struct sock *sk = sock->sk;
3078 struct packet_sock *po;
3079 struct packet_fanout *f;
3081 union tpacket_req_u req_u;
3089 mutex_lock(&net->packet.sklist_lock);
3090 sk_del_node_init_rcu(sk);
3091 mutex_unlock(&net->packet.sklist_lock);
3094 sock_prot_inuse_add(net, sk->sk_prot, -1);
3097 spin_lock(&po->bind_lock);
3098 unregister_prot_hook(sk, false);
3099 packet_cached_dev_reset(po);
3101 if (po->prot_hook.dev) {
3102 dev_put(po->prot_hook.dev);
3103 po->prot_hook.dev = NULL;
3105 spin_unlock(&po->bind_lock);
3107 packet_flush_mclist(sk);
3110 if (po->rx_ring.pg_vec) {
3111 memset(&req_u, 0, sizeof(req_u));
3112 packet_set_ring(sk, &req_u, 1, 0);
3115 if (po->tx_ring.pg_vec) {
3116 memset(&req_u, 0, sizeof(req_u));
3117 packet_set_ring(sk, &req_u, 1, 1);
3121 f = fanout_release(sk);
3125 kfree(po->rollover);
3127 fanout_release_data(f);
3131 * Now the socket is dead. No more input will appear.
3138 skb_queue_purge(&sk->sk_receive_queue);
3139 packet_free_pending(po);
3140 sk_refcnt_debug_release(sk);
3147 * Attach a packet hook.
3150 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3153 struct packet_sock *po = pkt_sk(sk);
3154 struct net_device *dev_curr;
3157 struct net_device *dev = NULL;
3159 bool unlisted = false;
3162 spin_lock(&po->bind_lock);
3174 dev = dev_get_by_name_rcu(sock_net(sk), name);
3179 } else if (ifindex) {
3180 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3190 proto_curr = po->prot_hook.type;
3191 dev_curr = po->prot_hook.dev;
3193 need_rehook = proto_curr != proto || dev_curr != dev;
3198 /* prevents packet_notifier() from calling
3199 * register_prot_hook()
3201 WRITE_ONCE(po->num, 0);
3202 __unregister_prot_hook(sk, true);
3204 dev_curr = po->prot_hook.dev;
3206 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3210 BUG_ON(po->running);
3211 WRITE_ONCE(po->num, proto);
3212 po->prot_hook.type = proto;
3214 if (unlikely(unlisted)) {
3216 po->prot_hook.dev = NULL;
3217 WRITE_ONCE(po->ifindex, -1);
3218 packet_cached_dev_reset(po);
3220 po->prot_hook.dev = dev;
3221 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3222 packet_cached_dev_assign(po, dev);
3228 if (proto == 0 || !need_rehook)
3231 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3232 register_prot_hook(sk);
3234 sk->sk_err = ENETDOWN;
3235 if (!sock_flag(sk, SOCK_DEAD))
3236 sk->sk_error_report(sk);
3241 spin_unlock(&po->bind_lock);
3247 * Bind a packet socket to a device
3250 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3253 struct sock *sk = sock->sk;
3254 char name[sizeof(uaddr->sa_data_min) + 1];
3260 if (addr_len != sizeof(struct sockaddr))
3262 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3265 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3266 name[sizeof(uaddr->sa_data_min)] = 0;
3268 return packet_do_bind(sk, name, 0, 0);
3271 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3273 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3274 struct sock *sk = sock->sk;
3280 if (addr_len < sizeof(struct sockaddr_ll))
3282 if (sll->sll_family != AF_PACKET)
3285 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3288 static struct proto packet_proto = {
3290 .owner = THIS_MODULE,
3291 .obj_size = sizeof(struct packet_sock),
3295 * Create a packet of type SOCK_PACKET.
3298 static int packet_create(struct net *net, struct socket *sock, int protocol,
3302 struct packet_sock *po;
3303 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3306 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3308 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3309 sock->type != SOCK_PACKET)
3310 return -ESOCKTNOSUPPORT;
3312 sock->state = SS_UNCONNECTED;
3315 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3319 sock->ops = &packet_ops;
3320 if (sock->type == SOCK_PACKET)
3321 sock->ops = &packet_ops_spkt;
3323 sock_init_data(sock, sk);
3326 init_completion(&po->skb_completion);
3327 sk->sk_family = PF_PACKET;
3329 po->xmit = dev_queue_xmit;
3331 err = packet_alloc_pending(po);
3335 packet_cached_dev_reset(po);
3337 sk->sk_destruct = packet_sock_destruct;
3338 sk_refcnt_debug_inc(sk);
3341 * Attach a protocol block
3344 spin_lock_init(&po->bind_lock);
3345 mutex_init(&po->pg_vec_lock);
3346 po->rollover = NULL;
3347 po->prot_hook.func = packet_rcv;
3349 if (sock->type == SOCK_PACKET)
3350 po->prot_hook.func = packet_rcv_spkt;
3352 po->prot_hook.af_packet_priv = sk;
3353 po->prot_hook.af_packet_net = sock_net(sk);
3356 po->prot_hook.type = proto;
3357 __register_prot_hook(sk);
3360 mutex_lock(&net->packet.sklist_lock);
3361 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3362 mutex_unlock(&net->packet.sklist_lock);
3365 sock_prot_inuse_add(net, &packet_proto, 1);
3376 * Pull a packet from our receive queue and hand it to the user.
3377 * If necessary we block.
3380 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3383 struct sock *sk = sock->sk;
3384 struct sk_buff *skb;
3386 int vnet_hdr_len = 0;
3387 unsigned int origlen = 0;
3390 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3394 /* What error should we return now? EUNATTACH? */
3395 if (pkt_sk(sk)->ifindex < 0)
3399 if (flags & MSG_ERRQUEUE) {
3400 err = sock_recv_errqueue(sk, msg, len,
3401 SOL_PACKET, PACKET_TX_TIMESTAMP);
3406 * Call the generic datagram receiver. This handles all sorts
3407 * of horrible races and re-entrancy so we can forget about it
3408 * in the protocol layers.
3410 * Now it will return ENETDOWN, if device have just gone down,
3411 * but then it will block.
3414 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3417 * An error occurred so return it. Because skb_recv_datagram()
3418 * handles the blocking we don't see and worry about blocking
3425 packet_rcv_try_clear_pressure(pkt_sk(sk));
3427 if (pkt_sk(sk)->has_vnet_hdr) {
3428 err = packet_rcv_vnet(msg, skb, &len);
3431 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3434 /* You lose any data beyond the buffer you gave. If it worries
3435 * a user program they can ask the device for its MTU
3441 msg->msg_flags |= MSG_TRUNC;
3444 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3448 if (sock->type != SOCK_PACKET) {
3449 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3451 /* Original length was stored in sockaddr_ll fields */
3452 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3453 sll->sll_family = AF_PACKET;
3454 sll->sll_protocol = skb->protocol;
3457 sock_recv_ts_and_drops(msg, sk, skb);
3459 if (msg->msg_name) {
3460 const size_t max_len = min(sizeof(skb->cb),
3461 sizeof(struct sockaddr_storage));
3464 /* If the address length field is there to be filled
3465 * in, we fill it in now.
3467 if (sock->type == SOCK_PACKET) {
3468 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3469 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3470 copy_len = msg->msg_namelen;
3472 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3474 msg->msg_namelen = sll->sll_halen +
3475 offsetof(struct sockaddr_ll, sll_addr);
3476 copy_len = msg->msg_namelen;
3477 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3478 memset(msg->msg_name +
3479 offsetof(struct sockaddr_ll, sll_addr),
3480 0, sizeof(sll->sll_addr));
3481 msg->msg_namelen = sizeof(struct sockaddr_ll);
3484 if (WARN_ON_ONCE(copy_len > max_len)) {
3486 msg->msg_namelen = copy_len;
3488 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3491 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3492 struct tpacket_auxdata aux;
3494 aux.tp_status = TP_STATUS_USER;
3495 if (skb->ip_summed == CHECKSUM_PARTIAL)
3496 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3497 else if (skb->pkt_type != PACKET_OUTGOING &&
3498 skb_csum_unnecessary(skb))
3499 aux.tp_status |= TP_STATUS_CSUM_VALID;
3501 aux.tp_len = origlen;
3502 aux.tp_snaplen = skb->len;
3504 aux.tp_net = skb_network_offset(skb);
3505 if (skb_vlan_tag_present(skb)) {
3506 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3507 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3508 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3510 aux.tp_vlan_tci = 0;
3511 aux.tp_vlan_tpid = 0;
3513 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3517 * Free or return the buffer as appropriate. Again this
3518 * hides all the races and re-entrancy issues from us.
3520 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3523 skb_free_datagram(sk, skb);
3528 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3531 struct net_device *dev;
3532 struct sock *sk = sock->sk;
3537 uaddr->sa_family = AF_PACKET;
3538 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3540 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3542 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3545 return sizeof(*uaddr);
3548 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3551 struct net_device *dev;
3552 struct sock *sk = sock->sk;
3553 struct packet_sock *po = pkt_sk(sk);
3554 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3560 ifindex = READ_ONCE(po->ifindex);
3561 sll->sll_family = AF_PACKET;
3562 sll->sll_ifindex = ifindex;
3563 sll->sll_protocol = READ_ONCE(po->num);
3564 sll->sll_pkttype = 0;
3566 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3568 sll->sll_hatype = dev->type;
3569 sll->sll_halen = dev->addr_len;
3570 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3572 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3577 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3580 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3584 case PACKET_MR_MULTICAST:
3585 if (i->alen != dev->addr_len)
3588 return dev_mc_add(dev, i->addr);
3590 return dev_mc_del(dev, i->addr);
3592 case PACKET_MR_PROMISC:
3593 return dev_set_promiscuity(dev, what);
3594 case PACKET_MR_ALLMULTI:
3595 return dev_set_allmulti(dev, what);
3596 case PACKET_MR_UNICAST:
3597 if (i->alen != dev->addr_len)
3600 return dev_uc_add(dev, i->addr);
3602 return dev_uc_del(dev, i->addr);
3610 static void packet_dev_mclist_delete(struct net_device *dev,
3611 struct packet_mclist **mlp)
3613 struct packet_mclist *ml;
3615 while ((ml = *mlp) != NULL) {
3616 if (ml->ifindex == dev->ifindex) {
3617 packet_dev_mc(dev, ml, -1);
3625 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3627 struct packet_sock *po = pkt_sk(sk);
3628 struct packet_mclist *ml, *i;
3629 struct net_device *dev;
3635 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3640 if (mreq->mr_alen > dev->addr_len)
3644 i = kmalloc(sizeof(*i), GFP_KERNEL);
3649 for (ml = po->mclist; ml; ml = ml->next) {
3650 if (ml->ifindex == mreq->mr_ifindex &&
3651 ml->type == mreq->mr_type &&
3652 ml->alen == mreq->mr_alen &&
3653 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3655 /* Free the new element ... */
3661 i->type = mreq->mr_type;
3662 i->ifindex = mreq->mr_ifindex;
3663 i->alen = mreq->mr_alen;
3664 memcpy(i->addr, mreq->mr_address, i->alen);
3665 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3667 i->next = po->mclist;
3669 err = packet_dev_mc(dev, i, 1);
3671 po->mclist = i->next;
3680 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3682 struct packet_mclist *ml, **mlp;
3686 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3687 if (ml->ifindex == mreq->mr_ifindex &&
3688 ml->type == mreq->mr_type &&
3689 ml->alen == mreq->mr_alen &&
3690 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3691 if (--ml->count == 0) {
3692 struct net_device *dev;
3694 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3696 packet_dev_mc(dev, ml, -1);
3706 static void packet_flush_mclist(struct sock *sk)
3708 struct packet_sock *po = pkt_sk(sk);
3709 struct packet_mclist *ml;
3715 while ((ml = po->mclist) != NULL) {
3716 struct net_device *dev;
3718 po->mclist = ml->next;
3719 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3721 packet_dev_mc(dev, ml, -1);
3728 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3729 unsigned int optlen)
3731 struct sock *sk = sock->sk;
3732 struct packet_sock *po = pkt_sk(sk);
3735 if (level != SOL_PACKET)
3736 return -ENOPROTOOPT;
3739 case PACKET_ADD_MEMBERSHIP:
3740 case PACKET_DROP_MEMBERSHIP:
3742 struct packet_mreq_max mreq;
3744 memset(&mreq, 0, sizeof(mreq));
3745 if (len < sizeof(struct packet_mreq))
3747 if (len > sizeof(mreq))
3749 if (copy_from_sockptr(&mreq, optval, len))
3751 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3753 if (optname == PACKET_ADD_MEMBERSHIP)
3754 ret = packet_mc_add(sk, &mreq);
3756 ret = packet_mc_drop(sk, &mreq);
3760 case PACKET_RX_RING:
3761 case PACKET_TX_RING:
3763 union tpacket_req_u req_u;
3767 switch (po->tp_version) {
3770 len = sizeof(req_u.req);
3774 len = sizeof(req_u.req3);
3780 if (copy_from_sockptr(&req_u.req, optval, len))
3783 ret = packet_set_ring(sk, &req_u, 0,
3784 optname == PACKET_TX_RING);
3789 case PACKET_COPY_THRESH:
3793 if (optlen != sizeof(val))
3795 if (copy_from_sockptr(&val, optval, sizeof(val)))
3798 pkt_sk(sk)->copy_thresh = val;
3801 case PACKET_VERSION:
3805 if (optlen != sizeof(val))
3807 if (copy_from_sockptr(&val, optval, sizeof(val)))
3818 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3821 po->tp_version = val;
3827 case PACKET_RESERVE:
3831 if (optlen != sizeof(val))
3833 if (copy_from_sockptr(&val, optval, sizeof(val)))
3838 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3841 po->tp_reserve = val;
3851 if (optlen != sizeof(val))
3853 if (copy_from_sockptr(&val, optval, sizeof(val)))
3857 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3860 po->tp_loss = !!val;
3866 case PACKET_AUXDATA:
3870 if (optlen < sizeof(val))
3872 if (copy_from_sockptr(&val, optval, sizeof(val)))
3875 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3878 case PACKET_ORIGDEV:
3882 if (optlen < sizeof(val))
3884 if (copy_from_sockptr(&val, optval, sizeof(val)))
3887 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3890 case PACKET_VNET_HDR:
3894 if (sock->type != SOCK_RAW)
3896 if (optlen < sizeof(val))
3898 if (copy_from_sockptr(&val, optval, sizeof(val)))
3902 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3905 po->has_vnet_hdr = !!val;
3911 case PACKET_TIMESTAMP:
3915 if (optlen != sizeof(val))
3917 if (copy_from_sockptr(&val, optval, sizeof(val)))
3920 po->tp_tstamp = val;
3925 struct fanout_args args = { 0 };
3927 if (optlen != sizeof(int) && optlen != sizeof(args))
3929 if (copy_from_sockptr(&args, optval, optlen))
3932 return fanout_add(sk, &args);
3934 case PACKET_FANOUT_DATA:
3936 /* Paired with the WRITE_ONCE() in fanout_add() */
3937 if (!READ_ONCE(po->fanout))
3940 return fanout_set_data(po, optval, optlen);
3942 case PACKET_IGNORE_OUTGOING:
3946 if (optlen != sizeof(val))
3948 if (copy_from_sockptr(&val, optval, sizeof(val)))
3950 if (val < 0 || val > 1)
3953 WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
3956 case PACKET_TX_HAS_OFF:
3960 if (optlen != sizeof(val))
3962 if (copy_from_sockptr(&val, optval, sizeof(val)))
3966 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3969 po->tp_tx_has_off = !!val;
3975 case PACKET_QDISC_BYPASS:
3979 if (optlen != sizeof(val))
3981 if (copy_from_sockptr(&val, optval, sizeof(val)))
3984 /* Paired with all lockless reads of po->xmit */
3985 WRITE_ONCE(po->xmit, val ? packet_direct_xmit : dev_queue_xmit);
3989 return -ENOPROTOOPT;
3993 static int packet_getsockopt(struct socket *sock, int level, int optname,
3994 char __user *optval, int __user *optlen)
3997 int val, lv = sizeof(val);
3998 struct sock *sk = sock->sk;
3999 struct packet_sock *po = pkt_sk(sk);
4001 union tpacket_stats_u st;
4002 struct tpacket_rollover_stats rstats;
4005 if (level != SOL_PACKET)
4006 return -ENOPROTOOPT;
4008 if (get_user(len, optlen))
4015 case PACKET_STATISTICS:
4016 spin_lock_bh(&sk->sk_receive_queue.lock);
4017 memcpy(&st, &po->stats, sizeof(st));
4018 memset(&po->stats, 0, sizeof(po->stats));
4019 spin_unlock_bh(&sk->sk_receive_queue.lock);
4020 drops = atomic_xchg(&po->tp_drops, 0);
4022 if (po->tp_version == TPACKET_V3) {
4023 lv = sizeof(struct tpacket_stats_v3);
4024 st.stats3.tp_drops = drops;
4025 st.stats3.tp_packets += drops;
4028 lv = sizeof(struct tpacket_stats);
4029 st.stats1.tp_drops = drops;
4030 st.stats1.tp_packets += drops;
4035 case PACKET_AUXDATA:
4036 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4038 case PACKET_ORIGDEV:
4039 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4041 case PACKET_VNET_HDR:
4042 val = po->has_vnet_hdr;
4044 case PACKET_VERSION:
4045 val = po->tp_version;
4048 if (len > sizeof(int))
4050 if (len < sizeof(int))
4052 if (copy_from_user(&val, optval, len))
4056 val = sizeof(struct tpacket_hdr);
4059 val = sizeof(struct tpacket2_hdr);
4062 val = sizeof(struct tpacket3_hdr);
4068 case PACKET_RESERVE:
4069 val = po->tp_reserve;
4074 case PACKET_TIMESTAMP:
4075 val = po->tp_tstamp;
4079 ((u32)po->fanout->id |
4080 ((u32)po->fanout->type << 16) |
4081 ((u32)po->fanout->flags << 24)) :
4084 case PACKET_IGNORE_OUTGOING:
4085 val = READ_ONCE(po->prot_hook.ignore_outgoing);
4087 case PACKET_ROLLOVER_STATS:
4090 rstats.tp_all = atomic_long_read(&po->rollover->num);
4091 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4092 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4094 lv = sizeof(rstats);
4096 case PACKET_TX_HAS_OFF:
4097 val = po->tp_tx_has_off;
4099 case PACKET_QDISC_BYPASS:
4100 val = packet_use_direct_xmit(po);
4103 return -ENOPROTOOPT;
4108 if (put_user(len, optlen))
4110 if (copy_to_user(optval, data, len))
4115 static int packet_notifier(struct notifier_block *this,
4116 unsigned long msg, void *ptr)
4119 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4120 struct net *net = dev_net(dev);
4123 sk_for_each_rcu(sk, &net->packet.sklist) {
4124 struct packet_sock *po = pkt_sk(sk);
4127 case NETDEV_UNREGISTER:
4129 packet_dev_mclist_delete(dev, &po->mclist);
4133 if (dev->ifindex == po->ifindex) {
4134 spin_lock(&po->bind_lock);
4136 __unregister_prot_hook(sk, false);
4137 sk->sk_err = ENETDOWN;
4138 if (!sock_flag(sk, SOCK_DEAD))
4139 sk->sk_error_report(sk);
4141 if (msg == NETDEV_UNREGISTER) {
4142 packet_cached_dev_reset(po);
4143 WRITE_ONCE(po->ifindex, -1);
4144 if (po->prot_hook.dev)
4145 dev_put(po->prot_hook.dev);
4146 po->prot_hook.dev = NULL;
4148 spin_unlock(&po->bind_lock);
4152 if (dev->ifindex == po->ifindex) {
4153 spin_lock(&po->bind_lock);
4155 register_prot_hook(sk);
4156 spin_unlock(&po->bind_lock);
4166 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4169 struct sock *sk = sock->sk;
4174 int amount = sk_wmem_alloc_get(sk);
4176 return put_user(amount, (int __user *)arg);
4180 struct sk_buff *skb;
4183 spin_lock_bh(&sk->sk_receive_queue.lock);
4184 skb = skb_peek(&sk->sk_receive_queue);
4187 spin_unlock_bh(&sk->sk_receive_queue.lock);
4188 return put_user(amount, (int __user *)arg);
4198 case SIOCGIFBRDADDR:
4199 case SIOCSIFBRDADDR:
4200 case SIOCGIFNETMASK:
4201 case SIOCSIFNETMASK:
4202 case SIOCGIFDSTADDR:
4203 case SIOCSIFDSTADDR:
4205 return inet_dgram_ops.ioctl(sock, cmd, arg);
4209 return -ENOIOCTLCMD;
4214 static __poll_t packet_poll(struct file *file, struct socket *sock,
4217 struct sock *sk = sock->sk;
4218 struct packet_sock *po = pkt_sk(sk);
4219 __poll_t mask = datagram_poll(file, sock, wait);
4221 spin_lock_bh(&sk->sk_receive_queue.lock);
4222 if (po->rx_ring.pg_vec) {
4223 if (!packet_previous_rx_frame(po, &po->rx_ring,
4225 mask |= EPOLLIN | EPOLLRDNORM;
4227 packet_rcv_try_clear_pressure(po);
4228 spin_unlock_bh(&sk->sk_receive_queue.lock);
4229 spin_lock_bh(&sk->sk_write_queue.lock);
4230 if (po->tx_ring.pg_vec) {
4231 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4232 mask |= EPOLLOUT | EPOLLWRNORM;
4234 spin_unlock_bh(&sk->sk_write_queue.lock);
4239 /* Dirty? Well, I still did not learn better way to account
4243 static void packet_mm_open(struct vm_area_struct *vma)
4245 struct file *file = vma->vm_file;
4246 struct socket *sock = file->private_data;
4247 struct sock *sk = sock->sk;
4250 atomic_long_inc(&pkt_sk(sk)->mapped);
4253 static void packet_mm_close(struct vm_area_struct *vma)
4255 struct file *file = vma->vm_file;
4256 struct socket *sock = file->private_data;
4257 struct sock *sk = sock->sk;
4260 atomic_long_dec(&pkt_sk(sk)->mapped);
4263 static const struct vm_operations_struct packet_mmap_ops = {
4264 .open = packet_mm_open,
4265 .close = packet_mm_close,
4268 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4273 for (i = 0; i < len; i++) {
4274 if (likely(pg_vec[i].buffer)) {
4275 if (is_vmalloc_addr(pg_vec[i].buffer))
4276 vfree(pg_vec[i].buffer);
4278 free_pages((unsigned long)pg_vec[i].buffer,
4280 pg_vec[i].buffer = NULL;
4286 static char *alloc_one_pg_vec_page(unsigned long order)
4289 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4290 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4292 buffer = (char *) __get_free_pages(gfp_flags, order);
4296 /* __get_free_pages failed, fall back to vmalloc */
4297 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4301 /* vmalloc failed, lets dig into swap here */
4302 gfp_flags &= ~__GFP_NORETRY;
4303 buffer = (char *) __get_free_pages(gfp_flags, order);
4307 /* complete and utter failure */
4311 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4313 unsigned int block_nr = req->tp_block_nr;
4317 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4318 if (unlikely(!pg_vec))
4321 for (i = 0; i < block_nr; i++) {
4322 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4323 if (unlikely(!pg_vec[i].buffer))
4324 goto out_free_pgvec;
4331 free_pg_vec(pg_vec, order, block_nr);
4336 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4337 int closing, int tx_ring)
4339 struct pgv *pg_vec = NULL;
4340 struct packet_sock *po = pkt_sk(sk);
4341 unsigned long *rx_owner_map = NULL;
4342 int was_running, order = 0;
4343 struct packet_ring_buffer *rb;
4344 struct sk_buff_head *rb_queue;
4347 /* Added to avoid minimal code churn */
4348 struct tpacket_req *req = &req_u->req;
4350 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4351 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4355 if (atomic_long_read(&po->mapped))
4357 if (packet_read_pending(rb))
4361 if (req->tp_block_nr) {
4362 unsigned int min_frame_size;
4364 /* Sanity tests and some calculations */
4366 if (unlikely(rb->pg_vec))
4369 switch (po->tp_version) {
4371 po->tp_hdrlen = TPACKET_HDRLEN;
4374 po->tp_hdrlen = TPACKET2_HDRLEN;
4377 po->tp_hdrlen = TPACKET3_HDRLEN;
4382 if (unlikely((int)req->tp_block_size <= 0))
4384 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4386 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4387 if (po->tp_version >= TPACKET_V3 &&
4388 req->tp_block_size <
4389 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4391 if (unlikely(req->tp_frame_size < min_frame_size))
4393 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4396 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4397 if (unlikely(rb->frames_per_block == 0))
4399 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4401 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4406 order = get_order(req->tp_block_size);
4407 pg_vec = alloc_pg_vec(req, order);
4408 if (unlikely(!pg_vec))
4410 switch (po->tp_version) {
4412 /* Block transmit is not supported yet */
4414 init_prb_bdqc(po, rb, pg_vec, req_u);
4416 struct tpacket_req3 *req3 = &req_u->req3;
4418 if (req3->tp_retire_blk_tov ||
4419 req3->tp_sizeof_priv ||
4420 req3->tp_feature_req_word) {
4422 goto out_free_pg_vec;
4428 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4429 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4431 goto out_free_pg_vec;
4439 if (unlikely(req->tp_frame_nr))
4444 /* Detach socket from network */
4445 spin_lock(&po->bind_lock);
4446 was_running = po->running;
4449 WRITE_ONCE(po->num, 0);
4450 __unregister_prot_hook(sk, false);
4452 spin_unlock(&po->bind_lock);
4457 mutex_lock(&po->pg_vec_lock);
4458 if (closing || atomic_long_read(&po->mapped) == 0) {
4460 spin_lock_bh(&rb_queue->lock);
4461 swap(rb->pg_vec, pg_vec);
4462 if (po->tp_version <= TPACKET_V2)
4463 swap(rb->rx_owner_map, rx_owner_map);
4464 rb->frame_max = (req->tp_frame_nr - 1);
4466 rb->frame_size = req->tp_frame_size;
4467 spin_unlock_bh(&rb_queue->lock);
4469 swap(rb->pg_vec_order, order);
4470 swap(rb->pg_vec_len, req->tp_block_nr);
4472 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4473 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4474 tpacket_rcv : packet_rcv;
4475 skb_queue_purge(rb_queue);
4476 if (atomic_long_read(&po->mapped))
4477 pr_err("packet_mmap: vma is busy: %ld\n",
4478 atomic_long_read(&po->mapped));
4480 mutex_unlock(&po->pg_vec_lock);
4482 spin_lock(&po->bind_lock);
4484 WRITE_ONCE(po->num, num);
4485 register_prot_hook(sk);
4487 spin_unlock(&po->bind_lock);
4488 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4489 /* Because we don't support block-based V3 on tx-ring */
4491 prb_shutdown_retire_blk_timer(po, rb_queue);
4496 bitmap_free(rx_owner_map);
4497 free_pg_vec(pg_vec, order, req->tp_block_nr);
4503 static int packet_mmap(struct file *file, struct socket *sock,
4504 struct vm_area_struct *vma)
4506 struct sock *sk = sock->sk;
4507 struct packet_sock *po = pkt_sk(sk);
4508 unsigned long size, expected_size;
4509 struct packet_ring_buffer *rb;
4510 unsigned long start;
4517 mutex_lock(&po->pg_vec_lock);
4520 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4522 expected_size += rb->pg_vec_len
4528 if (expected_size == 0)
4531 size = vma->vm_end - vma->vm_start;
4532 if (size != expected_size)
4535 start = vma->vm_start;
4536 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4537 if (rb->pg_vec == NULL)
4540 for (i = 0; i < rb->pg_vec_len; i++) {
4542 void *kaddr = rb->pg_vec[i].buffer;
4545 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4546 page = pgv_to_page(kaddr);
4547 err = vm_insert_page(vma, start, page);
4556 atomic_long_inc(&po->mapped);
4557 vma->vm_ops = &packet_mmap_ops;
4561 mutex_unlock(&po->pg_vec_lock);
4565 static const struct proto_ops packet_ops_spkt = {
4566 .family = PF_PACKET,
4567 .owner = THIS_MODULE,
4568 .release = packet_release,
4569 .bind = packet_bind_spkt,
4570 .connect = sock_no_connect,
4571 .socketpair = sock_no_socketpair,
4572 .accept = sock_no_accept,
4573 .getname = packet_getname_spkt,
4574 .poll = datagram_poll,
4575 .ioctl = packet_ioctl,
4576 .gettstamp = sock_gettstamp,
4577 .listen = sock_no_listen,
4578 .shutdown = sock_no_shutdown,
4579 .sendmsg = packet_sendmsg_spkt,
4580 .recvmsg = packet_recvmsg,
4581 .mmap = sock_no_mmap,
4582 .sendpage = sock_no_sendpage,
4585 static const struct proto_ops packet_ops = {
4586 .family = PF_PACKET,
4587 .owner = THIS_MODULE,
4588 .release = packet_release,
4589 .bind = packet_bind,
4590 .connect = sock_no_connect,
4591 .socketpair = sock_no_socketpair,
4592 .accept = sock_no_accept,
4593 .getname = packet_getname,
4594 .poll = packet_poll,
4595 .ioctl = packet_ioctl,
4596 .gettstamp = sock_gettstamp,
4597 .listen = sock_no_listen,
4598 .shutdown = sock_no_shutdown,
4599 .setsockopt = packet_setsockopt,
4600 .getsockopt = packet_getsockopt,
4601 .sendmsg = packet_sendmsg,
4602 .recvmsg = packet_recvmsg,
4603 .mmap = packet_mmap,
4604 .sendpage = sock_no_sendpage,
4607 static const struct net_proto_family packet_family_ops = {
4608 .family = PF_PACKET,
4609 .create = packet_create,
4610 .owner = THIS_MODULE,
4613 static struct notifier_block packet_netdev_notifier = {
4614 .notifier_call = packet_notifier,
4617 #ifdef CONFIG_PROC_FS
4619 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4622 struct net *net = seq_file_net(seq);
4625 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4628 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4630 struct net *net = seq_file_net(seq);
4631 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4634 static void packet_seq_stop(struct seq_file *seq, void *v)
4640 static int packet_seq_show(struct seq_file *seq, void *v)
4642 if (v == SEQ_START_TOKEN)
4643 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4645 struct sock *s = sk_entry(v);
4646 const struct packet_sock *po = pkt_sk(s);
4649 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4651 refcount_read(&s->sk_refcnt),
4653 ntohs(READ_ONCE(po->num)),
4654 READ_ONCE(po->ifindex),
4656 atomic_read(&s->sk_rmem_alloc),
4657 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4664 static const struct seq_operations packet_seq_ops = {
4665 .start = packet_seq_start,
4666 .next = packet_seq_next,
4667 .stop = packet_seq_stop,
4668 .show = packet_seq_show,
4672 static int __net_init packet_net_init(struct net *net)
4674 mutex_init(&net->packet.sklist_lock);
4675 INIT_HLIST_HEAD(&net->packet.sklist);
4677 #ifdef CONFIG_PROC_FS
4678 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4679 sizeof(struct seq_net_private)))
4681 #endif /* CONFIG_PROC_FS */
4686 static void __net_exit packet_net_exit(struct net *net)
4688 remove_proc_entry("packet", net->proc_net);
4689 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4692 static struct pernet_operations packet_net_ops = {
4693 .init = packet_net_init,
4694 .exit = packet_net_exit,
4698 static void __exit packet_exit(void)
4700 unregister_netdevice_notifier(&packet_netdev_notifier);
4701 unregister_pernet_subsys(&packet_net_ops);
4702 sock_unregister(PF_PACKET);
4703 proto_unregister(&packet_proto);
4706 static int __init packet_init(void)
4710 rc = proto_register(&packet_proto, 0);
4713 rc = sock_register(&packet_family_ops);
4716 rc = register_pernet_subsys(&packet_net_ops);
4719 rc = register_netdevice_notifier(&packet_netdev_notifier);
4726 unregister_pernet_subsys(&packet_net_ops);
4728 sock_unregister(PF_PACKET);
4730 proto_unregister(&packet_proto);
4735 module_init(packet_init);
4736 module_exit(packet_exit);
4737 MODULE_LICENSE("GPL");
4738 MODULE_ALIAS_NETPROTO(PF_PACKET);