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.
96 - If the device has no dev->header_ops->create, there is no LL header
103 needed_headroom to be (the real WiFi header length - the fake Ethernet
105 - packet socket receives packets with pulled ll header,
109 -----------
112 mac_header -> ll header
113 data -> data
116 mac_header -> ll header
117 data -> ll header
120 mac_header -> data
123 data -> data
126 mac_header -> data. ll header is invisible to us.
127 data -> data
135 ------------
137 dev->header_ops != NULL
138 mac_header -> ll header
139 data -> ll header
141 dev->header_ops == NULL (ll header is invisible to us)
142 mac_header -> data
143 data -> data
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)
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))
231 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
233 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
235 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
236 ((x)->kactive_blk_num+1) : 0)
251 dev = rcu_dereference(po->cached_dev); in packet_cached_dev_get()
262 rcu_assign_pointer(po->cached_dev, dev); in packet_cached_dev_assign()
267 RCU_INIT_POINTER(po->cached_dev, NULL); in packet_cached_dev_reset()
272 return po->xmit == packet_direct_xmit; in packet_use_direct_xmit()
277 struct net_device *dev = skb->dev; in packet_pick_tx_queue()
278 const struct net_device_ops *ops = dev->netdev_ops; in packet_pick_tx_queue()
283 skb->sender_cpu = cpu + 1; in packet_pick_tx_queue()
285 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); in packet_pick_tx_queue()
286 if (ops->ndo_select_queue) { in packet_pick_tx_queue()
287 queue_index = ops->ndo_select_queue(dev, skb, NULL); in packet_pick_tx_queue()
304 if (!po->running) { in __register_prot_hook()
305 if (po->fanout) in __register_prot_hook()
308 dev_add_pack(&po->prot_hook); in __register_prot_hook()
311 po->running = 1; in __register_prot_hook()
317 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); in register_prot_hook()
322 * the po->bind_lock and do a synchronize_net to make sure no
324 * of po->prot_hook. If the sync parameter is false, it is the
331 lockdep_assert_held_once(&po->bind_lock); in __unregister_prot_hook()
333 po->running = 0; in __unregister_prot_hook()
335 if (po->fanout) in __unregister_prot_hook()
338 __dev_remove_pack(&po->prot_hook); in __unregister_prot_hook()
343 spin_unlock(&po->bind_lock); in __unregister_prot_hook()
345 spin_lock(&po->bind_lock); in __unregister_prot_hook()
353 if (po->running) in unregister_prot_hook()
369 switch (po->tp_version) { in __packet_set_status()
371 h.h1->tp_status = status; in __packet_set_status()
372 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); in __packet_set_status()
375 h.h2->tp_status = status; in __packet_set_status()
376 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); in __packet_set_status()
379 h.h3->tp_status = status; in __packet_set_status()
380 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); in __packet_set_status()
397 switch (po->tp_version) { in __packet_get_status()
399 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); in __packet_get_status()
400 return h.h1->tp_status; in __packet_get_status()
402 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); in __packet_get_status()
403 return h.h2->tp_status; in __packet_get_status()
405 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); in __packet_get_status()
406 return h.h3->tp_status; in __packet_get_status()
421 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) in tpacket_get_timestamp()
425 ktime_to_timespec64_cond(skb->tstamp, ts)) in tpacket_get_timestamp()
438 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) in __packet_set_timestamp()
444 * all store the seconds in a 32-bit unsigned integer. in __packet_set_timestamp()
445 * If we create a version 4, that should have a 64-bit timestamp, in __packet_set_timestamp()
446 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit in __packet_set_timestamp()
449 switch (po->tp_version) { in __packet_set_timestamp()
451 h.h1->tp_sec = ts.tv_sec; in __packet_set_timestamp()
452 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; in __packet_set_timestamp()
455 h.h2->tp_sec = ts.tv_sec; in __packet_set_timestamp()
456 h.h2->tp_nsec = ts.tv_nsec; in __packet_set_timestamp()
459 h.h3->tp_sec = ts.tv_sec; in __packet_set_timestamp()
460 h.h3->tp_nsec = ts.tv_nsec; in __packet_set_timestamp()
468 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); in __packet_set_timestamp()
482 pg_vec_pos = position / rb->frames_per_block; in packet_lookup_frame()
483 frame_offset = position % rb->frames_per_block; in packet_lookup_frame()
485 h.raw = rb->pg_vec[pg_vec_pos].buffer + in packet_lookup_frame()
486 (frame_offset * rb->frame_size); in packet_lookup_frame()
498 return packet_lookup_frame(po, rb, rb->head, status); in packet_current_frame()
503 del_timer_sync(&pkc->retire_blk_timer); in prb_del_retire_blk_timer()
511 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_shutdown_retire_blk_timer()
513 spin_lock_bh(&rb_queue->lock); in prb_shutdown_retire_blk_timer()
514 pkc->delete_blk_timer = 1; in prb_shutdown_retire_blk_timer()
515 spin_unlock_bh(&rb_queue->lock); in prb_shutdown_retire_blk_timer()
524 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_setup_retire_blk_timer()
525 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, in prb_setup_retire_blk_timer()
527 pkc->retire_blk_timer.expires = jiffies; in prb_setup_retire_blk_timer()
539 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); in prb_calc_retire_blk_tmo()
570 p1->feature_req_word = req_u->req3.tp_feature_req_word; in prb_init_ft_ops()
583 p1->knxt_seq_num = 1; in init_prb_bdqc()
584 p1->pkbdq = pg_vec; in init_prb_bdqc()
586 p1->pkblk_start = pg_vec[0].buffer; in init_prb_bdqc()
587 p1->kblk_size = req_u->req3.tp_block_size; in init_prb_bdqc()
588 p1->knum_blocks = req_u->req3.tp_block_nr; in init_prb_bdqc()
589 p1->hdrlen = po->tp_hdrlen; in init_prb_bdqc()
590 p1->version = po->tp_version; in init_prb_bdqc()
591 p1->last_kactive_blk_num = 0; in init_prb_bdqc()
592 po->stats.stats3.tp_freeze_q_cnt = 0; in init_prb_bdqc()
593 if (req_u->req3.tp_retire_blk_tov) in init_prb_bdqc()
594 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; in init_prb_bdqc()
596 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, in init_prb_bdqc()
597 req_u->req3.tp_block_size); in init_prb_bdqc()
598 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); in init_prb_bdqc()
599 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; in init_prb_bdqc()
600 rwlock_init(&p1->blk_fill_in_prog_lock); in init_prb_bdqc()
602 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); in init_prb_bdqc()
613 mod_timer(&pkc->retire_blk_timer, in _prb_refresh_rx_retire_blk_timer()
614 jiffies + pkc->tov_in_jiffies); in _prb_refresh_rx_retire_blk_timer()
615 pkc->last_kactive_blk_num = pkc->kactive_blk_num; in _prb_refresh_rx_retire_blk_timer()
622 * on packet-by-packet basis.
624 * With a 1MB block-size, on a 1Gbps line, it will take
637 * a) line-speed and b) block-size.
645 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_retire_rx_blk_timer_expired()
649 spin_lock(&po->sk.sk_receive_queue.lock); in prb_retire_rx_blk_timer_expired()
654 if (unlikely(pkc->delete_blk_timer)) in prb_retire_rx_blk_timer_expired()
668 write_lock(&pkc->blk_fill_in_prog_lock); in prb_retire_rx_blk_timer_expired()
669 write_unlock(&pkc->blk_fill_in_prog_lock); in prb_retire_rx_blk_timer_expired()
672 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { in prb_retire_rx_blk_timer_expired()
684 /* Case 1. Queue was frozen because user-space was in prb_retire_rx_blk_timer_expired()
689 * Ok, user-space is still behind. in prb_retire_rx_blk_timer_expired()
694 /* Case 2. queue was frozen,user-space caught up, in prb_retire_rx_blk_timer_expired()
699 * Thawing/timer-refresh is a side effect. in prb_retire_rx_blk_timer_expired()
711 spin_unlock(&po->sk.sk_receive_queue.lock); in prb_retire_rx_blk_timer_expired()
727 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); in prb_flush_block()
764 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; in prb_close_block()
765 struct sock *sk = &po->sk; in prb_close_block()
767 if (atomic_read(&po->tp_drops)) in prb_close_block()
770 last_pkt = (struct tpacket3_hdr *)pkc1->prev; in prb_close_block()
771 last_pkt->tp_next_offset = 0; in prb_close_block()
775 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; in prb_close_block()
776 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; in prb_close_block()
778 /* Ok, we tmo'd - so get the current time. in prb_close_block()
785 h1->ts_last_pkt.ts_sec = ts.tv_sec; in prb_close_block()
786 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; in prb_close_block()
794 sk->sk_data_ready(sk); in prb_close_block()
796 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); in prb_close_block()
801 pkc->reset_pending_on_curr_blk = 0; in prb_thaw_queue()
815 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; in prb_open_block()
823 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; in prb_open_block()
825 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
829 h1->ts_first_pkt.ts_sec = ts.tv_sec; in prb_open_block()
830 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; in prb_open_block()
832 pkc1->pkblk_start = (char *)pbd1; in prb_open_block()
833 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
835 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
838 pbd1->version = pkc1->version; in prb_open_block()
839 pkc1->prev = pkc1->nxt_offset; in prb_open_block()
840 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; in prb_open_block()
853 * 4) user-space is either sleeping or processing block '0'.
854 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
855 * it will close block-7,loop around and try to fill block '0'.
856 * call-flow:
860 * |->(BLOCK_STATUS == USER) evaluates to true
861 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
866 * re-open block-0 in near future.
868 * case and __packet_lookup_frame_in_block will check if block-0
869 * is free and can now be re-used.
874 pkc->reset_pending_on_curr_blk = 1; in prb_freeze_queue()
875 po->stats.stats3.tp_freeze_q_cnt++; in prb_freeze_queue()
908 return (void *)pkc->nxt_offset; in prb_dispatch_next_block()
920 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't in prb_retire_current_block()
925 * the timer-handler already handled this case. in prb_retire_current_block()
929 write_lock(&pkc->blk_fill_in_prog_lock); in prb_retire_current_block()
930 write_unlock(&pkc->blk_fill_in_prog_lock); in prb_retire_current_block()
944 return pkc->reset_pending_on_curr_blk; in prb_queue_frozen()
948 __releases(&pkc->blk_fill_in_prog_lock) in prb_clear_blk_fill_status()
952 read_unlock(&pkc->blk_fill_in_prog_lock); in prb_clear_blk_fill_status()
958 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); in prb_fill_rxhash()
964 ppd->hv1.tp_rxhash = 0; in prb_clear_rxhash()
970 if (skb_vlan_tag_present(pkc->skb)) { in prb_fill_vlan_info()
971 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); in prb_fill_vlan_info()
972 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); in prb_fill_vlan_info()
973 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; in prb_fill_vlan_info()
975 ppd->hv1.tp_vlan_tci = 0; in prb_fill_vlan_info()
976 ppd->hv1.tp_vlan_tpid = 0; in prb_fill_vlan_info()
977 ppd->tp_status = TP_STATUS_AVAILABLE; in prb_fill_vlan_info()
984 ppd->hv1.tp_padding = 0; in prb_run_all_ft_ops()
987 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) in prb_run_all_ft_ops()
997 __acquires(&pkc->blk_fill_in_prog_lock) in prb_fill_curr_block()
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); in prb_fill_curr_block()
1003 pkc->prev = curr; in prb_fill_curr_block()
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); in prb_fill_curr_block()
1007 read_lock(&pkc->blk_fill_in_prog_lock); in prb_fill_curr_block()
1011 /* Assumes caller has the sk->rx_queue.lock */
1021 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in __packet_lookup_frame_in_block()
1028 * is still in_use by user-space. in __packet_lookup_frame_in_block()
1035 * Ok, the block was released by user-space. in __packet_lookup_frame_in_block()
1045 curr = pkc->nxt_offset; in __packet_lookup_frame_in_block()
1046 pkc->skb = skb; in __packet_lookup_frame_in_block()
1047 end = (char *)pbd + pkc->kblk_size; in __packet_lookup_frame_in_block()
1067 * No free blocks are available.user_space hasn't caught up yet. in __packet_lookup_frame_in_block()
1078 switch (po->tp_version) { in packet_current_rx_frame()
1081 curr = packet_lookup_frame(po, &po->rx_ring, in packet_current_rx_frame()
1082 po->rx_ring.head, status); in packet_current_rx_frame()
1109 if (rb->prb_bdqc.kactive_blk_num) in prb_previous_blk_num()
1110 prev = rb->prb_bdqc.kactive_blk_num-1; in prb_previous_blk_num()
1112 prev = rb->prb_bdqc.knum_blocks-1; in prb_previous_blk_num()
1129 if (po->tp_version <= TPACKET_V2) in packet_previous_rx_frame()
1138 switch (po->tp_version) { in packet_increment_rx_head()
1154 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; in packet_previous_frame()
1160 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; in packet_increment_head()
1165 this_cpu_inc(*rb->pending_refcnt); in packet_inc_pending()
1170 this_cpu_dec(*rb->pending_refcnt); in packet_dec_pending()
1179 if (rb->pending_refcnt == NULL) in packet_read_pending()
1183 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); in packet_read_pending()
1190 po->rx_ring.pending_refcnt = NULL; in packet_alloc_pending()
1192 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); in packet_alloc_pending()
1193 if (unlikely(po->tx_ring.pending_refcnt == NULL)) in packet_alloc_pending()
1194 return -ENOBUFS; in packet_alloc_pending()
1201 free_percpu(po->tx_ring.pending_refcnt); in packet_free_pending()
1213 len = READ_ONCE(po->rx_ring.frame_max) + 1; in __tpacket_has_room()
1214 idx = READ_ONCE(po->rx_ring.head); in __tpacket_has_room()
1218 idx -= len; in __tpacket_has_room()
1219 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); in __tpacket_has_room()
1226 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); in __tpacket_v3_has_room()
1227 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); in __tpacket_v3_has_room()
1231 idx -= len; in __tpacket_v3_has_room()
1232 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); in __tpacket_v3_has_room()
1238 const struct sock *sk = &po->sk; in __packet_rcv_has_room()
1241 if (po->prot_hook.func != tpacket_rcv) { in __packet_rcv_has_room()
1242 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); in __packet_rcv_has_room()
1243 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) in __packet_rcv_has_room()
1244 - (skb ? skb->truesize : 0); in __packet_rcv_has_room()
1254 if (po->tp_version == TPACKET_V3) { in __packet_rcv_has_room()
1276 if (READ_ONCE(po->pressure) != pressure) in packet_rcv_has_room()
1277 WRITE_ONCE(po->pressure, pressure); in packet_rcv_has_room()
1284 if (READ_ONCE(po->pressure) && in packet_rcv_try_clear_pressure()
1286 WRITE_ONCE(po->pressure, 0); in packet_rcv_try_clear_pressure()
1291 skb_queue_purge(&sk->sk_error_queue); in packet_sock_destruct()
1293 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); in packet_sock_destruct()
1294 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); in packet_sock_destruct()
1306 u32 *history = po->rollover->history; in fanout_flow_is_huge()
1335 unsigned int val = atomic_inc_return(&f->rr_cur); in fanout_demux_lb()
1362 po = pkt_sk(rcu_dereference(f->arr[idx])); in fanout_demux_rollover()
1372 i = j = min_t(int, po->rollover->sock, num - 1); in fanout_demux_rollover()
1374 po_next = pkt_sk(rcu_dereference(f->arr[i])); in fanout_demux_rollover()
1375 if (po_next != po_skip && !READ_ONCE(po_next->pressure) && in fanout_demux_rollover()
1378 po->rollover->sock = i; in fanout_demux_rollover()
1379 atomic_long_inc(&po->rollover->num); in fanout_demux_rollover()
1381 atomic_long_inc(&po->rollover->num_huge); in fanout_demux_rollover()
1389 atomic_long_inc(&po->rollover->num_failed); in fanout_demux_rollover()
1408 prog = rcu_dereference(f->bpf_prog); in fanout_demux_bpf()
1418 return f->flags & (flag >> 8); in fanout_has_flag()
1424 struct packet_fanout *f = pt->af_packet_priv; in packet_rcv_fanout()
1425 unsigned int num = READ_ONCE(f->num_members); in packet_rcv_fanout()
1426 struct net *net = read_pnet(&f->net); in packet_rcv_fanout()
1440 switch (f->type) { in packet_rcv_fanout()
1469 po = pkt_sk(rcu_dereference(f->arr[idx])); in packet_rcv_fanout()
1470 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); in packet_rcv_fanout()
1480 struct packet_fanout *f = po->fanout; in __fanout_link()
1482 spin_lock(&f->lock); in __fanout_link()
1483 rcu_assign_pointer(f->arr[f->num_members], sk); in __fanout_link()
1485 f->num_members++; in __fanout_link()
1486 if (f->num_members == 1) in __fanout_link()
1487 dev_add_pack(&f->prot_hook); in __fanout_link()
1488 spin_unlock(&f->lock); in __fanout_link()
1493 struct packet_fanout *f = po->fanout; in __fanout_unlink()
1496 spin_lock(&f->lock); in __fanout_unlink()
1497 for (i = 0; i < f->num_members; i++) { in __fanout_unlink()
1498 if (rcu_dereference_protected(f->arr[i], in __fanout_unlink()
1499 lockdep_is_held(&f->lock)) == sk) in __fanout_unlink()
1502 BUG_ON(i >= f->num_members); in __fanout_unlink()
1503 rcu_assign_pointer(f->arr[i], in __fanout_unlink()
1504 rcu_dereference_protected(f->arr[f->num_members - 1], in __fanout_unlink()
1505 lockdep_is_held(&f->lock))); in __fanout_unlink()
1506 f->num_members--; in __fanout_unlink()
1507 if (f->num_members == 0) in __fanout_unlink()
1508 __dev_remove_pack(&f->prot_hook); in __fanout_unlink()
1509 spin_unlock(&f->lock); in __fanout_unlink()
1514 if (sk->sk_family != PF_PACKET) in match_fanout_group()
1517 return ptype->af_packet_priv == pkt_sk(sk)->fanout; in match_fanout_group()
1522 switch (f->type) { in fanout_init_data()
1524 atomic_set(&f->rr_cur, 0); in fanout_init_data()
1528 RCU_INIT_POINTER(f->bpf_prog, NULL); in fanout_init_data()
1537 spin_lock(&f->lock); in __fanout_set_data_bpf()
1538 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); in __fanout_set_data_bpf()
1539 rcu_assign_pointer(f->bpf_prog, new); in __fanout_set_data_bpf()
1540 spin_unlock(&f->lock); in __fanout_set_data_bpf()
1555 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) in fanout_set_data_cbpf()
1556 return -EPERM; in fanout_set_data_cbpf()
1566 __fanout_set_data_bpf(po->fanout, new); in fanout_set_data_cbpf()
1576 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) in fanout_set_data_ebpf()
1577 return -EPERM; in fanout_set_data_ebpf()
1579 return -EINVAL; in fanout_set_data_ebpf()
1581 return -EFAULT; in fanout_set_data_ebpf()
1587 __fanout_set_data_bpf(po->fanout, new); in fanout_set_data_ebpf()
1594 switch (po->fanout->type) { in fanout_set_data()
1600 return -EINVAL; in fanout_set_data()
1606 switch (f->type) { in fanout_release_data()
1618 if (f->id == candidate_id && in __fanout_id_is_free()
1619 read_pnet(&f->net) == sock_net(sk)) { in __fanout_id_is_free()
1645 struct packet_rollover *rollover = NULL; in fanout_add() local
1647 u16 type_flags = args->type_flags; in fanout_add()
1651 u16 id = args->id; in fanout_add()
1657 return -EINVAL; in fanout_add()
1667 return -EINVAL; in fanout_add()
1672 err = -EALREADY; in fanout_add()
1673 if (po->fanout) in fanout_add()
1678 err = -ENOMEM; in fanout_add()
1679 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); in fanout_add()
1680 if (!rollover) in fanout_add()
1682 atomic_long_set(&rollover->num, 0); in fanout_add()
1683 atomic_long_set(&rollover->num_huge, 0); in fanout_add()
1684 atomic_long_set(&rollover->num_failed, 0); in fanout_add()
1689 err = -EINVAL; in fanout_add()
1693 err = -ENOMEM; in fanout_add()
1702 if (f->id == id && in fanout_add()
1703 read_pnet(&f->net) == sock_net(sk)) { in fanout_add()
1708 err = -EINVAL; in fanout_add()
1710 if (match->flags != flags) in fanout_add()
1712 if (args->max_num_members && in fanout_add()
1713 args->max_num_members != match->max_num_members) in fanout_add()
1716 if (args->max_num_members > PACKET_FANOUT_MAX) in fanout_add()
1718 if (!args->max_num_members) in fanout_add()
1720 args->max_num_members = 256; in fanout_add()
1721 err = -ENOMEM; in fanout_add()
1722 match = kvzalloc(struct_size(match, arr, args->max_num_members), in fanout_add()
1726 write_pnet(&match->net, sock_net(sk)); in fanout_add()
1727 match->id = id; in fanout_add()
1728 match->type = type; in fanout_add()
1729 match->flags = flags; in fanout_add()
1730 INIT_LIST_HEAD(&match->list); in fanout_add()
1731 spin_lock_init(&match->lock); in fanout_add()
1732 refcount_set(&match->sk_ref, 0); in fanout_add()
1734 match->prot_hook.type = po->prot_hook.type; in fanout_add()
1735 match->prot_hook.dev = po->prot_hook.dev; in fanout_add()
1736 match->prot_hook.func = packet_rcv_fanout; in fanout_add()
1737 match->prot_hook.af_packet_priv = match; in fanout_add()
1738 match->prot_hook.af_packet_net = read_pnet(&match->net); in fanout_add()
1739 match->prot_hook.id_match = match_fanout_group; in fanout_add()
1740 match->max_num_members = args->max_num_members; in fanout_add()
1741 list_add(&match->list, &fanout_list); in fanout_add()
1743 err = -EINVAL; in fanout_add()
1745 spin_lock(&po->bind_lock); in fanout_add()
1746 if (po->running && in fanout_add()
1747 match->type == type && in fanout_add()
1748 match->prot_hook.type == po->prot_hook.type && in fanout_add()
1749 match->prot_hook.dev == po->prot_hook.dev) { in fanout_add()
1750 err = -ENOSPC; in fanout_add()
1751 if (refcount_read(&match->sk_ref) < match->max_num_members) { in fanout_add()
1752 __dev_remove_pack(&po->prot_hook); in fanout_add()
1755 WRITE_ONCE(po->fanout, match); in fanout_add()
1757 po->rollover = rollover; in fanout_add()
1758 rollover = NULL; in fanout_add()
1759 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); in fanout_add()
1764 spin_unlock(&po->bind_lock); in fanout_add()
1766 if (err && !refcount_read(&match->sk_ref)) { in fanout_add()
1767 list_del(&match->list); in fanout_add()
1772 kfree(rollover); in fanout_add()
1777 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1778 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1788 f = po->fanout; in fanout_release()
1790 po->fanout = NULL; in fanout_release()
1792 if (refcount_dec_and_test(&f->sk_ref)) in fanout_release()
1793 list_del(&f->list); in fanout_release()
1805 /* Earlier code assumed this would be a VLAN pkt, double-check in packet_extra_vlan_len_allowed()
1809 if (unlikely(dev->type != ARPHRD_ETHER)) in packet_extra_vlan_len_allowed()
1813 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); in packet_extra_vlan_len_allowed()
1831 sk = pt->af_packet_priv; in packet_rcv_spkt()
1840 * For outgoing ones skb->data == skb_mac_header(skb) in packet_rcv_spkt()
1844 if (skb->pkt_type == PACKET_LOOPBACK) in packet_rcv_spkt()
1860 spkt = &PACKET_SKB_CB(skb)->sa.pkt; in packet_rcv_spkt()
1862 skb_push(skb, skb->data - skb_mac_header(skb)); in packet_rcv_spkt()
1868 spkt->spkt_family = dev->type; in packet_rcv_spkt()
1869 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); in packet_rcv_spkt()
1870 spkt->spkt_protocol = skb->protocol; in packet_rcv_spkt()
1888 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && in packet_parse_headers()
1889 sock->type == SOCK_RAW) { in packet_parse_headers()
1891 skb->protocol = dev_parse_header_protocol(skb); in packet_parse_headers()
1905 struct sock *sk = sock->sk; in packet_sendmsg_spkt()
1906 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); in packet_sendmsg_spkt()
1919 if (msg->msg_namelen < sizeof(struct sockaddr)) in packet_sendmsg_spkt()
1920 return -EINVAL; in packet_sendmsg_spkt()
1921 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) in packet_sendmsg_spkt()
1922 proto = saddr->spkt_protocol; in packet_sendmsg_spkt()
1924 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ in packet_sendmsg_spkt()
1930 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; in packet_sendmsg_spkt()
1933 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); in packet_sendmsg_spkt()
1934 err = -ENODEV; in packet_sendmsg_spkt()
1938 err = -ENETDOWN; in packet_sendmsg_spkt()
1939 if (!(dev->flags & IFF_UP)) in packet_sendmsg_spkt()
1949 err = -EPROTONOSUPPORT; in packet_sendmsg_spkt()
1955 err = -EMSGSIZE; in packet_sendmsg_spkt()
1956 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) in packet_sendmsg_spkt()
1961 int tlen = dev->needed_tailroom; in packet_sendmsg_spkt()
1962 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; in packet_sendmsg_spkt()
1967 return -ENOBUFS; in packet_sendmsg_spkt()
1977 skb->data -= hhlen; in packet_sendmsg_spkt()
1978 skb->tail -= hhlen; in packet_sendmsg_spkt()
1988 if (!dev_validate_header(dev, skb->data, len)) { in packet_sendmsg_spkt()
1989 err = -EINVAL; in packet_sendmsg_spkt()
1992 if (len > (dev->mtu + dev->hard_header_len + extra_len) && in packet_sendmsg_spkt()
1994 err = -EMSGSIZE; in packet_sendmsg_spkt()
1999 if (msg->msg_controllen) { in packet_sendmsg_spkt()
2005 skb->protocol = proto; in packet_sendmsg_spkt()
2006 skb->dev = dev; in packet_sendmsg_spkt()
2007 skb->priority = sk->sk_priority; in packet_sendmsg_spkt()
2008 skb->mark = sk->sk_mark; in packet_sendmsg_spkt()
2009 skb->tstamp = sockc.transmit_time; in packet_sendmsg_spkt()
2014 skb->no_fcs = 1; in packet_sendmsg_spkt()
2036 filter = rcu_dereference(sk->sk_filter); in run_filter()
2038 res = bpf_prog_run_clear_cb(filter->prog, skb); in run_filter()
2050 return -EINVAL; in packet_rcv_vnet()
2051 *len -= sizeof(vnet_hdr); in packet_rcv_vnet()
2054 return -EINVAL; in packet_rcv_vnet()
2063 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2064 * and skb->cb are mangled. It works because (and until) packets
2077 u8 *skb_head = skb->data; in packet_rcv()
2078 int skb_len = skb->len; in packet_rcv()
2082 if (skb->pkt_type == PACKET_LOOPBACK) in packet_rcv()
2085 sk = pt->af_packet_priv; in packet_rcv()
2091 skb->dev = dev; in packet_rcv()
2101 if (sk->sk_type != SOCK_DGRAM) in packet_rcv()
2102 skb_push(skb, skb->data - skb_mac_header(skb)); in packet_rcv()
2103 else if (skb->pkt_type == PACKET_OUTGOING) { in packet_rcv()
2109 snaplen = skb->len; in packet_rcv()
2117 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) in packet_rcv()
2125 if (skb_head != skb->data) { in packet_rcv()
2126 skb->data = skb_head; in packet_rcv()
2127 skb->len = skb_len; in packet_rcv()
2133 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); in packet_rcv()
2135 sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_rcv()
2136 sll->sll_hatype = dev->type; in packet_rcv()
2137 sll->sll_pkttype = skb->pkt_type; in packet_rcv()
2138 if (unlikely(po->origdev)) in packet_rcv()
2139 sll->sll_ifindex = orig_dev->ifindex; in packet_rcv()
2141 sll->sll_ifindex = dev->ifindex; in packet_rcv()
2143 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); in packet_rcv()
2145 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). in packet_rcv()
2148 PACKET_SKB_CB(skb)->sa.origlen = skb->len; in packet_rcv()
2154 skb->dev = NULL; in packet_rcv()
2160 spin_lock(&sk->sk_receive_queue.lock); in packet_rcv()
2161 po->stats.stats1.tp_packets++; in packet_rcv()
2163 __skb_queue_tail(&sk->sk_receive_queue, skb); in packet_rcv()
2164 spin_unlock(&sk->sk_receive_queue.lock); in packet_rcv()
2165 sk->sk_data_ready(sk); in packet_rcv()
2170 atomic_inc(&po->tp_drops); in packet_rcv()
2171 atomic_inc(&sk->sk_drops); in packet_rcv()
2174 if (skb_head != skb->data && skb_shared(skb)) { in packet_rcv()
2175 skb->data = skb_head; in packet_rcv()
2176 skb->len = skb_len; in packet_rcv()
2193 u8 *skb_head = skb->data; in tpacket_rcv()
2194 int skb_len = skb->len; in tpacket_rcv()
2213 if (skb->pkt_type == PACKET_LOOPBACK) in tpacket_rcv()
2216 sk = pt->af_packet_priv; in tpacket_rcv()
2223 if (sk->sk_type != SOCK_DGRAM) in tpacket_rcv()
2224 skb_push(skb, skb->data - skb_mac_header(skb)); in tpacket_rcv()
2225 else if (skb->pkt_type == PACKET_OUTGOING) { in tpacket_rcv()
2231 snaplen = skb->len; in tpacket_rcv()
2239 atomic_inc(&po->tp_drops); in tpacket_rcv()
2243 if (skb->ip_summed == CHECKSUM_PARTIAL) in tpacket_rcv()
2245 else if (skb->pkt_type != PACKET_OUTGOING && 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()
2281 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0, in tpacket_rcv()
2282 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll)); in tpacket_rcv()
2286 snaplen = po->rx_ring.frame_size - macoff; in tpacket_rcv()
2293 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { in tpacket_rcv()
2296 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; in tpacket_rcv()
2302 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; in tpacket_rcv()
2306 spin_lock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2312 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2313 slot_id = po->rx_ring.head; in tpacket_rcv()
2314 if (test_bit(slot_id, po->rx_ring.rx_owner_map)) in tpacket_rcv()
2316 __set_bit(slot_id, po->rx_ring.rx_owner_map); in tpacket_rcv()
2320 virtio_net_hdr_from_skb(skb, h.raw + macoff - in tpacket_rcv()
2323 if (po->tp_version == TPACKET_V3) in tpacket_rcv()
2324 prb_clear_blk_fill_status(&po->rx_ring); in tpacket_rcv()
2328 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2329 packet_increment_rx_head(po, &po->rx_ring); in tpacket_rcv()
2331 * LOSING will be reported till you read the stats, in tpacket_rcv()
2332 * because it's COR - Clear On Read. in tpacket_rcv()
2336 if (atomic_read(&po->tp_drops)) in tpacket_rcv()
2340 po->stats.stats1.tp_packets++; in tpacket_rcv()
2343 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); in tpacket_rcv()
2345 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2353 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE); in tpacket_rcv()
2359 switch (po->tp_version) { in tpacket_rcv()
2361 h.h1->tp_len = skb->len; in tpacket_rcv()
2362 h.h1->tp_snaplen = snaplen; in tpacket_rcv()
2363 h.h1->tp_mac = macoff; in tpacket_rcv()
2364 h.h1->tp_net = netoff; in tpacket_rcv()
2365 h.h1->tp_sec = ts.tv_sec; in tpacket_rcv()
2366 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; in tpacket_rcv()
2370 h.h2->tp_len = skb->len; in tpacket_rcv()
2371 h.h2->tp_snaplen = snaplen; in tpacket_rcv()
2372 h.h2->tp_mac = macoff; in tpacket_rcv()
2373 h.h2->tp_net = netoff; in tpacket_rcv()
2374 h.h2->tp_sec = ts.tv_sec; in tpacket_rcv()
2375 h.h2->tp_nsec = ts.tv_nsec; in tpacket_rcv()
2377 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); in tpacket_rcv()
2378 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); in tpacket_rcv()
2381 h.h2->tp_vlan_tci = 0; in tpacket_rcv()
2382 h.h2->tp_vlan_tpid = 0; in tpacket_rcv()
2384 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); in tpacket_rcv()
2391 h.h3->tp_status |= status; in tpacket_rcv()
2392 h.h3->tp_len = skb->len; in tpacket_rcv()
2393 h.h3->tp_snaplen = snaplen; in tpacket_rcv()
2394 h.h3->tp_mac = macoff; in tpacket_rcv()
2395 h.h3->tp_net = netoff; in tpacket_rcv()
2396 h.h3->tp_sec = ts.tv_sec; in tpacket_rcv()
2397 h.h3->tp_nsec = ts.tv_nsec; in tpacket_rcv()
2398 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); in tpacket_rcv()
2406 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); in tpacket_rcv()
2407 sll->sll_family = AF_PACKET; in tpacket_rcv()
2408 sll->sll_hatype = dev->type; in tpacket_rcv()
2409 sll->sll_protocol = skb->protocol; in tpacket_rcv()
2410 sll->sll_pkttype = skb->pkt_type; in tpacket_rcv()
2411 if (unlikely(po->origdev)) in tpacket_rcv()
2412 sll->sll_ifindex = orig_dev->ifindex; in tpacket_rcv()
2414 sll->sll_ifindex = dev->ifindex; in tpacket_rcv()
2419 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2431 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2432 spin_lock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2434 __clear_bit(slot_id, po->rx_ring.rx_owner_map); in tpacket_rcv()
2435 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2436 sk->sk_data_ready(sk); in tpacket_rcv()
2437 } else if (po->tp_version == TPACKET_V3) { in tpacket_rcv()
2438 prb_clear_blk_fill_status(&po->rx_ring); in tpacket_rcv()
2442 if (skb_head != skb->data && skb_shared(skb)) { in tpacket_rcv()
2443 skb->data = skb_head; in tpacket_rcv()
2444 skb->len = skb_len; in tpacket_rcv()
2454 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2455 atomic_inc(&po->tp_drops); in tpacket_rcv()
2458 sk->sk_data_ready(sk); in tpacket_rcv()
2465 struct packet_sock *po = pkt_sk(skb->sk); in tpacket_destruct_skb()
2467 if (likely(po->tx_ring.pg_vec)) { in tpacket_destruct_skb()
2472 packet_dec_pending(&po->tx_ring); in tpacket_destruct_skb()
2477 if (!packet_read_pending(&po->tx_ring)) in tpacket_destruct_skb()
2478 complete(&po->skb_completion); in tpacket_destruct_skb()
2486 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && in __packet_snd_vnet_parse()
2487 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + in __packet_snd_vnet_parse()
2488 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > in __packet_snd_vnet_parse()
2489 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) in __packet_snd_vnet_parse()
2490 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), in __packet_snd_vnet_parse()
2491 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + in __packet_snd_vnet_parse()
2492 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); in __packet_snd_vnet_parse()
2494 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) in __packet_snd_vnet_parse()
2495 return -EINVAL; in __packet_snd_vnet_parse()
2504 return -EINVAL; in packet_snd_vnet_parse()
2505 *len -= sizeof(*vnet_hdr); in packet_snd_vnet_parse()
2507 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) in packet_snd_vnet_parse()
2508 return -EFAULT; in packet_snd_vnet_parse()
2520 struct socket *sock = po->sk.sk_socket; in tpacket_fill_skb()
2526 skb->protocol = proto; in tpacket_fill_skb()
2527 skb->dev = dev; in tpacket_fill_skb()
2528 skb->priority = po->sk.sk_priority; in tpacket_fill_skb()
2529 skb->mark = po->sk.sk_mark; in tpacket_fill_skb()
2530 skb->tstamp = sockc->transmit_time; in tpacket_fill_skb()
2531 skb_setup_tx_timestamp(skb, sockc->tsflags); in tpacket_fill_skb()
2539 if (sock->type == SOCK_DGRAM) { in tpacket_fill_skb()
2543 return -EINVAL; in tpacket_fill_skb()
2547 skb_push(skb, dev->hard_header_len); in tpacket_fill_skb()
2548 skb_put(skb, copylen - dev->hard_header_len); in tpacket_fill_skb()
2552 if (!dev_validate_header(dev, skb->data, hdrlen)) in tpacket_fill_skb()
2553 return -EINVAL; in tpacket_fill_skb()
2556 to_write -= hdrlen; in tpacket_fill_skb()
2560 len_max = PAGE_SIZE - offset; in tpacket_fill_skb()
2563 skb->data_len = to_write; in tpacket_fill_skb()
2564 skb->len += to_write; in tpacket_fill_skb()
2565 skb->truesize += to_write; in tpacket_fill_skb()
2566 refcount_add(to_write, &po->sk.sk_wmem_alloc); in tpacket_fill_skb()
2569 nr_frags = skb_shinfo(skb)->nr_frags; in tpacket_fill_skb()
2574 return -EFAULT; in tpacket_fill_skb()
2582 to_write -= len; in tpacket_fill_skb()
2601 switch (po->tp_version) { in tpacket_parse_header()
2603 if (ph.h3->tp_next_offset != 0) { in tpacket_parse_header()
2605 return -EINVAL; in tpacket_parse_header()
2607 tp_len = ph.h3->tp_len; in tpacket_parse_header()
2610 tp_len = ph.h2->tp_len; in tpacket_parse_header()
2613 tp_len = ph.h1->tp_len; in tpacket_parse_header()
2618 return -EMSGSIZE; in tpacket_parse_header()
2621 if (unlikely(po->tp_tx_has_off)) { in tpacket_parse_header()
2624 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); in tpacket_parse_header()
2625 off_max = po->tx_ring.frame_size - tp_len; in tpacket_parse_header()
2626 if (po->sk.sk_type == SOCK_DGRAM) { in tpacket_parse_header()
2627 switch (po->tp_version) { in tpacket_parse_header()
2629 off = ph.h3->tp_net; in tpacket_parse_header()
2632 off = ph.h2->tp_net; in tpacket_parse_header()
2635 off = ph.h1->tp_net; in tpacket_parse_header()
2639 switch (po->tp_version) { in tpacket_parse_header()
2641 off = ph.h3->tp_mac; in tpacket_parse_header()
2644 off = ph.h2->tp_mac; in tpacket_parse_header()
2647 off = ph.h1->tp_mac; in tpacket_parse_header()
2652 return -EINVAL; in tpacket_parse_header()
2654 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); in tpacket_parse_header()
2670 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); in tpacket_snd()
2671 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); in tpacket_snd()
2680 mutex_lock(&po->pg_vec_lock); in tpacket_snd()
2685 if (unlikely(!po->tx_ring.pg_vec)) { in tpacket_snd()
2686 err = -EBUSY; in tpacket_snd()
2691 proto = READ_ONCE(po->num); in tpacket_snd()
2693 err = -EINVAL; in tpacket_snd()
2694 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) in tpacket_snd()
2696 if (msg->msg_namelen < (saddr->sll_halen in tpacket_snd()
2700 proto = saddr->sll_protocol; in tpacket_snd()
2701 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); in tpacket_snd()
2702 if (po->sk.sk_socket->type == SOCK_DGRAM) { in tpacket_snd()
2703 if (dev && msg->msg_namelen < dev->addr_len + in tpacket_snd()
2706 addr = saddr->sll_addr; in tpacket_snd()
2710 err = -ENXIO; in tpacket_snd()
2713 err = -ENETDOWN; in tpacket_snd()
2714 if (unlikely(!(dev->flags & IFF_UP))) in tpacket_snd()
2717 sockcm_init(&sockc, &po->sk); in tpacket_snd()
2718 if (msg->msg_controllen) { in tpacket_snd()
2719 err = sock_cmsg_send(&po->sk, msg, &sockc); in tpacket_snd()
2724 if (po->sk.sk_socket->type == SOCK_RAW) in tpacket_snd()
2725 reserve = dev->hard_header_len; in tpacket_snd()
2726 size_max = po->tx_ring.frame_size in tpacket_snd()
2727 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); in tpacket_snd()
2729 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) in tpacket_snd()
2730 size_max = dev->mtu + reserve + VLAN_HLEN; in tpacket_snd()
2732 reinit_completion(&po->skb_completion); in tpacket_snd()
2735 ph = packet_current_frame(po, &po->tx_ring, in tpacket_snd()
2739 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); in tpacket_snd()
2740 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); in tpacket_snd()
2742 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; in tpacket_snd()
2757 tlen = dev->needed_tailroom; in tpacket_snd()
2758 if (po->has_vnet_hdr) { in tpacket_snd()
2761 tp_len -= sizeof(*vnet_hdr); in tpacket_snd()
2764 tp_len = -EINVAL; in tpacket_snd()
2768 vnet_hdr->hdr_len); in tpacket_snd()
2770 copylen = max_t(int, copylen, dev->hard_header_len); in tpacket_snd()
2771 skb = sock_alloc_send_skb(&po->sk, in tpacket_snd()
2773 (copylen - dev->hard_header_len), in tpacket_snd()
2785 tp_len > dev->mtu + reserve && in tpacket_snd()
2786 !po->has_vnet_hdr && in tpacket_snd()
2788 tp_len = -EMSGSIZE; in tpacket_snd()
2792 if (po->tp_loss) { in tpacket_snd()
2795 packet_increment_head(&po->tx_ring); in tpacket_snd()
2805 if (po->has_vnet_hdr) { in tpacket_snd()
2807 tp_len = -EINVAL; in tpacket_snd()
2813 skb->destructor = tpacket_destruct_skb; in tpacket_snd()
2815 packet_inc_pending(&po->tx_ring); in tpacket_snd()
2818 err = po->xmit(skb); in tpacket_snd()
2834 packet_increment_head(&po->tx_ring); in tpacket_snd()
2838 * to call it as it's per_cpu variable, but in fast-path in tpacket_snd()
2839 * we already short-circuit the loop with the first in tpacket_snd()
2843 (need_wait && packet_read_pending(&po->tx_ring)))); in tpacket_snd()
2854 mutex_unlock(&po->pg_vec_lock); in tpacket_snd()
2869 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, in packet_alloc_skb()
2876 skb->data_len = len - linear; in packet_alloc_skb()
2877 skb->len += len - linear; in packet_alloc_skb()
2884 struct sock *sk = sock->sk; in packet_snd()
2885 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); in packet_snd()
2905 proto = READ_ONCE(po->num); in packet_snd()
2907 err = -EINVAL; in packet_snd()
2908 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) in packet_snd()
2910 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) in packet_snd()
2912 proto = saddr->sll_protocol; in packet_snd()
2913 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); in packet_snd()
2914 if (sock->type == SOCK_DGRAM) { in packet_snd()
2915 if (dev && msg->msg_namelen < dev->addr_len + in packet_snd()
2918 addr = saddr->sll_addr; in packet_snd()
2922 err = -ENXIO; in packet_snd()
2925 err = -ENETDOWN; in packet_snd()
2926 if (unlikely(!(dev->flags & IFF_UP))) in packet_snd()
2930 sockc.mark = sk->sk_mark; in packet_snd()
2931 if (msg->msg_controllen) { in packet_snd()
2937 if (sock->type == SOCK_RAW) in packet_snd()
2938 reserve = dev->hard_header_len; in packet_snd()
2939 if (po->has_vnet_hdr) { in packet_snd()
2948 err = -EPROTONOSUPPORT; in packet_snd()
2954 err = -EMSGSIZE; in packet_snd()
2956 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) in packet_snd()
2959 err = -ENOBUFS; in packet_snd()
2961 tlen = dev->needed_tailroom; in packet_snd()
2963 linear = max(linear, min_t(int, len, dev->hard_header_len)); in packet_snd()
2965 msg->msg_flags & MSG_DONTWAIT, &err); in packet_snd()
2971 err = -EINVAL; in packet_snd()
2972 if (sock->type == SOCK_DGRAM) { in packet_snd()
2977 skb_reserve(skb, -reserve); in packet_snd()
2979 dev->min_header_len != dev->hard_header_len) in packet_snd()
2983 /* Returns -EFAULT on error */ in packet_snd()
2984 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); in packet_snd()
2988 if ((sock->type == SOCK_RAW && in packet_snd()
2989 !dev_validate_header(dev, skb->data, len)) || !skb->len) { in packet_snd()
2990 err = -EINVAL; in packet_snd()
2996 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && in packet_snd()
2998 err = -EMSGSIZE; in packet_snd()
3002 skb->protocol = proto; in packet_snd()
3003 skb->dev = dev; in packet_snd()
3004 skb->priority = sk->sk_priority; in packet_snd()
3005 skb->mark = sockc.mark; in packet_snd()
3006 skb->tstamp = sockc.transmit_time; in packet_snd()
3019 skb->no_fcs = 1; in packet_snd()
3021 err = po->xmit(skb); in packet_snd()
3044 struct sock *sk = sock->sk; in packet_sendmsg()
3050 if (data_race(po->tx_ring.pg_vec)) in packet_sendmsg()
3063 struct sock *sk = sock->sk; in packet_release()
3075 mutex_lock(&net->packet.sklist_lock); in packet_release()
3077 mutex_unlock(&net->packet.sklist_lock); in packet_release()
3080 sock_prot_inuse_add(net, sk->sk_prot, -1); in packet_release()
3083 spin_lock(&po->bind_lock); in packet_release()
3087 if (po->prot_hook.dev) { in packet_release()
3088 dev_put(po->prot_hook.dev); in packet_release()
3089 po->prot_hook.dev = NULL; in packet_release()
3091 spin_unlock(&po->bind_lock); in packet_release()
3096 if (po->rx_ring.pg_vec) { in packet_release()
3101 if (po->tx_ring.pg_vec) { in packet_release()
3111 kfree(po->rollover); in packet_release()
3117 * Now the socket is dead. No more input will appear. in packet_release()
3120 sock->sk = NULL; in packet_release()
3124 skb_queue_purge(&sk->sk_receive_queue); in packet_release()
3148 spin_lock(&po->bind_lock); in packet_do_bind()
3151 if (po->fanout) { in packet_do_bind()
3152 ret = -EINVAL; in packet_do_bind()
3159 ret = -ENODEV; in packet_do_bind()
3165 ret = -ENODEV; in packet_do_bind()
3173 proto_curr = po->prot_hook.type; in packet_do_bind()
3174 dev_curr = po->prot_hook.dev; in packet_do_bind()
3179 if (po->running) { in packet_do_bind()
3184 WRITE_ONCE(po->num, 0); in packet_do_bind()
3187 dev_curr = po->prot_hook.dev; in packet_do_bind()
3190 dev->ifindex); in packet_do_bind()
3193 BUG_ON(po->running); in packet_do_bind()
3194 WRITE_ONCE(po->num, proto); in packet_do_bind()
3195 po->prot_hook.type = proto; in packet_do_bind()
3199 po->prot_hook.dev = NULL; in packet_do_bind()
3200 WRITE_ONCE(po->ifindex, -1); in packet_do_bind()
3203 po->prot_hook.dev = dev; in packet_do_bind()
3204 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); in packet_do_bind()
3214 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { in packet_do_bind()
3217 sk->sk_err = ENETDOWN; in packet_do_bind()
3219 sk->sk_error_report(sk); in packet_do_bind()
3224 spin_unlock(&po->bind_lock); in packet_do_bind()
3236 struct sock *sk = sock->sk; in packet_bind_spkt()
3237 char name[sizeof(uaddr->sa_data) + 1]; in packet_bind_spkt()
3244 return -EINVAL; in packet_bind_spkt()
3245 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be in packet_bind_spkt()
3246 * zero-terminated. in packet_bind_spkt()
3248 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); in packet_bind_spkt()
3249 name[sizeof(uaddr->sa_data)] = 0; in packet_bind_spkt()
3251 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); in packet_bind_spkt()
3257 struct sock *sk = sock->sk; in packet_bind()
3264 return -EINVAL; in packet_bind()
3265 if (sll->sll_family != AF_PACKET) in packet_bind()
3266 return -EINVAL; in packet_bind()
3268 return packet_do_bind(sk, NULL, sll->sll_ifindex, in packet_bind()
3269 sll->sll_protocol ? : pkt_sk(sk)->num); in packet_bind()
3290 if (!ns_capable(net->user_ns, CAP_NET_RAW)) in packet_create()
3291 return -EPERM; in packet_create()
3292 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && in packet_create()
3293 sock->type != SOCK_PACKET) in packet_create()
3294 return -ESOCKTNOSUPPORT; in packet_create()
3296 sock->state = SS_UNCONNECTED; in packet_create()
3298 err = -ENOBUFS; in packet_create()
3303 sock->ops = &packet_ops; in packet_create()
3304 if (sock->type == SOCK_PACKET) in packet_create()
3305 sock->ops = &packet_ops_spkt; in packet_create()
3310 init_completion(&po->skb_completion); in packet_create()
3311 sk->sk_family = PF_PACKET; in packet_create()
3312 po->num = proto; in packet_create()
3313 po->xmit = dev_queue_xmit; in packet_create()
3321 sk->sk_destruct = packet_sock_destruct; in packet_create()
3328 spin_lock_init(&po->bind_lock); in packet_create()
3329 mutex_init(&po->pg_vec_lock); in packet_create()
3330 po->rollover = NULL; in packet_create()
3331 po->prot_hook.func = packet_rcv; in packet_create()
3333 if (sock->type == SOCK_PACKET) in packet_create()
3334 po->prot_hook.func = packet_rcv_spkt; in packet_create()
3336 po->prot_hook.af_packet_priv = sk; in packet_create()
3337 po->prot_hook.af_packet_net = sock_net(sk); in packet_create()
3340 po->prot_hook.type = proto; in packet_create()
3344 mutex_lock(&net->packet.sklist_lock); in packet_create()
3345 sk_add_node_tail_rcu(sk, &net->packet.sklist); in packet_create()
3346 mutex_unlock(&net->packet.sklist_lock); in packet_create()
3367 struct sock *sk = sock->sk; in packet_recvmsg()
3373 err = -EINVAL; in packet_recvmsg()
3379 if (pkt_sk(sk)->ifindex < 0) in packet_recvmsg()
3380 return -ENODEV; in packet_recvmsg()
3391 * of horrible races and re-entrancy so we can forget about it in packet_recvmsg()
3411 if (pkt_sk(sk)->has_vnet_hdr) { in packet_recvmsg()
3422 copied = skb->len; in packet_recvmsg()
3425 msg->msg_flags |= MSG_TRUNC; in packet_recvmsg()
3432 if (sock->type != SOCK_PACKET) { in packet_recvmsg()
3433 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3436 origlen = PACKET_SKB_CB(skb)->sa.origlen; in packet_recvmsg()
3437 sll->sll_family = AF_PACKET; in packet_recvmsg()
3438 sll->sll_protocol = skb->protocol; in packet_recvmsg()
3443 if (msg->msg_name) { in packet_recvmsg()
3444 const size_t max_len = min(sizeof(skb->cb), in packet_recvmsg()
3451 if (sock->type == SOCK_PACKET) { in packet_recvmsg()
3453 msg->msg_namelen = sizeof(struct sockaddr_pkt); in packet_recvmsg()
3454 copy_len = msg->msg_namelen; in packet_recvmsg()
3456 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3458 msg->msg_namelen = sll->sll_halen + in packet_recvmsg()
3460 copy_len = msg->msg_namelen; in packet_recvmsg()
3461 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { in packet_recvmsg()
3462 memset(msg->msg_name + in packet_recvmsg()
3464 0, sizeof(sll->sll_addr)); in packet_recvmsg()
3465 msg->msg_namelen = sizeof(struct sockaddr_ll); in packet_recvmsg()
3470 msg->msg_namelen = copy_len; in packet_recvmsg()
3472 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); in packet_recvmsg()
3475 if (pkt_sk(sk)->auxdata) { in packet_recvmsg()
3479 if (skb->ip_summed == CHECKSUM_PARTIAL) in packet_recvmsg()
3481 else if (skb->pkt_type != PACKET_OUTGOING && in packet_recvmsg()
3486 aux.tp_snaplen = skb->len; in packet_recvmsg()
3491 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); in packet_recvmsg()
3502 * hides all the races and re-entrancy issues from us. in packet_recvmsg()
3504 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); in packet_recvmsg()
3516 struct sock *sk = sock->sk; in packet_getname_spkt()
3519 return -EOPNOTSUPP; in packet_getname_spkt()
3521 uaddr->sa_family = AF_PACKET; in packet_getname_spkt()
3522 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); in packet_getname_spkt()
3524 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); in packet_getname_spkt()
3526 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); in packet_getname_spkt()
3536 struct sock *sk = sock->sk; in packet_getname()
3542 return -EOPNOTSUPP; in packet_getname()
3544 ifindex = READ_ONCE(po->ifindex); in packet_getname()
3545 sll->sll_family = AF_PACKET; in packet_getname()
3546 sll->sll_ifindex = ifindex; in packet_getname()
3547 sll->sll_protocol = READ_ONCE(po->num); in packet_getname()
3548 sll->sll_pkttype = 0; in packet_getname()
3552 sll->sll_hatype = dev->type; in packet_getname()
3553 sll->sll_halen = dev->addr_len; in packet_getname()
3554 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); in packet_getname()
3556 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ in packet_getname()
3557 sll->sll_halen = 0; in packet_getname()
3561 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; in packet_getname()
3567 switch (i->type) { in packet_dev_mc()
3569 if (i->alen != dev->addr_len) in packet_dev_mc()
3570 return -EINVAL; in packet_dev_mc()
3572 return dev_mc_add(dev, i->addr); in packet_dev_mc()
3574 return dev_mc_del(dev, i->addr); in packet_dev_mc()
3581 if (i->alen != dev->addr_len) in packet_dev_mc()
3582 return -EINVAL; in packet_dev_mc()
3584 return dev_uc_add(dev, i->addr); in packet_dev_mc()
3586 return dev_uc_del(dev, i->addr); in packet_dev_mc()
3600 if (ml->ifindex == dev->ifindex) { in packet_dev_mclist_delete()
3601 packet_dev_mc(dev, ml, -1); in packet_dev_mclist_delete()
3602 *mlp = ml->next; in packet_dev_mclist_delete()
3605 mlp = &ml->next; in packet_dev_mclist_delete()
3618 err = -ENODEV; in packet_mc_add()
3619 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); in packet_mc_add()
3623 err = -EINVAL; in packet_mc_add()
3624 if (mreq->mr_alen > dev->addr_len) in packet_mc_add()
3627 err = -ENOBUFS; in packet_mc_add()
3633 for (ml = po->mclist; ml; ml = ml->next) { in packet_mc_add()
3634 if (ml->ifindex == mreq->mr_ifindex && in packet_mc_add()
3635 ml->type == mreq->mr_type && in packet_mc_add()
3636 ml->alen == mreq->mr_alen && in packet_mc_add()
3637 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { in packet_mc_add()
3638 ml->count++; in packet_mc_add()
3645 i->type = mreq->mr_type; in packet_mc_add()
3646 i->ifindex = mreq->mr_ifindex; in packet_mc_add()
3647 i->alen = mreq->mr_alen; in packet_mc_add()
3648 memcpy(i->addr, mreq->mr_address, i->alen); in packet_mc_add()
3649 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); in packet_mc_add()
3650 i->count = 1; in packet_mc_add()
3651 i->next = po->mclist; in packet_mc_add()
3652 po->mclist = i; in packet_mc_add()
3655 po->mclist = i->next; in packet_mc_add()
3670 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { in packet_mc_drop()
3671 if (ml->ifindex == mreq->mr_ifindex && in packet_mc_drop()
3672 ml->type == mreq->mr_type && in packet_mc_drop()
3673 ml->alen == mreq->mr_alen && in packet_mc_drop()
3674 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { in packet_mc_drop()
3675 if (--ml->count == 0) { in packet_mc_drop()
3677 *mlp = ml->next; in packet_mc_drop()
3678 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); in packet_mc_drop()
3680 packet_dev_mc(dev, ml, -1); in packet_mc_drop()
3695 if (!po->mclist) in packet_flush_mclist()
3699 while ((ml = po->mclist) != NULL) { in packet_flush_mclist()
3702 po->mclist = ml->next; in packet_flush_mclist()
3703 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); in packet_flush_mclist()
3705 packet_dev_mc(dev, ml, -1); in packet_flush_mclist()
3715 struct sock *sk = sock->sk; in packet_setsockopt()
3720 return -ENOPROTOOPT; in packet_setsockopt()
3730 return -EINVAL; in packet_setsockopt()
3734 return -EFAULT; in packet_setsockopt()
3736 return -EINVAL; in packet_setsockopt()
3751 switch (po->tp_version) { in packet_setsockopt()
3762 ret = -EINVAL; in packet_setsockopt()
3765 ret = -EFAULT; in packet_setsockopt()
3778 return -EINVAL; in packet_setsockopt()
3780 return -EFAULT; in packet_setsockopt()
3782 pkt_sk(sk)->copy_thresh = val; in packet_setsockopt()
3790 return -EINVAL; in packet_setsockopt()
3792 return -EFAULT; in packet_setsockopt()
3799 return -EINVAL; in packet_setsockopt()
3802 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3803 ret = -EBUSY; in packet_setsockopt()
3805 po->tp_version = val; in packet_setsockopt()
3816 return -EINVAL; in packet_setsockopt()
3818 return -EFAULT; in packet_setsockopt()
3820 return -EINVAL; in packet_setsockopt()
3822 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3823 ret = -EBUSY; in packet_setsockopt()
3825 po->tp_reserve = val; in packet_setsockopt()
3836 return -EINVAL; in packet_setsockopt()
3838 return -EFAULT; in packet_setsockopt()
3841 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3842 ret = -EBUSY; in packet_setsockopt()
3844 po->tp_loss = !!val; in packet_setsockopt()
3855 return -EINVAL; in packet_setsockopt()
3857 return -EFAULT; in packet_setsockopt()
3860 po->auxdata = !!val; in packet_setsockopt()
3869 return -EINVAL; in packet_setsockopt()
3871 return -EFAULT; in packet_setsockopt()
3874 po->origdev = !!val; in packet_setsockopt()
3882 if (sock->type != SOCK_RAW) in packet_setsockopt()
3883 return -EINVAL; in packet_setsockopt()
3885 return -EINVAL; in packet_setsockopt()
3887 return -EFAULT; in packet_setsockopt()
3890 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3891 ret = -EBUSY; in packet_setsockopt()
3893 po->has_vnet_hdr = !!val; in packet_setsockopt()
3904 return -EINVAL; in packet_setsockopt()
3906 return -EFAULT; in packet_setsockopt()
3908 po->tp_tstamp = val; in packet_setsockopt()
3916 return -EINVAL; in packet_setsockopt()
3918 return -EFAULT; in packet_setsockopt()
3925 if (!READ_ONCE(po->fanout)) in packet_setsockopt()
3926 return -EINVAL; in packet_setsockopt()
3935 return -EINVAL; in packet_setsockopt()
3937 return -EFAULT; in packet_setsockopt()
3939 return -EINVAL; in packet_setsockopt()
3941 po->prot_hook.ignore_outgoing = !!val; in packet_setsockopt()
3949 return -EINVAL; in packet_setsockopt()
3951 return -EFAULT; in packet_setsockopt()
3954 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3955 ret = -EBUSY; in packet_setsockopt()
3957 po->tp_tx_has_off = !!val; in packet_setsockopt()
3968 return -EINVAL; in packet_setsockopt()
3970 return -EFAULT; in packet_setsockopt()
3972 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; in packet_setsockopt()
3976 return -ENOPROTOOPT; in packet_setsockopt()
3985 struct sock *sk = sock->sk; in packet_getsockopt()
3993 return -ENOPROTOOPT; in packet_getsockopt()
3996 return -EFAULT; in packet_getsockopt()
3999 return -EINVAL; in packet_getsockopt()
4003 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_getsockopt()
4004 memcpy(&st, &po->stats, sizeof(st)); in packet_getsockopt()
4005 memset(&po->stats, 0, sizeof(po->stats)); in packet_getsockopt()
4006 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_getsockopt()
4007 drops = atomic_xchg(&po->tp_drops, 0); in packet_getsockopt()
4009 if (po->tp_version == TPACKET_V3) { in packet_getsockopt()
4023 val = po->auxdata; in packet_getsockopt()
4026 val = po->origdev; in packet_getsockopt()
4029 val = po->has_vnet_hdr; in packet_getsockopt()
4032 val = po->tp_version; in packet_getsockopt()
4038 return -EINVAL; in packet_getsockopt()
4040 return -EFAULT; in packet_getsockopt()
4052 return -EINVAL; in packet_getsockopt()
4056 val = po->tp_reserve; in packet_getsockopt()
4059 val = po->tp_loss; in packet_getsockopt()
4062 val = po->tp_tstamp; in packet_getsockopt()
4065 val = (po->fanout ? in packet_getsockopt()
4066 ((u32)po->fanout->id | in packet_getsockopt()
4067 ((u32)po->fanout->type << 16) | in packet_getsockopt()
4068 ((u32)po->fanout->flags << 24)) : in packet_getsockopt()
4072 val = po->prot_hook.ignore_outgoing; in packet_getsockopt()
4075 if (!po->rollover) in packet_getsockopt()
4076 return -EINVAL; in packet_getsockopt()
4077 rstats.tp_all = atomic_long_read(&po->rollover->num); in packet_getsockopt()
4078 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); in packet_getsockopt()
4079 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); in packet_getsockopt()
4084 val = po->tp_tx_has_off; in packet_getsockopt()
4090 return -ENOPROTOOPT; in packet_getsockopt()
4096 return -EFAULT; in packet_getsockopt()
4098 return -EFAULT; in packet_getsockopt()
4110 sk_for_each_rcu(sk, &net->packet.sklist) { in packet_notifier()
4115 if (po->mclist) in packet_notifier()
4116 packet_dev_mclist_delete(dev, &po->mclist); in packet_notifier()
4120 if (dev->ifindex == po->ifindex) { in packet_notifier()
4121 spin_lock(&po->bind_lock); in packet_notifier()
4122 if (po->running) { in packet_notifier()
4124 sk->sk_err = ENETDOWN; in packet_notifier()
4126 sk->sk_error_report(sk); in packet_notifier()
4130 WRITE_ONCE(po->ifindex, -1); in packet_notifier()
4131 if (po->prot_hook.dev) in packet_notifier()
4132 dev_put(po->prot_hook.dev); in packet_notifier()
4133 po->prot_hook.dev = NULL; in packet_notifier()
4135 spin_unlock(&po->bind_lock); in packet_notifier()
4139 if (dev->ifindex == po->ifindex) { in packet_notifier()
4140 spin_lock(&po->bind_lock); in packet_notifier()
4141 if (po->num) in packet_notifier()
4143 spin_unlock(&po->bind_lock); in packet_notifier()
4156 struct sock *sk = sock->sk; in packet_ioctl()
4170 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_ioctl()
4171 skb = skb_peek(&sk->sk_receive_queue); in packet_ioctl()
4173 amount = skb->len; in packet_ioctl()
4174 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_ioctl()
4196 return -ENOIOCTLCMD; in packet_ioctl()
4204 struct sock *sk = sock->sk; in packet_poll()
4208 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_poll()
4209 if (po->rx_ring.pg_vec) { in packet_poll()
4210 if (!packet_previous_rx_frame(po, &po->rx_ring, in packet_poll()
4215 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_poll()
4216 spin_lock_bh(&sk->sk_write_queue.lock); in packet_poll()
4217 if (po->tx_ring.pg_vec) { in packet_poll()
4218 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) in packet_poll()
4221 spin_unlock_bh(&sk->sk_write_queue.lock); in packet_poll()
4232 struct file *file = vma->vm_file; in packet_mm_open()
4233 struct socket *sock = file->private_data; in packet_mm_open()
4234 struct sock *sk = sock->sk; in packet_mm_open()
4237 atomic_inc(&pkt_sk(sk)->mapped); in packet_mm_open()
4242 struct file *file = vma->vm_file; in packet_mm_close()
4243 struct socket *sock = file->private_data; in packet_mm_close()
4244 struct sock *sk = sock->sk; in packet_mm_close()
4247 atomic_dec(&pkt_sk(sk)->mapped); in packet_mm_close()
4300 unsigned int block_nr = req->tp_block_nr; in alloc_pg_vec()
4335 struct tpacket_req *req = &req_u->req; in packet_set_ring()
4337 rb = tx_ring ? &po->tx_ring : &po->rx_ring; in packet_set_ring()
4338 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; in packet_set_ring()
4340 err = -EBUSY; in packet_set_ring()
4342 if (atomic_read(&po->mapped)) in packet_set_ring()
4348 if (req->tp_block_nr) { in packet_set_ring()
4352 err = -EBUSY; in packet_set_ring()
4353 if (unlikely(rb->pg_vec)) in packet_set_ring()
4356 switch (po->tp_version) { in packet_set_ring()
4358 po->tp_hdrlen = TPACKET_HDRLEN; in packet_set_ring()
4361 po->tp_hdrlen = TPACKET2_HDRLEN; in packet_set_ring()
4364 po->tp_hdrlen = TPACKET3_HDRLEN; in packet_set_ring()
4368 err = -EINVAL; in packet_set_ring()
4369 if (unlikely((int)req->tp_block_size <= 0)) in packet_set_ring()
4371 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) in packet_set_ring()
4373 min_frame_size = po->tp_hdrlen + po->tp_reserve; in packet_set_ring()
4374 if (po->tp_version >= TPACKET_V3 && in packet_set_ring()
4375 req->tp_block_size < in packet_set_ring()
4376 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) in packet_set_ring()
4378 if (unlikely(req->tp_frame_size < min_frame_size)) in packet_set_ring()
4380 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) in packet_set_ring()
4383 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; in packet_set_ring()
4384 if (unlikely(rb->frames_per_block == 0)) in packet_set_ring()
4386 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) in packet_set_ring()
4388 if (unlikely((rb->frames_per_block * req->tp_block_nr) != in packet_set_ring()
4389 req->tp_frame_nr)) in packet_set_ring()
4392 err = -ENOMEM; in packet_set_ring()
4393 order = get_order(req->tp_block_size); in packet_set_ring()
4397 switch (po->tp_version) { in packet_set_ring()
4403 struct tpacket_req3 *req3 = &req_u->req3; in packet_set_ring()
4405 if (req3->tp_retire_blk_tov || in packet_set_ring()
4406 req3->tp_sizeof_priv || in packet_set_ring()
4407 req3->tp_feature_req_word) { in packet_set_ring()
4408 err = -EINVAL; in packet_set_ring()
4415 rx_owner_map = bitmap_alloc(req->tp_frame_nr, in packet_set_ring()
4425 err = -EINVAL; in packet_set_ring()
4426 if (unlikely(req->tp_frame_nr)) in packet_set_ring()
4432 spin_lock(&po->bind_lock); in packet_set_ring()
4433 was_running = po->running; in packet_set_ring()
4434 num = po->num; in packet_set_ring()
4436 WRITE_ONCE(po->num, 0); in packet_set_ring()
4439 spin_unlock(&po->bind_lock); in packet_set_ring()
4443 err = -EBUSY; in packet_set_ring()
4444 mutex_lock(&po->pg_vec_lock); in packet_set_ring()
4445 if (closing || atomic_read(&po->mapped) == 0) { in packet_set_ring()
4447 spin_lock_bh(&rb_queue->lock); in packet_set_ring()
4448 swap(rb->pg_vec, pg_vec); in packet_set_ring()
4449 if (po->tp_version <= TPACKET_V2) in packet_set_ring()
4450 swap(rb->rx_owner_map, rx_owner_map); in packet_set_ring()
4451 rb->frame_max = (req->tp_frame_nr - 1); in packet_set_ring()
4452 rb->head = 0; in packet_set_ring()
4453 rb->frame_size = req->tp_frame_size; in packet_set_ring()
4454 spin_unlock_bh(&rb_queue->lock); in packet_set_ring()
4456 swap(rb->pg_vec_order, order); in packet_set_ring()
4457 swap(rb->pg_vec_len, req->tp_block_nr); in packet_set_ring()
4459 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; in packet_set_ring()
4460 po->prot_hook.func = (po->rx_ring.pg_vec) ? in packet_set_ring()
4463 if (atomic_read(&po->mapped)) in packet_set_ring()
4465 atomic_read(&po->mapped)); in packet_set_ring()
4467 mutex_unlock(&po->pg_vec_lock); in packet_set_ring()
4469 spin_lock(&po->bind_lock); in packet_set_ring()
4471 WRITE_ONCE(po->num, num); in packet_set_ring()
4474 spin_unlock(&po->bind_lock); in packet_set_ring()
4475 if (pg_vec && (po->tp_version > TPACKET_V2)) { in packet_set_ring()
4476 /* Because we don't support block-based V3 on tx-ring */ in packet_set_ring()
4484 free_pg_vec(pg_vec, order, req->tp_block_nr); in packet_set_ring()
4493 struct sock *sk = sock->sk; in packet_mmap()
4498 int err = -EINVAL; in packet_mmap()
4501 if (vma->vm_pgoff) in packet_mmap()
4502 return -EINVAL; in packet_mmap()
4504 mutex_lock(&po->pg_vec_lock); in packet_mmap()
4507 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { in packet_mmap()
4508 if (rb->pg_vec) { in packet_mmap()
4509 expected_size += rb->pg_vec_len in packet_mmap()
4510 * rb->pg_vec_pages in packet_mmap()
4518 size = vma->vm_end - vma->vm_start; in packet_mmap()
4522 start = vma->vm_start; in packet_mmap()
4523 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { in packet_mmap()
4524 if (rb->pg_vec == NULL) in packet_mmap()
4527 for (i = 0; i < rb->pg_vec_len; i++) { in packet_mmap()
4529 void *kaddr = rb->pg_vec[i].buffer; in packet_mmap()
4532 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { in packet_mmap()
4543 atomic_inc(&po->mapped); in packet_mmap()
4544 vma->vm_ops = &packet_mmap_ops; in packet_mmap()
4548 mutex_unlock(&po->pg_vec_lock); in packet_mmap()
4612 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); in packet_seq_start()
4618 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); in packet_seq_next()
4636 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", in packet_seq_show()
4638 refcount_read(&s->sk_refcnt), in packet_seq_show()
4639 s->sk_type, in packet_seq_show()
4640 ntohs(READ_ONCE(po->num)), in packet_seq_show()
4641 READ_ONCE(po->ifindex), in packet_seq_show()
4642 po->running, in packet_seq_show()
4643 atomic_read(&s->sk_rmem_alloc), in packet_seq_show()
4661 mutex_init(&net->packet.sklist_lock); in packet_net_init()
4662 INIT_HLIST_HEAD(&net->packet.sklist); in packet_net_init()
4665 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, in packet_net_init()
4667 return -ENOMEM; in packet_net_init()
4675 remove_proc_entry("packet", net->proc_net); in packet_net_exit()
4676 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); in packet_net_exit()