Lines Matching +full:no +full:- +full:read +full:- +full:rollover
1 // SPDX-License-Identifier: GPL-2.0-or-later
7 * PACKET - implements raw packet sockets.
15 * Alan Cox : new skbuff lists, look ma no backlogs!
18 * added. Also fixed the peek/read crash
22 * Alan Cox : Re-commented the code.
27 * dubious gcc output. Can you read
30 * Alan Cox : New buffers. Use sk->mac.raw.
99 - If the device has no dev->header_ops->create, there is no LL header
106 needed_headroom to be (the real WiFi header length - the fake Ethernet
108 - packet socket receives packets with pulled ll header,
112 -----------
115 mac_header -> ll header
116 data -> data
119 mac_header -> ll header
120 data -> ll header
123 mac_header -> data
126 data -> data
129 mac_header -> data. ll header is invisible to us.
130 data -> data
138 ------------
141 mac_header -> ll header
142 data -> ll header
145 mac_header -> data
146 data -> data
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
230 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
232 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
234 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
236 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
238 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
239 ((x)->kactive_blk_num+1) : 0)
254 dev = rcu_dereference(po->cached_dev); in packet_cached_dev_get()
264 rcu_assign_pointer(po->cached_dev, dev); in packet_cached_dev_assign()
269 RCU_INIT_POINTER(po->cached_dev, NULL); in packet_cached_dev_reset()
274 return po->xmit == packet_direct_xmit; in packet_use_direct_xmit()
279 struct net_device *dev = skb->dev; in packet_pick_tx_queue()
280 const struct net_device_ops *ops = dev->netdev_ops; in packet_pick_tx_queue()
285 skb->sender_cpu = cpu + 1; in packet_pick_tx_queue()
287 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); in packet_pick_tx_queue()
288 if (ops->ndo_select_queue) { in packet_pick_tx_queue()
289 queue_index = ops->ndo_select_queue(dev, skb, NULL); in packet_pick_tx_queue()
306 if (!po->running) { in __register_prot_hook()
307 if (po->fanout) in __register_prot_hook()
310 dev_add_pack(&po->prot_hook); in __register_prot_hook()
313 po->running = 1; in __register_prot_hook()
319 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); in register_prot_hook()
324 * the po->bind_lock and do a synchronize_net to make sure no
326 * of po->prot_hook. If the sync parameter is false, it is the
333 lockdep_assert_held_once(&po->bind_lock); in __unregister_prot_hook()
335 po->running = 0; in __unregister_prot_hook()
337 if (po->fanout) in __unregister_prot_hook()
340 __dev_remove_pack(&po->prot_hook); in __unregister_prot_hook()
345 spin_unlock(&po->bind_lock); in __unregister_prot_hook()
347 spin_lock(&po->bind_lock); in __unregister_prot_hook()
355 if (po->running) in unregister_prot_hook()
371 switch (po->tp_version) { in __packet_set_status()
373 h.h1->tp_status = status; in __packet_set_status()
374 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); in __packet_set_status()
377 h.h2->tp_status = status; in __packet_set_status()
378 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); in __packet_set_status()
381 h.h3->tp_status = status; in __packet_set_status()
382 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); in __packet_set_status()
399 switch (po->tp_version) { in __packet_get_status()
401 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); in __packet_get_status()
402 return h.h1->tp_status; in __packet_get_status()
404 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); in __packet_get_status()
405 return h.h2->tp_status; in __packet_get_status()
407 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); in __packet_get_status()
408 return h.h3->tp_status; in __packet_get_status()
423 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) in tpacket_get_timestamp()
427 ktime_to_timespec64_cond(skb->tstamp, ts)) in tpacket_get_timestamp()
440 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) in __packet_set_timestamp()
446 * all store the seconds in a 32-bit unsigned integer. in __packet_set_timestamp()
447 * If we create a version 4, that should have a 64-bit timestamp, in __packet_set_timestamp()
448 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit in __packet_set_timestamp()
451 switch (po->tp_version) { in __packet_set_timestamp()
453 h.h1->tp_sec = ts.tv_sec; in __packet_set_timestamp()
454 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; in __packet_set_timestamp()
457 h.h2->tp_sec = ts.tv_sec; in __packet_set_timestamp()
458 h.h2->tp_nsec = ts.tv_nsec; in __packet_set_timestamp()
461 h.h3->tp_sec = ts.tv_sec; in __packet_set_timestamp()
462 h.h3->tp_nsec = ts.tv_nsec; in __packet_set_timestamp()
470 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); in __packet_set_timestamp()
484 pg_vec_pos = position / rb->frames_per_block; in packet_lookup_frame()
485 frame_offset = position % rb->frames_per_block; in packet_lookup_frame()
487 h.raw = rb->pg_vec[pg_vec_pos].buffer + in packet_lookup_frame()
488 (frame_offset * rb->frame_size); in packet_lookup_frame()
500 return packet_lookup_frame(po, rb, rb->head, status); in packet_current_frame()
505 del_timer_sync(&pkc->retire_blk_timer); in prb_del_retire_blk_timer()
513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_shutdown_retire_blk_timer()
515 spin_lock_bh(&rb_queue->lock); in prb_shutdown_retire_blk_timer()
516 pkc->delete_blk_timer = 1; in prb_shutdown_retire_blk_timer()
517 spin_unlock_bh(&rb_queue->lock); in prb_shutdown_retire_blk_timer()
526 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_setup_retire_blk_timer()
527 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, in prb_setup_retire_blk_timer()
529 pkc->retire_blk_timer.expires = jiffies; in prb_setup_retire_blk_timer()
541 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); in prb_calc_retire_blk_tmo()
572 p1->feature_req_word = req_u->req3.tp_feature_req_word; in prb_init_ft_ops()
585 p1->knxt_seq_num = 1; in init_prb_bdqc()
586 p1->pkbdq = pg_vec; in init_prb_bdqc()
588 p1->pkblk_start = pg_vec[0].buffer; in init_prb_bdqc()
589 p1->kblk_size = req_u->req3.tp_block_size; in init_prb_bdqc()
590 p1->knum_blocks = req_u->req3.tp_block_nr; in init_prb_bdqc()
591 p1->hdrlen = po->tp_hdrlen; in init_prb_bdqc()
592 p1->version = po->tp_version; in init_prb_bdqc()
593 p1->last_kactive_blk_num = 0; in init_prb_bdqc()
594 po->stats.stats3.tp_freeze_q_cnt = 0; in init_prb_bdqc()
595 if (req_u->req3.tp_retire_blk_tov) in init_prb_bdqc()
596 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; in init_prb_bdqc()
598 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, in init_prb_bdqc()
599 req_u->req3.tp_block_size); in init_prb_bdqc()
600 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); in init_prb_bdqc()
601 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; in init_prb_bdqc()
602 rwlock_init(&p1->blk_fill_in_prog_lock); in init_prb_bdqc()
604 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); in init_prb_bdqc()
615 mod_timer(&pkc->retire_blk_timer, in _prb_refresh_rx_retire_blk_timer()
616 jiffies + pkc->tov_in_jiffies); in _prb_refresh_rx_retire_blk_timer()
617 pkc->last_kactive_blk_num = pkc->kactive_blk_num; in _prb_refresh_rx_retire_blk_timer()
624 * on packet-by-packet basis.
626 * With a 1MB block-size, on a 1Gbps line, it will take
639 * a) line-speed and b) block-size.
647 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_retire_rx_blk_timer_expired()
651 spin_lock(&po->sk.sk_receive_queue.lock); in prb_retire_rx_blk_timer_expired()
656 if (unlikely(pkc->delete_blk_timer)) in prb_retire_rx_blk_timer_expired()
670 write_lock(&pkc->blk_fill_in_prog_lock); in prb_retire_rx_blk_timer_expired()
671 write_unlock(&pkc->blk_fill_in_prog_lock); in prb_retire_rx_blk_timer_expired()
674 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { in prb_retire_rx_blk_timer_expired()
686 /* Case 1. Queue was frozen because user-space was in prb_retire_rx_blk_timer_expired()
691 * Ok, user-space is still behind. in prb_retire_rx_blk_timer_expired()
696 /* Case 2. queue was frozen,user-space caught up, in prb_retire_rx_blk_timer_expired()
701 * Thawing/timer-refresh is a side effect. in prb_retire_rx_blk_timer_expired()
713 spin_unlock(&po->sk.sk_receive_queue.lock); in prb_retire_rx_blk_timer_expired()
729 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); in prb_flush_block()
766 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; in prb_close_block()
767 struct sock *sk = &po->sk; in prb_close_block()
769 if (atomic_read(&po->tp_drops)) in prb_close_block()
772 last_pkt = (struct tpacket3_hdr *)pkc1->prev; in prb_close_block()
773 last_pkt->tp_next_offset = 0; in prb_close_block()
777 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; in prb_close_block()
778 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; in prb_close_block()
780 /* Ok, we tmo'd - so get the current time. in prb_close_block()
787 h1->ts_last_pkt.ts_sec = ts.tv_sec; in prb_close_block()
788 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; in prb_close_block()
796 sk->sk_data_ready(sk); in prb_close_block()
798 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); in prb_close_block()
803 pkc->reset_pending_on_curr_blk = 0; in prb_thaw_queue()
817 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; in prb_open_block()
825 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; in prb_open_block()
827 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
831 h1->ts_first_pkt.ts_sec = ts.tv_sec; in prb_open_block()
832 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; in prb_open_block()
834 pkc1->pkblk_start = (char *)pbd1; in prb_open_block()
835 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
837 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
840 pbd1->version = pkc1->version; in prb_open_block()
841 pkc1->prev = pkc1->nxt_offset; in prb_open_block()
842 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; in prb_open_block()
855 * 4) user-space is either sleeping or processing block '0'.
856 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
857 * it will close block-7,loop around and try to fill block '0'.
858 * call-flow:
862 * |->(BLOCK_STATUS == USER) evaluates to true
863 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
868 * re-open block-0 in near future.
870 * case and __packet_lookup_frame_in_block will check if block-0
871 * is free and can now be re-used.
876 pkc->reset_pending_on_curr_blk = 1; in prb_freeze_queue()
877 po->stats.stats3.tp_freeze_q_cnt++; in prb_freeze_queue()
910 return (void *)pkc->nxt_offset; in prb_dispatch_next_block()
922 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't in prb_retire_current_block()
927 * the timer-handler already handled this case. in prb_retire_current_block()
931 write_lock(&pkc->blk_fill_in_prog_lock); in prb_retire_current_block()
932 write_unlock(&pkc->blk_fill_in_prog_lock); in prb_retire_current_block()
946 return pkc->reset_pending_on_curr_blk; in prb_queue_frozen()
950 __releases(&pkc->blk_fill_in_prog_lock) in prb_clear_blk_fill_status()
954 read_unlock(&pkc->blk_fill_in_prog_lock); in prb_clear_blk_fill_status()
960 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); in prb_fill_rxhash()
966 ppd->hv1.tp_rxhash = 0; in prb_clear_rxhash()
972 if (skb_vlan_tag_present(pkc->skb)) { in prb_fill_vlan_info()
973 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); in prb_fill_vlan_info()
974 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); in prb_fill_vlan_info()
975 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; in prb_fill_vlan_info()
977 ppd->hv1.tp_vlan_tci = 0; in prb_fill_vlan_info()
978 ppd->hv1.tp_vlan_tpid = 0; in prb_fill_vlan_info()
979 ppd->tp_status = TP_STATUS_AVAILABLE; in prb_fill_vlan_info()
986 ppd->hv1.tp_padding = 0; in prb_run_all_ft_ops()
989 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) in prb_run_all_ft_ops()
999 __acquires(&pkc->blk_fill_in_prog_lock) in prb_fill_curr_block()
1004 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); in prb_fill_curr_block()
1005 pkc->prev = curr; in prb_fill_curr_block()
1006 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); in prb_fill_curr_block()
1009 read_lock(&pkc->blk_fill_in_prog_lock); in prb_fill_curr_block()
1013 /* Assumes caller has the sk->rx_queue.lock */
1023 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in __packet_lookup_frame_in_block()
1030 * is still in_use by user-space. in __packet_lookup_frame_in_block()
1037 * Ok, the block was released by user-space. in __packet_lookup_frame_in_block()
1047 curr = pkc->nxt_offset; in __packet_lookup_frame_in_block()
1048 pkc->skb = skb; in __packet_lookup_frame_in_block()
1049 end = (char *)pbd + pkc->kblk_size; in __packet_lookup_frame_in_block()
1069 * No free blocks are available.user_space hasn't caught up yet. in __packet_lookup_frame_in_block()
1080 switch (po->tp_version) { in packet_current_rx_frame()
1083 curr = packet_lookup_frame(po, &po->rx_ring, in packet_current_rx_frame()
1084 po->rx_ring.head, status); in packet_current_rx_frame()
1111 if (rb->prb_bdqc.kactive_blk_num) in prb_previous_blk_num()
1112 prev = rb->prb_bdqc.kactive_blk_num-1; in prb_previous_blk_num()
1114 prev = rb->prb_bdqc.knum_blocks-1; in prb_previous_blk_num()
1131 if (po->tp_version <= TPACKET_V2) in packet_previous_rx_frame()
1140 switch (po->tp_version) { in packet_increment_rx_head()
1156 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; in packet_previous_frame()
1162 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; in packet_increment_head()
1167 this_cpu_inc(*rb->pending_refcnt); in packet_inc_pending()
1172 this_cpu_dec(*rb->pending_refcnt); in packet_dec_pending()
1181 if (rb->pending_refcnt == NULL) in packet_read_pending()
1185 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); in packet_read_pending()
1192 po->rx_ring.pending_refcnt = NULL; in packet_alloc_pending()
1194 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); in packet_alloc_pending()
1195 if (unlikely(po->tx_ring.pending_refcnt == NULL)) in packet_alloc_pending()
1196 return -ENOBUFS; in packet_alloc_pending()
1203 free_percpu(po->tx_ring.pending_refcnt); in packet_free_pending()
1215 len = READ_ONCE(po->rx_ring.frame_max) + 1; in __tpacket_has_room()
1216 idx = READ_ONCE(po->rx_ring.head); in __tpacket_has_room()
1220 idx -= len; in __tpacket_has_room()
1221 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); in __tpacket_has_room()
1228 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); in __tpacket_v3_has_room()
1229 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); in __tpacket_v3_has_room()
1233 idx -= len; in __tpacket_v3_has_room()
1234 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); in __tpacket_v3_has_room()
1240 const struct sock *sk = &po->sk; in __packet_rcv_has_room()
1243 if (po->prot_hook.func != tpacket_rcv) { in __packet_rcv_has_room()
1244 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); in __packet_rcv_has_room()
1245 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) in __packet_rcv_has_room()
1246 - (skb ? skb->truesize : 0); in __packet_rcv_has_room()
1256 if (po->tp_version == TPACKET_V3) { in __packet_rcv_has_room()
1278 if (READ_ONCE(po->pressure) != pressure) in packet_rcv_has_room()
1279 WRITE_ONCE(po->pressure, pressure); in packet_rcv_has_room()
1286 if (READ_ONCE(po->pressure) && in packet_rcv_try_clear_pressure()
1288 WRITE_ONCE(po->pressure, 0); in packet_rcv_try_clear_pressure()
1293 skb_queue_purge(&sk->sk_error_queue); in packet_sock_destruct()
1295 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); in packet_sock_destruct()
1296 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); in packet_sock_destruct()
1308 u32 *history = po->rollover->history; in fanout_flow_is_huge()
1337 unsigned int val = atomic_inc_return(&f->rr_cur); in fanout_demux_lb()
1364 po = pkt_sk(rcu_dereference(f->arr[idx])); in fanout_demux_rollover()
1374 i = j = min_t(int, po->rollover->sock, num - 1); in fanout_demux_rollover()
1376 po_next = pkt_sk(rcu_dereference(f->arr[i])); in fanout_demux_rollover()
1377 if (po_next != po_skip && !READ_ONCE(po_next->pressure) && in fanout_demux_rollover()
1380 po->rollover->sock = i; in fanout_demux_rollover()
1381 atomic_long_inc(&po->rollover->num); in fanout_demux_rollover()
1383 atomic_long_inc(&po->rollover->num_huge); in fanout_demux_rollover()
1391 atomic_long_inc(&po->rollover->num_failed); in fanout_demux_rollover()
1410 prog = rcu_dereference(f->bpf_prog); in fanout_demux_bpf()
1420 return f->flags & (flag >> 8); in fanout_has_flag()
1426 struct packet_fanout *f = pt->af_packet_priv; in packet_rcv_fanout()
1427 unsigned int num = READ_ONCE(f->num_members); in packet_rcv_fanout()
1428 struct net *net = read_pnet(&f->net); in packet_rcv_fanout()
1442 switch (f->type) { in packet_rcv_fanout()
1471 po = pkt_sk(rcu_dereference(f->arr[idx])); in packet_rcv_fanout()
1472 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); in packet_rcv_fanout()
1482 struct packet_fanout *f = po->fanout; in __fanout_link()
1484 spin_lock(&f->lock); in __fanout_link()
1485 rcu_assign_pointer(f->arr[f->num_members], sk); in __fanout_link()
1487 f->num_members++; in __fanout_link()
1488 if (f->num_members == 1) in __fanout_link()
1489 dev_add_pack(&f->prot_hook); in __fanout_link()
1490 spin_unlock(&f->lock); in __fanout_link()
1495 struct packet_fanout *f = po->fanout; in __fanout_unlink()
1498 spin_lock(&f->lock); in __fanout_unlink()
1499 for (i = 0; i < f->num_members; i++) { in __fanout_unlink()
1500 if (rcu_dereference_protected(f->arr[i], in __fanout_unlink()
1501 lockdep_is_held(&f->lock)) == sk) in __fanout_unlink()
1504 BUG_ON(i >= f->num_members); in __fanout_unlink()
1505 rcu_assign_pointer(f->arr[i], in __fanout_unlink()
1506 rcu_dereference_protected(f->arr[f->num_members - 1], in __fanout_unlink()
1507 lockdep_is_held(&f->lock))); in __fanout_unlink()
1508 f->num_members--; in __fanout_unlink()
1509 if (f->num_members == 0) in __fanout_unlink()
1510 __dev_remove_pack(&f->prot_hook); in __fanout_unlink()
1511 spin_unlock(&f->lock); in __fanout_unlink()
1516 if (sk->sk_family != PF_PACKET) in match_fanout_group()
1519 return ptype->af_packet_priv == pkt_sk(sk)->fanout; in match_fanout_group()
1524 switch (f->type) { in fanout_init_data()
1526 atomic_set(&f->rr_cur, 0); in fanout_init_data()
1530 RCU_INIT_POINTER(f->bpf_prog, NULL); in fanout_init_data()
1539 spin_lock(&f->lock); in __fanout_set_data_bpf()
1540 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); in __fanout_set_data_bpf()
1541 rcu_assign_pointer(f->bpf_prog, new); in __fanout_set_data_bpf()
1542 spin_unlock(&f->lock); in __fanout_set_data_bpf()
1557 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) in fanout_set_data_cbpf()
1558 return -EPERM; in fanout_set_data_cbpf()
1568 __fanout_set_data_bpf(po->fanout, new); in fanout_set_data_cbpf()
1578 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) in fanout_set_data_ebpf()
1579 return -EPERM; in fanout_set_data_ebpf()
1581 return -EINVAL; in fanout_set_data_ebpf()
1583 return -EFAULT; in fanout_set_data_ebpf()
1589 __fanout_set_data_bpf(po->fanout, new); in fanout_set_data_ebpf()
1596 switch (po->fanout->type) { in fanout_set_data()
1602 return -EINVAL; in fanout_set_data()
1608 switch (f->type) { in fanout_release_data()
1620 if (f->id == candidate_id && in __fanout_id_is_free()
1621 read_pnet(&f->net) == sock_net(sk)) { in __fanout_id_is_free()
1647 struct packet_rollover *rollover = NULL; in fanout_add() local
1649 u16 type_flags = args->type_flags; in fanout_add()
1653 u16 id = args->id; in fanout_add()
1659 return -EINVAL; in fanout_add()
1670 return -EINVAL; in fanout_add()
1675 err = -EALREADY; in fanout_add()
1676 if (po->fanout) in fanout_add()
1681 err = -ENOMEM; in fanout_add()
1682 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); in fanout_add()
1683 if (!rollover) in fanout_add()
1685 atomic_long_set(&rollover->num, 0); in fanout_add()
1686 atomic_long_set(&rollover->num_huge, 0); in fanout_add()
1687 atomic_long_set(&rollover->num_failed, 0); in fanout_add()
1692 err = -EINVAL; in fanout_add()
1696 err = -ENOMEM; in fanout_add()
1705 if (f->id == id && in fanout_add()
1706 read_pnet(&f->net) == sock_net(sk)) { in fanout_add()
1711 err = -EINVAL; in fanout_add()
1713 if (match->flags != flags) in fanout_add()
1715 if (args->max_num_members && in fanout_add()
1716 args->max_num_members != match->max_num_members) in fanout_add()
1719 if (args->max_num_members > PACKET_FANOUT_MAX) in fanout_add()
1721 if (!args->max_num_members) in fanout_add()
1723 args->max_num_members = 256; in fanout_add()
1724 err = -ENOMEM; in fanout_add()
1725 match = kvzalloc(struct_size(match, arr, args->max_num_members), in fanout_add()
1729 write_pnet(&match->net, sock_net(sk)); in fanout_add()
1730 match->id = id; in fanout_add()
1731 match->type = type; in fanout_add()
1732 match->flags = flags; in fanout_add()
1733 INIT_LIST_HEAD(&match->list); in fanout_add()
1734 spin_lock_init(&match->lock); in fanout_add()
1735 refcount_set(&match->sk_ref, 0); in fanout_add()
1737 match->prot_hook.type = po->prot_hook.type; in fanout_add()
1738 match->prot_hook.dev = po->prot_hook.dev; in fanout_add()
1739 match->prot_hook.func = packet_rcv_fanout; in fanout_add()
1740 match->prot_hook.af_packet_priv = match; in fanout_add()
1741 match->prot_hook.id_match = match_fanout_group; in fanout_add()
1742 match->max_num_members = args->max_num_members; in fanout_add()
1743 list_add(&match->list, &fanout_list); in fanout_add()
1745 err = -EINVAL; in fanout_add()
1747 spin_lock(&po->bind_lock); in fanout_add()
1748 if (po->running && in fanout_add()
1749 match->type == type && in fanout_add()
1750 match->prot_hook.type == po->prot_hook.type && in fanout_add()
1751 match->prot_hook.dev == po->prot_hook.dev) { in fanout_add()
1752 err = -ENOSPC; in fanout_add()
1753 if (refcount_read(&match->sk_ref) < match->max_num_members) { in fanout_add()
1754 __dev_remove_pack(&po->prot_hook); in fanout_add()
1755 po->fanout = match; in fanout_add()
1756 po->rollover = rollover; in fanout_add()
1757 rollover = NULL; in fanout_add()
1758 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); in fanout_add()
1763 spin_unlock(&po->bind_lock); in fanout_add()
1765 if (err && !refcount_read(&match->sk_ref)) { in fanout_add()
1766 list_del(&match->list); in fanout_add()
1771 kfree(rollover); in fanout_add()
1776 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1777 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1787 f = po->fanout; in fanout_release()
1789 po->fanout = NULL; in fanout_release()
1791 if (refcount_dec_and_test(&f->sk_ref)) in fanout_release()
1792 list_del(&f->list); in fanout_release()
1804 /* Earlier code assumed this would be a VLAN pkt, double-check in packet_extra_vlan_len_allowed()
1808 if (unlikely(dev->type != ARPHRD_ETHER)) in packet_extra_vlan_len_allowed()
1812 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); in packet_extra_vlan_len_allowed()
1830 sk = pt->af_packet_priv; in packet_rcv_spkt()
1839 * For outgoing ones skb->data == skb_mac_header(skb) in packet_rcv_spkt()
1843 if (skb->pkt_type == PACKET_LOOPBACK) in packet_rcv_spkt()
1859 spkt = &PACKET_SKB_CB(skb)->sa.pkt; in packet_rcv_spkt()
1861 skb_push(skb, skb->data - skb_mac_header(skb)); in packet_rcv_spkt()
1867 spkt->spkt_family = dev->type; in packet_rcv_spkt()
1868 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); in packet_rcv_spkt()
1869 spkt->spkt_protocol = skb->protocol; in packet_rcv_spkt()
1887 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && in packet_parse_headers()
1888 sock->type == SOCK_RAW) { in packet_parse_headers()
1890 skb->protocol = dev_parse_header_protocol(skb); in packet_parse_headers()
1904 struct sock *sk = sock->sk; in packet_sendmsg_spkt()
1905 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); in packet_sendmsg_spkt()
1918 if (msg->msg_namelen < sizeof(struct sockaddr)) in packet_sendmsg_spkt()
1919 return -EINVAL; in packet_sendmsg_spkt()
1920 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) in packet_sendmsg_spkt()
1921 proto = saddr->spkt_protocol; in packet_sendmsg_spkt()
1923 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ in packet_sendmsg_spkt()
1929 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; in packet_sendmsg_spkt()
1932 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); in packet_sendmsg_spkt()
1933 err = -ENODEV; in packet_sendmsg_spkt()
1937 err = -ENETDOWN; in packet_sendmsg_spkt()
1938 if (!(dev->flags & IFF_UP)) in packet_sendmsg_spkt()
1948 err = -EPROTONOSUPPORT; in packet_sendmsg_spkt()
1954 err = -EMSGSIZE; in packet_sendmsg_spkt()
1955 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) in packet_sendmsg_spkt()
1960 int tlen = dev->needed_tailroom; in packet_sendmsg_spkt()
1961 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; in packet_sendmsg_spkt()
1966 return -ENOBUFS; in packet_sendmsg_spkt()
1976 skb->data -= hhlen; in packet_sendmsg_spkt()
1977 skb->tail -= hhlen; in packet_sendmsg_spkt()
1987 if (!dev_validate_header(dev, skb->data, len)) { in packet_sendmsg_spkt()
1988 err = -EINVAL; in packet_sendmsg_spkt()
1991 if (len > (dev->mtu + dev->hard_header_len + extra_len) && in packet_sendmsg_spkt()
1993 err = -EMSGSIZE; in packet_sendmsg_spkt()
1998 if (msg->msg_controllen) { in packet_sendmsg_spkt()
2004 skb->protocol = proto; in packet_sendmsg_spkt()
2005 skb->dev = dev; in packet_sendmsg_spkt()
2006 skb->priority = sk->sk_priority; in packet_sendmsg_spkt()
2007 skb->mark = sk->sk_mark; in packet_sendmsg_spkt()
2008 skb->tstamp = sockc.transmit_time; in packet_sendmsg_spkt()
2013 skb->no_fcs = 1; in packet_sendmsg_spkt()
2035 filter = rcu_dereference(sk->sk_filter); in run_filter()
2037 res = bpf_prog_run_clear_cb(filter->prog, skb); in run_filter()
2049 return -EINVAL; in packet_rcv_vnet()
2050 *len -= sizeof(vnet_hdr); in packet_rcv_vnet()
2053 return -EINVAL; in packet_rcv_vnet()
2062 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2063 * and skb->cb are mangled. It works because (and until) packets
2076 u8 *skb_head = skb->data; in packet_rcv()
2077 int skb_len = skb->len; in packet_rcv()
2081 if (skb->pkt_type == PACKET_LOOPBACK) in packet_rcv()
2084 sk = pt->af_packet_priv; in packet_rcv()
2090 skb->dev = dev; in packet_rcv()
2100 if (sk->sk_type != SOCK_DGRAM) in packet_rcv()
2101 skb_push(skb, skb->data - skb_mac_header(skb)); in packet_rcv()
2102 else if (skb->pkt_type == PACKET_OUTGOING) { in packet_rcv()
2108 snaplen = skb->len; in packet_rcv()
2116 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) in packet_rcv()
2124 if (skb_head != skb->data) { in packet_rcv()
2125 skb->data = skb_head; in packet_rcv()
2126 skb->len = skb_len; in packet_rcv()
2132 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); in packet_rcv()
2134 sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_rcv()
2135 sll->sll_hatype = dev->type; in packet_rcv()
2136 sll->sll_pkttype = skb->pkt_type; in packet_rcv()
2137 if (unlikely(po->origdev)) in packet_rcv()
2138 sll->sll_ifindex = orig_dev->ifindex; in packet_rcv()
2140 sll->sll_ifindex = dev->ifindex; in packet_rcv()
2142 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); in packet_rcv()
2144 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). in packet_rcv()
2147 PACKET_SKB_CB(skb)->sa.origlen = skb->len; in packet_rcv()
2153 skb->dev = NULL; in packet_rcv()
2159 spin_lock(&sk->sk_receive_queue.lock); in packet_rcv()
2160 po->stats.stats1.tp_packets++; in packet_rcv()
2162 __skb_queue_tail(&sk->sk_receive_queue, skb); in packet_rcv()
2163 spin_unlock(&sk->sk_receive_queue.lock); in packet_rcv()
2164 sk->sk_data_ready(sk); in packet_rcv()
2169 atomic_inc(&po->tp_drops); in packet_rcv()
2170 atomic_inc(&sk->sk_drops); in packet_rcv()
2173 if (skb_head != skb->data && skb_shared(skb)) { in packet_rcv()
2174 skb->data = skb_head; in packet_rcv()
2175 skb->len = skb_len; in packet_rcv()
2192 u8 *skb_head = skb->data; in tpacket_rcv()
2193 int skb_len = skb->len; in tpacket_rcv()
2212 if (skb->pkt_type == PACKET_LOOPBACK) in tpacket_rcv()
2215 sk = pt->af_packet_priv; in tpacket_rcv()
2222 if (sk->sk_type != SOCK_DGRAM) in tpacket_rcv()
2223 skb_push(skb, skb->data - skb_mac_header(skb)); in tpacket_rcv()
2224 else if (skb->pkt_type == PACKET_OUTGOING) { in tpacket_rcv()
2230 snaplen = skb->len; in tpacket_rcv()
2238 atomic_inc(&po->tp_drops); in tpacket_rcv()
2242 if (skb->ip_summed == CHECKSUM_PARTIAL) in tpacket_rcv()
2244 else if (skb->pkt_type != PACKET_OUTGOING && in tpacket_rcv()
2245 (skb->ip_summed == CHECKSUM_COMPLETE || in tpacket_rcv()
2252 if (sk->sk_type == SOCK_DGRAM) { in tpacket_rcv()
2253 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + in tpacket_rcv()
2254 po->tp_reserve; in tpacket_rcv()
2257 netoff = TPACKET_ALIGN(po->tp_hdrlen + in tpacket_rcv()
2259 po->tp_reserve; in tpacket_rcv()
2260 if (po->has_vnet_hdr) { in tpacket_rcv()
2264 macoff = netoff - maclen; in tpacket_rcv()
2267 atomic_inc(&po->tp_drops); in tpacket_rcv()
2270 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2271 if (macoff + snaplen > po->rx_ring.frame_size) { in tpacket_rcv()
2272 if (po->copy_thresh && in tpacket_rcv()
2273 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { in tpacket_rcv()
2278 skb_head = skb->data; in tpacket_rcv()
2283 snaplen = po->rx_ring.frame_size - macoff; in tpacket_rcv()
2290 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { in tpacket_rcv()
2293 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; in tpacket_rcv()
2299 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; in tpacket_rcv()
2303 spin_lock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2309 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2310 slot_id = po->rx_ring.head; in tpacket_rcv()
2311 if (test_bit(slot_id, po->rx_ring.rx_owner_map)) in tpacket_rcv()
2313 __set_bit(slot_id, po->rx_ring.rx_owner_map); in tpacket_rcv()
2317 virtio_net_hdr_from_skb(skb, h.raw + macoff - in tpacket_rcv()
2320 if (po->tp_version == TPACKET_V3) in tpacket_rcv()
2321 prb_clear_blk_fill_status(&po->rx_ring); in tpacket_rcv()
2325 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2326 packet_increment_rx_head(po, &po->rx_ring); in tpacket_rcv()
2328 * LOSING will be reported till you read the stats, in tpacket_rcv()
2329 * because it's COR - Clear On Read. in tpacket_rcv()
2333 if (atomic_read(&po->tp_drops)) in tpacket_rcv()
2337 po->stats.stats1.tp_packets++; in tpacket_rcv()
2340 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); in tpacket_rcv()
2342 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2350 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE); in tpacket_rcv()
2356 switch (po->tp_version) { in tpacket_rcv()
2358 h.h1->tp_len = skb->len; in tpacket_rcv()
2359 h.h1->tp_snaplen = snaplen; in tpacket_rcv()
2360 h.h1->tp_mac = macoff; in tpacket_rcv()
2361 h.h1->tp_net = netoff; in tpacket_rcv()
2362 h.h1->tp_sec = ts.tv_sec; in tpacket_rcv()
2363 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; in tpacket_rcv()
2367 h.h2->tp_len = skb->len; in tpacket_rcv()
2368 h.h2->tp_snaplen = snaplen; in tpacket_rcv()
2369 h.h2->tp_mac = macoff; in tpacket_rcv()
2370 h.h2->tp_net = netoff; in tpacket_rcv()
2371 h.h2->tp_sec = ts.tv_sec; in tpacket_rcv()
2372 h.h2->tp_nsec = ts.tv_nsec; in tpacket_rcv()
2374 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); in tpacket_rcv()
2375 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); in tpacket_rcv()
2378 h.h2->tp_vlan_tci = 0; in tpacket_rcv()
2379 h.h2->tp_vlan_tpid = 0; in tpacket_rcv()
2381 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); in tpacket_rcv()
2388 h.h3->tp_status |= status; in tpacket_rcv()
2389 h.h3->tp_len = skb->len; in tpacket_rcv()
2390 h.h3->tp_snaplen = snaplen; in tpacket_rcv()
2391 h.h3->tp_mac = macoff; in tpacket_rcv()
2392 h.h3->tp_net = netoff; in tpacket_rcv()
2393 h.h3->tp_sec = ts.tv_sec; in tpacket_rcv()
2394 h.h3->tp_nsec = ts.tv_nsec; in tpacket_rcv()
2395 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); in tpacket_rcv()
2403 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); in tpacket_rcv()
2404 sll->sll_family = AF_PACKET; in tpacket_rcv()
2405 sll->sll_hatype = dev->type; in tpacket_rcv()
2406 sll->sll_protocol = skb->protocol; in tpacket_rcv()
2407 sll->sll_pkttype = skb->pkt_type; in tpacket_rcv()
2408 if (unlikely(po->origdev)) in tpacket_rcv()
2409 sll->sll_ifindex = orig_dev->ifindex; in tpacket_rcv()
2411 sll->sll_ifindex = dev->ifindex; in tpacket_rcv()
2416 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2428 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2429 spin_lock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2431 __clear_bit(slot_id, po->rx_ring.rx_owner_map); in tpacket_rcv()
2432 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2433 sk->sk_data_ready(sk); in tpacket_rcv()
2434 } else if (po->tp_version == TPACKET_V3) { in tpacket_rcv()
2435 prb_clear_blk_fill_status(&po->rx_ring); in tpacket_rcv()
2439 if (skb_head != skb->data && skb_shared(skb)) { in tpacket_rcv()
2440 skb->data = skb_head; in tpacket_rcv()
2441 skb->len = skb_len; in tpacket_rcv()
2451 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2452 atomic_inc(&po->tp_drops); in tpacket_rcv()
2455 sk->sk_data_ready(sk); in tpacket_rcv()
2462 struct packet_sock *po = pkt_sk(skb->sk); in tpacket_destruct_skb()
2464 if (likely(po->tx_ring.pg_vec)) { in tpacket_destruct_skb()
2469 packet_dec_pending(&po->tx_ring); in tpacket_destruct_skb()
2474 if (!packet_read_pending(&po->tx_ring)) in tpacket_destruct_skb()
2475 complete(&po->skb_completion); in tpacket_destruct_skb()
2483 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && in __packet_snd_vnet_parse()
2484 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + in __packet_snd_vnet_parse()
2485 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > in __packet_snd_vnet_parse()
2486 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) in __packet_snd_vnet_parse()
2487 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), in __packet_snd_vnet_parse()
2488 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + in __packet_snd_vnet_parse()
2489 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); in __packet_snd_vnet_parse()
2491 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) in __packet_snd_vnet_parse()
2492 return -EINVAL; in __packet_snd_vnet_parse()
2501 return -EINVAL; in packet_snd_vnet_parse()
2502 *len -= sizeof(*vnet_hdr); in packet_snd_vnet_parse()
2504 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) in packet_snd_vnet_parse()
2505 return -EFAULT; in packet_snd_vnet_parse()
2517 struct socket *sock = po->sk.sk_socket; in tpacket_fill_skb()
2523 skb->protocol = proto; in tpacket_fill_skb()
2524 skb->dev = dev; in tpacket_fill_skb()
2525 skb->priority = po->sk.sk_priority; in tpacket_fill_skb()
2526 skb->mark = po->sk.sk_mark; in tpacket_fill_skb()
2527 skb->tstamp = sockc->transmit_time; in tpacket_fill_skb()
2528 skb_setup_tx_timestamp(skb, sockc->tsflags); in tpacket_fill_skb()
2536 if (sock->type == SOCK_DGRAM) { in tpacket_fill_skb()
2540 return -EINVAL; in tpacket_fill_skb()
2544 skb_push(skb, dev->hard_header_len); in tpacket_fill_skb()
2545 skb_put(skb, copylen - dev->hard_header_len); in tpacket_fill_skb()
2549 if (!dev_validate_header(dev, skb->data, hdrlen)) in tpacket_fill_skb()
2550 return -EINVAL; in tpacket_fill_skb()
2553 to_write -= hdrlen; in tpacket_fill_skb()
2557 len_max = PAGE_SIZE - offset; in tpacket_fill_skb()
2560 skb->data_len = to_write; in tpacket_fill_skb()
2561 skb->len += to_write; in tpacket_fill_skb()
2562 skb->truesize += to_write; in tpacket_fill_skb()
2563 refcount_add(to_write, &po->sk.sk_wmem_alloc); in tpacket_fill_skb()
2566 nr_frags = skb_shinfo(skb)->nr_frags; in tpacket_fill_skb()
2571 return -EFAULT; in tpacket_fill_skb()
2579 to_write -= len; in tpacket_fill_skb()
2598 switch (po->tp_version) { in tpacket_parse_header()
2600 if (ph.h3->tp_next_offset != 0) { in tpacket_parse_header()
2602 return -EINVAL; in tpacket_parse_header()
2604 tp_len = ph.h3->tp_len; in tpacket_parse_header()
2607 tp_len = ph.h2->tp_len; in tpacket_parse_header()
2610 tp_len = ph.h1->tp_len; in tpacket_parse_header()
2615 return -EMSGSIZE; in tpacket_parse_header()
2618 if (unlikely(po->tp_tx_has_off)) { in tpacket_parse_header()
2621 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); in tpacket_parse_header()
2622 off_max = po->tx_ring.frame_size - tp_len; in tpacket_parse_header()
2623 if (po->sk.sk_type == SOCK_DGRAM) { in tpacket_parse_header()
2624 switch (po->tp_version) { in tpacket_parse_header()
2626 off = ph.h3->tp_net; in tpacket_parse_header()
2629 off = ph.h2->tp_net; in tpacket_parse_header()
2632 off = ph.h1->tp_net; in tpacket_parse_header()
2636 switch (po->tp_version) { in tpacket_parse_header()
2638 off = ph.h3->tp_mac; in tpacket_parse_header()
2641 off = ph.h2->tp_mac; in tpacket_parse_header()
2644 off = ph.h1->tp_mac; in tpacket_parse_header()
2649 return -EINVAL; in tpacket_parse_header()
2651 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); in tpacket_parse_header()
2667 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); in tpacket_snd()
2668 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); in tpacket_snd()
2677 mutex_lock(&po->pg_vec_lock); in tpacket_snd()
2682 if (unlikely(!po->tx_ring.pg_vec)) { in tpacket_snd()
2683 err = -EBUSY; in tpacket_snd()
2688 proto = READ_ONCE(po->num); in tpacket_snd()
2690 err = -EINVAL; in tpacket_snd()
2691 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) in tpacket_snd()
2693 if (msg->msg_namelen < (saddr->sll_halen in tpacket_snd()
2697 proto = saddr->sll_protocol; in tpacket_snd()
2698 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); in tpacket_snd()
2699 if (po->sk.sk_socket->type == SOCK_DGRAM) { in tpacket_snd()
2700 if (dev && msg->msg_namelen < dev->addr_len + in tpacket_snd()
2703 addr = saddr->sll_addr; in tpacket_snd()
2707 err = -ENXIO; in tpacket_snd()
2710 err = -ENETDOWN; in tpacket_snd()
2711 if (unlikely(!(dev->flags & IFF_UP))) in tpacket_snd()
2714 sockcm_init(&sockc, &po->sk); in tpacket_snd()
2715 if (msg->msg_controllen) { in tpacket_snd()
2716 err = sock_cmsg_send(&po->sk, msg, &sockc); in tpacket_snd()
2721 if (po->sk.sk_socket->type == SOCK_RAW) in tpacket_snd()
2722 reserve = dev->hard_header_len; in tpacket_snd()
2723 size_max = po->tx_ring.frame_size in tpacket_snd()
2724 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); in tpacket_snd()
2726 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) in tpacket_snd()
2727 size_max = dev->mtu + reserve + VLAN_HLEN; in tpacket_snd()
2729 reinit_completion(&po->skb_completion); in tpacket_snd()
2732 ph = packet_current_frame(po, &po->tx_ring, in tpacket_snd()
2736 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); in tpacket_snd()
2737 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); in tpacket_snd()
2739 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; in tpacket_snd()
2754 tlen = dev->needed_tailroom; in tpacket_snd()
2755 if (po->has_vnet_hdr) { in tpacket_snd()
2758 tp_len -= sizeof(*vnet_hdr); in tpacket_snd()
2761 tp_len = -EINVAL; in tpacket_snd()
2765 vnet_hdr->hdr_len); in tpacket_snd()
2767 copylen = max_t(int, copylen, dev->hard_header_len); in tpacket_snd()
2768 skb = sock_alloc_send_skb(&po->sk, in tpacket_snd()
2770 (copylen - dev->hard_header_len), in tpacket_snd()
2782 tp_len > dev->mtu + reserve && in tpacket_snd()
2783 !po->has_vnet_hdr && in tpacket_snd()
2785 tp_len = -EMSGSIZE; in tpacket_snd()
2789 if (po->tp_loss) { in tpacket_snd()
2792 packet_increment_head(&po->tx_ring); in tpacket_snd()
2802 if (po->has_vnet_hdr) { in tpacket_snd()
2804 tp_len = -EINVAL; in tpacket_snd()
2810 skb->destructor = tpacket_destruct_skb; in tpacket_snd()
2812 packet_inc_pending(&po->tx_ring); in tpacket_snd()
2815 err = po->xmit(skb); in tpacket_snd()
2830 packet_increment_head(&po->tx_ring); in tpacket_snd()
2834 * to call it as it's per_cpu variable, but in fast-path in tpacket_snd()
2835 * we already short-circuit the loop with the first in tpacket_snd()
2839 (need_wait && packet_read_pending(&po->tx_ring)))); in tpacket_snd()
2850 mutex_unlock(&po->pg_vec_lock); in tpacket_snd()
2865 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, in packet_alloc_skb()
2872 skb->data_len = len - linear; in packet_alloc_skb()
2873 skb->len += len - linear; in packet_alloc_skb()
2880 struct sock *sk = sock->sk; in packet_snd()
2881 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); in packet_snd()
2901 proto = READ_ONCE(po->num); in packet_snd()
2903 err = -EINVAL; in packet_snd()
2904 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) in packet_snd()
2906 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) in packet_snd()
2908 proto = saddr->sll_protocol; in packet_snd()
2909 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); in packet_snd()
2910 if (sock->type == SOCK_DGRAM) { in packet_snd()
2911 if (dev && msg->msg_namelen < dev->addr_len + in packet_snd()
2914 addr = saddr->sll_addr; in packet_snd()
2918 err = -ENXIO; in packet_snd()
2921 err = -ENETDOWN; in packet_snd()
2922 if (unlikely(!(dev->flags & IFF_UP))) in packet_snd()
2926 sockc.mark = sk->sk_mark; in packet_snd()
2927 if (msg->msg_controllen) { in packet_snd()
2933 if (sock->type == SOCK_RAW) in packet_snd()
2934 reserve = dev->hard_header_len; in packet_snd()
2935 if (po->has_vnet_hdr) { in packet_snd()
2944 err = -EPROTONOSUPPORT; in packet_snd()
2950 err = -EMSGSIZE; in packet_snd()
2952 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) in packet_snd()
2955 err = -ENOBUFS; in packet_snd()
2957 tlen = dev->needed_tailroom; in packet_snd()
2959 linear = max(linear, min_t(int, len, dev->hard_header_len)); in packet_snd()
2961 msg->msg_flags & MSG_DONTWAIT, &err); in packet_snd()
2967 err = -EINVAL; in packet_snd()
2968 if (sock->type == SOCK_DGRAM) { in packet_snd()
2973 skb_reserve(skb, -reserve); in packet_snd()
2975 dev->min_header_len != dev->hard_header_len) in packet_snd()
2979 /* Returns -EFAULT on error */ in packet_snd()
2980 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); in packet_snd()
2984 if (sock->type == SOCK_RAW && in packet_snd()
2985 !dev_validate_header(dev, skb->data, len)) { in packet_snd()
2986 err = -EINVAL; in packet_snd()
2992 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && in packet_snd()
2994 err = -EMSGSIZE; in packet_snd()
2998 skb->protocol = proto; in packet_snd()
2999 skb->dev = dev; in packet_snd()
3000 skb->priority = sk->sk_priority; in packet_snd()
3001 skb->mark = sockc.mark; in packet_snd()
3002 skb->tstamp = sockc.transmit_time; in packet_snd()
3015 skb->no_fcs = 1; in packet_snd()
3017 err = po->xmit(skb); in packet_snd()
3035 struct sock *sk = sock->sk; in packet_sendmsg()
3041 if (data_race(po->tx_ring.pg_vec)) in packet_sendmsg()
3054 struct sock *sk = sock->sk; in packet_release()
3066 mutex_lock(&net->packet.sklist_lock); in packet_release()
3068 mutex_unlock(&net->packet.sklist_lock); in packet_release()
3071 sock_prot_inuse_add(net, sk->sk_prot, -1); in packet_release()
3074 spin_lock(&po->bind_lock); in packet_release()
3078 if (po->prot_hook.dev) { in packet_release()
3079 dev_put(po->prot_hook.dev); in packet_release()
3080 po->prot_hook.dev = NULL; in packet_release()
3082 spin_unlock(&po->bind_lock); in packet_release()
3087 if (po->rx_ring.pg_vec) { in packet_release()
3092 if (po->tx_ring.pg_vec) { in packet_release()
3102 kfree(po->rollover); in packet_release()
3108 * Now the socket is dead. No more input will appear. in packet_release()
3111 sock->sk = NULL; in packet_release()
3115 skb_queue_purge(&sk->sk_receive_queue); in packet_release()
3139 spin_lock(&po->bind_lock); in packet_do_bind()
3142 if (po->fanout) { in packet_do_bind()
3143 ret = -EINVAL; in packet_do_bind()
3150 ret = -ENODEV; in packet_do_bind()
3156 ret = -ENODEV; in packet_do_bind()
3163 proto_curr = po->prot_hook.type; in packet_do_bind()
3164 dev_curr = po->prot_hook.dev; in packet_do_bind()
3169 if (po->running) { in packet_do_bind()
3174 WRITE_ONCE(po->num, 0); in packet_do_bind()
3177 dev_curr = po->prot_hook.dev; in packet_do_bind()
3180 dev->ifindex); in packet_do_bind()
3183 BUG_ON(po->running); in packet_do_bind()
3184 WRITE_ONCE(po->num, proto); in packet_do_bind()
3185 po->prot_hook.type = proto; in packet_do_bind()
3189 po->prot_hook.dev = NULL; in packet_do_bind()
3190 WRITE_ONCE(po->ifindex, -1); in packet_do_bind()
3193 po->prot_hook.dev = dev; in packet_do_bind()
3194 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); in packet_do_bind()
3203 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { in packet_do_bind()
3206 sk->sk_err = ENETDOWN; in packet_do_bind()
3213 spin_unlock(&po->bind_lock); in packet_do_bind()
3225 struct sock *sk = sock->sk; in packet_bind_spkt()
3226 char name[sizeof(uaddr->sa_data) + 1]; in packet_bind_spkt()
3233 return -EINVAL; in packet_bind_spkt()
3234 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be in packet_bind_spkt()
3235 * zero-terminated. in packet_bind_spkt()
3237 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); in packet_bind_spkt()
3238 name[sizeof(uaddr->sa_data)] = 0; in packet_bind_spkt()
3240 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); in packet_bind_spkt()
3246 struct sock *sk = sock->sk; in packet_bind()
3253 return -EINVAL; in packet_bind()
3254 if (sll->sll_family != AF_PACKET) in packet_bind()
3255 return -EINVAL; in packet_bind()
3257 return packet_do_bind(sk, NULL, sll->sll_ifindex, in packet_bind()
3258 sll->sll_protocol ? : pkt_sk(sk)->num); in packet_bind()
3279 if (!ns_capable(net->user_ns, CAP_NET_RAW)) in packet_create()
3280 return -EPERM; in packet_create()
3281 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && in packet_create()
3282 sock->type != SOCK_PACKET) in packet_create()
3283 return -ESOCKTNOSUPPORT; in packet_create()
3285 sock->state = SS_UNCONNECTED; in packet_create()
3287 err = -ENOBUFS; in packet_create()
3292 sock->ops = &packet_ops; in packet_create()
3293 if (sock->type == SOCK_PACKET) in packet_create()
3294 sock->ops = &packet_ops_spkt; in packet_create()
3299 init_completion(&po->skb_completion); in packet_create()
3300 sk->sk_family = PF_PACKET; in packet_create()
3301 po->num = proto; in packet_create()
3302 po->xmit = dev_queue_xmit; in packet_create()
3310 sk->sk_destruct = packet_sock_destruct; in packet_create()
3317 spin_lock_init(&po->bind_lock); in packet_create()
3318 mutex_init(&po->pg_vec_lock); in packet_create()
3319 po->rollover = NULL; in packet_create()
3320 po->prot_hook.func = packet_rcv; in packet_create()
3322 if (sock->type == SOCK_PACKET) in packet_create()
3323 po->prot_hook.func = packet_rcv_spkt; in packet_create()
3325 po->prot_hook.af_packet_priv = sk; in packet_create()
3328 po->prot_hook.type = proto; in packet_create()
3332 mutex_lock(&net->packet.sklist_lock); in packet_create()
3333 sk_add_node_tail_rcu(sk, &net->packet.sklist); in packet_create()
3334 mutex_unlock(&net->packet.sklist_lock); in packet_create()
3355 struct sock *sk = sock->sk; in packet_recvmsg()
3361 err = -EINVAL; in packet_recvmsg()
3367 if (pkt_sk(sk)->ifindex < 0) in packet_recvmsg()
3368 return -ENODEV; in packet_recvmsg()
3379 * of horrible races and re-entrancy so we can forget about it in packet_recvmsg()
3399 if (pkt_sk(sk)->has_vnet_hdr) { in packet_recvmsg()
3410 copied = skb->len; in packet_recvmsg()
3413 msg->msg_flags |= MSG_TRUNC; in packet_recvmsg()
3420 if (sock->type != SOCK_PACKET) { in packet_recvmsg()
3421 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3424 origlen = PACKET_SKB_CB(skb)->sa.origlen; in packet_recvmsg()
3425 sll->sll_family = AF_PACKET; in packet_recvmsg()
3426 sll->sll_protocol = skb->protocol; in packet_recvmsg()
3431 if (msg->msg_name) { in packet_recvmsg()
3437 if (sock->type == SOCK_PACKET) { in packet_recvmsg()
3439 msg->msg_namelen = sizeof(struct sockaddr_pkt); in packet_recvmsg()
3440 copy_len = msg->msg_namelen; in packet_recvmsg()
3442 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3444 msg->msg_namelen = sll->sll_halen + in packet_recvmsg()
3446 copy_len = msg->msg_namelen; in packet_recvmsg()
3447 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { in packet_recvmsg()
3448 memset(msg->msg_name + in packet_recvmsg()
3450 0, sizeof(sll->sll_addr)); in packet_recvmsg()
3451 msg->msg_namelen = sizeof(struct sockaddr_ll); in packet_recvmsg()
3454 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); in packet_recvmsg()
3457 if (pkt_sk(sk)->auxdata) { in packet_recvmsg()
3461 if (skb->ip_summed == CHECKSUM_PARTIAL) in packet_recvmsg()
3463 else if (skb->pkt_type != PACKET_OUTGOING && in packet_recvmsg()
3464 (skb->ip_summed == CHECKSUM_COMPLETE || in packet_recvmsg()
3469 aux.tp_snaplen = skb->len; in packet_recvmsg()
3474 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); in packet_recvmsg()
3485 * hides all the races and re-entrancy issues from us. in packet_recvmsg()
3487 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); in packet_recvmsg()
3499 struct sock *sk = sock->sk; in packet_getname_spkt()
3502 return -EOPNOTSUPP; in packet_getname_spkt()
3504 uaddr->sa_family = AF_PACKET; in packet_getname_spkt()
3505 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); in packet_getname_spkt()
3507 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); in packet_getname_spkt()
3509 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); in packet_getname_spkt()
3519 struct sock *sk = sock->sk; in packet_getname()
3525 return -EOPNOTSUPP; in packet_getname()
3527 ifindex = READ_ONCE(po->ifindex); in packet_getname()
3528 sll->sll_family = AF_PACKET; in packet_getname()
3529 sll->sll_ifindex = ifindex; in packet_getname()
3530 sll->sll_protocol = READ_ONCE(po->num); in packet_getname()
3531 sll->sll_pkttype = 0; in packet_getname()
3535 sll->sll_hatype = dev->type; in packet_getname()
3536 sll->sll_halen = dev->addr_len; in packet_getname()
3537 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); in packet_getname()
3539 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ in packet_getname()
3540 sll->sll_halen = 0; in packet_getname()
3544 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; in packet_getname()
3550 switch (i->type) { in packet_dev_mc()
3552 if (i->alen != dev->addr_len) in packet_dev_mc()
3553 return -EINVAL; in packet_dev_mc()
3555 return dev_mc_add(dev, i->addr); in packet_dev_mc()
3557 return dev_mc_del(dev, i->addr); in packet_dev_mc()
3564 if (i->alen != dev->addr_len) in packet_dev_mc()
3565 return -EINVAL; in packet_dev_mc()
3567 return dev_uc_add(dev, i->addr); in packet_dev_mc()
3569 return dev_uc_del(dev, i->addr); in packet_dev_mc()
3583 if (ml->ifindex == dev->ifindex) { in packet_dev_mclist_delete()
3584 packet_dev_mc(dev, ml, -1); in packet_dev_mclist_delete()
3585 *mlp = ml->next; in packet_dev_mclist_delete()
3588 mlp = &ml->next; in packet_dev_mclist_delete()
3601 err = -ENODEV; in packet_mc_add()
3602 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); in packet_mc_add()
3606 err = -EINVAL; in packet_mc_add()
3607 if (mreq->mr_alen > dev->addr_len) in packet_mc_add()
3610 err = -ENOBUFS; in packet_mc_add()
3616 for (ml = po->mclist; ml; ml = ml->next) { in packet_mc_add()
3617 if (ml->ifindex == mreq->mr_ifindex && in packet_mc_add()
3618 ml->type == mreq->mr_type && in packet_mc_add()
3619 ml->alen == mreq->mr_alen && in packet_mc_add()
3620 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { in packet_mc_add()
3621 ml->count++; in packet_mc_add()
3628 i->type = mreq->mr_type; in packet_mc_add()
3629 i->ifindex = mreq->mr_ifindex; in packet_mc_add()
3630 i->alen = mreq->mr_alen; in packet_mc_add()
3631 memcpy(i->addr, mreq->mr_address, i->alen); in packet_mc_add()
3632 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); in packet_mc_add()
3633 i->count = 1; in packet_mc_add()
3634 i->next = po->mclist; in packet_mc_add()
3635 po->mclist = i; in packet_mc_add()
3638 po->mclist = i->next; in packet_mc_add()
3653 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { in packet_mc_drop()
3654 if (ml->ifindex == mreq->mr_ifindex && in packet_mc_drop()
3655 ml->type == mreq->mr_type && in packet_mc_drop()
3656 ml->alen == mreq->mr_alen && in packet_mc_drop()
3657 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { in packet_mc_drop()
3658 if (--ml->count == 0) { in packet_mc_drop()
3660 *mlp = ml->next; in packet_mc_drop()
3661 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); in packet_mc_drop()
3663 packet_dev_mc(dev, ml, -1); in packet_mc_drop()
3678 if (!po->mclist) in packet_flush_mclist()
3682 while ((ml = po->mclist) != NULL) { in packet_flush_mclist()
3685 po->mclist = ml->next; in packet_flush_mclist()
3686 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); in packet_flush_mclist()
3688 packet_dev_mc(dev, ml, -1); in packet_flush_mclist()
3698 struct sock *sk = sock->sk; in packet_setsockopt()
3703 return -ENOPROTOOPT; in packet_setsockopt()
3713 return -EINVAL; in packet_setsockopt()
3717 return -EFAULT; in packet_setsockopt()
3719 return -EINVAL; in packet_setsockopt()
3734 switch (po->tp_version) { in packet_setsockopt()
3745 ret = -EINVAL; in packet_setsockopt()
3748 ret = -EFAULT; in packet_setsockopt()
3761 return -EINVAL; in packet_setsockopt()
3763 return -EFAULT; in packet_setsockopt()
3765 pkt_sk(sk)->copy_thresh = val; in packet_setsockopt()
3773 return -EINVAL; in packet_setsockopt()
3775 return -EFAULT; in packet_setsockopt()
3782 return -EINVAL; in packet_setsockopt()
3785 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3786 ret = -EBUSY; in packet_setsockopt()
3788 po->tp_version = val; in packet_setsockopt()
3799 return -EINVAL; in packet_setsockopt()
3801 return -EFAULT; in packet_setsockopt()
3803 return -EINVAL; in packet_setsockopt()
3805 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3806 ret = -EBUSY; in packet_setsockopt()
3808 po->tp_reserve = val; in packet_setsockopt()
3819 return -EINVAL; in packet_setsockopt()
3821 return -EFAULT; in packet_setsockopt()
3824 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3825 ret = -EBUSY; in packet_setsockopt()
3827 po->tp_loss = !!val; in packet_setsockopt()
3838 return -EINVAL; in packet_setsockopt()
3840 return -EFAULT; in packet_setsockopt()
3843 po->auxdata = !!val; in packet_setsockopt()
3852 return -EINVAL; in packet_setsockopt()
3854 return -EFAULT; in packet_setsockopt()
3857 po->origdev = !!val; in packet_setsockopt()
3865 if (sock->type != SOCK_RAW) in packet_setsockopt()
3866 return -EINVAL; in packet_setsockopt()
3868 return -EINVAL; in packet_setsockopt()
3870 return -EFAULT; in packet_setsockopt()
3873 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3874 ret = -EBUSY; in packet_setsockopt()
3876 po->has_vnet_hdr = !!val; in packet_setsockopt()
3887 return -EINVAL; in packet_setsockopt()
3889 return -EFAULT; in packet_setsockopt()
3891 po->tp_tstamp = val; in packet_setsockopt()
3899 return -EINVAL; in packet_setsockopt()
3901 return -EFAULT; in packet_setsockopt()
3907 if (!po->fanout) in packet_setsockopt()
3908 return -EINVAL; in packet_setsockopt()
3917 return -EINVAL; in packet_setsockopt()
3919 return -EFAULT; in packet_setsockopt()
3921 return -EINVAL; in packet_setsockopt()
3923 po->prot_hook.ignore_outgoing = !!val; in packet_setsockopt()
3931 return -EINVAL; in packet_setsockopt()
3933 return -EFAULT; in packet_setsockopt()
3936 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec) in packet_setsockopt()
3937 po->tp_tx_has_off = !!val; in packet_setsockopt()
3947 return -EINVAL; in packet_setsockopt()
3949 return -EFAULT; in packet_setsockopt()
3951 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; in packet_setsockopt()
3955 return -ENOPROTOOPT; in packet_setsockopt()
3964 struct sock *sk = sock->sk; in packet_getsockopt()
3972 return -ENOPROTOOPT; in packet_getsockopt()
3975 return -EFAULT; in packet_getsockopt()
3978 return -EINVAL; in packet_getsockopt()
3982 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_getsockopt()
3983 memcpy(&st, &po->stats, sizeof(st)); in packet_getsockopt()
3984 memset(&po->stats, 0, sizeof(po->stats)); in packet_getsockopt()
3985 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_getsockopt()
3986 drops = atomic_xchg(&po->tp_drops, 0); in packet_getsockopt()
3988 if (po->tp_version == TPACKET_V3) { in packet_getsockopt()
4002 val = po->auxdata; in packet_getsockopt()
4005 val = po->origdev; in packet_getsockopt()
4008 val = po->has_vnet_hdr; in packet_getsockopt()
4011 val = po->tp_version; in packet_getsockopt()
4017 return -EINVAL; in packet_getsockopt()
4019 return -EFAULT; in packet_getsockopt()
4031 return -EINVAL; in packet_getsockopt()
4035 val = po->tp_reserve; in packet_getsockopt()
4038 val = po->tp_loss; in packet_getsockopt()
4041 val = po->tp_tstamp; in packet_getsockopt()
4044 val = (po->fanout ? in packet_getsockopt()
4045 ((u32)po->fanout->id | in packet_getsockopt()
4046 ((u32)po->fanout->type << 16) | in packet_getsockopt()
4047 ((u32)po->fanout->flags << 24)) : in packet_getsockopt()
4051 val = po->prot_hook.ignore_outgoing; in packet_getsockopt()
4054 if (!po->rollover) in packet_getsockopt()
4055 return -EINVAL; in packet_getsockopt()
4056 rstats.tp_all = atomic_long_read(&po->rollover->num); in packet_getsockopt()
4057 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); in packet_getsockopt()
4058 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); in packet_getsockopt()
4063 val = po->tp_tx_has_off; in packet_getsockopt()
4069 return -ENOPROTOOPT; in packet_getsockopt()
4075 return -EFAULT; in packet_getsockopt()
4077 return -EFAULT; in packet_getsockopt()
4089 sk_for_each_rcu(sk, &net->packet.sklist) { in packet_notifier()
4094 if (po->mclist) in packet_notifier()
4095 packet_dev_mclist_delete(dev, &po->mclist); in packet_notifier()
4099 if (dev->ifindex == po->ifindex) { in packet_notifier()
4100 spin_lock(&po->bind_lock); in packet_notifier()
4101 if (po->running) { in packet_notifier()
4103 sk->sk_err = ENETDOWN; in packet_notifier()
4109 WRITE_ONCE(po->ifindex, -1); in packet_notifier()
4110 dev_put(po->prot_hook.dev); in packet_notifier()
4111 po->prot_hook.dev = NULL; in packet_notifier()
4113 spin_unlock(&po->bind_lock); in packet_notifier()
4117 if (dev->ifindex == po->ifindex) { in packet_notifier()
4118 spin_lock(&po->bind_lock); in packet_notifier()
4119 if (po->num) in packet_notifier()
4121 spin_unlock(&po->bind_lock); in packet_notifier()
4134 struct sock *sk = sock->sk; in packet_ioctl()
4148 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_ioctl()
4149 skb = skb_peek(&sk->sk_receive_queue); in packet_ioctl()
4151 amount = skb->len; in packet_ioctl()
4152 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_ioctl()
4174 return -ENOIOCTLCMD; in packet_ioctl()
4182 struct sock *sk = sock->sk; in packet_poll()
4186 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_poll()
4187 if (po->rx_ring.pg_vec) { in packet_poll()
4188 if (!packet_previous_rx_frame(po, &po->rx_ring, in packet_poll()
4193 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_poll()
4194 spin_lock_bh(&sk->sk_write_queue.lock); in packet_poll()
4195 if (po->tx_ring.pg_vec) { in packet_poll()
4196 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) in packet_poll()
4199 spin_unlock_bh(&sk->sk_write_queue.lock); in packet_poll()
4210 struct file *file = vma->vm_file; in packet_mm_open()
4211 struct socket *sock = file->private_data; in packet_mm_open()
4212 struct sock *sk = sock->sk; in packet_mm_open()
4215 atomic_inc(&pkt_sk(sk)->mapped); in packet_mm_open()
4220 struct file *file = vma->vm_file; in packet_mm_close()
4221 struct socket *sock = file->private_data; in packet_mm_close()
4222 struct sock *sk = sock->sk; in packet_mm_close()
4225 atomic_dec(&pkt_sk(sk)->mapped); in packet_mm_close()
4278 unsigned int block_nr = req->tp_block_nr; in alloc_pg_vec()
4313 struct tpacket_req *req = &req_u->req; in packet_set_ring()
4315 rb = tx_ring ? &po->tx_ring : &po->rx_ring; in packet_set_ring()
4316 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; in packet_set_ring()
4318 err = -EBUSY; in packet_set_ring()
4320 if (atomic_read(&po->mapped)) in packet_set_ring()
4326 if (req->tp_block_nr) { in packet_set_ring()
4330 err = -EBUSY; in packet_set_ring()
4331 if (unlikely(rb->pg_vec)) in packet_set_ring()
4334 switch (po->tp_version) { in packet_set_ring()
4336 po->tp_hdrlen = TPACKET_HDRLEN; in packet_set_ring()
4339 po->tp_hdrlen = TPACKET2_HDRLEN; in packet_set_ring()
4342 po->tp_hdrlen = TPACKET3_HDRLEN; in packet_set_ring()
4346 err = -EINVAL; in packet_set_ring()
4347 if (unlikely((int)req->tp_block_size <= 0)) in packet_set_ring()
4349 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) in packet_set_ring()
4351 min_frame_size = po->tp_hdrlen + po->tp_reserve; in packet_set_ring()
4352 if (po->tp_version >= TPACKET_V3 && in packet_set_ring()
4353 req->tp_block_size < in packet_set_ring()
4354 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) in packet_set_ring()
4356 if (unlikely(req->tp_frame_size < min_frame_size)) in packet_set_ring()
4358 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) in packet_set_ring()
4361 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; in packet_set_ring()
4362 if (unlikely(rb->frames_per_block == 0)) in packet_set_ring()
4364 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) in packet_set_ring()
4366 if (unlikely((rb->frames_per_block * req->tp_block_nr) != in packet_set_ring()
4367 req->tp_frame_nr)) in packet_set_ring()
4370 err = -ENOMEM; in packet_set_ring()
4371 order = get_order(req->tp_block_size); in packet_set_ring()
4375 switch (po->tp_version) { in packet_set_ring()
4381 struct tpacket_req3 *req3 = &req_u->req3; in packet_set_ring()
4383 if (req3->tp_retire_blk_tov || in packet_set_ring()
4384 req3->tp_sizeof_priv || in packet_set_ring()
4385 req3->tp_feature_req_word) { in packet_set_ring()
4386 err = -EINVAL; in packet_set_ring()
4393 rx_owner_map = bitmap_alloc(req->tp_frame_nr, in packet_set_ring()
4403 err = -EINVAL; in packet_set_ring()
4404 if (unlikely(req->tp_frame_nr)) in packet_set_ring()
4410 spin_lock(&po->bind_lock); in packet_set_ring()
4411 was_running = po->running; in packet_set_ring()
4412 num = po->num; in packet_set_ring()
4414 WRITE_ONCE(po->num, 0); in packet_set_ring()
4417 spin_unlock(&po->bind_lock); in packet_set_ring()
4421 err = -EBUSY; in packet_set_ring()
4422 mutex_lock(&po->pg_vec_lock); in packet_set_ring()
4423 if (closing || atomic_read(&po->mapped) == 0) { in packet_set_ring()
4425 spin_lock_bh(&rb_queue->lock); in packet_set_ring()
4426 swap(rb->pg_vec, pg_vec); in packet_set_ring()
4427 if (po->tp_version <= TPACKET_V2) in packet_set_ring()
4428 swap(rb->rx_owner_map, rx_owner_map); in packet_set_ring()
4429 rb->frame_max = (req->tp_frame_nr - 1); in packet_set_ring()
4430 rb->head = 0; in packet_set_ring()
4431 rb->frame_size = req->tp_frame_size; in packet_set_ring()
4432 spin_unlock_bh(&rb_queue->lock); in packet_set_ring()
4434 swap(rb->pg_vec_order, order); in packet_set_ring()
4435 swap(rb->pg_vec_len, req->tp_block_nr); in packet_set_ring()
4437 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; in packet_set_ring()
4438 po->prot_hook.func = (po->rx_ring.pg_vec) ? in packet_set_ring()
4441 if (atomic_read(&po->mapped)) in packet_set_ring()
4443 atomic_read(&po->mapped)); in packet_set_ring()
4445 mutex_unlock(&po->pg_vec_lock); in packet_set_ring()
4447 spin_lock(&po->bind_lock); in packet_set_ring()
4449 WRITE_ONCE(po->num, num); in packet_set_ring()
4452 spin_unlock(&po->bind_lock); in packet_set_ring()
4453 if (pg_vec && (po->tp_version > TPACKET_V2)) { in packet_set_ring()
4454 /* Because we don't support block-based V3 on tx-ring */ in packet_set_ring()
4462 free_pg_vec(pg_vec, order, req->tp_block_nr); in packet_set_ring()
4470 struct sock *sk = sock->sk; in packet_mmap()
4475 int err = -EINVAL; in packet_mmap()
4478 if (vma->vm_pgoff) in packet_mmap()
4479 return -EINVAL; in packet_mmap()
4481 mutex_lock(&po->pg_vec_lock); in packet_mmap()
4484 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { in packet_mmap()
4485 if (rb->pg_vec) { in packet_mmap()
4486 expected_size += rb->pg_vec_len in packet_mmap()
4487 * rb->pg_vec_pages in packet_mmap()
4495 size = vma->vm_end - vma->vm_start; in packet_mmap()
4499 start = vma->vm_start; in packet_mmap()
4500 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { in packet_mmap()
4501 if (rb->pg_vec == NULL) in packet_mmap()
4504 for (i = 0; i < rb->pg_vec_len; i++) { in packet_mmap()
4506 void *kaddr = rb->pg_vec[i].buffer; in packet_mmap()
4509 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { in packet_mmap()
4520 atomic_inc(&po->mapped); in packet_mmap()
4521 vma->vm_ops = &packet_mmap_ops; in packet_mmap()
4525 mutex_unlock(&po->pg_vec_lock); in packet_mmap()
4589 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); in packet_seq_start()
4595 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); in packet_seq_next()
4609 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk"); in packet_seq_show()
4615 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", in packet_seq_show()
4617 refcount_read(&s->sk_refcnt), in packet_seq_show()
4618 s->sk_type, in packet_seq_show()
4619 ntohs(READ_ONCE(po->num)), in packet_seq_show()
4620 READ_ONCE(po->ifindex), in packet_seq_show()
4621 po->running, in packet_seq_show()
4622 atomic_read(&s->sk_rmem_alloc), in packet_seq_show()
4640 mutex_init(&net->packet.sklist_lock); in packet_net_init()
4641 INIT_HLIST_HEAD(&net->packet.sklist); in packet_net_init()
4644 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, in packet_net_init()
4646 return -ENOMEM; in packet_net_init()
4654 remove_proc_entry("packet", net->proc_net); in packet_net_exit()
4655 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); in packet_net_exit()