1.3.2

@rafl rafl released this Jan 17, 2018 · 6 commits to master since this release

  • Allocate memory for MMDB_entry_data_list_s structs in separate chunks
    rather than one large chunk. This simplifies accessing memory in
    MMDB_get_entry_data_list() and increases performance. It builds on the
    changes in 1.3.0 and 1.3.1.
  • We no longer export data_pool_* symbols. These are internal functions
    but we were previously exporting them. Pull request by Faidon Liambotis.
    GitHub #162.
  • Build with POSIX.1-2008 by default if the system supports it. This allows
    use of open() with O_CLOEXEC. We retain support for systems that
    provide only POSIX.1-2001.
  • Open the database with the O_CLOEXEC flag if the system provides it.
    This avoids cases where we could leak fds when called in multi-threaded
    programs that fork() and exec(). Original report and PR by Brandon L
    Black.
  • Added a test to ensure we export only intended symbols (e.g. MMDB_*).

1.3.1

@horgh horgh released this Nov 24, 2017 · 45 commits to master since this release

  • Fix build problems related to rpl_malloc(). Pull request by Rainer
    Gerhards. GitHub #152.
  • Fix a race to set and read data in a field on the MMDB_s struct
    (ipv4_start_node). GitHub #153.
  • Fix cases of invalid memory access when using
    MMDB_get_entry_data_list(). This was introduced in 1.3.0 and occurred
    when performing large lookups. GitHub #153.

1.3.0

@horgh horgh released this Nov 10, 2017 · 65 commits to master since this release

  • Perform fewer memory allocations in MMDB_get_entry_data_list(). This
    significantly improves its performance. GitHub #147.
  • Fix mmdblookup's build epoch reporting on some systems. Big endian
    systems with a 32-bit time_t no longer show a database build date of
    1970-01-01 00:00:00. Pull request by Rainer Jung. GitHub #143.

1.2.1

@oschwald oschwald released this May 15, 2017 · 84 commits to master since this release

  • Use autoconf to check the system's endianness rather than trying to do this
    with compiler-defined macros like __BYTE_ORDER__. Apparently this didn't
    work properly on a Sparc system. GitHub #120.
  • Several compiler warnings on Visual C++ were fixed. Pull request by Marcel
    Raad. GitHub #130.
  • Fix segmentation faults found in MMDB_open() using afl-fuzz. This
    occurred on corrupt databases that had a data pointer large enough to
    cause an integer overflow when doing bound checking. Reported by Ryan
    Whitworth. GitHub #140.
  • Add --disable-tests option to configure. Pull request by Fabrice
    Fontaine. GitHub #136.

1.2.0 - Extended API for Iteration of Database

@oschwald oschwald released this Mar 23, 2016 · 119 commits to master since this release

  • Four additional fields were added to the end of the MMDB_search_node_s
    struct returned by MMDB_read_node. These fields allow the user to iterate
    through the search tree without making undocumented assumptions about how
    this library works internally and without knowing the specific details of
    the database format. GitHub #110.

Minor Bug Fixes

@rafl rafl released this Mar 20, 2016 · 124 commits to master since this release

1.1.5 - 2016-03-20

  • Previously, reading a database with a pointer in the metadata would cause an
    MMDB_INVALID_METADATA_ERROR to be returned. This was due to an invalid
    offset being used when calculating the pointer. The data_section and
    metadata_section fields now both point to the beginning of the data
    section. Previously, data_section pointed to the beginning of the data
    separator. This will not affect anyone using only documented fields from
    MMDB_s.
  • MMDB_lookup_sockaddr will set mmdb_error to
    MMDB_IPV6_LOOKUP_IN_IPV4_DATABASE_ERROR if an IPv6 sockaddr is looked up
    in an IPv4-only database. Previously only MMDB_lookup_string would set
    this error code.
  • When resolving an address, this library now relies on getaddrinfo to
    determine the address family rather than trying to guess it itself.

Packaging fixes

@autarch autarch released this Jan 6, 2016 · 142 commits to master since this release

1.1.4 - 2016-01-06

  • Packaging fixes. The 1.1.3 tarball release contained a lot of extra junk in the t/ directory.

More overflow and malloc checks

@autarch autarch released this Jan 5, 2016 · 149 commits to master since this release

This tarball has since been removed because it was not cleanly packaged. Use 1.1.4 or newer instead.

  • Added several additional checks to make sure that we don't attempt to read
    past the end of the databases's data section. Implemented by Tobias
    Stoeckmann. GitHub #103.
  • When searching for the database metadata, there was a bug that caused the
    code to think it had found valid metadata when none existed. In addition,
    this could lead to an attempt to read past the end of the database
    entirely. Finally, if there are multiple metadata markers in the database,
    we treat the final one as the start of the metdata, instead of the first.
    Implemented by Tobias Stoeckmann. GitHub #102.
  • Don't attempt to mmap a file that is too large to be mmapped on the
    system. Implemented by Tobias Stoeckmann. GitHub #101.
  • Added a missing out of memory check when reading a file's
    metadata. Implemented by Tobias Stoeckmann. GitHub #101.
  • Added several additional checks to make sure that we never attempt to
    malloc more than SIZE_MAX memory, which would lead to integer
    overflow. This could only happen with pathological databases. Implemented by
    Tobias Stoeckmann. GitHub #101.

Improved handling of corrupt databses and pkg-config support

@oschwald oschwald released this Nov 16, 2015 · 170 commits to master since this release

  • IMPORTANT: This release includes a number of important security fixes. Among these fixes is improved validation of the database metadata. Unfortunately, MaxMind GeoIP2 and GeoLite2 databases created earlier than January 28, 2014 had an invalid data type for the record_size in the metadata. Previously these databases worked on little endian machines with libmaxminddb but did not work on big endian machines. Due to increased safety checks when reading the file, these databases will no longer work on any platform. If you are using one of these databases, we recommend that you upgrade to the latest GeoLite2 or GeoIP2 database
  • Added pkg-config support. If your system supports it, then running make install now installs a libmaxminddb.pc file for pkgconfig. Implemented by
    Jan Vcelak.
  • Several segmentation faults found with afl-fuzz were fixed. These were
    caused by missing bounds checking and missing verification of data type.
  • MMDB_get_entry_data_list will now fail on data structures with a depth
    greater than 512 and data structures that are cyclic. This should not
    affect any known MaxMind DB in production. All databases produced by
    MaxMind have a depth of less than five.