On 11/5/2024 3:12 AM, Pablo Neira Ayuso wrote: > Hi, > > There is another issue in this recent series, see below. > > This needs another follow up. Dan Carpenter sent a patch: https://lore.kernel.org/lkml/b226a01a-2545-4b67-9cc6-59cfa0ffabbc@xxxxxxxxxxxxxxxx/ > > Thanks. > > On Sat, Nov 02, 2024 at 12:21:01PM +0300, Dan Carpenter wrote: >> Hello Casey Schaufler, >> >> Commit 95a3c11eb670 ("lsm: replace context+len with lsm_context") >> from Oct 23, 2024 (linux-next), leads to the following Smatch static >> checker warning: >> >> net/netfilter/nfnetlink_queue.c:646 nfqnl_build_packet_message() >> warn: always true condition '(seclen >= 0) => (0-u32max >= 0)' >> >> net/netfilter/nfnetlink_queue.c:813 nfqnl_build_packet_message() >> warn: always true condition '(seclen >= 0) => (0-u32max >= 0)' >> >> net/netfilter/nfnetlink_queue.c:822 nfqnl_build_packet_message() >> warn: always true condition '(seclen >= 0) => (0-u32max >= 0)' >> >> net/netfilter/nfnetlink_queue.c >> 551 static struct sk_buff * >> 552 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue, >> 553 struct nf_queue_entry *entry, >> 554 __be32 **packet_id_ptr) >> 555 { >> 556 size_t size; >> 557 size_t data_len = 0, cap_len = 0; >> 558 unsigned int hlen = 0; >> 559 struct sk_buff *skb; >> 560 struct nlattr *nla; >> 561 struct nfqnl_msg_packet_hdr *pmsg; >> 562 struct nlmsghdr *nlh; >> 563 struct sk_buff *entskb = entry->skb; >> 564 struct net_device *indev; >> 565 struct net_device *outdev; >> 566 struct nf_conn *ct = NULL; >> 567 enum ip_conntrack_info ctinfo = 0; >> 568 const struct nfnl_ct_hook *nfnl_ct; >> 569 bool csum_verify; >> 570 struct lsm_context ctx; >> 571 u32 seclen = 0; >> ^^^^^^^^^^ >> >> 572 ktime_t tstamp; >> 573 >> 574 size = nlmsg_total_size(sizeof(struct nfgenmsg)) >> 575 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr)) >> 576 + nla_total_size(sizeof(u_int32_t)) /* ifindex */ >> 577 + nla_total_size(sizeof(u_int32_t)) /* ifindex */ >> 578 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) >> 579 + nla_total_size(sizeof(u_int32_t)) /* ifindex */ >> 580 + nla_total_size(sizeof(u_int32_t)) /* ifindex */ >> 581 #endif >> 582 + nla_total_size(sizeof(u_int32_t)) /* mark */ >> 583 + nla_total_size(sizeof(u_int32_t)) /* priority */ >> 584 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw)) >> 585 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */ >> 586 #if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID) >> 587 + nla_total_size(sizeof(u_int32_t)) /* classid */ >> 588 #endif >> 589 + nla_total_size(sizeof(u_int32_t)); /* cap_len */ >> 590 >> 591 tstamp = skb_tstamp_cond(entskb, false); >> 592 if (tstamp) >> 593 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp)); >> 594 >> 595 size += nfqnl_get_bridge_size(entry); >> 596 >> 597 if (entry->state.hook <= NF_INET_FORWARD || >> 598 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL)) >> 599 csum_verify = !skb_csum_unnecessary(entskb); >> 600 else >> 601 csum_verify = false; >> 602 >> 603 outdev = entry->state.out; >> 604 >> 605 switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) { >> 606 case NFQNL_COPY_META: >> 607 case NFQNL_COPY_NONE: >> 608 break; >> 609 >> 610 case NFQNL_COPY_PACKET: >> 611 if (!(queue->flags & NFQA_CFG_F_GSO) && >> 612 entskb->ip_summed == CHECKSUM_PARTIAL && >> 613 nf_queue_checksum_help(entskb)) >> 614 return NULL; >> 615 >> 616 data_len = READ_ONCE(queue->copy_range); >> 617 if (data_len > entskb->len) >> 618 data_len = entskb->len; >> 619 >> 620 hlen = skb_zerocopy_headlen(entskb); >> 621 hlen = min_t(unsigned int, hlen, data_len); >> 622 size += sizeof(struct nlattr) + hlen; >> 623 cap_len = entskb->len; >> 624 break; >> 625 } >> 626 >> 627 nfnl_ct = rcu_dereference(nfnl_ct_hook); >> 628 >> 629 #if IS_ENABLED(CONFIG_NF_CONNTRACK) >> 630 if (queue->flags & NFQA_CFG_F_CONNTRACK) { >> 631 if (nfnl_ct != NULL) { >> 632 ct = nf_ct_get(entskb, &ctinfo); >> 633 if (ct != NULL) >> 634 size += nfnl_ct->build_size(ct); >> 635 } >> 636 } >> 637 #endif >> 638 >> 639 if (queue->flags & NFQA_CFG_F_UID_GID) { >> 640 size += (nla_total_size(sizeof(u_int32_t)) /* uid */ >> 641 + nla_total_size(sizeof(u_int32_t))); /* gid */ >> 642 } >> 643 >> 644 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) { >> 645 seclen = nfqnl_get_sk_secctx(entskb, &ctx); >> >> nfqnl_get_sk_secctx() returns negative error codes. It needs to be changed ot >> int as well instead of u32. >> >> 646 if (seclen >= 0) >> 647 size += nla_total_size(seclen); >> 648 } >> 649 >> 650 skb = alloc_skb(size, GFP_ATOMIC); >> 651 if (!skb) { >> 652 skb_tx_error(entskb); >> 653 goto nlmsg_failure; >> 654 } >> 655 >> 656 nlh = nfnl_msg_put(skb, 0, 0, >> 657 nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET), >> 658 0, entry->state.pf, NFNETLINK_V0, >> 659 htons(queue->queue_num)); >> 660 if (!nlh) { >> 661 skb_tx_error(entskb); >> 662 kfree_skb(skb); >> 663 goto nlmsg_failure; >> 664 } >> 665 >> 666 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg)); >> 667 pmsg = nla_data(nla); >> 668 pmsg->hw_protocol = entskb->protocol; >> 669 pmsg->hook = entry->state.hook; >> 670 *packet_id_ptr = &pmsg->packet_id; >> 671 >> 672 indev = entry->state.in; >> 673 if (indev) { >> 674 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER) >> 675 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex))) >> 676 goto nla_put_failure; >> 677 #else >> 678 if (entry->state.pf == PF_BRIDGE) { >> 679 /* Case 1: indev is physical input device, we need to >> 680 * look for bridge group (when called from >> 681 * netfilter_bridge) */ >> 682 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV, >> 683 htonl(indev->ifindex)) || >> 684 /* this is the bridge group "brX" */ >> 685 /* rcu_read_lock()ed by __nf_queue */ >> 686 nla_put_be32(skb, NFQA_IFINDEX_INDEV, >> 687 htonl(br_port_get_rcu(indev)->br->dev->ifindex))) >> 688 goto nla_put_failure; >> 689 } else { >> 690 int physinif; >> 691 >> 692 /* Case 2: indev is bridge group, we need to look for >> 693 * physical device (when called from ipv4) */ >> 694 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, >> 695 htonl(indev->ifindex))) >> 696 goto nla_put_failure; >> 697 >> 698 physinif = nf_bridge_get_physinif(entskb); >> 699 if (physinif && >> 700 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV, >> 701 htonl(physinif))) >> 702 goto nla_put_failure; >> 703 } >> 704 #endif >> 705 } >> 706 >> 707 if (outdev) { >> 708 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER) >> 709 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex))) >> 710 goto nla_put_failure; >> 711 #else >> 712 if (entry->state.pf == PF_BRIDGE) { >> 713 /* Case 1: outdev is physical output device, we need to >> 714 * look for bridge group (when called from >> 715 * netfilter_bridge) */ >> 716 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV, >> 717 htonl(outdev->ifindex)) || >> 718 /* this is the bridge group "brX" */ >> 719 /* rcu_read_lock()ed by __nf_queue */ >> 720 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, >> 721 htonl(br_port_get_rcu(outdev)->br->dev->ifindex))) >> 722 goto nla_put_failure; >> 723 } else { >> 724 int physoutif; >> 725 >> 726 /* Case 2: outdev is bridge group, we need to look for >> 727 * physical output device (when called from ipv4) */ >> 728 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, >> 729 htonl(outdev->ifindex))) >> 730 goto nla_put_failure; >> 731 >> 732 physoutif = nf_bridge_get_physoutif(entskb); >> 733 if (physoutif && >> 734 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV, >> 735 htonl(physoutif))) >> 736 goto nla_put_failure; >> 737 } >> 738 #endif >> 739 } >> 740 >> 741 if (entskb->mark && >> 742 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark))) >> 743 goto nla_put_failure; >> 744 >> 745 if (entskb->priority && >> 746 nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority))) >> 747 goto nla_put_failure; >> 748 >> 749 if (indev && entskb->dev && >> 750 skb_mac_header_was_set(entskb) && >> 751 skb_mac_header_len(entskb) != 0) { >> 752 struct nfqnl_msg_packet_hw phw; >> 753 int len; >> 754 >> 755 memset(&phw, 0, sizeof(phw)); >> 756 len = dev_parse_header(entskb, phw.hw_addr); >> 757 if (len) { >> 758 phw.hw_addrlen = htons(len); >> 759 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw)) >> 760 goto nla_put_failure; >> 761 } >> 762 } >> 763 >> 764 if (nfqnl_put_bridge(entry, skb) < 0) >> 765 goto nla_put_failure; >> 766 >> 767 if (entry->state.hook <= NF_INET_FORWARD && tstamp) { >> 768 struct nfqnl_msg_packet_timestamp ts; >> 769 struct timespec64 kts = ktime_to_timespec64(tstamp); >> 770 >> 771 ts.sec = cpu_to_be64(kts.tv_sec); >> 772 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC); >> 773 >> 774 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts)) >> 775 goto nla_put_failure; >> 776 } >> 777 >> 778 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk && >> 779 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0) >> 780 goto nla_put_failure; >> 781 >> 782 if (nfqnl_put_sk_classid(skb, entskb->sk) < 0) >> 783 goto nla_put_failure; >> 784 >> 785 if (seclen && nla_put(skb, NFQA_SECCTX, ctx.len, ctx.context)) >> ^^^^^^ >> This doesn't look right. >> >> >> 786 goto nla_put_failure; >> 787 >> 788 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0) >> 789 goto nla_put_failure; >> 790 >> 791 if (cap_len > data_len && >> 792 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len))) >> 793 goto nla_put_failure; >> 794 >> 795 if (nfqnl_put_packet_info(skb, entskb, csum_verify)) >> 796 goto nla_put_failure; >> 797 >> 798 if (data_len) { >> 799 struct nlattr *nla; >> 800 >> 801 if (skb_tailroom(skb) < sizeof(*nla) + hlen) >> 802 goto nla_put_failure; >> 803 >> 804 nla = skb_put(skb, sizeof(*nla)); >> 805 nla->nla_type = NFQA_PAYLOAD; >> 806 nla->nla_len = nla_attr_size(data_len); >> 807 >> 808 if (skb_zerocopy(skb, entskb, data_len, hlen)) >> 809 goto nla_put_failure; >> 810 } >> 811 >> 812 nlh->nlmsg_len = skb->len; >> 813 if (seclen >= 0) >> ^^^^^^^^^^^ >> >> 814 security_release_secctx(&ctx); >> 815 return skb; >> 816 >> 817 nla_put_failure: >> 818 skb_tx_error(entskb); >> 819 kfree_skb(skb); >> 820 net_err_ratelimited("nf_queue: error creating packet message\n"); >> 821 nlmsg_failure: >> 822 if (seclen >= 0) >> ^^^^^^^^^^^ >> >> 823 security_release_secctx(&ctx); >> 824 return NULL; >> 825 } >> >> >> >> regards, >> dan carpenter >>
