Builder libpcap-solaris10-sparcv9 Build #3245

Results:

Build successful

SourceStamp:

BuildSlave:

Reason:

The SingleBranchScheduler scheduler named 'schedule-libpcap-solaris10-sparcv9' triggered this build

Steps and Logfiles:

1. git update ( 25 secs )
   1. stdio
2. shell Set developer mode ( 0 secs )
   1. stdio
3. shell_1 './autogen.sh' ( 21 secs )
   1. stdio
4. shell_2 './configure' ( 58 secs )
   1. stdio
   2. config.log
5. shell_3 'gmake' ( 47 secs )
   1. stdio
6. shell_4 'gmake check' ( 14 mins, 13 secs )
   1. stdio
7. shell_5 'rm -rf ...' ( 0 secs )
   1. stdio
8. shell_6 'gmake install ...' ( 3 secs )
   1. stdio
9. trigger triggered trigger-tcpdump-solaris10-sparcv9 ( 11 mins, 35 secs )
   1. - no logs -
   2. tcpdump-solaris10-sparcv9 #6105

Build Properties:

Forced Build Properties:

Responsible Users:

1. Denis Ovsienko

   denisohnoyoudont@ovsienko.info

Timing:

All Changes:

:

1. Change #255530

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 14 Jan 2026 22:11:10
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	809629356687e7b50b9d85068648c3fa09dd3d7f

   Comments

   Separate DECnet and IPv4 code paths better.
   The only reason for gen_ncode() to be aware of DECnet is that DECnet
   address syntax grammar-wise is a subset of IPv4 address syntax, so it is
   a common entry point for both.  There is no reason for gen_host() to be
   aware of DECnet because otherwise the function uses gen_hostop(), which
   operates in IPv4 space.
   
   To make it clear that the two code paths diverge early, make the
   DECnet-specific parts of gen_ncode() and gen_host() a new function,
   gen_dnhost().  In TESTrun add two reject tests and update various
   comments to show the new code paths.
   
   In pcapint_atodn() make the output argument 16-bit to match the address
   size and return 1 on success for consistency with other functions that
   parse network addresses.  Update translatetest accordingly.

   Changed files

   • gencode.c
   • nametoaddr.c
   • nametoaddr.h
   • testprogs/TESTrun
   • testprogs/translatetest.c

2. Change #255531

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 14 Jan 2026 22:11:12
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	8041eb9f2219b029c9b18d56fe1097c1f97479b7

   Comments

   Remove some duplicate tests.
   As it turns out, a small subset of tests in commit 1f3648a duplicated
   what was already covered with the tests in commit a5090c6, albeit using
   different labels.  Lose one of the copies.

   Changed files

   • testprogs/TESTrun

3. Change #255532

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 14 Jan 2026 22:11:12
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	1daf61c4a59c380a0a025f21cec7cac099c60eec

   Comments

   Fix a bug in the earlier DLT_PFLOG optimization.
   My earlier commit bea13af introduced a bug: when in the Q_ARP and Q_RARP
   cases the conditional early return skips the invocation of gen_hostop(),
   it removes one unwanted side effect, but also another side effect, which
   is the wanted validation of dir qualifier.  This caused a failure to
   reject filter expressions that have "arp" or "rarp" as the proto
   qualifier, a WLAN dir qualifier, "host" or "net" as the type qualifier
   and an ID that is either an IPv4 address or an IPv4 CIDR or a network
   name or a hostname that resolves to at least one IPv4 address, for
   example:
   
   $ filtertest -O IPV4 'arp addr1 net 10.0.0.0/8'
   (000) ld       #0x1
   (001) jeq      #0x0             jt 2	jf 3
   (002) ret      #262144
   (003) ret      #0
   
   To fix this, in gen_host(), gen_host6(), gen_dnhost() and gen_acode()
   reject all WLAN dir qualifiers early.  In gen_hostop(), gen_hostop6(),
   gen_dnhostop() and gen_mac8host() convert the existing default cases to
   generate an internal error because now these default cases should never
   match.  In gen_mac48host() do both unless the code path diverges to
   gen_wlanhostop():
   
   $ filtertest -O IPV4 'arp addr1 net 10.0.0.0/8'
   filtertest: 'addr1' is valid for 802.11 syntax only
   
   The [ARP/RARP-specific] regression did not fail any of the existing
   tests, which correspond to the Q_IP code path in gen_host(), which
   always invoked gen_hostop(), so in TESTrun expand the reject tests
   combinatorial space to cover more code paths.

   Changed files

   • gencode.c
   • testprogs/TESTrun

4. Change #255533

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 14 Jan 2026 22:11:12
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	5462648e501d59aebf3e0eb8b85d39ece25c74f1

   Comments

   Start testing more than one IPv4/IPv6 address per name.
   Introduce three new hostnames under host357.libpcap.test and add accept
   filter tests for "host" to show the resulting filter programs.

   Changed files

   • testprogs/TESTrun

5. Change #255534

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Thu 15 Jan 2026 13:28:29
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	8bb792bcb34f539447ae4c4a6a2e5f63bdefe5e4

   Comments

   Refine a few IPv4/IPv6-related functions.
   In gen_host() factor [effectively] common code out of the switch block.
   Ibid., instead of deriving the context from a type qualifier make it an
   argument.  Idem in gen_host6(), and convert a switch to an if.  In
   gen_host46_byname() and elsewhere supply the context to gen_host() and
   gen_host6(), and make it a bit more descriptive.  Convert a memset() to
   a static const.  Update the tests.
   
   Unlike gen_hostop(), in gen_hostop6() the address offsets are constant,
   so remove these from the function signature.  Make IPv6, IPv4, ARP and
   RARP address offsets named constants, constify various function
   arguments and use u_char for dir and proto qualifiers.

   Changed files

   • gencode.c
   • testprogs/TESTrun

6. Change #255535

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Thu 15 Jan 2026 13:29:24
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	5bf820d6accb9fe67b8455a24fcdfc37e3012361

   Comments

   Test link protocol once per AF, not per address.
   In gen_host() and gen_host6() instead of one address and one netmask
   take two vectors and their size and generate a link protocol check once
   per vector instead of once per address.  In gen_host46_byname() derive
   such vectors from the linked list and then process one vector per
   invocation.  This follows the logic in commits 2559282 and 6f8a28c
   earlier and produces shorter unoptimized programs when a hostname
   resolves to more than one A/AAAA address.  Update the tests.

   Changed files

   • CHANGES
   • gencode.c
   • testprogs/TESTrun

7. Change #255536

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Thu 15 Jan 2026 16:35:47
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	2ceba355be7bcd6ac58c9ad1a96de5d8cf1374a0

   Comments

   Sort resolved IPv4 and IPv6 addresses.
   DNS resolvers tend to shuffle multiple records of the same type (in this
   case A and AAAA) for the same hostname.  The difference in the ordering
   unnecessarily complicates comparison of filter programs in general and
   the tests in particular, and is negligible otherwise, so sort IPv4 and
   IPv6 addresses and remove one more (see commit a516cf3) external effect
   from this scope.  Update the tests.  This addresses all of the remaining
   problem space discussed in GH #1574.

   Changed files

   • CHANGES
   • gencode.c
   • testprogs/TESTrun

8. Change #255537

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Sun 18 Jan 2026 15:51:44
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	cc0e534f3d987393bd2a90fb5d1fe7bd944813fc

   Comments

   In "gateway" negate the host(s), but not the protocol.
   "[arp|ip|rarp] gateway <name>" since tcpdump 2.0 (before libpcap 0.0)
   has been the same as "link host <name> and [arp|ip|arp] host not
   <name>".  It is not obvious that the latter is the same as "link host
   <name> and not [arp|ip|arp] host <name>" and that the latter applies the
   negation to both the address(es) and the link protocol.  This is why,
   for example, "ip gateway" besides the expected forwarded IPv4 packets
   also matches any non-IPv4 protocols (ARP, IPv6, RARP, LLDP, LACP, STP,
   etc.) so long as the source or the destination MAC-48 address matches.
   
   On the one hand, nobody has reported this bug before.  On the other
   hand, "gateway" became mutually exclusive with IPv6 support in commit
   7fe3c11 (i.e. in every libpcap release since 0.5 in January 2000), and
   IPv6 support became enabled by default in commit bc54719 (i.e. in every
   libpcap release since 1.0.0 in October 2008).  Because since commit
   a027fdc the primitive became always enabled, it makes the most sense to
   fix it before it becomes available again in libpcap 1.11.0.
   
   To fix this, in gen_gateway() instead of negating the result of
   gen_host46_byname() pass the request to negate the host(s) to the
   function , let it propagate the request to gen_hosts() and gen_hosts6()
   and there handle it correctly.  Update the tests as required, in
   particular, fix all results marked as "BUG" in the gateway apply filter
   tests.  This addresses most of the problem space discussed in GH #1571.

   Changed files

   • CHANGES
   • gencode.c
   • pcap-filter.manmisc.in
   • testprogs/TESTrun

9. Change #255538

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Sun 18 Jan 2026 15:51:49
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	19c0d1671e70ae15bf2e544dcf79351e6d095016

   Comments

   Reject "gateway" within MPLS, VXLAN or Geneve.
   The solution space of this primitive addresses the problem space of
   filtering standard IPv4 traffic on a broadcast network segment, where
   MAC-48 and IPv4 addresses relate in a certain way.
   
   The "mpls" primitive became available in libpcap 0.9.1 (May 2005) -- a
   few years before "gateway" became disabled by default.  However, in an
   MPLS packet the MAC-48 address precedes the label stack and the switched
   frame follows it, so even if the frame carries raw IPv4, the IPv4
   addresses are likely unrelated to the Ethernet segment and the LSR's own
   IPv4 addresses.  The "vxlan" and "geneve" primitives became available
   after "gateway" became disabled by default.
   
   Since "gateway" never worked correctly until the previous commit in the
   first place, backward compatibility is not a factor in this particular
   case, so avoid reintroducing this space for confusion whilst
   reintroducing "gateway".  This addresses the remaining part of the
   problem space in GH #1571.

   Changed files

   • CHANGES
   • gencode.c
   • pcap-filter.manmisc.in
   • testprogs/TESTrun