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.
101 - If the device has no dev->header_ops->create, there is no LL header
108 needed_headroom to be (the real WiFi header length - the fake Ethernet
110 - packet socket receives packets with pulled ll header,
114 -----------
117 mac_header -> ll header
118 data -> data
121 mac_header -> ll header
122 data -> ll header
125 mac_header -> data
128 data -> data
131 mac_header -> data. ll header is invisible to us.
132 data -> data
140 ------------
143 mac_header -> ll header
144 data -> ll header
147 mac_header -> data
148 data -> data
183 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
184 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
185 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
186 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
187 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
188 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
254 next = skb->next; in nf_hook_direct_egress()
257 if (!nf_hook_egress(skb, &rc, skb->dev)) in nf_hook_direct_egress()
263 tail->next = skb; in nf_hook_direct_egress()
293 dev = rcu_dereference(po->cached_dev); in packet_cached_dev_get()
303 rcu_assign_pointer(po->cached_dev, dev); in packet_cached_dev_assign()
308 RCU_INIT_POINTER(po->cached_dev, NULL); in packet_cached_dev_reset()
313 struct net_device *dev = skb->dev; in packet_pick_tx_queue()
314 const struct net_device_ops *ops = dev->netdev_ops; in packet_pick_tx_queue()
319 skb->sender_cpu = cpu + 1; in packet_pick_tx_queue()
321 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); in packet_pick_tx_queue()
322 if (ops->ndo_select_queue) { in packet_pick_tx_queue()
323 queue_index = ops->ndo_select_queue(dev, skb, NULL); in packet_pick_tx_queue()
341 if (po->fanout) in __register_prot_hook()
344 dev_add_pack(&po->prot_hook); in __register_prot_hook()
353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); in register_prot_hook()
358 * the po->bind_lock and do a synchronize_net to make sure no
360 * of po->prot_hook. If the sync parameter is false, it is the
367 lockdep_assert_held_once(&po->bind_lock); in __unregister_prot_hook()
371 if (po->fanout) in __unregister_prot_hook()
374 __dev_remove_pack(&po->prot_hook); in __unregister_prot_hook()
379 spin_unlock(&po->bind_lock); in __unregister_prot_hook()
381 spin_lock(&po->bind_lock); in __unregister_prot_hook()
407 switch (po->tp_version) { in __packet_set_status()
409 WRITE_ONCE(h.h1->tp_status, status); in __packet_set_status()
410 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); in __packet_set_status()
413 WRITE_ONCE(h.h2->tp_status, status); in __packet_set_status()
414 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); in __packet_set_status()
417 WRITE_ONCE(h.h3->tp_status, status); in __packet_set_status()
418 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); in __packet_set_status()
437 switch (po->tp_version) { in __packet_get_status()
439 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); in __packet_get_status()
440 return READ_ONCE(h.h1->tp_status); in __packet_get_status()
442 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); in __packet_get_status()
443 return READ_ONCE(h.h2->tp_status); in __packet_get_status()
445 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); in __packet_get_status()
446 return READ_ONCE(h.h3->tp_status); in __packet_get_status()
461 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) in tpacket_get_timestamp()
478 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp)))) in __packet_set_timestamp()
484 * all store the seconds in a 32-bit unsigned integer. in __packet_set_timestamp()
485 * If we create a version 4, that should have a 64-bit timestamp, in __packet_set_timestamp()
486 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit in __packet_set_timestamp()
489 switch (po->tp_version) { in __packet_set_timestamp()
491 h.h1->tp_sec = ts.tv_sec; in __packet_set_timestamp()
492 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; in __packet_set_timestamp()
495 h.h2->tp_sec = ts.tv_sec; in __packet_set_timestamp()
496 h.h2->tp_nsec = ts.tv_nsec; in __packet_set_timestamp()
499 h.h3->tp_sec = ts.tv_sec; in __packet_set_timestamp()
500 h.h3->tp_nsec = ts.tv_nsec; in __packet_set_timestamp()
508 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); in __packet_set_timestamp()
522 pg_vec_pos = position / rb->frames_per_block; in packet_lookup_frame()
523 frame_offset = position % rb->frames_per_block; in packet_lookup_frame()
525 h.raw = rb->pg_vec[pg_vec_pos].buffer + in packet_lookup_frame()
526 (frame_offset * rb->frame_size); in packet_lookup_frame()
538 return packet_lookup_frame(po, rb, rb->head, status); in packet_current_frame()
551 * header is at the hard_header_len offset in non-variable in vlan_get_tci()
556 if (dev->min_header_len == dev->hard_header_len) in vlan_get_tci()
557 header_len = dev->hard_header_len; in vlan_get_tci()
559 header_len = dev->min_header_len; in vlan_get_tci()
568 return ntohs(vh->h_vlan_TCI); in vlan_get_tci()
573 __be16 proto = skb->protocol; in vlan_get_protocol_dgram()
584 del_timer_sync(&pkc->retire_blk_timer); in prb_del_retire_blk_timer()
592 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_shutdown_retire_blk_timer()
594 spin_lock_bh(&rb_queue->lock); in prb_shutdown_retire_blk_timer()
595 pkc->delete_blk_timer = 1; in prb_shutdown_retire_blk_timer()
596 spin_unlock_bh(&rb_queue->lock); in prb_shutdown_retire_blk_timer()
605 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_setup_retire_blk_timer()
606 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, in prb_setup_retire_blk_timer()
608 pkc->retire_blk_timer.expires = jiffies; in prb_setup_retire_blk_timer()
620 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); in prb_calc_retire_blk_tmo()
651 p1->feature_req_word = req_u->req3.tp_feature_req_word; in prb_init_ft_ops()
664 p1->knxt_seq_num = 1; in init_prb_bdqc()
665 p1->pkbdq = pg_vec; in init_prb_bdqc()
667 p1->pkblk_start = pg_vec[0].buffer; in init_prb_bdqc()
668 p1->kblk_size = req_u->req3.tp_block_size; in init_prb_bdqc()
669 p1->knum_blocks = req_u->req3.tp_block_nr; in init_prb_bdqc()
670 p1->hdrlen = po->tp_hdrlen; in init_prb_bdqc()
671 p1->version = po->tp_version; in init_prb_bdqc()
672 p1->last_kactive_blk_num = 0; in init_prb_bdqc()
673 po->stats.stats3.tp_freeze_q_cnt = 0; in init_prb_bdqc()
674 if (req_u->req3.tp_retire_blk_tov) in init_prb_bdqc()
675 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; in init_prb_bdqc()
677 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, in init_prb_bdqc()
678 req_u->req3.tp_block_size); in init_prb_bdqc()
679 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); in init_prb_bdqc()
680 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; in init_prb_bdqc()
681 rwlock_init(&p1->blk_fill_in_prog_lock); in init_prb_bdqc()
683 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); in init_prb_bdqc()
694 mod_timer(&pkc->retire_blk_timer, in _prb_refresh_rx_retire_blk_timer()
695 jiffies + pkc->tov_in_jiffies); in _prb_refresh_rx_retire_blk_timer()
696 pkc->last_kactive_blk_num = pkc->kactive_blk_num; in _prb_refresh_rx_retire_blk_timer()
703 * on packet-by-packet basis.
705 * With a 1MB block-size, on a 1Gbps line, it will take
718 * a) line-speed and b) block-size.
726 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in prb_retire_rx_blk_timer_expired()
730 spin_lock(&po->sk.sk_receive_queue.lock); in prb_retire_rx_blk_timer_expired()
735 if (unlikely(pkc->delete_blk_timer)) in prb_retire_rx_blk_timer_expired()
749 write_lock(&pkc->blk_fill_in_prog_lock); in prb_retire_rx_blk_timer_expired()
750 write_unlock(&pkc->blk_fill_in_prog_lock); in prb_retire_rx_blk_timer_expired()
753 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { in prb_retire_rx_blk_timer_expired()
765 /* Case 1. Queue was frozen because user-space was in prb_retire_rx_blk_timer_expired()
770 * Ok, user-space is still behind. in prb_retire_rx_blk_timer_expired()
775 /* Case 2. queue was frozen,user-space caught up, in prb_retire_rx_blk_timer_expired()
780 * Thawing/timer-refresh is a side effect. in prb_retire_rx_blk_timer_expired()
792 spin_unlock(&po->sk.sk_receive_queue.lock); in prb_retire_rx_blk_timer_expired()
808 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); in prb_flush_block()
845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; in prb_close_block()
846 struct sock *sk = &po->sk; in prb_close_block()
848 if (atomic_read(&po->tp_drops)) in prb_close_block()
851 last_pkt = (struct tpacket3_hdr *)pkc1->prev; in prb_close_block()
852 last_pkt->tp_next_offset = 0; in prb_close_block()
856 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; in prb_close_block()
857 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; in prb_close_block()
859 /* Ok, we tmo'd - so get the current time. in prb_close_block()
866 h1->ts_last_pkt.ts_sec = ts.tv_sec; in prb_close_block()
867 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; in prb_close_block()
875 sk->sk_data_ready(sk); in prb_close_block()
877 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); in prb_close_block()
882 pkc->reset_pending_on_curr_blk = 0; in prb_thaw_queue()
896 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; in prb_open_block()
904 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; in prb_open_block()
906 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
910 h1->ts_first_pkt.ts_sec = ts.tv_sec; in prb_open_block()
911 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; in prb_open_block()
913 pkc1->pkblk_start = (char *)pbd1; in prb_open_block()
914 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
916 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); in prb_open_block()
919 pbd1->version = pkc1->version; in prb_open_block()
920 pkc1->prev = pkc1->nxt_offset; in prb_open_block()
921 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; in prb_open_block()
934 * 4) user-space is either sleeping or processing block '0'.
935 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
936 * it will close block-7,loop around and try to fill block '0'.
937 * call-flow:
941 * |->(BLOCK_STATUS == USER) evaluates to true
942 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
947 * re-open block-0 in near future.
949 * case and __packet_lookup_frame_in_block will check if block-0
950 * is free and can now be re-used.
955 pkc->reset_pending_on_curr_blk = 1; in prb_freeze_queue()
956 po->stats.stats3.tp_freeze_q_cnt++; in prb_freeze_queue()
989 return (void *)pkc->nxt_offset; in prb_dispatch_next_block()
1001 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't in prb_retire_current_block()
1006 * the timer-handler already handled this case. in prb_retire_current_block()
1010 write_lock(&pkc->blk_fill_in_prog_lock); in prb_retire_current_block()
1011 write_unlock(&pkc->blk_fill_in_prog_lock); in prb_retire_current_block()
1025 return pkc->reset_pending_on_curr_blk; in prb_queue_frozen()
1029 __releases(&pkc->blk_fill_in_prog_lock) in prb_clear_blk_fill_status()
1033 read_unlock(&pkc->blk_fill_in_prog_lock); in prb_clear_blk_fill_status()
1039 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); in prb_fill_rxhash()
1045 ppd->hv1.tp_rxhash = 0; in prb_clear_rxhash()
1053 if (skb_vlan_tag_present(pkc->skb)) { in prb_fill_vlan_info()
1054 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); in prb_fill_vlan_info()
1055 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); in prb_fill_vlan_info()
1056 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; in prb_fill_vlan_info()
1057 } else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) { in prb_fill_vlan_info()
1058 ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev); in prb_fill_vlan_info()
1059 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol); in prb_fill_vlan_info()
1060 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; in prb_fill_vlan_info()
1062 ppd->hv1.tp_vlan_tci = 0; in prb_fill_vlan_info()
1063 ppd->hv1.tp_vlan_tpid = 0; in prb_fill_vlan_info()
1064 ppd->tp_status = TP_STATUS_AVAILABLE; in prb_fill_vlan_info()
1071 ppd->hv1.tp_padding = 0; in prb_run_all_ft_ops()
1074 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) in prb_run_all_ft_ops()
1084 __acquires(&pkc->blk_fill_in_prog_lock) in prb_fill_curr_block()
1089 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); in prb_fill_curr_block()
1090 pkc->prev = curr; in prb_fill_curr_block()
1091 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); in prb_fill_curr_block()
1094 read_lock(&pkc->blk_fill_in_prog_lock); in prb_fill_curr_block()
1098 /* Assumes caller has the sk->rx_queue.lock */
1108 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); in __packet_lookup_frame_in_block()
1115 * is still in_use by user-space. in __packet_lookup_frame_in_block()
1122 * Ok, the block was released by user-space. in __packet_lookup_frame_in_block()
1132 curr = pkc->nxt_offset; in __packet_lookup_frame_in_block()
1133 pkc->skb = skb; in __packet_lookup_frame_in_block()
1134 end = (char *)pbd + pkc->kblk_size; in __packet_lookup_frame_in_block()
1154 * No free blocks are available.user_space hasn't caught up yet. in __packet_lookup_frame_in_block()
1165 switch (po->tp_version) { in packet_current_rx_frame()
1168 curr = packet_lookup_frame(po, &po->rx_ring, in packet_current_rx_frame()
1169 po->rx_ring.head, status); in packet_current_rx_frame()
1196 if (rb->prb_bdqc.kactive_blk_num) in prb_previous_blk_num()
1197 prev = rb->prb_bdqc.kactive_blk_num-1; in prb_previous_blk_num()
1199 prev = rb->prb_bdqc.knum_blocks-1; in prb_previous_blk_num()
1216 if (po->tp_version <= TPACKET_V2) in packet_previous_rx_frame()
1225 switch (po->tp_version) { in packet_increment_rx_head()
1241 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; in packet_previous_frame()
1247 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; in packet_increment_head()
1252 this_cpu_inc(*rb->pending_refcnt); in packet_inc_pending()
1257 this_cpu_dec(*rb->pending_refcnt); in packet_dec_pending()
1266 if (rb->pending_refcnt == NULL) in packet_read_pending()
1270 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); in packet_read_pending()
1277 po->rx_ring.pending_refcnt = NULL; in packet_alloc_pending()
1279 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); in packet_alloc_pending()
1280 if (unlikely(po->tx_ring.pending_refcnt == NULL)) in packet_alloc_pending()
1281 return -ENOBUFS; in packet_alloc_pending()
1288 free_percpu(po->tx_ring.pending_refcnt); in packet_free_pending()
1300 len = READ_ONCE(po->rx_ring.frame_max) + 1; in __tpacket_has_room()
1301 idx = READ_ONCE(po->rx_ring.head); in __tpacket_has_room()
1305 idx -= len; in __tpacket_has_room()
1306 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); in __tpacket_has_room()
1313 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); in __tpacket_v3_has_room()
1314 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); in __tpacket_v3_has_room()
1318 idx -= len; in __tpacket_v3_has_room()
1319 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); in __tpacket_v3_has_room()
1325 const struct sock *sk = &po->sk; in __packet_rcv_has_room()
1328 if (po->prot_hook.func != tpacket_rcv) { in __packet_rcv_has_room()
1329 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); in __packet_rcv_has_room()
1330 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) in __packet_rcv_has_room()
1331 - (skb ? skb->truesize : 0); in __packet_rcv_has_room()
1341 if (po->tp_version == TPACKET_V3) { in __packet_rcv_has_room()
1379 skb_queue_purge(&sk->sk_error_queue); in packet_sock_destruct()
1381 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); in packet_sock_destruct()
1382 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); in packet_sock_destruct()
1392 u32 *history = po->rollover->history; in fanout_flow_is_huge()
1421 unsigned int val = atomic_inc_return(&f->rr_cur); in fanout_demux_lb()
1448 po = pkt_sk(rcu_dereference(f->arr[idx])); in fanout_demux_rollover()
1458 i = j = min_t(int, po->rollover->sock, num - 1); in fanout_demux_rollover()
1460 po_next = pkt_sk(rcu_dereference(f->arr[i])); in fanout_demux_rollover()
1465 po->rollover->sock = i; in fanout_demux_rollover()
1466 atomic_long_inc(&po->rollover->num); in fanout_demux_rollover()
1468 atomic_long_inc(&po->rollover->num_huge); in fanout_demux_rollover()
1476 atomic_long_inc(&po->rollover->num_failed); in fanout_demux_rollover()
1495 prog = rcu_dereference(f->bpf_prog); in fanout_demux_bpf()
1505 return f->flags & (flag >> 8); in fanout_has_flag()
1511 struct packet_fanout *f = pt->af_packet_priv; in packet_rcv_fanout()
1512 unsigned int num = READ_ONCE(f->num_members); in packet_rcv_fanout()
1513 struct net *net = read_pnet(&f->net); in packet_rcv_fanout()
1527 switch (f->type) { in packet_rcv_fanout()
1556 po = pkt_sk(rcu_dereference(f->arr[idx])); in packet_rcv_fanout()
1557 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); in packet_rcv_fanout()
1567 struct packet_fanout *f = po->fanout; in __fanout_link()
1569 spin_lock(&f->lock); in __fanout_link()
1570 rcu_assign_pointer(f->arr[f->num_members], sk); in __fanout_link()
1572 f->num_members++; in __fanout_link()
1573 if (f->num_members == 1) in __fanout_link()
1574 dev_add_pack(&f->prot_hook); in __fanout_link()
1575 spin_unlock(&f->lock); in __fanout_link()
1580 struct packet_fanout *f = po->fanout; in __fanout_unlink()
1583 spin_lock(&f->lock); in __fanout_unlink()
1584 for (i = 0; i < f->num_members; i++) { in __fanout_unlink()
1585 if (rcu_dereference_protected(f->arr[i], in __fanout_unlink()
1586 lockdep_is_held(&f->lock)) == sk) in __fanout_unlink()
1589 BUG_ON(i >= f->num_members); in __fanout_unlink()
1590 rcu_assign_pointer(f->arr[i], in __fanout_unlink()
1591 rcu_dereference_protected(f->arr[f->num_members - 1], in __fanout_unlink()
1592 lockdep_is_held(&f->lock))); in __fanout_unlink()
1593 f->num_members--; in __fanout_unlink()
1594 if (f->num_members == 0) in __fanout_unlink()
1595 __dev_remove_pack(&f->prot_hook); in __fanout_unlink()
1596 spin_unlock(&f->lock); in __fanout_unlink()
1601 if (sk->sk_family != PF_PACKET) in match_fanout_group()
1604 return ptype->af_packet_priv == pkt_sk(sk)->fanout; in match_fanout_group()
1609 switch (f->type) { in fanout_init_data()
1611 atomic_set(&f->rr_cur, 0); in fanout_init_data()
1615 RCU_INIT_POINTER(f->bpf_prog, NULL); in fanout_init_data()
1624 spin_lock(&f->lock); in __fanout_set_data_bpf()
1625 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); in __fanout_set_data_bpf()
1626 rcu_assign_pointer(f->bpf_prog, new); in __fanout_set_data_bpf()
1627 spin_unlock(&f->lock); in __fanout_set_data_bpf()
1642 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) in fanout_set_data_cbpf()
1643 return -EPERM; in fanout_set_data_cbpf()
1653 __fanout_set_data_bpf(po->fanout, new); in fanout_set_data_cbpf()
1663 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) in fanout_set_data_ebpf()
1664 return -EPERM; in fanout_set_data_ebpf()
1666 return -EINVAL; in fanout_set_data_ebpf()
1668 return -EFAULT; in fanout_set_data_ebpf()
1674 __fanout_set_data_bpf(po->fanout, new); in fanout_set_data_ebpf()
1681 switch (po->fanout->type) { in fanout_set_data()
1687 return -EINVAL; in fanout_set_data()
1693 switch (f->type) { in fanout_release_data()
1705 if (f->id == candidate_id && in __fanout_id_is_free()
1706 read_pnet(&f->net) == sock_net(sk)) { in __fanout_id_is_free()
1732 struct packet_rollover *rollover = NULL; in fanout_add() local
1734 u16 type_flags = args->type_flags; in fanout_add()
1738 u16 id = args->id; in fanout_add()
1744 return -EINVAL; in fanout_add()
1755 return -EINVAL; in fanout_add()
1760 err = -EALREADY; in fanout_add()
1761 if (po->fanout) in fanout_add()
1766 err = -ENOMEM; in fanout_add()
1767 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); in fanout_add()
1768 if (!rollover) in fanout_add()
1770 atomic_long_set(&rollover->num, 0); in fanout_add()
1771 atomic_long_set(&rollover->num_huge, 0); in fanout_add()
1772 atomic_long_set(&rollover->num_failed, 0); in fanout_add()
1777 err = -EINVAL; in fanout_add()
1781 err = -ENOMEM; in fanout_add()
1790 if (f->id == id && in fanout_add()
1791 read_pnet(&f->net) == sock_net(sk)) { in fanout_add()
1796 err = -EINVAL; in fanout_add()
1798 if (match->flags != flags) in fanout_add()
1800 if (args->max_num_members && in fanout_add()
1801 args->max_num_members != match->max_num_members) in fanout_add()
1804 if (args->max_num_members > PACKET_FANOUT_MAX) in fanout_add()
1806 if (!args->max_num_members) in fanout_add()
1808 args->max_num_members = 256; in fanout_add()
1809 err = -ENOMEM; in fanout_add()
1810 match = kvzalloc(struct_size(match, arr, args->max_num_members), in fanout_add()
1814 write_pnet(&match->net, sock_net(sk)); in fanout_add()
1815 match->id = id; in fanout_add()
1816 match->type = type; in fanout_add()
1817 match->flags = flags; in fanout_add()
1818 INIT_LIST_HEAD(&match->list); in fanout_add()
1819 spin_lock_init(&match->lock); in fanout_add()
1820 refcount_set(&match->sk_ref, 0); in fanout_add()
1822 match->prot_hook.type = po->prot_hook.type; in fanout_add()
1823 match->prot_hook.dev = po->prot_hook.dev; in fanout_add()
1824 match->prot_hook.func = packet_rcv_fanout; in fanout_add()
1825 match->prot_hook.af_packet_priv = match; in fanout_add()
1826 match->prot_hook.af_packet_net = read_pnet(&match->net); in fanout_add()
1827 match->prot_hook.id_match = match_fanout_group; in fanout_add()
1828 match->max_num_members = args->max_num_members; in fanout_add()
1829 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING; in fanout_add()
1830 list_add(&match->list, &fanout_list); in fanout_add()
1832 err = -EINVAL; in fanout_add()
1834 spin_lock(&po->bind_lock); in fanout_add()
1836 match->type == type && in fanout_add()
1837 match->prot_hook.type == po->prot_hook.type && in fanout_add()
1838 match->prot_hook.dev == po->prot_hook.dev) { in fanout_add()
1839 err = -ENOSPC; in fanout_add()
1840 if (refcount_read(&match->sk_ref) < match->max_num_members) { in fanout_add()
1841 __dev_remove_pack(&po->prot_hook); in fanout_add()
1844 WRITE_ONCE(po->fanout, match); in fanout_add()
1846 po->rollover = rollover; in fanout_add()
1847 rollover = NULL; in fanout_add()
1848 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); in fanout_add()
1853 spin_unlock(&po->bind_lock); in fanout_add()
1855 if (err && !refcount_read(&match->sk_ref)) { in fanout_add()
1856 list_del(&match->list); in fanout_add()
1861 kfree(rollover); in fanout_add()
1866 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1867 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1877 f = po->fanout; in fanout_release()
1879 po->fanout = NULL; in fanout_release()
1881 if (refcount_dec_and_test(&f->sk_ref)) in fanout_release()
1882 list_del(&f->list); in fanout_release()
1894 /* Earlier code assumed this would be a VLAN pkt, double-check in packet_extra_vlan_len_allowed()
1898 if (unlikely(dev->type != ARPHRD_ETHER)) in packet_extra_vlan_len_allowed()
1902 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); in packet_extra_vlan_len_allowed()
1920 sk = pt->af_packet_priv; in packet_rcv_spkt()
1929 * For outgoing ones skb->data == skb_mac_header(skb) in packet_rcv_spkt()
1933 if (skb->pkt_type == PACKET_LOOPBACK) in packet_rcv_spkt()
1949 spkt = &PACKET_SKB_CB(skb)->sa.pkt; in packet_rcv_spkt()
1951 skb_push(skb, skb->data - skb_mac_header(skb)); in packet_rcv_spkt()
1957 spkt->spkt_family = dev->type; in packet_rcv_spkt()
1958 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); in packet_rcv_spkt()
1959 spkt->spkt_protocol = skb->protocol; in packet_rcv_spkt()
1979 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && in packet_parse_headers()
1980 sock->type == SOCK_RAW) { in packet_parse_headers()
1982 skb->protocol = dev_parse_header_protocol(skb); in packet_parse_headers()
1986 if (likely(skb->dev->type == ARPHRD_ETHER) && in packet_parse_headers()
1987 eth_type_vlan(skb->protocol) && in packet_parse_headers()
1988 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) in packet_parse_headers()
2002 struct sock *sk = sock->sk; in packet_sendmsg_spkt()
2003 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); in packet_sendmsg_spkt()
2016 if (msg->msg_namelen < sizeof(struct sockaddr)) in packet_sendmsg_spkt()
2017 return -EINVAL; in packet_sendmsg_spkt()
2018 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) in packet_sendmsg_spkt()
2019 proto = saddr->spkt_protocol; in packet_sendmsg_spkt()
2021 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ in packet_sendmsg_spkt()
2027 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; in packet_sendmsg_spkt()
2030 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); in packet_sendmsg_spkt()
2031 err = -ENODEV; in packet_sendmsg_spkt()
2035 err = -ENETDOWN; in packet_sendmsg_spkt()
2036 if (!(dev->flags & IFF_UP)) in packet_sendmsg_spkt()
2046 err = -EPROTONOSUPPORT; in packet_sendmsg_spkt()
2052 err = -EMSGSIZE; in packet_sendmsg_spkt()
2053 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) in packet_sendmsg_spkt()
2058 int tlen = dev->needed_tailroom; in packet_sendmsg_spkt()
2059 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; in packet_sendmsg_spkt()
2064 return -ENOBUFS; in packet_sendmsg_spkt()
2074 skb->data -= hhlen; in packet_sendmsg_spkt()
2075 skb->tail -= hhlen; in packet_sendmsg_spkt()
2085 if (!dev_validate_header(dev, skb->data, len) || !skb->len) { in packet_sendmsg_spkt()
2086 err = -EINVAL; in packet_sendmsg_spkt()
2089 if (len > (dev->mtu + dev->hard_header_len + extra_len) && in packet_sendmsg_spkt()
2091 err = -EMSGSIZE; in packet_sendmsg_spkt()
2096 if (msg->msg_controllen) { in packet_sendmsg_spkt()
2102 skb->protocol = proto; in packet_sendmsg_spkt()
2103 skb->dev = dev; in packet_sendmsg_spkt()
2104 skb->priority = READ_ONCE(sk->sk_priority); in packet_sendmsg_spkt()
2105 skb->mark = READ_ONCE(sk->sk_mark); in packet_sendmsg_spkt()
2106 skb->tstamp = sockc.transmit_time; in packet_sendmsg_spkt()
2111 skb->no_fcs = 1; in packet_sendmsg_spkt()
2133 filter = rcu_dereference(sk->sk_filter); in run_filter()
2135 res = bpf_prog_run_clear_cb(filter->prog, skb); in run_filter()
2147 return -EINVAL; in packet_rcv_vnet()
2148 *len -= vnet_hdr_sz; in packet_rcv_vnet()
2151 return -EINVAL; in packet_rcv_vnet()
2160 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2161 * and skb->cb are mangled. It works because (and until) packets
2174 u8 *skb_head = skb->data; in packet_rcv()
2175 int skb_len = skb->len; in packet_rcv()
2179 if (skb->pkt_type == PACKET_LOOPBACK) in packet_rcv()
2182 sk = pt->af_packet_priv; in packet_rcv()
2188 skb->dev = dev; in packet_rcv()
2198 if (sk->sk_type != SOCK_DGRAM) in packet_rcv()
2199 skb_push(skb, skb->data - skb_mac_header(skb)); in packet_rcv()
2200 else if (skb->pkt_type == PACKET_OUTGOING) { in packet_rcv()
2206 snaplen = skb->len; in packet_rcv()
2214 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) in packet_rcv()
2222 if (skb_head != skb->data) { in packet_rcv()
2223 skb->data = skb_head; in packet_rcv()
2224 skb->len = skb_len; in packet_rcv()
2230 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); in packet_rcv()
2232 sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_rcv()
2233 sll->sll_hatype = dev->type; in packet_rcv()
2234 sll->sll_pkttype = skb->pkt_type; in packet_rcv()
2236 sll->sll_ifindex = orig_dev->ifindex; in packet_rcv()
2238 sll->sll_ifindex = dev->ifindex; in packet_rcv()
2240 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); in packet_rcv()
2242 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). in packet_rcv()
2245 PACKET_SKB_CB(skb)->sa.origlen = skb->len; in packet_rcv()
2251 skb->dev = NULL; in packet_rcv()
2257 spin_lock(&sk->sk_receive_queue.lock); in packet_rcv()
2258 po->stats.stats1.tp_packets++; in packet_rcv()
2261 __skb_queue_tail(&sk->sk_receive_queue, skb); in packet_rcv()
2262 spin_unlock(&sk->sk_receive_queue.lock); in packet_rcv()
2263 sk->sk_data_ready(sk); in packet_rcv()
2268 atomic_inc(&po->tp_drops); in packet_rcv()
2269 atomic_inc(&sk->sk_drops); in packet_rcv()
2272 if (skb_head != skb->data && skb_shared(skb)) { in packet_rcv()
2273 skb->data = skb_head; in packet_rcv()
2274 skb->len = skb_len; in packet_rcv()
2291 u8 *skb_head = skb->data; in tpacket_rcv()
2292 int skb_len = skb->len; in tpacket_rcv()
2311 if (skb->pkt_type == PACKET_LOOPBACK) in tpacket_rcv()
2314 sk = pt->af_packet_priv; in tpacket_rcv()
2321 if (sk->sk_type != SOCK_DGRAM) in tpacket_rcv()
2322 skb_push(skb, skb->data - skb_mac_header(skb)); in tpacket_rcv()
2323 else if (skb->pkt_type == PACKET_OUTGOING) { in tpacket_rcv()
2329 snaplen = skb->len; in tpacket_rcv()
2337 atomic_inc(&po->tp_drops); in tpacket_rcv()
2341 if (skb->ip_summed == CHECKSUM_PARTIAL) in tpacket_rcv()
2343 else if (skb->pkt_type != PACKET_OUTGOING && in tpacket_rcv()
2352 if (sk->sk_type == SOCK_DGRAM) { in tpacket_rcv()
2353 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + in tpacket_rcv()
2354 po->tp_reserve; in tpacket_rcv()
2357 netoff = TPACKET_ALIGN(po->tp_hdrlen + in tpacket_rcv()
2359 po->tp_reserve; in tpacket_rcv()
2360 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); in tpacket_rcv()
2363 macoff = netoff - maclen; in tpacket_rcv()
2366 atomic_inc(&po->tp_drops); in tpacket_rcv()
2369 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2370 if (macoff + snaplen > po->rx_ring.frame_size) { in tpacket_rcv()
2371 if (po->copy_thresh && in tpacket_rcv()
2372 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { in tpacket_rcv()
2377 skb_head = skb->data; in tpacket_rcv()
2380 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0, in tpacket_rcv()
2381 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll)); in tpacket_rcv()
2385 snaplen = po->rx_ring.frame_size - macoff; in tpacket_rcv()
2392 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { in tpacket_rcv()
2395 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; in tpacket_rcv()
2401 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; in tpacket_rcv()
2405 spin_lock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2411 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2412 slot_id = po->rx_ring.head; in tpacket_rcv()
2413 if (test_bit(slot_id, po->rx_ring.rx_owner_map)) in tpacket_rcv()
2415 __set_bit(slot_id, po->rx_ring.rx_owner_map); in tpacket_rcv()
2419 virtio_net_hdr_from_skb(skb, h.raw + macoff - in tpacket_rcv()
2422 if (po->tp_version == TPACKET_V3) in tpacket_rcv()
2423 prb_clear_blk_fill_status(&po->rx_ring); in tpacket_rcv()
2427 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2428 packet_increment_rx_head(po, &po->rx_ring); in tpacket_rcv()
2430 * LOSING will be reported till you read the stats, in tpacket_rcv()
2431 * because it's COR - Clear On Read. in tpacket_rcv()
2435 if (atomic_read(&po->tp_drops)) in tpacket_rcv()
2439 po->stats.stats1.tp_packets++; in tpacket_rcv()
2443 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); in tpacket_rcv()
2445 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2453 READ_ONCE(po->tp_tstamp) | in tpacket_rcv()
2460 switch (po->tp_version) { in tpacket_rcv()
2462 h.h1->tp_len = skb->len; in tpacket_rcv()
2463 h.h1->tp_snaplen = snaplen; in tpacket_rcv()
2464 h.h1->tp_mac = macoff; in tpacket_rcv()
2465 h.h1->tp_net = netoff; in tpacket_rcv()
2466 h.h1->tp_sec = ts.tv_sec; in tpacket_rcv()
2467 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; in tpacket_rcv()
2471 h.h2->tp_len = skb->len; in tpacket_rcv()
2472 h.h2->tp_snaplen = snaplen; in tpacket_rcv()
2473 h.h2->tp_mac = macoff; in tpacket_rcv()
2474 h.h2->tp_net = netoff; in tpacket_rcv()
2475 h.h2->tp_sec = ts.tv_sec; in tpacket_rcv()
2476 h.h2->tp_nsec = ts.tv_nsec; in tpacket_rcv()
2478 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); in tpacket_rcv()
2479 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); in tpacket_rcv()
2481 } else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) { in tpacket_rcv()
2482 h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev); in tpacket_rcv()
2483 h.h2->tp_vlan_tpid = ntohs(skb->protocol); in tpacket_rcv()
2486 h.h2->tp_vlan_tci = 0; in tpacket_rcv()
2487 h.h2->tp_vlan_tpid = 0; in tpacket_rcv()
2489 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); in tpacket_rcv()
2496 h.h3->tp_status |= status; in tpacket_rcv()
2497 h.h3->tp_len = skb->len; in tpacket_rcv()
2498 h.h3->tp_snaplen = snaplen; in tpacket_rcv()
2499 h.h3->tp_mac = macoff; in tpacket_rcv()
2500 h.h3->tp_net = netoff; in tpacket_rcv()
2501 h.h3->tp_sec = ts.tv_sec; in tpacket_rcv()
2502 h.h3->tp_nsec = ts.tv_nsec; in tpacket_rcv()
2503 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); in tpacket_rcv()
2511 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); in tpacket_rcv()
2512 sll->sll_family = AF_PACKET; in tpacket_rcv()
2513 sll->sll_hatype = dev->type; in tpacket_rcv()
2514 sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ? in tpacket_rcv()
2515 vlan_get_protocol_dgram(skb) : skb->protocol; in tpacket_rcv()
2516 sll->sll_pkttype = skb->pkt_type; in tpacket_rcv()
2518 sll->sll_ifindex = orig_dev->ifindex; in tpacket_rcv()
2520 sll->sll_ifindex = dev->ifindex; in tpacket_rcv()
2525 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2537 if (po->tp_version <= TPACKET_V2) { in tpacket_rcv()
2538 spin_lock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2540 __clear_bit(slot_id, po->rx_ring.rx_owner_map); in tpacket_rcv()
2541 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2542 sk->sk_data_ready(sk); in tpacket_rcv()
2543 } else if (po->tp_version == TPACKET_V3) { in tpacket_rcv()
2544 prb_clear_blk_fill_status(&po->rx_ring); in tpacket_rcv()
2548 if (skb_head != skb->data && skb_shared(skb)) { in tpacket_rcv()
2549 skb->data = skb_head; in tpacket_rcv()
2550 skb->len = skb_len; in tpacket_rcv()
2560 spin_unlock(&sk->sk_receive_queue.lock); in tpacket_rcv()
2561 atomic_inc(&po->tp_drops); in tpacket_rcv()
2564 sk->sk_data_ready(sk); in tpacket_rcv()
2571 struct packet_sock *po = pkt_sk(skb->sk); in tpacket_destruct_skb()
2573 if (likely(po->tx_ring.pg_vec)) { in tpacket_destruct_skb()
2578 packet_dec_pending(&po->tx_ring); in tpacket_destruct_skb()
2583 complete(&po->skb_completion); in tpacket_destruct_skb()
2591 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && in __packet_snd_vnet_parse()
2592 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + in __packet_snd_vnet_parse()
2593 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > in __packet_snd_vnet_parse()
2594 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) in __packet_snd_vnet_parse()
2595 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), in __packet_snd_vnet_parse()
2596 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + in __packet_snd_vnet_parse()
2597 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); in __packet_snd_vnet_parse()
2599 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) in __packet_snd_vnet_parse()
2600 return -EINVAL; in __packet_snd_vnet_parse()
2611 return -EINVAL; in packet_snd_vnet_parse()
2612 *len -= vnet_hdr_sz; in packet_snd_vnet_parse()
2614 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) in packet_snd_vnet_parse()
2615 return -EFAULT; in packet_snd_vnet_parse()
2623 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr)); in packet_snd_vnet_parse()
2635 struct socket *sock = po->sk.sk_socket; in tpacket_fill_skb()
2641 skb->protocol = proto; in tpacket_fill_skb()
2642 skb->dev = dev; in tpacket_fill_skb()
2643 skb->priority = READ_ONCE(po->sk.sk_priority); in tpacket_fill_skb()
2644 skb->mark = READ_ONCE(po->sk.sk_mark); in tpacket_fill_skb()
2645 skb->tstamp = sockc->transmit_time; in tpacket_fill_skb()
2646 skb_setup_tx_timestamp(skb, sockc->tsflags); in tpacket_fill_skb()
2654 if (sock->type == SOCK_DGRAM) { in tpacket_fill_skb()
2658 return -EINVAL; in tpacket_fill_skb()
2662 skb_push(skb, dev->hard_header_len); in tpacket_fill_skb()
2663 skb_put(skb, copylen - dev->hard_header_len); in tpacket_fill_skb()
2667 if (!dev_validate_header(dev, skb->data, hdrlen)) in tpacket_fill_skb()
2668 return -EINVAL; in tpacket_fill_skb()
2671 to_write -= hdrlen; in tpacket_fill_skb()
2675 len_max = PAGE_SIZE - offset; in tpacket_fill_skb()
2678 skb->data_len = to_write; in tpacket_fill_skb()
2679 skb->len += to_write; in tpacket_fill_skb()
2680 skb->truesize += to_write; in tpacket_fill_skb()
2681 refcount_add(to_write, &po->sk.sk_wmem_alloc); in tpacket_fill_skb()
2684 nr_frags = skb_shinfo(skb)->nr_frags; in tpacket_fill_skb()
2689 return -EFAULT; in tpacket_fill_skb()
2697 to_write -= len; in tpacket_fill_skb()
2716 switch (po->tp_version) { in tpacket_parse_header()
2718 if (ph.h3->tp_next_offset != 0) { in tpacket_parse_header()
2720 return -EINVAL; in tpacket_parse_header()
2722 tp_len = ph.h3->tp_len; in tpacket_parse_header()
2725 tp_len = ph.h2->tp_len; in tpacket_parse_header()
2728 tp_len = ph.h1->tp_len; in tpacket_parse_header()
2733 return -EMSGSIZE; in tpacket_parse_header()
2739 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); in tpacket_parse_header()
2740 off_max = po->tx_ring.frame_size - tp_len; in tpacket_parse_header()
2741 if (po->sk.sk_type == SOCK_DGRAM) { in tpacket_parse_header()
2742 switch (po->tp_version) { in tpacket_parse_header()
2744 off = ph.h3->tp_net; in tpacket_parse_header()
2747 off = ph.h2->tp_net; in tpacket_parse_header()
2750 off = ph.h1->tp_net; in tpacket_parse_header()
2754 switch (po->tp_version) { in tpacket_parse_header()
2756 off = ph.h3->tp_mac; in tpacket_parse_header()
2759 off = ph.h2->tp_mac; in tpacket_parse_header()
2762 off = ph.h1->tp_mac; in tpacket_parse_header()
2767 return -EINVAL; in tpacket_parse_header()
2769 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); in tpacket_parse_header()
2785 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); in tpacket_snd()
2786 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); in tpacket_snd()
2787 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); in tpacket_snd()
2796 mutex_lock(&po->pg_vec_lock); in tpacket_snd()
2801 if (unlikely(!po->tx_ring.pg_vec)) { in tpacket_snd()
2802 err = -EBUSY; in tpacket_snd()
2807 proto = READ_ONCE(po->num); in tpacket_snd()
2809 err = -EINVAL; in tpacket_snd()
2810 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) in tpacket_snd()
2812 if (msg->msg_namelen < (saddr->sll_halen in tpacket_snd()
2816 proto = saddr->sll_protocol; in tpacket_snd()
2817 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); in tpacket_snd()
2818 if (po->sk.sk_socket->type == SOCK_DGRAM) { in tpacket_snd()
2819 if (dev && msg->msg_namelen < dev->addr_len + in tpacket_snd()
2822 addr = saddr->sll_addr; in tpacket_snd()
2826 err = -ENXIO; in tpacket_snd()
2829 err = -ENETDOWN; in tpacket_snd()
2830 if (unlikely(!(dev->flags & IFF_UP))) in tpacket_snd()
2833 sockcm_init(&sockc, &po->sk); in tpacket_snd()
2834 if (msg->msg_controllen) { in tpacket_snd()
2835 err = sock_cmsg_send(&po->sk, msg, &sockc); in tpacket_snd()
2840 if (po->sk.sk_socket->type == SOCK_RAW) in tpacket_snd()
2841 reserve = dev->hard_header_len; in tpacket_snd()
2842 size_max = po->tx_ring.frame_size in tpacket_snd()
2843 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); in tpacket_snd()
2845 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz) in tpacket_snd()
2846 size_max = dev->mtu + reserve + VLAN_HLEN; in tpacket_snd()
2848 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); in tpacket_snd()
2849 reinit_completion(&po->skb_completion); in tpacket_snd()
2852 ph = packet_current_frame(po, &po->tx_ring, in tpacket_snd()
2857 * fast-path we don't have to call it, only when ph in tpacket_snd()
2860 if (need_wait && packet_read_pending(&po->tx_ring)) { in tpacket_snd()
2861 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); in tpacket_snd()
2863 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; in tpacket_snd()
2879 tlen = dev->needed_tailroom; in tpacket_snd()
2883 tp_len -= vnet_hdr_sz; in tpacket_snd()
2886 tp_len = -EINVAL; in tpacket_snd()
2890 vnet_hdr->hdr_len); in tpacket_snd()
2892 copylen = max_t(int, copylen, dev->hard_header_len); in tpacket_snd()
2893 skb = sock_alloc_send_skb(&po->sk, in tpacket_snd()
2895 (copylen - dev->hard_header_len), in tpacket_snd()
2907 tp_len > dev->mtu + reserve && in tpacket_snd()
2910 tp_len = -EMSGSIZE; in tpacket_snd()
2917 packet_increment_head(&po->tx_ring); in tpacket_snd()
2929 tp_len = -EINVAL; in tpacket_snd()
2935 skb->destructor = tpacket_destruct_skb; in tpacket_snd()
2937 packet_inc_pending(&po->tx_ring); in tpacket_snd()
2956 packet_increment_head(&po->tx_ring); in tpacket_snd()
2969 mutex_unlock(&po->pg_vec_lock); in tpacket_snd()
2984 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) in packet_alloc_skb()
2985 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER); in packet_alloc_skb()
2986 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, in packet_alloc_skb()
2993 skb->data_len = len - linear; in packet_alloc_skb()
2994 skb->len += len - linear; in packet_alloc_skb()
3001 struct sock *sk = sock->sk; in packet_snd()
3002 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); in packet_snd()
3012 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); in packet_snd()
3022 proto = READ_ONCE(po->num); in packet_snd()
3024 err = -EINVAL; in packet_snd()
3025 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) in packet_snd()
3027 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) in packet_snd()
3029 proto = saddr->sll_protocol; in packet_snd()
3030 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); in packet_snd()
3031 if (sock->type == SOCK_DGRAM) { in packet_snd()
3032 if (dev && msg->msg_namelen < dev->addr_len + in packet_snd()
3035 addr = saddr->sll_addr; in packet_snd()
3039 err = -ENXIO; in packet_snd()
3042 err = -ENETDOWN; in packet_snd()
3043 if (unlikely(!(dev->flags & IFF_UP))) in packet_snd()
3047 sockc.mark = READ_ONCE(sk->sk_mark); in packet_snd()
3048 if (msg->msg_controllen) { in packet_snd()
3054 if (sock->type == SOCK_RAW) in packet_snd()
3055 reserve = dev->hard_header_len; in packet_snd()
3064 err = -EPROTONOSUPPORT; in packet_snd()
3070 err = -EMSGSIZE; in packet_snd()
3072 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) in packet_snd()
3075 err = -ENOBUFS; in packet_snd()
3077 tlen = dev->needed_tailroom; in packet_snd()
3079 linear = max(linear, min_t(int, len, dev->hard_header_len)); in packet_snd()
3081 msg->msg_flags & MSG_DONTWAIT, &err); in packet_snd()
3087 err = -EINVAL; in packet_snd()
3088 if (sock->type == SOCK_DGRAM) { in packet_snd()
3093 skb_reserve(skb, -reserve); in packet_snd()
3095 dev->min_header_len != dev->hard_header_len) in packet_snd()
3099 /* Returns -EFAULT on error */ in packet_snd()
3100 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); in packet_snd()
3104 if ((sock->type == SOCK_RAW && in packet_snd()
3105 !dev_validate_header(dev, skb->data, len)) || !skb->len) { in packet_snd()
3106 err = -EINVAL; in packet_snd()
3112 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && in packet_snd()
3114 err = -EMSGSIZE; in packet_snd()
3118 skb->protocol = proto; in packet_snd()
3119 skb->dev = dev; in packet_snd()
3120 skb->priority = READ_ONCE(sk->sk_priority); in packet_snd()
3121 skb->mark = sockc.mark; in packet_snd()
3122 skb->tstamp = sockc.transmit_time; in packet_snd()
3125 skb->no_fcs = 1; in packet_snd()
3160 struct sock *sk = sock->sk; in packet_sendmsg()
3166 if (data_race(po->tx_ring.pg_vec)) in packet_sendmsg()
3179 struct sock *sk = sock->sk; in packet_release()
3191 mutex_lock(&net->packet.sklist_lock); in packet_release()
3193 mutex_unlock(&net->packet.sklist_lock); in packet_release()
3195 sock_prot_inuse_add(net, sk->sk_prot, -1); in packet_release()
3197 spin_lock(&po->bind_lock); in packet_release()
3201 if (po->prot_hook.dev) { in packet_release()
3202 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); in packet_release()
3203 po->prot_hook.dev = NULL; in packet_release()
3205 spin_unlock(&po->bind_lock); in packet_release()
3210 if (po->rx_ring.pg_vec) { in packet_release()
3215 if (po->tx_ring.pg_vec) { in packet_release()
3225 kfree(po->rollover); in packet_release()
3231 * Now the socket is dead. No more input will appear. in packet_release()
3234 sock->sk = NULL; in packet_release()
3238 skb_queue_purge(&sk->sk_receive_queue); in packet_release()
3259 spin_lock(&po->bind_lock); in packet_do_bind()
3261 proto = po->num; in packet_do_bind()
3265 if (po->fanout) { in packet_do_bind()
3266 ret = -EINVAL; in packet_do_bind()
3273 ret = -ENODEV; in packet_do_bind()
3279 ret = -ENODEV; in packet_do_bind()
3284 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev; in packet_do_bind()
3293 WRITE_ONCE(po->num, 0); in packet_do_bind()
3298 dev->ifindex); in packet_do_bind()
3302 WRITE_ONCE(po->num, proto); in packet_do_bind()
3303 po->prot_hook.type = proto; in packet_do_bind()
3305 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); in packet_do_bind()
3308 po->prot_hook.dev = NULL; in packet_do_bind()
3309 WRITE_ONCE(po->ifindex, -1); in packet_do_bind()
3312 netdev_hold(dev, &po->prot_hook.dev_tracker, in packet_do_bind()
3314 po->prot_hook.dev = dev; in packet_do_bind()
3315 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); in packet_do_bind()
3324 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { in packet_do_bind()
3327 sk->sk_err = ENETDOWN; in packet_do_bind()
3334 spin_unlock(&po->bind_lock); in packet_do_bind()
3346 struct sock *sk = sock->sk; in packet_bind_spkt()
3347 char name[sizeof(uaddr->sa_data_min) + 1]; in packet_bind_spkt()
3354 return -EINVAL; in packet_bind_spkt()
3355 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be in packet_bind_spkt()
3356 * zero-terminated. in packet_bind_spkt()
3358 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min)); in packet_bind_spkt()
3359 name[sizeof(uaddr->sa_data_min)] = 0; in packet_bind_spkt()
3367 struct sock *sk = sock->sk; in packet_bind()
3374 return -EINVAL; in packet_bind()
3375 if (sll->sll_family != AF_PACKET) in packet_bind()
3376 return -EINVAL; in packet_bind()
3378 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol); in packet_bind()
3399 if (!ns_capable(net->user_ns, CAP_NET_RAW)) in packet_create()
3400 return -EPERM; in packet_create()
3401 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && in packet_create()
3402 sock->type != SOCK_PACKET) in packet_create()
3403 return -ESOCKTNOSUPPORT; in packet_create()
3405 sock->state = SS_UNCONNECTED; in packet_create()
3407 err = -ENOBUFS; in packet_create()
3412 sock->ops = &packet_ops; in packet_create()
3413 if (sock->type == SOCK_PACKET) in packet_create()
3414 sock->ops = &packet_ops_spkt; in packet_create()
3423 init_completion(&po->skb_completion); in packet_create()
3424 sk->sk_family = PF_PACKET; in packet_create()
3425 po->num = proto; in packet_create()
3429 sk->sk_destruct = packet_sock_destruct; in packet_create()
3435 spin_lock_init(&po->bind_lock); in packet_create()
3436 mutex_init(&po->pg_vec_lock); in packet_create()
3437 po->rollover = NULL; in packet_create()
3438 po->prot_hook.func = packet_rcv; in packet_create()
3440 if (sock->type == SOCK_PACKET) in packet_create()
3441 po->prot_hook.func = packet_rcv_spkt; in packet_create()
3443 po->prot_hook.af_packet_priv = sk; in packet_create()
3444 po->prot_hook.af_packet_net = sock_net(sk); in packet_create()
3447 po->prot_hook.type = proto; in packet_create()
3451 mutex_lock(&net->packet.sklist_lock); in packet_create()
3452 sk_add_node_tail_rcu(sk, &net->packet.sklist); in packet_create()
3453 mutex_unlock(&net->packet.sklist_lock); in packet_create()
3472 struct sock *sk = sock->sk; in packet_recvmsg()
3475 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz); in packet_recvmsg()
3478 err = -EINVAL; in packet_recvmsg()
3484 if (pkt_sk(sk)->ifindex < 0) in packet_recvmsg()
3485 return -ENODEV; in packet_recvmsg()
3496 * of horrible races and re-entrancy so we can forget about it in packet_recvmsg()
3526 copied = skb->len; in packet_recvmsg()
3529 msg->msg_flags |= MSG_TRUNC; in packet_recvmsg()
3536 if (sock->type != SOCK_PACKET) { in packet_recvmsg()
3537 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3540 origlen = PACKET_SKB_CB(skb)->sa.origlen; in packet_recvmsg()
3541 sll->sll_family = AF_PACKET; in packet_recvmsg()
3542 sll->sll_protocol = (sock->type == SOCK_DGRAM) ? in packet_recvmsg()
3543 vlan_get_protocol_dgram(skb) : skb->protocol; in packet_recvmsg()
3548 if (msg->msg_name) { in packet_recvmsg()
3549 const size_t max_len = min(sizeof(skb->cb), in packet_recvmsg()
3556 if (sock->type == SOCK_PACKET) { in packet_recvmsg()
3558 msg->msg_namelen = sizeof(struct sockaddr_pkt); in packet_recvmsg()
3559 copy_len = msg->msg_namelen; in packet_recvmsg()
3561 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3563 msg->msg_namelen = sll->sll_halen + in packet_recvmsg()
3565 copy_len = msg->msg_namelen; in packet_recvmsg()
3566 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { in packet_recvmsg()
3567 memset(msg->msg_name + in packet_recvmsg()
3569 0, sizeof(sll->sll_addr)); in packet_recvmsg()
3570 msg->msg_namelen = sizeof(struct sockaddr_ll); in packet_recvmsg()
3575 msg->msg_namelen = copy_len; in packet_recvmsg()
3577 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); in packet_recvmsg()
3584 if (skb->ip_summed == CHECKSUM_PARTIAL) in packet_recvmsg()
3586 else if (skb->pkt_type != PACKET_OUTGOING && in packet_recvmsg()
3593 aux.tp_snaplen = skb->len; in packet_recvmsg()
3598 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); in packet_recvmsg()
3600 } else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) { in packet_recvmsg()
3601 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; in packet_recvmsg()
3605 dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex); in packet_recvmsg()
3608 aux.tp_vlan_tpid = ntohs(skb->protocol); in packet_recvmsg()
3624 * hides all the races and re-entrancy issues from us. in packet_recvmsg()
3626 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); in packet_recvmsg()
3638 struct sock *sk = sock->sk; in packet_getname_spkt()
3641 return -EOPNOTSUPP; in packet_getname_spkt()
3643 uaddr->sa_family = AF_PACKET; in packet_getname_spkt()
3644 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min)); in packet_getname_spkt()
3646 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); in packet_getname_spkt()
3648 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min)); in packet_getname_spkt()
3658 struct sock *sk = sock->sk; in packet_getname()
3664 return -EOPNOTSUPP; in packet_getname()
3666 ifindex = READ_ONCE(po->ifindex); in packet_getname()
3667 sll->sll_family = AF_PACKET; in packet_getname()
3668 sll->sll_ifindex = ifindex; in packet_getname()
3669 sll->sll_protocol = READ_ONCE(po->num); in packet_getname()
3670 sll->sll_pkttype = 0; in packet_getname()
3674 sll->sll_hatype = dev->type; in packet_getname()
3675 sll->sll_halen = dev->addr_len; in packet_getname()
3678 memcpy(((struct sockaddr_storage *)sll)->__data + in packet_getname()
3679 offsetof(struct sockaddr_ll, sll_addr) - in packet_getname()
3681 dev->dev_addr, dev->addr_len); in packet_getname()
3683 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ in packet_getname()
3684 sll->sll_halen = 0; in packet_getname()
3688 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; in packet_getname()
3694 switch (i->type) { in packet_dev_mc()
3696 if (i->alen != dev->addr_len) in packet_dev_mc()
3697 return -EINVAL; in packet_dev_mc()
3699 return dev_mc_add(dev, i->addr); in packet_dev_mc()
3701 return dev_mc_del(dev, i->addr); in packet_dev_mc()
3708 if (i->alen != dev->addr_len) in packet_dev_mc()
3709 return -EINVAL; in packet_dev_mc()
3711 return dev_uc_add(dev, i->addr); in packet_dev_mc()
3713 return dev_uc_del(dev, i->addr); in packet_dev_mc()
3727 if (ml->ifindex == dev->ifindex) { in packet_dev_mclist_delete()
3728 packet_dev_mc(dev, ml, -1); in packet_dev_mclist_delete()
3729 *mlp = ml->next; in packet_dev_mclist_delete()
3732 mlp = &ml->next; in packet_dev_mclist_delete()
3745 err = -ENODEV; in packet_mc_add()
3746 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); in packet_mc_add()
3750 err = -EINVAL; in packet_mc_add()
3751 if (mreq->mr_alen > dev->addr_len) in packet_mc_add()
3754 err = -ENOBUFS; in packet_mc_add()
3760 for (ml = po->mclist; ml; ml = ml->next) { in packet_mc_add()
3761 if (ml->ifindex == mreq->mr_ifindex && in packet_mc_add()
3762 ml->type == mreq->mr_type && in packet_mc_add()
3763 ml->alen == mreq->mr_alen && in packet_mc_add()
3764 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { in packet_mc_add()
3765 ml->count++; in packet_mc_add()
3772 i->type = mreq->mr_type; in packet_mc_add()
3773 i->ifindex = mreq->mr_ifindex; in packet_mc_add()
3774 i->alen = mreq->mr_alen; in packet_mc_add()
3775 memcpy(i->addr, mreq->mr_address, i->alen); in packet_mc_add()
3776 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); in packet_mc_add()
3777 i->count = 1; in packet_mc_add()
3778 i->next = po->mclist; in packet_mc_add()
3779 po->mclist = i; in packet_mc_add()
3782 po->mclist = i->next; in packet_mc_add()
3797 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { in packet_mc_drop()
3798 if (ml->ifindex == mreq->mr_ifindex && in packet_mc_drop()
3799 ml->type == mreq->mr_type && in packet_mc_drop()
3800 ml->alen == mreq->mr_alen && in packet_mc_drop()
3801 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { in packet_mc_drop()
3802 if (--ml->count == 0) { in packet_mc_drop()
3804 *mlp = ml->next; in packet_mc_drop()
3805 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); in packet_mc_drop()
3807 packet_dev_mc(dev, ml, -1); in packet_mc_drop()
3822 if (!po->mclist) in packet_flush_mclist()
3826 while ((ml = po->mclist) != NULL) { in packet_flush_mclist()
3829 po->mclist = ml->next; in packet_flush_mclist()
3830 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); in packet_flush_mclist()
3832 packet_dev_mc(dev, ml, -1); in packet_flush_mclist()
3842 struct sock *sk = sock->sk; in packet_setsockopt()
3847 return -ENOPROTOOPT; in packet_setsockopt()
3857 return -EINVAL; in packet_setsockopt()
3861 return -EFAULT; in packet_setsockopt()
3863 return -EINVAL; in packet_setsockopt()
3876 ret = -EINVAL; in packet_setsockopt()
3878 switch (po->tp_version) { in packet_setsockopt()
3885 -EINVAL : 0; in packet_setsockopt()
3893 -EINVAL : 0; in packet_setsockopt()
3907 return -EINVAL; in packet_setsockopt()
3909 return -EFAULT; in packet_setsockopt()
3911 pkt_sk(sk)->copy_thresh = val; in packet_setsockopt()
3919 return -EINVAL; in packet_setsockopt()
3921 return -EFAULT; in packet_setsockopt()
3928 return -EINVAL; in packet_setsockopt()
3931 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3932 ret = -EBUSY; in packet_setsockopt()
3934 po->tp_version = val; in packet_setsockopt()
3945 return -EINVAL; in packet_setsockopt()
3947 return -EFAULT; in packet_setsockopt()
3949 return -EINVAL; in packet_setsockopt()
3951 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3952 ret = -EBUSY; in packet_setsockopt()
3954 po->tp_reserve = val; in packet_setsockopt()
3965 return -EINVAL; in packet_setsockopt()
3967 return -EFAULT; in packet_setsockopt()
3970 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
3971 ret = -EBUSY; in packet_setsockopt()
3984 return -EINVAL; in packet_setsockopt()
3986 return -EFAULT; in packet_setsockopt()
3996 return -EINVAL; in packet_setsockopt()
3998 return -EFAULT; in packet_setsockopt()
4008 if (sock->type != SOCK_RAW) in packet_setsockopt()
4009 return -EINVAL; in packet_setsockopt()
4011 return -EINVAL; in packet_setsockopt()
4013 return -EFAULT; in packet_setsockopt()
4018 return -EINVAL; in packet_setsockopt()
4024 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { in packet_setsockopt()
4025 ret = -EBUSY; in packet_setsockopt()
4027 WRITE_ONCE(po->vnet_hdr_sz, hdr_len); in packet_setsockopt()
4038 return -EINVAL; in packet_setsockopt()
4040 return -EFAULT; in packet_setsockopt()
4042 WRITE_ONCE(po->tp_tstamp, val); in packet_setsockopt()
4050 return -EINVAL; in packet_setsockopt()
4052 return -EFAULT; in packet_setsockopt()
4059 if (!READ_ONCE(po->fanout)) in packet_setsockopt()
4060 return -EINVAL; in packet_setsockopt()
4069 return -EINVAL; in packet_setsockopt()
4071 return -EFAULT; in packet_setsockopt()
4073 return -EINVAL; in packet_setsockopt()
4075 WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val); in packet_setsockopt()
4083 return -EINVAL; in packet_setsockopt()
4085 return -EFAULT; in packet_setsockopt()
4088 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec) in packet_setsockopt()
4099 return -EINVAL; in packet_setsockopt()
4101 return -EFAULT; in packet_setsockopt()
4107 return -ENOPROTOOPT; in packet_setsockopt()
4116 struct sock *sk = sock->sk; in packet_getsockopt()
4124 return -ENOPROTOOPT; in packet_getsockopt()
4127 return -EFAULT; in packet_getsockopt()
4130 return -EINVAL; in packet_getsockopt()
4134 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_getsockopt()
4135 memcpy(&st, &po->stats, sizeof(st)); in packet_getsockopt()
4136 memset(&po->stats, 0, sizeof(po->stats)); in packet_getsockopt()
4137 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_getsockopt()
4138 drops = atomic_xchg(&po->tp_drops, 0); in packet_getsockopt()
4140 if (po->tp_version == TPACKET_V3) { in packet_getsockopt()
4160 val = !!READ_ONCE(po->vnet_hdr_sz); in packet_getsockopt()
4163 val = READ_ONCE(po->vnet_hdr_sz); in packet_getsockopt()
4166 val = po->tp_version; in packet_getsockopt()
4172 return -EINVAL; in packet_getsockopt()
4174 return -EFAULT; in packet_getsockopt()
4186 return -EINVAL; in packet_getsockopt()
4190 val = po->tp_reserve; in packet_getsockopt()
4196 val = READ_ONCE(po->tp_tstamp); in packet_getsockopt()
4199 val = (po->fanout ? in packet_getsockopt()
4200 ((u32)po->fanout->id | in packet_getsockopt()
4201 ((u32)po->fanout->type << 16) | in packet_getsockopt()
4202 ((u32)po->fanout->flags << 24)) : in packet_getsockopt()
4206 val = READ_ONCE(po->prot_hook.ignore_outgoing); in packet_getsockopt()
4209 if (!po->rollover) in packet_getsockopt()
4210 return -EINVAL; in packet_getsockopt()
4211 rstats.tp_all = atomic_long_read(&po->rollover->num); in packet_getsockopt()
4212 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); in packet_getsockopt()
4213 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); in packet_getsockopt()
4224 return -ENOPROTOOPT; in packet_getsockopt()
4230 return -EFAULT; in packet_getsockopt()
4232 return -EFAULT; in packet_getsockopt()
4244 sk_for_each_rcu(sk, &net->packet.sklist) { in packet_notifier()
4249 if (po->mclist) in packet_notifier()
4250 packet_dev_mclist_delete(dev, &po->mclist); in packet_notifier()
4254 if (dev->ifindex == po->ifindex) { in packet_notifier()
4255 spin_lock(&po->bind_lock); in packet_notifier()
4258 sk->sk_err = ENETDOWN; in packet_notifier()
4264 WRITE_ONCE(po->ifindex, -1); in packet_notifier()
4265 netdev_put(po->prot_hook.dev, in packet_notifier()
4266 &po->prot_hook.dev_tracker); in packet_notifier()
4267 po->prot_hook.dev = NULL; in packet_notifier()
4269 spin_unlock(&po->bind_lock); in packet_notifier()
4273 if (dev->ifindex == po->ifindex) { in packet_notifier()
4274 spin_lock(&po->bind_lock); in packet_notifier()
4275 if (po->num) in packet_notifier()
4277 spin_unlock(&po->bind_lock); in packet_notifier()
4290 struct sock *sk = sock->sk; in packet_ioctl()
4304 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_ioctl()
4305 skb = skb_peek(&sk->sk_receive_queue); in packet_ioctl()
4307 amount = skb->len; in packet_ioctl()
4308 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_ioctl()
4330 return -ENOIOCTLCMD; in packet_ioctl()
4338 struct sock *sk = sock->sk; in packet_poll()
4342 spin_lock_bh(&sk->sk_receive_queue.lock); in packet_poll()
4343 if (po->rx_ring.pg_vec) { in packet_poll()
4344 if (!packet_previous_rx_frame(po, &po->rx_ring, in packet_poll()
4349 spin_unlock_bh(&sk->sk_receive_queue.lock); in packet_poll()
4350 spin_lock_bh(&sk->sk_write_queue.lock); in packet_poll()
4351 if (po->tx_ring.pg_vec) { in packet_poll()
4352 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) in packet_poll()
4355 spin_unlock_bh(&sk->sk_write_queue.lock); in packet_poll()
4366 struct file *file = vma->vm_file; in packet_mm_open()
4367 struct socket *sock = file->private_data; in packet_mm_open()
4368 struct sock *sk = sock->sk; in packet_mm_open()
4371 atomic_long_inc(&pkt_sk(sk)->mapped); in packet_mm_open()
4376 struct file *file = vma->vm_file; in packet_mm_close()
4377 struct socket *sock = file->private_data; in packet_mm_close()
4378 struct sock *sk = sock->sk; in packet_mm_close()
4381 atomic_long_dec(&pkt_sk(sk)->mapped); in packet_mm_close()
4434 unsigned int block_nr = req->tp_block_nr; in alloc_pg_vec()
4469 struct tpacket_req *req = &req_u->req; in packet_set_ring()
4471 rb = tx_ring ? &po->tx_ring : &po->rx_ring; in packet_set_ring()
4472 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; in packet_set_ring()
4474 err = -EBUSY; in packet_set_ring()
4476 if (atomic_long_read(&po->mapped)) in packet_set_ring()
4482 if (req->tp_block_nr) { in packet_set_ring()
4486 err = -EBUSY; in packet_set_ring()
4487 if (unlikely(rb->pg_vec)) in packet_set_ring()
4490 switch (po->tp_version) { in packet_set_ring()
4492 po->tp_hdrlen = TPACKET_HDRLEN; in packet_set_ring()
4495 po->tp_hdrlen = TPACKET2_HDRLEN; in packet_set_ring()
4498 po->tp_hdrlen = TPACKET3_HDRLEN; in packet_set_ring()
4502 err = -EINVAL; in packet_set_ring()
4503 if (unlikely((int)req->tp_block_size <= 0)) in packet_set_ring()
4505 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) in packet_set_ring()
4507 min_frame_size = po->tp_hdrlen + po->tp_reserve; in packet_set_ring()
4508 if (po->tp_version >= TPACKET_V3 && in packet_set_ring()
4509 req->tp_block_size < in packet_set_ring()
4510 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) in packet_set_ring()
4512 if (unlikely(req->tp_frame_size < min_frame_size)) in packet_set_ring()
4514 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) in packet_set_ring()
4517 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; in packet_set_ring()
4518 if (unlikely(rb->frames_per_block == 0)) in packet_set_ring()
4520 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) in packet_set_ring()
4522 if (unlikely((rb->frames_per_block * req->tp_block_nr) != in packet_set_ring()
4523 req->tp_frame_nr)) in packet_set_ring()
4526 err = -ENOMEM; in packet_set_ring()
4527 order = get_order(req->tp_block_size); in packet_set_ring()
4531 switch (po->tp_version) { in packet_set_ring()
4537 struct tpacket_req3 *req3 = &req_u->req3; in packet_set_ring()
4539 if (req3->tp_retire_blk_tov || in packet_set_ring()
4540 req3->tp_sizeof_priv || in packet_set_ring()
4541 req3->tp_feature_req_word) { in packet_set_ring()
4542 err = -EINVAL; in packet_set_ring()
4549 rx_owner_map = bitmap_alloc(req->tp_frame_nr, in packet_set_ring()
4559 err = -EINVAL; in packet_set_ring()
4560 if (unlikely(req->tp_frame_nr)) in packet_set_ring()
4566 spin_lock(&po->bind_lock); in packet_set_ring()
4568 num = po->num; in packet_set_ring()
4570 WRITE_ONCE(po->num, 0); in packet_set_ring()
4573 spin_unlock(&po->bind_lock); in packet_set_ring()
4577 err = -EBUSY; in packet_set_ring()
4578 mutex_lock(&po->pg_vec_lock); in packet_set_ring()
4579 if (closing || atomic_long_read(&po->mapped) == 0) { in packet_set_ring()
4581 spin_lock_bh(&rb_queue->lock); in packet_set_ring()
4582 swap(rb->pg_vec, pg_vec); in packet_set_ring()
4583 if (po->tp_version <= TPACKET_V2) in packet_set_ring()
4584 swap(rb->rx_owner_map, rx_owner_map); in packet_set_ring()
4585 rb->frame_max = (req->tp_frame_nr - 1); in packet_set_ring()
4586 rb->head = 0; in packet_set_ring()
4587 rb->frame_size = req->tp_frame_size; in packet_set_ring()
4588 spin_unlock_bh(&rb_queue->lock); in packet_set_ring()
4590 swap(rb->pg_vec_order, order); in packet_set_ring()
4591 swap(rb->pg_vec_len, req->tp_block_nr); in packet_set_ring()
4593 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; in packet_set_ring()
4594 po->prot_hook.func = (po->rx_ring.pg_vec) ? in packet_set_ring()
4597 if (atomic_long_read(&po->mapped)) in packet_set_ring()
4599 atomic_long_read(&po->mapped)); in packet_set_ring()
4601 mutex_unlock(&po->pg_vec_lock); in packet_set_ring()
4603 spin_lock(&po->bind_lock); in packet_set_ring()
4605 WRITE_ONCE(po->num, num); in packet_set_ring()
4608 spin_unlock(&po->bind_lock); in packet_set_ring()
4609 if (pg_vec && (po->tp_version > TPACKET_V2)) { in packet_set_ring()
4610 /* Because we don't support block-based V3 on tx-ring */ in packet_set_ring()
4618 free_pg_vec(pg_vec, order, req->tp_block_nr); in packet_set_ring()
4627 struct sock *sk = sock->sk; in packet_mmap()
4632 int err = -EINVAL; in packet_mmap()
4635 if (vma->vm_pgoff) in packet_mmap()
4636 return -EINVAL; in packet_mmap()
4638 mutex_lock(&po->pg_vec_lock); in packet_mmap()
4641 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { in packet_mmap()
4642 if (rb->pg_vec) { in packet_mmap()
4643 expected_size += rb->pg_vec_len in packet_mmap()
4644 * rb->pg_vec_pages in packet_mmap()
4652 size = vma->vm_end - vma->vm_start; in packet_mmap()
4656 start = vma->vm_start; in packet_mmap()
4657 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { in packet_mmap()
4658 if (rb->pg_vec == NULL) in packet_mmap()
4661 for (i = 0; i < rb->pg_vec_len; i++) { in packet_mmap()
4663 void *kaddr = rb->pg_vec[i].buffer; in packet_mmap()
4666 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { in packet_mmap()
4677 atomic_long_inc(&po->mapped); in packet_mmap()
4678 vma->vm_ops = &packet_mmap_ops; in packet_mmap()
4682 mutex_unlock(&po->pg_vec_lock); in packet_mmap()
4744 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); in packet_seq_start()
4750 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); in packet_seq_next()
4764 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk"); in packet_seq_show()
4770 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", in packet_seq_show()
4772 refcount_read(&s->sk_refcnt), in packet_seq_show()
4773 s->sk_type, in packet_seq_show()
4774 ntohs(READ_ONCE(po->num)), in packet_seq_show()
4775 READ_ONCE(po->ifindex), in packet_seq_show()
4777 atomic_read(&s->sk_rmem_alloc), in packet_seq_show()
4795 mutex_init(&net->packet.sklist_lock); in packet_net_init()
4796 INIT_HLIST_HEAD(&net->packet.sklist); in packet_net_init()
4799 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, in packet_net_init()
4801 return -ENOMEM; in packet_net_init()
4809 remove_proc_entry("packet", net->proc_net); in packet_net_exit()
4810 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); in packet_net_exit()