pmacct (Promiscuous mode IP Accounting package) v0.11.4
pmacct is Copyright (C) 2003-2007 by Paolo Lucente

0.11.4 -- 25-04-2007
  + support for TCP flags has been introduced. Flags are ORed on a
    per-aggregate basis (same as what NetFlow does on a per-flow
    basis). The 'aggregate' directive now supports the 'tcpflags'
    keyword. SQL tables v7 have also been introduced in order to
    support the feature inside the SQL plugins. 
  + 'nfacctd_sql_log' directive is being introduced. In nfacctd,
    it makes SQL plugins to use a) NetFlow's First Switched value
    as "stamp_inserted" timestamp and b) Last Switched value as
    "stamp_updated" timestamp. Then, a) by not aggregating flows
    and b) not making use of timeslots, this directive allows to
    log singular flows in the SQL database.
  + sfprobe and nfprobe plugins are now able to propagate tags to
    remote collectors through sFlow v5 and NetFlow v9 protocols. 
    The 'tag' key must be appended to sfprobe/nfprobe 'aggregate'
    config directives.
  + pmacct memory client is now able to output either TopN bytes,
    flows or packets statistics. The feature is enabled by a new
    '-T' commandline switch.
  + The Pre-Tagging map is now dynamically allocated and a new
    'pre_tag_map_entries' config directive allows to set the size
    of the map. Its default value (384) should be suitable for
    most common scenarios.
  ! Bugfix in nfprobe plugin: struct cb_ctxt was not initialized
    thus causing the application to exit prematurely (thinking it
    finished available memory). Thanks to Elio Eraseo for fixing
    the issue.
  ! Some misplaced defines were preventing 0.11.3 code to compile
    smoothly on OpenBSD boxes. Thanks to Dmitry Moshkov for fixing
    it.
  ! Bugfix in SQL handlers, MY_count_ip_proto_handler(): an array
    boundary was not properly checked and could cause the daemon
    to SEGV receiving certain packets. Thanks to Dmitry Frolov for 
    debugging and fixing the issue.
  ! NF_counters_renormalize_handler() renormalizes sampled NetFlow
    v5 flows. It now checks whether a positive Sampling Rate value
    is defined rather than looking for the Sampling Mode. It makes
    the feature working on Juniper routers. Thanks once again to
    Inge Bjornvall Arnesen.

0.11.3 -- 31-01-2007
  + 'aggregate_filter' directive now supports multiple pcap-style
    filters, comma separated. This, in turn, allows to bind up to
    128 filters to each activated plugin.
  + nfacctd and sfacctd turn-back time when restarting the daemon
    has been significantly improved by both creating new listening
    sockets with SO_REUSEADDR option and disassociating them first
    thing on receiving SIGINT signal. 
  + A new threaded version of pmacctd stream classification engine  
    is being introduced. Code status is experimental and disabled
    by default; it could be enabled by providing --enable-threads
    at configure time. Many thanks to Francois Deppierraz and Eneo
    Tecnologia for contributing this useful piece of code. 
  + A new 'flow_handling_threads' configuration directive allows to
    set the number of threads of the stream classification engine,
    by default 10.
  + A couple new '[ns]facctd_disable_checks' config directives aim
    to disable health checks over incoming NetFlow/sFlow streams
    (ie. in cases of non-standard vendor's implementations). Many 
    thanks to Andrey Chernomyrdin for his patch.
  ! sfv245_check_status() was running checks (ie. verify sequence
    numbers) using sender's IP address. More correctly, it has to
    look at the Agent Address field included in sFlow datagrams.
    Many thanks to Juraj Sucik for spotting the issue. 
  ! nfprobe plugin was not compiling properly in conjunction with
    --disable-l2 configure switch. Many thanks to Inge Bjornvall
    Arnesen for submitting the patch.
  ! sfacctd: fixed a bug which was preventing 'aggregate_filter'
    to match values properly in src_port, dst_port, ip proto and
    tos fields. Thanks to Chris Fletcher for spotting the issue.
  ! SQL cache: fixed a bug preventing safe actions to take place
    correctly. It has arisen in version 0.11.2 and hadn't severe
    impact.

0.11.2 -- 28-11-2006
  + 'sql_max_writers' configuration directive is being introduced:
    sets the maximum number of concurrent writer processes the SQL
    plugin can fire, allowing the daemon to degrade gracefully in
    case of major database unavailibility.
  + 'sql_history_since_epoch' is being introduced: enables the use
    of timestamps (stamp_inserted, stamp_updated) in the standard
    seconds since the Epoch format as an alternative to the default
    date-time format. 
  + 'sql_aggressive_classification' behaviour is changed: simpler
    more effective. It now operates by delaying cache-to-DB purge
    of unknown traffic streams - which would still have chances to
    be correctly classified - for a few 'sql_refresh_time' slots.
    The old mechanism was making use of negative UPDATE queries.
  + The way SQL writer processes are spawned by the SQL plugin has
    slightly changed in order to better exploit fork()'s copy-on-
    write behaviour: the writer now is mostly read-only while the
    plugin does most write operations before spawning the writer.
  ! The list of environment variables passed to the SQL triggers,
    'sql_trigger_exec', has been updated.
  ! Fixed a bug related to sequence number checks for NetFlow v5
    datagrams. Thanks very much to Peter Nixon for reporting it.

0.11.1 -- 25-10-2006
  + PostgreSQL plugin: 'sql_use_copy' configuration directive has
    been introduced; instructs the plugin to build non-UPDATE SQL
    queries using COPY (in place of INSERT). While providing same
    functionalities of INSERT, COPY is more efficient. It requires
    'sql_dont_try_update' to be enabled. Thanks to Arturas Lapiene
    for his support during the development.
  + nfprobe plugin: support for IPv4 ToS/DSCP, IPv6 CoS and MPLS
    top-most label has been introduced. 
  ! Some alignment issues concerning both pkt_extras structure and
    Core process to Plugins memory rings have been fixed. Daemons
    are now reported to be running ok on MIPS/SPARC architectures.
    Many thanks to Michal Krzysztofowicz for his strong support.
  ! sfprobe plugin: a maximum default limit of 256 bytes is set
    on packet payload copy when building Flow Samples in pmacctd
    (ie. if capturing full packets through libpcap, we don't want
    them to be entirely copied into sFlow datagrams).
  ! Sanity checks now take place when processing 'sql_refresh_time'
    values and error messages are thrown out.
  ! Fixes have been committed to IPv6 code in xflow_status.c as it
    was not compiling properly on both Solaris and IRIX.

0.11.0 -- 27-09-2006
  + NetFlow v5 sampling and renormalization are now supported:
    a) 'nfacctd' is able to renormalize bytes/packets counters and
    apply Pre-Tagging basing on the sampling rate specified in the
    datagram; b) 'sampling_rate' config key applies to 'nfprobe'
    plugin which is now able to generate sampling informations.
  + 'nfacctd' and 'sfacctd' are now able to give out informations
    about the status of active NetFlow/sFlow streams in terms of
    good/bad/missing datagrams. Whenever an anomaly happens (ie.
    missing or bad packets) a detailed message is logged; overral
    reports are logged by sending SIGUSR1 signals to the daemon. 
  + 'logfile' configuration directive is introduced: it allows to
    log directly to custom files. This adds to console and syslog
    logging options.
  ! Old renormalization structure, renorm_table, has been dropped;
    the new one, which applies to both NetFlow and sFlow, is tied
    into the brand new xflow_status_table structure. 
  ! When 'nfacctd_as_new' was not in use, NetFlow v5 src_as/dst_as
    values were erroneously swapped. Thanks to Thomas Stegbauer
    for reporting the bug.
  ! Incorrect timeout value for poll() has been fixed in 'sfprobe'
    plugin. It was leading the plugin to take too much resources.
  ! 'nfprobe' plugin was inserting jumps while generating sequence
    numbers.
  ! 'nfprobe' plugin behaviour in handling 'networks_file' content
    has been changed and now equals 'sfprobe': IP addresses which
    are not belonging to known networks/ASNs are no longer zeroed. 
  ! 'sfprobe' was not generating correct sample_pool values.
    
0.11.0rc3 -- 30-08-2006
  + 'sfprobe' plugin can now transport packet/flow classification
    tags inside sFlow v5 datagrams. Then, such tags can be read by
    the sFlow collector, sfacctd. 
  + 'sfprobe' plugin is able to encapsulate basic Extended Gateway
    informations (src_as, dst_as) into sFlow v5 datagrams starting
    from a Networks File - networks_file configuration directive.
  + 'nfprobe' now supports network data coming from libpcap/tcpdump
    style savefile ('pcap_savefile', -I). 
  + pmacctd is now able to capture packets from DLT_NULL, which is
    BSD loopback encapsulation link type. Thanks to Gert Burger for
    his support.
  + Sampling layer has been improved: it's now able to sample flows
    from NetFlow datagrams (not only packets arriving through sFlow
    or libpcap); 'sfprobe' sampling layer has been tied into this
    mechanism and as a result, 'sfprobe_sampling_rate' is now an
    alias for 'sampling_rate' and its default value is 1 (ie. no
    sampling). This change will benefit 'sfprobe' in terms of better
    efficiency.
  + A new 'pmacctd_flow_buffer_buckets' directive defines the number
    of buckets of the Flow Buffer. This value has to scale to higher
    power of 2 accordingly to the link traffic rate and is useful
    when packet classification is enabled. Many thanks for testing,
    debugging and support go to Steve Cliffe. 
  + A new 'sql_locking_style' directive allows to choose among two
    types of locking: "table" (default) and "row". More details are
    in the CONFIG-KEYS document. "row" locking has to be considered
    as experimental. Many thanks go to Aaron Glenn and Peter Nixon
    for their close support, work and thoughts.
  ! IPv6 support is now working; it was broken in 0.11.0rc2; thanks
    to Nigel Roberts for signalling and fixing the issue. 
  ! Fixed a few issues concerning the building system and related to 
    the introduction of some new subtrees. Thanks to Kirill Ponomarew
    and Peter Nixon for signalling them.
  ! Fixed some signal()-related issues when running the package under
    DragonflyBSD. Being fork of FreeBSD 4.x, it needs same cautions. 
    Thanks to Aaron Glenn for his support.

0.11.0rc2 -- 08-08-2006
  + 'nfprobe' plugin can now transport packet/flow classification
    tags inside NetFlow v9 datagrams, using custom field type 200.
    Then, such tags can be read by the NetFlow collector, nfacctd.  
  + 'nfprobe' plugin has now ability to select a Engine Type/Engine 
    ID through a newly introduced 'nfprobe_engine' config directive.
    It will mainly allow a collector to distinguish between distinct
    probe instances originating from the same IP address.
  + 'nfprobe' plugin now can automagically select different NetFlow
    v9 template IDs, useful when multiple 'nfprobe' plugins run as
    part of the same daemon instance. 
  + 'sfprobe' plugin is now able to redistribute NetFlow flows into
    sFlow samples. This adds to sFlow -> sFlow and libpcap -> sFlow.
  + A new data structure to pass extended data to specific plugins
    has been added. It is placed on the ring, next to pkt_data. It
    is meant to pass extra data to plugins and, same time, avoiding
    to inflate the main data structure.
  ! Wrong arguments were injected into a recently introduced Log()
    call in plugin_hooks.c; it's now fixed: under certain conditions,
    this was generating SEGV at startup while using 'sfprobe' plugin.  
  ! Updated documentation; examples and quickstart guides for using
    pmacct as both emitter and collector of NetFlow and sFlow have 
    been added.
  - Hooks to compile pmacct the no-mmap() style have been removed.

0.11.0rc1 -- 20-07-2006
  + pmacct DAEMONS ARE NOW ABLE TO CREATE AND EXPORT NETFLOW PACKETS:
    a new 'nfprobe' plugin is available and allows to create NetFlow
    v1/v5/v9 datagrams and export them to a IPv4/IPv6 collector. The
    work is based on softflowd 0.9.7 software. A set of configuration
    directives allows to tune timeouts (nfprobe_timeouts), cache size
    (nfprobe_maxflows), collector parameters (nfprobe_receiver), TTL
    value (nfprobe_hoplimit) and NetFlow version of the datagrams to
    be exported (nfprobe_version). Many thanks to Ivan A. Beveridge,
    Peter Nixon and Sven Anderson for their support and thoughts and
    to Damien Miller, author of softflowd.
  + pmacct DAEMONS ARE NOW ABLE TO CREATE AND EXPORT SFLOW PACKETS:  
    a new 'sfprobe' plugin is available and allows to create sFlow 
    v5 datagrams and export them to a IPv4 collector. The work is
    based on InMon sFlow Agent 5.6 software. A set of configuration
    directives allows to tune sampling rate (sfprobe_sampling_rate),
    sFlow agent IP address (sfprobe_agentip), collector parameters
    (sfprobe_receiver) and agentSubId value (sfprobe_agentsubid). 
    Many thanks to InMon for their software and Ivan A. Beveridge
    for his support.
  ! An incorrect pointer to the received packet was preventing Pre-
    Tagging filters to work correctly against DLT_LINUX_SLL links.
    Many thanks to Zhuang Yuyao for reporting the issue.
  ! Proper checks on protocol number were missing in pmacct client
    program, allowing to look further the bounds of the _protocols
    array. Many thanks to Denis N. Voituk for patching the issue.

0.10.3 -- 21-06-2006 
  + New Pre-Tagging key 'label': mark the rule with label's value.
    Labels don't need to be unique: when jumping, the first matching
    label wins. 
  + New Pre-Tagging key 'jeq': Jump on EQual. Jumps to the supplied
    label in case of rule match. Before jumping, the tagged flow is
    returned to active plugins, as it happens for any regular match
    (set return=false to change this). In case of multiple matches
    for a signle flow, plugins showing 'tag' key inside 'aggregate'
    directive will receive each tagged copy; plugins not receiving
    tags will still receive unique copy of the flow. 
    sFlow and NetFlow are usually uni-directional, ie. ingress-only
    or egress-only (to avoid duplicates). Meaningful application of
    JEQs is tagging flows two times: by incoming interface and by
    outgoing one. Only forward jumps are allowed. "next" is reserved
    label and causes to jump to the next rule. Many thanks to Aaron 
    Glenn for brainstormings about this point. 
  + New Pre-Tagging key 'return': if set to 'true' (which is default
    behaviour) returns the current packet/flow to active plugins, in
    case of match. If switched to 'false', it will prevent this to
    happen. It might be thought either as an extra filtering layer
    (bound to explicit Pre-Tagging rules) or (also in conjunction with
    'stack') as a way to add flexibility to JEQs.
  + New Pre-Tagging key 'stack': actually '+' (ie. sum symbol) is the
    unique supported value. This key makes sense only if JEQs are in
    use. When matching, accumulate IDs, using the specified operator/
    function. For example, usually <tag>=<currentID>. By setting
    'stack=+' you will be able to get <tag>=<previousID + currentID>.
  ! Pre-Tagging table now supports a maximum of 384 rules. Because
    of the newly introduced flow alteration features, tables are
    no longer internally re-ordered. However, IPv4 and IPv6 stacks
    are still segregated each other.

0.10.2 -- 16-05-2006
  + A new '-l' option is supported by pmacct client tool: it allows
    to enable locking of the memory table explicitely, when serving
    the requested operation. 
  + Pre-Tagging infrastructure is now featuring negations for almost
    all supported keys with the exclusion of id, ip and filter. To
    negate, the '-' (minus symbol) need to be prepended; eg.: id=X
    ip=Y in=-1 means tag with X, data received from Net/sFlow agent
    with IP address Y and not coming from interface 1.
  + pre_tag_filter config directive is now featuring same negation
    capabilities as Pre-Tagging infrastructure.
  + Q16 added to FAQS document: a sum of tips for running smoothly
    SQL tables. Many thanks to Wim Kerkhoff and Sven Anderson for
    bringing up the points. 

0.10.1 -- 18-04-2006
  + AS numbers and IP addresses are no more multiplexed into the same
    field. This ends the limitation of being unable to have both data
    types in the same table (which could be useful for troubleshooting
    purposes, for example). A new SQL table version, v6, is introduced
    in order to support this new data model in all SQL plugins.  
  ! Minor fixes to PostgreSQL table schemas, v2 to v5: a) the 'vlan'
    field was erroneously missing from primary keys, slowing down
    INSERT and UPDATE queries; b) primary keys were identified as
    'acct_pk', thus not allowing multiple tables of different version
    to share the same database; now constraint name is: 'acct_vX_pk',
    with X being the version number. Many thanks to Sven Anderson for
    catching the a)
  ! An alignment issue has been catched when the etheraddr_string() 
    gets called from count_src|dst_mac_handlers() in sql_handlers.c
    This seems to be closely connected to a similar trouble catched
    by Daniel Streicher on x86_64 recently.
  ! Fixed an issue with mask_elem() in server.c . Both src|dst_net
    primitives were not (positively, ie. copied back when required)
    masked.  

0.10.0 -- 22-03-2006
  + Collectors (ie. pmacctd) are now compiled exporting full Dynamic
    Symbol Table. This allows shared object (SO) classifiers to call
    routines included in the collector code. Moreover, a small set
    of library functions - specifically aimed to deal smoothly with
    the classifiers' table - are now included in the collector code:
    pmct_un|register(), pmct_find_first|last_free(), pmct_isfree(),
    pmct_get() and pmct_get_num_entries(). For further reading, take
    a look to README.developers document in classifiers tarball.
  + Classifiers table, which is the linked-list structure containing
    all the active classifiers (RE + SO), is now loaded into a shared
    memory segment, allowing plugins to keep updated about changes to
    the table. Furthermore, the table is now dynamically allocated at
    runtime, allowing an arbitrary number of classifiers to be loaded
    via the new 'classifier_table_num' configuration directive.
  + Pre-Tagging infrastructure adds two new primitives to tag network
    traffic: src_as and dst_as, the source and destination Autonomous
    System Number (ASN). In pmacctd they work against a Network Map
    ('networks_file' configuration directive). In nfacctd and sfacctd
    they work against both sFlow/NetFlow ASN fields and Network Maps.
    Many thanks to Aaron Glenn for his strong support.
  ! PostgreSQL plugin and pmpgplay no more make use of EXCLUSIVE LOCKS 
    whenever the sql_dont_try_update directive is activated. We assume
    there is no need for them in a INSERTs-only framework as integrity
    of data is still guaranteed by transactions. The patch has been
    contributed by Jamie Wilkinson, many thanks ! 
  ! Commandline switches and a configuration file should cohexist and
    the formers need to take precedence over the latter, if required.
    This is a rather standard (and definitely more flexible) approach;
    before this release they were mutual exclusive. Read UPGRADE notes
    at this propo. Thanks for the suggestion to Ivan A. Beveridge.
  ! Some glibc functions (noticeably syslog()) rely upon a rather non-
    standard "extern char *__progname" pointer. Now, its existence is
    properly checked at configuration time. On Linux, setproctitle()
    was causing plugin name/type to get cutted down in messages sent
    to the syslog facility. Thanks to Karl Latiss for his bug report.
  ! Solved a bug involving the load of IPv6 entries from Networks Maps.
    It was causing the count of such entries to be always zero.

0.10.0rc3 -- 01-03-2006
  + Aapplication layer (L7) classification capabilities of pmacctd have
    been improved: shared object (SO) classifiers have been introduced;
    they are loaded runtime through dlopen(). pmacct offers them support
    for contexts (informations gathered - by the same classifier - from
    previous packets either in the same uni-directional flow or in the
    reverse one), private memory areas and lower layer header pointers,
    resulting in extra flexibility. Some examples can be found at the
    webpage: http://www.ba.cnr.it/~paolo/pmacct/classification/ 
  + 'classifier_tentatives' configuration key has been added: it allows
    to customize the number of tentatives made in order to classify a
    flow. The default number is five, which has proven to be ok but for
    certain types of classification it might result restrictive.
  + 'pmacctd_conntrack_buffer_size' configuration key has been added: it
    (intuitively) defines the size for the connection tracking buffer.
  + Support for Token Ring (IEEE 802.5) interfaces has been introduced
    in pmacctd. Many thanks to Flavio Piccolo for his strong support. 
  + 'savefile_wait' (-W commandline) configuration key has been added: if 
    set to true causes pmacctd to not return but wait to be killed after
    being finished with the supplied savefile. Useful when pushing data
    from a tcpdump/ethereal tracefile into a memory table (ie. to build
    graphs).
  ! An erroneous replacement of dst with src in mask_elem() was causing
    queries like "pmacct -c dst_host -M|-N <IP address>" to return zero
    counters. Thanks to Ryan Sleevi for signalling the weird behaviour. 
  ! Management of the connection tracking buffer has been changed: now,
    a successful search frees the matched entry instead of moving it in
    a chain of stale entries, available for quick reuse.
  ! Error logging of SQL plugins has been somewhat improved: now, error
    messages returned by the SQL software are forwarded to sql_db_error() 
    This will definitely allow to exit from the obscure crypticism of
    some generic error strings.

0.10.0rc2 -- 14-02-2006
  + CONNECTION TRACKING modules has been introduced into pmacctd: they are
    C routines that hint IP address/port couples for upcoming data streams
    as signalled by one of the parties into the control channel whenever
    is not possible to go with a RE classificator. Conntrack modules for
    FTP, SIP and RTSP protocols are included.
  + 'pidfile' directive way of work has been improved: firstly, whenever 
    a collector shuts down nicely, it now removes its pidfile. Secondly,
    active plugins now create a pidfile too: it takes the following form:
    <pidfile>-<plugin type>.<plugin name>. Thanks to Ivan A. Beveridge
    for sharing his thoughts at this propo.
  ! Minor fixes to the classification engine: TCP packets with no payload
    are not considered useful classification tentatives; a new flow can
    inherit the class of his reverse flow whenever it's still reasonably
    valid.
  ! Solved a segmentation fault issue affecting the classificator engine,
    whenever the 'snaplen' directive was not specified. Thanks to Flavio
    Piccolo for signalling it.
  ! Fixed a bug in the PostgreSQL plugin: it appeared in 0.10.0rc1 and was
    uniquely related to the newly introduced negative UPDATE SQL query.  
  ! INTERNALS has been updated with few notes about the new classification
    and connection tracking features. 

0.10.0rc1 -- 24-01-2006
  + PACKET CLASSIFICATION capabilities have been introduced into pmacctd:
    the implemented approach is fully extensible: classification patterns
    are based on regular expressions (RE), human-readable, must be placed
    into a common directory and have a .pat file extension. Many patterns
    for widespread protocols are available at L7-filter project homepage. 
    To support this feature, a new 'classifiers' configuration directive
    has been added. It expects full path to a spool directory containing
    the patterns.
  + A new 'sql_aggressive_classification' directive has been added aswell:
    it allows to move unclassified packets even in the case they are no
    more cached by the SQL plugin. This aggressive policy works by firing
    negative UPDATE SQL queries that, whenever successful, are followed
    by positive ones charging the extra packets to their final class.
  ! Input and Output interface fields (Pre-Tagging) have been set to be  
    32 bits wide. While NetFlow is ok with 16 bits, some sFlow agents are
    used to bigger integer values in order to identify their interfaces.
    The fix is courtesy of Aaron Glenn. Thank you. 
  ! Flow filtering troubles have been noticed while handling MPLS-tagged
    flows inside NetFlow v9 datagrams. Thanks to Nitzan Tzelniker for his
    cooperation in solving the issue.
  ! A new exit_all() routine now handles nicely fatal errors detected by
    the Core Process, after plugins creation. It avoids leaving orphan
    plugins after the Core Process shutdown.

0.9.6 -- 27-Dec-2005
  + Support for 'sql_multi_values' has been introduced into the new SQLite
    3.x plugin. It allows to chain multiple INSERT queries into a single
    SQL statement. The idea is that inserting many rows at the same time
    is much faster than using separate single-row statements.
  ! MySQL plugin fix: AS numbers were sent to the database unquoted while 
    the corresponding field was declared as CHAR. By correctly wrapping AS
    numbers, a major performance increase (expecially when UPDATE queries
    are spawned) has been confirmed. Many thanks to Inge Bjørnvall Arnesen
    for discovering, signalling and solving the issue.
  ! MySQL plugin fix: multi-values INSERT queries have been optimized by
    pushing out of the queue purging loop the proper handling for the EOQ
    event.
  ! The introduction of the intermidiate SQL layer in the 0.9.5 version
    choked the dynamic SQL table creation capability. This has been fixed.
    Thanks to Vitalij Brajchuk for promptly signalling the issue. 
  ! The 'pidfile' configuration key has got incorrectly disabled in both 
    nfacctd and sfacctd. Thanks to Aaron Glenn for signalling the issue.
  ! The 'daemonize' (-D) configuration key was incorrectly disabling the
    signal handlers from the Core Process once backgrounded. As a result
    the daemon was not listening for incoming SIGINTs. Again, many thanks 
    go to Aaron Glenn.
 
0.9.5 -- 07-Dec-2005
  + PMACCT OPENS TO SQLITE 3.x: a fully featured SQLite, version 3.x only,
    plugin has been introduced; SQLite is a small C library that implements
    a self-contained, embeddable, zero-configuration SQL (almost all SQL92)
    database engine. The plugin is LOCK-based and supports the "recovery
    mode" via an alternate database action. Expecially suitable for tiny
    and embedded environments. The plugin can be fired using the keyword
    'sqlite3'. See CONFIG-KEYS and EXAMPLES for further informations. 
  + A new SQL layer - common to MySQL, PostgreSQL and SQLite plugins - has
    been introduced. It's largely callback-based and results in a major
    architectural change: it sits below the specific SQL code (facing the
    Core Process's abstraction layer) and will (hopefully) help in reducing
    potential bugs and will allow for a quick implementation of new SQL
    plugins.
  ! A bug concerning the setup of insert callback functions for summed (in
    + out) IPv6 traffic has been fixed. The issue was affecting all SQL
    plugins. 
  ! A bug concerning the handling of MPLS labels has been fixed in pmacctd.
    Many thanks to Gregoire Tourres and Frontier Online for their support.

0.9.4p1 -- 14-Nov-2005
  ! Minor bugfix in pretag.c: a wrongly placed memcpy() was preventing the
    code to be compiled by gcc 2.x . Many thanks to Kirill Ponomarew and
    Kris Kennaway for signalling the issue.
  ! Fixed an alignment issue revealed in the query_header structure; it has
    been noticed only under some circumstances: '--enable-64bit' enabled,
    64bit platform and gcc 3.x . Many thanks to Aaron Glenn for his strong
    support in solving the issue.

0.9.4 -- 08-Nov-2005
  + Hot map reload has been introduced. Maps now can be modified and then
    reloaded without having to stop the daemon. SIGUSR2 has been reserved for
    this use. The feature applies to Pre-Tagging map (pre_tag_map), Networks
    map (networks_file) and Ports map (ports_file). It is enabled by default
    and might be disabled via the new 'refresh_maps' configuration directive.
    Further details are in CONFIG-KEYS. 
  ! Some major issues have been solved in the processing of libpcap-format
    savefiles. Some output inconsistencies were caused by a corruption of the
    pcap file handler; bufferization is now enabled by default and the last
    buffer is correctly processed. Many thanks go to Amir Plivatsky for his
    strong support.  
  ! 'sql_table_schema' directive: in read_SQLquery_from_file() the strchr()
    has been replaced by strrchr() allowing to chain more SQL statements as
    part of the SQL table creation. This results useful, for example, to do
    CREATE INDEX after CREATE TABLE. The patch is courtesy of Dmitriy Nikulin. 
  ! SIGTERM signal is now handled properly to ensure a better compatibility
    of all pmacct daemons under the daemontools framework. The patch is
    courtesy of David C. Maple.
  ! Memory plugin: some issues caused by the mix of not compatible compilation
    parameters have been fixed. Now the pmacct client now correctly returns a
    warning message if: counters are of different size (32bit vs 64bit) or IP
    addresses are of different size (IPv4-only vs IPv6-enabled packages). 
  ! Print plugin, few bugfixes: the handling of the data ring shared with the
    Core Process was not optimal; it has been rewritten. P_exit() routine was
    not correctly clearing cached data. 

0.9.3 -- 11-Oct-2005
  + IPv4/IPv6 multicast support has been introduced in the NetFlow (nfacctd)
    and the sFlow (sfacctd) daemons. A maximum of 20 multicast groups may be
    joined by a single daemon instance. Groups can be defined by using the two
    sister configuration keys: nfacctd_mcast_groups and sfacctd_mcast_groups. 
  + sfacctd: a new 'sfacctd_renormalize' config key allows to automatically 
    renormalize byte/packet counters value basing on informations acquired
    from the sFlow datagram. In particular, it allows to deal with scenarios
    in which multiple interfaces have been configured at different sampling
    rates. It also calculates an effective sampling rate which could differ
    from the configured one - expecially at high rates - because of various
    losses. Such estimated rate is then used for renormalization purposes.
    Many thanks go to Arnaud De-Bermingham and Ovanet for the strong support
    offered during the development.
  + sfacctd: a new 'sampling_rate' keyword is supported into the Pre-Tagging 
    layer. It allows to tag aggregates - generated from sFlow datagrams - on
    a sampling rate basis. 
  + setproctitle() calls have been introduced (quite conservatively) and are
    actually supported on Linux and BSDs. The process title is rewritten in
    the aim of giving the user more informations about the running processes
    (that is, it's not intended to be just a cosmetic stuff).
  ! sql_preprocess tier was suffering a bug: actions (eg. usrf, adjb), even
    if defined, were totally ignored if no checks were defined aswell. Many
    thanks to Draschl Clemens for signalling the issue.
  ! Some minor bugs have been catched around sfacctd and fixed accordingly.
    Again, many thanks to Arnaud De-Bermingham.

0.9.2 -- 14-Sep-2005
  + A new 'usrf' keyword is now supported into the 'sql_preprocess' tier: it
    allows to apply a generic uniform renormalization factor to counters. Its
    use is particularly suitable for use in conjunction with uniform sampling
    methods (for example simple random - e.g. sFlow, 'sampling_rate' directive
    or simple systematic - e.g. sampled NetFlow by Cisco and Juniper). 
  + A new 'adjb' keyword is now supported into the 'sql_preprocess' tier: it
    allows to add (or subtract in case of negative value) 'adjb' bytes to the
    bytes counter. This comes useful when fixed lower (link, llc, etc.) layer 
    sizes need to be included into the bytes counter (as explained by the Q7
    in the updated FAQS document).
  + A new '--enable-64bit' configuration switch allows to compile the package
    with byte/packet/flow counters of 64bit (instead of the usual 32bit ones). 
  ! The sampling algorithm endorsed by the 'sampling_rate' feature has been 
    enhanced to a simple randomic one (it was a simple systematic). 
  ! Some static memory structures are now declared as constants allowing to
    save memory space (given the multi-process architecture) and offering an
    overral better efficiency. The patch is courtesy of Andreas Mohr. Thanks.
  ! Some noisy compiler warnings have been troubleshooted along with some minor
    code cleanups; the contribution is from Jamie Wilkinson. Thanks.
  ! Some unaligned pointer issues have been solved. 

0.9.1 -- 16-Aug-2005
  + Probabilistic, flow size dependent sampling has been introduced into the
    'sql_preprocess' tier via the new 'fss' keyword: it is computed against
    the bytes counter and returns renormalized results. Aggregates which have
    collected more than the 'fss' threshold in the last time window are
    sampled. Those under the threshold are sampled with probability p(bytes). 
    For further details read the CONFIG-KEYS and the paper:
    - N.G. Duffield, C. Lund, M. Thorup, "Charging from sampled network usage"
      http://www.research.att.com/~duffield/pubs/DLT01-usage.pdf
  + Probabilistic sampling under hard resource constraints has been introduced
    into the 'sql_preprocess' tier via the new 'fsrc' keyword: it is computed
    against the bytes counter and returns renormalized results. The method
    selects only 'fsrc' flows from the set of the flows collected during the
    last time window, providing an unbiasied estimate of the real bytes counter.
    For further details read the CONFIG-KEYS and the paper:
    - N.G. Duffield, C. Lund, M. Thorup, "Flow Sampling Under Hard Resource Constraints"
      http://www.research.att.com/~duffield/pubs/DLT03-constrained.pdf
  + A new 'networks_mask' configuration directive has been introduced: it
    allows to specify a network mask - in bits - to be applied apply to src_net
    and dst_net primitives. The mask is applied before evaluating the content of 
    'networks_file' (if any).
  + Added a new signal handler for SIGUSR1 in pmacctd: a 'killall -USR1 pmacctd'
    now returns a few statistics via either console or syslog; the syslog level
    reserved for such purpose is the NOTICE. 
  ! sfacctd: an issue regarding non-IP packets has been fixed: some of them
    (mainly ARPs) were incorrectly reported. Now they are properly filtered out. 
  ! A minor memory leak has been fixed; it was affecting running instances of
    pmacctd, nfacctd and sfacctd with multiple plugins attached. Now resources
    are properly recollected. 

0.9.0 -- 25-Jul-2005
  + PMACCT OPENS TO sFlow: support for the sFlow v2/v4/v5 protocol has been
    introduced and a new daemon 'sfacctd' has been added. The implementation
    includes support for BGP, MPLS, VLANs, IPv4, IPv6 along with packet tagging,
    filtering and aggregation capabilities. 'sfacctd' makes use of Flow Samples
    exported by a sFlow agent while Counter Samples are skipped and the MIB is
    ignored. All actually supported backends are available for storage: MySQL,
    PostgreSQL and In-Memory tables. http://www.sflow.org/products/network.php
    lists the network equipments supporting the sFlow protocol.
  + A new commandline option '-L' is now supported by 'nfacctd' and 'sfacctd';
    it allows to specify an IPv4/IPv6 address where to bind the daemon. It is
    the equivalent for the 'nfacctd_ip' and 'sfacctd_ip' configuration directives. 
  ! The NetFlow v9 MPLS stack handler has been fixed; it now also sticks the BoS
    bit (Bottom of the Stack) to the last processed label. This makes the flow
    compliant to BPF filters compiled by the newly released libpcap 0.9.3. 
  ! Some Tru64 compilation issues related to the ip_flow.[c|h] files have been
    solved. 
  ! Some configuration tests have been added; u_intXX_t definitions are tested
    and fixed (whenever possible, ie. uintXX_t types are available). Particularly
    useful on Solaris and IRIX platforms.
  ! Configuration hints for MySQL headers have been enhanced. This will ease the
    compilation of pmacct against MySQL library either from a precompiled binary
    distribution or from the FreeBSD ports. Many hhanks for the bug report go to
    John Von Essen. 
  ! NetFlow v8 source/destination AS handlers have been fixed. 

0.8.8 -- 27-Jun-2005
  + Added IP flows support in pmacctd (release 0.8.5 has seen its introduction
    in nfacctd) for both IPv4 and IPv6 handlers. To enable flows accounting,
    the 'aggregate' directive now supports a new 'flows' keyword. The SQL table
    v4 has to be used in order to support this feature in both SQL plugins.
  + A new 'sum_mac' aggregation method has been added (this is in addition to
    the already consolidated ones: 'sum_host', 'sum_net', 'sum_as', 'sum_port').
    Sum is intended to be the total traffic (inbound traffic summed to outbound
    one) produced by a specific MAC address.
  + Two new configuration directives have been introduced in order to set an 
    upper bound to the growth of the fragment (default: 4Mb) and flow (default:
    16Mb) buffers: 'pmacctd_frag_buffer_size', 'pmacctd_flows_buffer_size'.
  + A new configuration directive 'pmacctd_flow_lifetime' has been added and 
    defines how long a flow could remain inactive (ie. no packets belonging to
    such flow are received) before considering it expired (default: 60 secs).
    This is part of the pmacctd IP flows support.
  + Console/syslog feedbacks about either generic errors or malformed packets
    have been greatly enhanced. Along with the cause of the message, now any
    generated message contains either the plugin name/type or the configuration
    file that is causing it. 
  ! nfacctd: when IPv6 is enabled (on non-BSD systems) the daemon now listens
    by default on a IPv6 socket getting rid of the v4-in-v6 mapping feature which
    helps in receiving NetFlow datagrams from both IPv4 and IPv6 agents. A new
    configure script switch --enable-v4-mapped is aimed to turn manually on/off
    the feature.
  ! Fixed an issue with the SIGCHLD handling routine on FreeBSD 4.x systems. It
    was causing the sudden creation of zombie processes because of the not correct
    retirement of exited childs. Many thanks for his bug report and strong support
    go to John Von Essen.
  ! Fixed an endianess issue regarding Solaris/x86 platforms caused by not proper
    preprocessor tests. Many thanks to Imre Csatlos for his bug report. 
  ! Fixed the default schema for the PostgreSQL table v4. The 'flows' field was
    lacking of the 'DEFAULT 0' modifier; it was causing some troubles expecially
    when such tables were used in conjunction with the 'sql_optimize_clauses'
    directive. Many thanks for his bug report and strong support go to Anik Rahman. 

0.8.7 -- 14-Jun-2005
  + pmacctd: MPLS support has been introduced. MPLS (on ethernet and ppp links)
    and MPLS-over-VLAN (ethernet only) packets are now supported and passed to
    upper layer routines. Filtering and tagging (Pre-Tagging) packets basing on
    MPLS labels is also supported. Recent libpcap is required (ie, CVS versions
    >= 06-06-2005 are highly adviceable because of the support for MPLS label
    hierarchies like "mpls 100000 and mpls 1024" that will match packets with 
    an outer label of 100000 and an inner label of 1024).  
  + nfacctd: VLAN and MAC addresses support for NetFlow v9 has been introduced.
    Each of them is mapped to its respective primitive (vlan, src_mac, dst_mac);
    filtering and tagging (Pre-Tagging) IPv4/IPv6 flows basing on them is also
    supported.
  + nfacctd: filtering and tagging (Pre-Tagging) IPv4/IPv6 flows basing on MPLS
    labels has been introduced (read the above notes regarding libpcap version
    requirements).
  + A new packet capturing size option has been added to pmacctd ('snaplen'
    configuration directive; '-L' commandline). It allows to change the default
    portion of the packet captured by the daemon. It results useful to cope 
    with not fixed protocol stacks (ie, the MPLS stack).
  + pmacctd: CHDLC support has been introduced. IPv4, IPv6 and MPLS packets are
    supported on this link layer protocol.
  ! Cleanups have been added to the NetFlow packet processing cycle. They are
    mainly aimed to ensure that no stale data is read from circular buffers 
    when processing NetFlow v8/v9 packets. 
  ! The NetFlow v9 VLAN handling routine was missing a ntohs() call, resulting
    in an ncorrect VLAN id on little endian architectures.  
  ! ether_aton()/ether_ntoa() routines were generating segmentation faults on
    x86_64 architectures. They have been replaced by a new handmade couple:
    etheraddr_string()/string_etheraddr(). Many thanks to Daniel Streicher for
    the bug report.  

0.8.6 -- 23-May-2005
  + The support for dynamic SQL tables has been introduced through the use of
    the following variables in the 'sql_table' directive: %d (the day of the
    month), %H (hours using an 24 hours clock), %m (month number), %M (minutes),
    %w (the day of the week as a decimal number), %W (week number in the current
    year) and %Y (the current year). This enables, for example, substitutions
    like the following ones:
      'acct_v4_%Y%m%d_%H%M'	==> 	'acct_v4_20050519_1500'
      'acct_v4_%w'		==> 	'acct_v4_05'
  + A new 'sql_table_schema' configuration directive has been added in order
    to allow the automatic creation of dynamic tables. It expects as value the
    full pathname to a file containing the schema to be used for table creation.
    An  example of the schema follows:
      CREATE TABLE acct_v4_%Y%m%d_%H%M (
        ... PostgreSQL/MySQL specific schema ...
      );
  + Support for MySQL multi-values INSERT clauses has been added. Inserting
    many rows in a single shot has proven to be much faster (many times faster
    in some cases) than using separate single INSERT statements. A new
    'sql_multi_values' configuration directive has been added to enable this
    feature. Its value is intended to be the size (in bytes) of the multi-values
    buffer. Out of the box, MySQL >= 4.0.x supports values up to 1024000 (1Mb).
    Because it does not require any changes on server side, people using MySQL
    are strongly encouraged to give it a try.
  + A new '--disable-l2' configure option has been added. It is aimed to compile 
    pmacct without support for Layer-2 stuff: MAC addresses and VLANs. This
    option - along with some more optimizations to memory structures done in this
    same release - have produced memory savings up to 25% compared to previous
    versions. 
  ! Recovery code for PostgreSQL plugin has been slightly revised and fixed.

0.8.5 -- 04-May-2005
  + Added IP flows counter support in nfacctd, the NetFlow accounting daemon,
    in addition to the packets and bytes ones. To enable flows accounting, the
    'aggregate' directive now supports a new 'flows' keyword. A new SQL table
    version, v4, has been also introduced to support this feature in both SQL
    plugins. 
  + 'sql_preprocess' directive have been strongly improved by the addition of
    new keywords to handle thresholds. This preprocessing feature is aimed to 
    process aggregates (via a comma-separated list of conditionals and checks)
    before they are pulled to the DB, thus resulting in a powerful selection
    tier; whether the check is meet, the aggregate goes on its way to the DB;
    the new thresholds are: maxp (maximum number of packets), maxb (maximum bytes
    transferred), minf/maxf (minimum/maximum number of flows), minbpp/maxbbp
    (minimum/maximum bytes per packet average value), minppf/maxppf (minimum/
    maximum packets per flow average value). 
  + Added a new 'sql_preprocess_type' directive; the values allowed are 'any'
    or 'all', with 'any' as default value. It is intended to be the connective
    whether 'sql_preprocess' contains multiple checks. 'any' requires that an
    aggregate has to match just one of the checks in order to be valid; 'all'
    requires a match against all of the checks instead.
  + Added the ability to instruct a BPF filter against the ToS field of a NetFlow
    packet.  
  ! Minor optimizations on the 'sql_preprocess' handler chain.

0.8.4 -- 14-Apr-2005
  + Added support for NetFlow v7/v8. The Version 7 (v7) format is exclusively
    supported by Cisco Catalyst series switches equipped with a NetFlow feature
    card (NFFC). v7 is not compatible with Cisco routers. The Version 8 (v8)
    format adds (with respect to older v5/v7 versions) router-based aggregation
    schemes.
  + Added the chance to tag packets basing on NetFlow v8 aggregation type field. 
    As the keyword suggests, it will work successfully just when processing  
    NetFlow v8 packets. Useful to split - backend side - data per aggregation
    type. 
  + pmacct client now is able to ask for the '0' (that is, untagged packets) tag
    value. Moreover, all 'sum' aggregations (sum_host, sum_net, sum_as, sum_port)
    can now be associated with both Pre/Post-Tagging.  
  ! Fixed a serious memory leak located in the routines for handling NetFlow v9
    templates. While the bug was needing certain conditions to manifest, anyone
    using NetFlow v9 is strongly encouraged to upgrade to this version. All
    previous versions were affected.
  ! Some gcc4 compliance issues have been solved. The source code is known to
    work fine on amd64 architectures. Thanks very much to Marcelo Goes for his
    patch.
  ! Engine Type/Engine ID fields were not correctly evaluated when using NetFlow
    v5 and Pre-Tagging. The issue has been fixed.
  ! Long comments in the Ports Definition File were causing some incorrect error
    messages. However it seems the file were processed correctly. Thanks to Bruno
    Mattarollo for signalling the issue.
  ! Minor fix to plugins hooking code. The reception of sparse SIGCHLD signals 
    were causing the poll() to return. The impact was null. The issue has been
    fixed by ignoring such signals. 

0.8.3 -- 29-Mar-2005
  + Pre-Tagging capabilities have been further enhanced: captured traffic can
    be now marked basing on the NetFlow nexthop/BGP nexthop fields. While the
    old NetFlow versions (v1, v5) carry an unique 'nexthop' field, NetFlow v9
    supports them into two distinguished fields. 
  + Packet/flows tagging is now explicit, gaining more flexibility: a new 'tag'
    keyword has been added to the 'aggregate' directive. It causes the traffic
    to be actually marked; the 'pre_tag_map' and 'post_tag' directives now just
    evaluate the tag to be assigned. Read further details about this topic in
    the UPGRADE document.
  + The 'pre_tag_filter' directive now accepts 0 (zero) as valid value: we have 
    to remember that zero is not a valid tag; hence, its support allows to split
    or filter untagged traffic from tagged one. 
  + Documentation has been expanded: a new FAQS entry now describes few and easy
    tweaks needed to replace the bytes counter type from u_int32_t to u_int64_t
    throughout the code (provided that the OS supports this type); it's useful
    in conjunction with the In-Memory plugin while exposed to very sustained
    traffic loads. A new FAQS entry describes the first efforts aimed to integrate
    pmacctd with popular flow-tools software by the way of the flow-export tool.
    A new UPGRADE document has been also created. 
  ! pmacct client was handling counters returned by the '-N' switch as signed
    integers, which is not correct. The issue has been fixed. Many thanks to
    Tobias Bengtsson for signalling it. 
  ! Two new routines file_lock()/file_unlock() have replaced the flock() calls
    because they were preventing the pmacct code to compile on Solaris. Basing
    over hints collected at configure time, the routines enable either the flock()
    or fcntl() code. Many thanks to Jan Baumann for signalling and solving the
    issue. 


0.8.2 -- 08-Mar-2005
  + Pre-Tagging capabilities have been enhanced: now, a Pre Tag Map allows to
    mark either packets or flows basing on the outcome of a BPF filter. Because
    of this new feature, Pre-tagging has been introduced in 'pmacctd' too.
    Pre-tagging was already allowing 'nfacctd' to translate some NetFlow packet
    fields (exporting agent IP address, Input/Output interface, Engine type and
    Engine ID) into an ID (also referred as 'tag'), a small number in the range
    1-65535. 
  + A new 'pmacctd_force_frag_handling' configuration directive has been added;
    it aims to support 'pmacctd' Pre-Tagging operations: whether the BPF filter
    requires tag assignation based on transport layer primitives (e.g. src port
    or dst port), this directive ensures the right tag is stamped to fragmented
    traffic too. 
  + Pre Tag filtering (which can be enabled via 'pre_tag_filter' configuration
    directive) allows to filter aggregates basing on the previously evaluated 
    ID: whether it matches with at least one of the filter values, the aggregate
    is delivered to the plugin. It has been enhanced by allowing to assign more 
    tags to a specific plugin. 
  + pmacctd: a new feature to read libpcap savefiles has been added; it can be
    enabled either via the 'pcap_savefile' configuration directive or the '-I'
    commandline switch. Files need to be already closed and correctly finalized
    in order to be read successfully. Many thanks to Rafael Portillo for proposing
    the idea.
  + pmacct client tool supports a new 'tag' keyword as value for the '-c' switch:
    it allows to query the daemon requesting a match against aggregate tags.  
  + pmacct client: the behaviour of the '-N' switch (which makes the client to
    return a counter onto the screen suitable for data injection in tools like MRTG,
    Cacti, RRDtool, etc.), has been enhanced: it was already allowing to ask data
    from the daemon but basing only on exact matches. This concept has now extended,
    adding both wildcarding of specific fields and partial matches. Furthermore,
    when multiple requests are encapsulated into a single query, their results are
    by default splitted (that is, each request has its result); a newly introduced
    '-S' switch now allows to sum multiple results into a single counter.
  ! Bugfix: proper checks for the existence of a 'pre_tag_map' file were bypassed
    under certain conditions; however, this erroneous behaviour was not causing any
    serious issue. The correct behaviour is to quit and report the problem to the
    user. 
  ! The sampling rate algorithm has been fixed from a minor issue: it was returning
    not expected results when 'sampling_rate: 1'. It now works as expected. Thanks
    to David C. Maple for his extensive support in gaining a better understanding
    of the problem. 

0.8.1p1 -- 22-Feb-2005
  ! 'sum_host' and 'sum_net' compound primitives have been fixed in order to
    work with IPv6 addresses.
  ! In-Memory Plugin: client queries spotted with both '-r' (reset counters) and
    '-N' (exact match, print counters only) switches enabled were causing the
    daemon to crash whether no entries were found. The problem has been fixed.
    Many thanks to Zach Chambers for signalling the issue.
  ! In-Memory Plugin: client queries spotted with either '-M' or '-N' switches
    enabled were failing to match actual data when either 'sum_host', 'sum_net' 
    or 'sum_as' primitives were in use. The issue has been fixed. 
  ! The modulo function applied to NetFlow v9 Template Cache has been enhanced
    in order to deal correctly with export agents having an IPv6 address. 
  ! Networks/AS definition file: a new check has been added in order to verify 
    whether network prefix/network mask pairs are compatible: if they are not,
    the mask is applied to the prefix. 
  ! Documentation has been expanded and revised.

0.8.1 -- 25-Jan-2005
  + Accounting and aggregation over DSCP, IPv4 ToS field and IPv6 traffic class
    field have been introduced ('aggregate' directive, 'tos' value): these fields
    are actually widely used to implement Layer-3 QoS policies by defining new
    classes of service (most noticeably 'Less than Best Effort' and 'Premium IP').
    MySQL and PostgreSQL tables v3 (third version) have been introduced (they
    contain an additional 4-bytes 'tos' field) to support the new Layer-3 QoS
    accounting.
  + nfacctd core process has been slightly optimized: each flow is encapsulated
    (thus, copied field-by-field) into a BPF-suitable structure only if one or
    more plugins actually require BPF filtering ('aggregate_filter' directive).
    Otherwise, if either filtering is not required or all requested filters fail
    to compile, the copy is skipped. 
  + 'pmacct', pmacct client tool: '-e' commandline option (which meaning is:
    full memory table erase) now might be supplied in conjunction with other
    options (thus avoiding the short time delays involved by two consecutive
    queries, ask-then-erase, which may also lead to small losses).
    The new implemented mechanism works as follow: queries over actual data
    (if any) are served before; the table is locked, new aggregates are queued
    until the erasure finishes (it may take seconds if the table is large enough);
    the table is unlocked; the queue of aggregates is processed and all normal
    operations are resumed. Many thanks to Piotr Gackiewicz for the valuable
    exchange of ideas.
  ! Bug fixed in nfacctd: source and destination AS numbers were incorrectly
    read from NetFlow packets. Thanks to Piotr Gackiewicz for his support. 
  ! Bug fixed in pmacct client: while retrieving the whole table content was
    displaying espected data, asking just for 'dst_as' field was resulting in
    no results instead. Thanks, once more, to Piotr Gackiewicz. 


0.8.0 -- 12-Jan-2005
  + PMACCT OPENS TO IPv6: IPv6 support has been introduced in both 'pmacctd' 
    and 'nfacctd' daemons. Because it requires larger memory structures to
    store its addresses, IPv6 support has been disabled by default. It could
    be enabled at configure time via '--enable-ipv6' switch. All filtering,
    tagging and mapping functions already support IPv6 addresses. Some notes 
    about IPv6 and SQL table schema have been dropped into README.IPv6 file,
    sql section of the tarball. 
  + PMACCT OPENS TO NetFlow v9: support for the template-based Cisco NetFlow
    v9 export protocol has been added. NetFlow v1/v5 were already supported.
    'nfacctd' may now be bound to an IPv6 interface and is able to read both
    IPv4 and IPv6 data flowsets. A single 'nfacctd' instance may read flows
    of different versions and coming from multiple exporting agents. Source
    and destination MAC addresses and VLAN tags are supported in addition to
    the primitives already supported in v1/v5 (source/destination IP addresses,
    AS, ports and IP protocol). Templates are cached and refreshed as soon as
    they are resent by the exporting agent.  
  + Pre Tag map ('pre_tag_map' configuration key), which allows to assign a
    small integer (ID) to an incoming flow basing on NetFlow auxiliar data,
    now may apply tags basing also over Engine Type (it provides uniqueness
    with respect to the routing engine on the exporting device) and Engine
    ID (it provides uniqueness with respect to the particular line card or
    VIP on the exporting device) fields. Incoming and Outcoming interfaces
    were already supported. See 'pretag.map.example' into tarball examples
    section and CONFIG-KEYS document for further details. 
  + Raw protocol (DLT_RAW) routine has been added; it usually allows to read
    data from tunnels and sitX devices (used for IPv6-in-IPv4 encapsulation).
  + Some tests for architecture endianess, CPU type and MMU unaligned memory
    access capability have been added. A small and rough (yes, they work the
    hard way) set of unaligned copy functions have been added. They are aimed
    to be introduced through the code, however first tests over MIPS R10000
    and Alpha EV67 (21264A) have shown positive results.  
  ! PPPoE and VLAN layer handling routines have been slightly revised for some
    additional checks. 
  ! Given the fairly good portability reported from the mmap() code introduced
    through the whole 0.7.x development stage, the use of shared memory segments
    is now enabled by default. The configure switch '--enable-mmap' has been
    replaced by '--disable-mmap'. 
  ! 'pmacct' client tool: because of the IPv6 addresses introduction, separator 
    character for multiple queries (commandline) have been changed to from
    ':' to ';'. 
  ! 'nfacctd': '-F' commandline switch was listed into available options list,
    but getopt() stanza was missing, thus returning an invalid option message.
    Thanks to Chris Koutras for his support in fixing the issue.
  ! Some variable assignations were causing lvalue errors with gcc 4.0. Thanks
    to Andreas Jochens for his support in signalling and solving the problem. 

0.7.9 -- 21-Dec-2004
  + A new data pre-processor has been introduced in both SQL plugins: it
    allows to filter out data (via conditionals, checks and actions) during
    a cache-to-DB purging event, before building SQL queries; this way, for
    example, aggregates which have accounted just a few packets or bytes may
    be either discarded or saved through the recovery mechanism (if enabled).
    The small set of preprocessing directives is reported into CONFIG-KEYS
    document. 
  + Some new environment variables are now available when firing a trigger
    from SQL plugins: $EFFECTIVE_ELEM_NUMBER reports the effective number
    of aggregates (that is, excluding those filtered out at preprocessing
    time) encapsulated in SQL queries; $TOTAL_ELEM_NUMBER reports the total
    number of aggregates instead.
    $INSERT_QUERIES_NUMBER and $UPDATE_QUERIES_NUMBER returns respectively
    the number of aggregates being successfully encapsulated into INSERT
    and UPDATE queries. $ELAPSED_TIME reports the time took to complete
    the last purging event. For further details and the list of supported
    environment variables take a look to TRIGGER_VARS document.
  + Some additions to both logfile players: a new '-n' switch allows to play
    N elements; this way, arbitrary portions of the file may be played using
    '-n' in conjunction with the (already existing) '-o' switch which allows
    to read the logfile starting at a specified offset. New switches '-H',
    '-D', '-T', '-U', '-P' have been introduced to override SQL parameters
    like hostname, DB, table, user and password. The '-t -d' combination
    (test only, debug) now allows to print over the screen the content of
    the logfile.
  + Logfiles size is now limited to a maximum of 2Gb, thus avoiding issues
    connected to the 32bit declaration of off_t. While many OS implment a
    solution to the problem, seems there are few chances to solve it in a
    portable way. When the maximum size is hit the old logfile is rotated
    appending to its filename a trailing small integer ( in a way similar
    to logrotate) and a fresh one is started.
  ! Logfile players: '-s' switch, which was allowing to play one element
    a time, has been superseded. Its current equivalent is: '-n 1'. 
  ! The file opening algorithm has been slightly changed in SQL plugins:
    flock() follows shortly the fopen() and all subsequent operations and
    evaluations are thus strictly serialized. freopen() is avoided. 

0.7.8 -- 02-Dec-2004
  + Recovery logfile structure has been enhanced. Following the logfile
    header has been created a new template structure. Templates will avoid 
    the issue of being not able to read old logfiles because of changes to
    internal data structures. Templates are made of an header and a number
    of entries, each describing a single field of the following data.
    Both players, pmmyplay and pmpgplay, are able to parse logfiles basing
    over the template description. Backward logfile compatibility is broken.
  + Execcutable triggering mechanism (from SQL plugins) has been enhanced:
    some status informations (eg. stats of the last purging event) are now
    passed to the trigged executable in the form of environment variables.
    The list of supported variables has been summarized into TRIGGER_VARS
    document. The mechanism allows to spawn executables for post-processsing
    operations at arbitrary timeframes.
  + Support for 'temporary' devices (like PPP and maybe PCMCIA cards too)
    has been introduced. A new configuration directive 'interface_wait' (or
    '-w' commandline) instructs pmacctd to wait for the listening device to 
    become available. It works both when in startup phase and when already
    into main loop. A big thanks to Andre Berger for his support.
  ! ppp_handler() routine, which is in charge to handle PPP packets, have
    been totally rewritten. Thanks, again, to Andre Berger for his support.
  ! All link layer handling routines have been revised; some extra checks
    have been added to overcome issues caused from malicious handcrafted
    packets. 
  ! Some time handling and timeout issues have been revised into PostgreSQL
    plugin code. They were affecting only the triggering mechanism. 
  ! Fixed an execv() bug into MY_Exec() and PG_Exec(). It was causing the
    not correct execution of triggers. Now, a zeroed argv parameter is
    passed to the function. The problem has been verified on FreeBSD. 

0.7.7 -- 16-Nov-2004
  + Added two new aggregation primitives: 'src_as' and 'dst_as'. They allow
    accounting based over Autonomous System number; 'pmacctd' requires AS
    numbers to be supplied into a 'networks_file' configuration directive
    (which allows to specify the path to a networks definition file);
    'nfacctd' may either look up AS numbers from the networks definition file
    or read them from each NetFlow flow (this is default). 'nfacctd_as_new'
    key could be used to switch 'nfacctd' behaviour.
  + Added some new aggregation modes: 'sum_net', 'sum_as', 'sum_port' ('sum'
    which is actually an alias for 'sum_host' has been already introduced
    early). Sum is intended to be the total traffic (that is, inbound plus
    outbound traffic amounts) for each entry.
  + Added another aggregation primitive: 'none'. It does not make use of any
    primitive: it allows to see total bytes and packets transferred through
    an interface. 
  + The definition of a 'networks_file' enables network lookup: hosts inside
    defined networks are ok; hosts outside them are 'zeroed'. This behaviour
    may now also be applied to 'src_host', 'dst_host' and 'sum_host'. Under
    certain conditions (eg. when using only host/net/as primitives and defined 
    networks comprise all transiting hosts) it may be seen an alternative
    way to filter data. 
  ! 'frontend'/'backend' PostgreSQL plugin operations have been obsoleted.
    'unified'/'typed' operations have been introduced instead. See 'sql_data'
    description, CONFIG-KEYS document, for further informations.
  ! Optimizations have been applied to: core process, the newly introduced
    cache code (see 0.7.6) and in-memory table plugin. 
  ! Fixed some string handling routines: trim_all_spaces(), mark_columns()
  ! Solved a potential race condition which was affecting write_pid_file() 

0.7.6 -- 27-Oct-2004
  + Many changes has been introduced on 'pmacct' client side. '-m' switch
    (which output was suitable as MRTG input) has been obsoleted (though it
    will continue to work for next few releases). A new '-N' switch has
    been added: it returns counter value, suitable for integration with
    either RRDtool or MRTG. 
  + Support for batch queries have also been added into pmacct client. It
    allows to join up to 4096 requests into a single query. Requests could
    either be concatenated commandline or read from a file (more details are
    in FAQS and EXAMPLES). Batch queries allow to handle efficiently high number
    of requests in a single shot (for example to timely feed data to a large
    amount of graphs). 
  + Still pmacct client: '-r' switch, which already allows to reset counters
    for matched entries, now it also applies to group of matches (also referred
    as partial matches). 
  + New scripts have been added into the examples tree which show how to
    integrate memory and SQL plugins with RRDtool, MRTG and GNUplot. 
  + Memory plugin (IMT) has been further enhanced; each query from pmacct
    client is now evaluated and if involves just a short ride through the
    memory structure, it is served by the plugin itself without spawning a
    new child process. Batch queries support and reordering of fragmented
    queries have also been added.  
  + New cache has been introduced in both SQL plugins; its layout is still
    an hash structure but it now features also chains, allocation, reuse and
    retirement of chained nodes. It also sports a LRU list of nodes which eases
    node handling. The new solution avoids the creation of a collision queue,
    ensuring uniqueness of data placed onto the queries queue. While this
    already greatly benefits a directive like 'sql_dont_try_update', it also
    opens new chances for post-processing operations of queries queue. 

0.7.5 -- 14-Oct-2004
  + Introduced support for the definition of a 'known ports' list, when
    either 'src_port' or 'dst_port' primitives are in use. Known ports
    will get written into the backend; unknown ports will be simply zeroed.
    It could be enabled via 'ports_file' configuration key or '-o' commandline
    switch.
  + Introduced support for weekly and monthly counters breakdown; hourly, 
    minutely and daily were already supported. New breakdowns could be
    enabled via 'w' and 'M' words in 'sql_history' and related configuration
    keys.  
  + Added a '-i' commandline switch to both 'pmmyplay' and 'pmpgplay' to
    avoid UPDATE SQL queries and skip directly to INSERT ones. Many thanks 
    to Jamie Wilkinson. 
  ! 'pmmyplay' and 'pmpgplay' code has been optimized and updated; some 
    pieces of locking and transactional code were included into the inner
    loop. A big thanks goes to Wim Kerkhoff and Jamie Wilkinson.
  ! Networks aggregation code has been revised and optimized; a direct-mapped
    cache has been introduced to store (and search) last search results
    from the networks table. A binary search algorithm, though optimized,
    over the table has still been preferred over alternative approaches
    (hash, tries).

0.7.4 -- 30-Sep-2004
  + Enhanced packet tagging support; it's now broken in Pre-Tagging and
    Post-Tagging; Pre-Tagging allows 'nfacctd' to assign an ID to a flow
    evaluating an arbitrary combination of supported NetFlow packet
    fields (actually: IP address, Input Interface, Output Interface); the
    Pre-Tagging map is global; Pre-Tag is applied as soon as each flow
    is processed; Post-Tagging allows both 'nfacctd' and 'pmacctd' to
    assign an ID to packets using a supplied value; Post-Tagging could be
    either global or local to a single plugin (and more plugins may tag
    differently); Post-Tag is applied as a last action before the packet
    is sent to the plugin. 'nfacctd_id_map' and 'pmacctd_id' configuration
    keys are now obsolete; 'pre_tag_map' and 'post_tag' are introduced to
    replace them. 
  + Added support for Pre-Tag filtering; it allows to filter packets basing
    on their Pre-Tag value. The filter is evaluated after Pre-Tagging but
    before Post-Tagging; it adds to BPF filtering support ('aggregate_filter'
    configuration key); 'pre_tag_filter' configuration key is introduced.
  + Added support for Packet Sampling; the current implementation bases on
    a simple systematic algorithm; the new 'sampling_rate' configuration
    key expects a positive integer value >= 1 which is the ratio of the
    packets to be sampled (translates in: pick only 1 out of N packets).
    The key is either global or local (meaning that each plugin could apply
    different sampling rates).
  ! Fixed a bug which was causing crashes in both 'pmacctd' and 'nfacctd'
    when '-r' parameter was specified commandline. Thanks to Ali Nikham 
    for his support. 

0.7.3 -- 31-Aug-2004
  + Added support for both Netflow 'input interface' and 'output interface'
    fields. These two fields are contained in each flow record inside a
    NetFlow packet. It works through ID mapping (read below).
  + The ID map file syntax has been enhanced to allow greater flexibility
    in ID assignation to packets; example: 'id=1 ip=192.168.1.1 in=3
    out=5'; the above line will cause the 'ID' 1 to be assigned to
    flows exported by a NetFlow agent (for example a router) which IP
    address is '192.168.1.1' and transiting from interface '3' to interface
    '5'. 
  + In-memory table operations have been enhanced when using shared memory;
    a new reset flag has been added to avoid race conditions.
  ! Configuration lines are no more limited to some fixed maximum length
    but are allocated dynamically; this to overcome the need for long
    configuration lines to declare arbitrary filters and plugin's list.
    Thanks to Jerry Ji for his support.
  ! Configuration handlers, which are responsible to parse and validate
    values for each configuration key, have been rewritten on the way
    for a better portability. 
  ! Signal handler routines have been changed to better accomodate SysV
    semantics. 
  ! Fixed shared memory mmap() operations on IRIX and SunOS; a further
    test checks for either 'MAP_ANON' or 'MAP_ANONYMOUS' definitions; in
    case of negative outcome, mmap() will use '/dev/zero'. 
  ! Packet handlers have been revised and optimized.
  ! Some optimizations have been added when using shared memory; write()
    function has been usually called to signal the arrival of each new packet,
    through the core process/plugin control channel; now it does so if and
    only if the plugin, on the other side, is actually blocking over a poll();
    because of sequence numbers guarantee, data is directly written into
    shared memory segment.

0.7.2p1 -- 08-Aug-2004
  ! Multiple fixes in plugin's configuration post checks; negative outcome
    of some checks was leading to clear misbehaviours. Versions affected
    are >= 0.7.0 . A big thanks goes to Alexandra Walford for her support.

0.7.2 -- 02-Aug-2004
  + VLAN accounting has been added. The new 'vlan' keyword is supported as
    argument of both '-c' commandline switch and 'aggregate' configuration
    key.
  + Distributed accounting support has been added. It could be enabled into
    'pmacctd' via 'pmacctd_id' configuration key and into 'nfacctd' via the
    'nfacctd_id_file' configuration key. While 'pmacctd_id' key expects as
    value a small integer, 'nfacctd_id_file' expects a path to a file which
    contains the mapping: 'IP address of the router (exporting Newflow) ->
    small integer'. This scheme ease tasks such as keeping track of who has
    generated what data and either cluster or keep disjoint data coming from
    different sources when using a SQL database as backend. 
  + Introduced SQL table version 2. The SQL schema is the same as existing
    tables with the following additions: support for distributed accounting;
    support for VLAN accounting.
  + Added MAC addresses query capabilties to pmacct client. 
  + Added '-r' commandline switch to pmacct client. It can only be used in
    conjunction with '-m' or '-M' switches. It allows to reset packet and
    bytes counters of the retrieved record.
  ! Exit codes have been fixed in both 'pmacctd' and 'nfacctd'. Thanks to
    Jerry Ji for his signallation. 
  ! Fixed a problem when retrieving data from memory table: sometimes null
    data (without any error message) was returned to the client; the problem
    has been successfully reproduced only on FreeBSD 5.1: after an accept()
    call, the socket being returned inherits same flags of the listening
    socket, this case non-blocking flag. Thanks to Nicolas Deffayet for his
    support. 
  ! Revised PostgreSQL creation script. 

0.7.1 -- 14-Jul-2004
  + Added shared memory implementation; core process, now, could push data
    into a shared memory segment and then signal arrival of new data to the
    plugin. Shared memory support could be enabled via '--enable-mmap' switch
    at configuration time.
  + Strongly enhanced gathering capabilities of pmacct client; pmacct client
    is used to fetch data from memory plugin; it is, now, able to ask exact
    or partial matches via '-M' switch and return a readable listing output.
    MRTG export capabilities, full table fetch and table status query are
    still supported. 
  + Introduced SQL table versioning. It could be enabled via 'sql_table_version'
    configuration switch. It will enable to build new SQL tables (for example
    adding new aggregation methods) while allowing who is not interested in new
    setups to work with old tables. 
  + Added checks for packet capture type; informations acquired are later used
    for better handling pcap interface. 
  ! Fixed some issues concerning pmacctd VLAN and PPPOE code.
  ! Fixed a mmap() issue on Tru64 systems.
  ! Fixed some minor poll() misbehaviours in MySQL, PgSQL and print plugins;
    they were not correctly handled. 
  

0.7.0p1 -- 13-Jul-2004
  ! Fixes in cache code; affects MySQL, PgSQL and print plugins.

0.7.0 -- 01-Jul-2004
  + PMACCT OPENS TO NETFLOW: a new network daemon, nfacctd, is introduced:
    nfacctd listens for Netflow V1/V5 packets; is able to apply BPF filters
    and to aggregate packets; it's then able to either save data in a memory
    table, MySQL or PostgreSQL database or simply output packets on the screen.
    It can read timestamps from Netflow packets in msecs, seconds or ignore
    them generating new timestamps; a simple allow table mechanism allows
    to silently discard Netflow packets not generated by a list of trusted
    hosts.  
  + Strongly enhanced IP fragmentation handling in pmacctd.
  + Added new checks into the building systems; new hints when it searches 
    for libraries and headers; initial tests for C compilers capabilities
    have been added. 
  + Works to let pmacct run on IRIX platforms continue; some issues with
    MipsPRO compiler have been solved; added proper compilation flags/hints.
    SIGCHLD is now properly handled and child processes are correctly retired.
    (a thank for his support goes to Joerg Behrens) 
  + First, timidous, introduction of mmap() calls in memory plugin; they need
    to be enabled with '--enable-mmap' flag at configure time. 
  ! Fixed a potential deadlock issue in PostgreSQL plugin; changed locking
    mechanism. (a big thank to Wim Kerkhoff)
  ! Fixed an issue concerning networks aggregation on Tru64 systems.

0.6.4p1 -- 01-Jun-2004
  ! Fixed an issue with cache aliasing in MySQL and PostgreSQL plugins.
    Other plugins are not affected; this potential issue affects only
    version 0.6.4, not previous ones. Anyone using these plugins with 
    0.6.4 is strongly encouraged to upgrade to 0.6.4p1. 

0.6.4 -- 27-May-2004
  + Added chance to launch executables from both SQL plugins at arbitrary 
    time intervals to ease data post-processing tasks. Two new keys are
    available: 'sql_trigger_exec' and 'sql_trigger_time'. If any interval
    is supplied the specified executable is triggered every time data is
    purged from the cache. 
  + Added a new 'print' plugin. Enabling it, data is pulled at regular
    intervals to stdout in a way similar to cflowd's 'flow-print'.
    tool. New config keys are 'print_refresh_time', 'print_cache_entries'
    and 'print_markers'. This last key enables the print of start/end
    markers each time the cache is purged.  
  + Added 'sql_dont_try_update' switch to avoid UPDATE queries to the DB
    and skip directly to INSERT ones. Performance gains has been noticed
    when UPDATEs are not necessary (eg. when using timeslots to break up
    counters and sql_history = sql_refresh_time).
    Thanks to Jamie Wilkinson. 
  + Optimized use of transactions in PostgreSQL plugin; in the new scheme
    is built a single big transaction for each cache purge process. This
    leads to good performance gains; recovery mechanisms have been modified 
    to overcome whole transaction trashing. Many thanks to James Gregory
    and Jamie Wilkinson.
  ! Enhanced debug messages output when specific error conditions are returned
    by the DB. 
  ! Fixed a potential counters overflow issue in both MySQL and PgSQL
    plugins cache.
  ! Fixed preprocessor definitions issue: LOCK_UN, LOCK_EX are undeclared
    on IRIX and Solaris. Thanks to Wilhelm Greiner for the fix.

0.6.3 -- 27-Apr-2004
  + Added support for full libpcap-style filtering capabilities inside
    pmacctd. This allows to bind arbitrary filters to each plugin (in
    addition to already existing chance to apply them to the listening
    interface via 'pcap_filter' configuraiton key). The config key to
    specify these new filters is 'aggregate_filter'.  
  + Strongly improved networks definition file handling; now the file is
    parsed and organized as a hierarchical tree in memory. This allows to
    recognize and support networks-in-networks. 
  + Initial optimizations has been done over the code produced in last
    few months. 
  + Preprocessor definitions has been added to some part of the code, to
    allow pmacctd compile over IRIX. It has been reported to work over a
    IRIX64 6.5.23 box. Thanks to Wilhelm Greiner for his efforts.
  + Added flock() protected access to recovery logfiles.
  ! Fixed an ugly SEGV issue detected in both 0.6.2's logfile player tools.

0.6.2 -- 14-Apr-2004
  + Added support for networks aggregation. Two new primitives has
    been added 'src_net' and 'dst_net' to be used in conjunction with
    a network's definitions file (path is supplied via 'networks_file'
    configuration key). An example of this file is in the examples/
    directory. 
    When this aggregation is enabled, IP addresses are compared against
    the networks table; then the matching network will get written to
    the backend; if any match occurs a '0.0.0.0' is written.
    A really big thank goes to Martin Anderberg for his strong support
    during last weeks. 
  + pipe() has been thrown away; socketpair() has been introduced to
    set up a communication channel between pmacctd core process and
    plugins.
  + Added 'plugin_pipe_size' configuration key to adjust queue depth 
    (size) beween core process and plugins. A default value is set by
    operating system; it could not suffice when handling heavy traffic
    loads. Added also a specific error string when pipe gets filled.
  + Added 'plugin_buffer_size' configuration key to enable chances to
    bufferize data to be sent to plugins. When under great loads this
    helps in preventing high CPU usage and excessive pressure over
    kernel.
  + SQL plugins aliasing behaviour has been changed; when no free space
    for new data is found and old data has to be pulled out, it's now
    actually written to the DB but it's inserted in a new 'collision
    queue'. This new queue is purged together with the 'queries queue'.
    See INTERNALS for further details. 
  + SQL plugins cache behaviour has been changed by a direct-mapped
    one to a 3-ways associative to get better scores when searching
    free space for new data. See INTERNALS for further details.
  + Added 'sql_cache_entries' configuration key to adjust bucket's
    number of SQL plugin cache. As every hashed structure, a prime
    number of buckets is advisable to get better dispersion of data
    through the table.
  ! Fixed a malloc() SEGV issue in in-memory table plugin first
    noticed with gcc 3.3.3 (Debian 20040320) and glibc 2.3.2.
  ! Fixed a SEGV issue carried with last release. Improved handling
    of communication channels between core process and plugins. 
  ! Uniformed plugin's handling of signals; now sending a SIGINT to
    all pmacctd processes causes it to flush caches and exit nicely.
  ! Updated documentation; still no man page.

0.6.1 -- 24-Mar-2004
  + A new concept has been introduced: plugin names. A name could
    be assigned to each running plugin allowing to run more
    instances of the same plugin type; each one is configurable
    with global or 'named' keys. Take a look to examples for
    further info. 
  + Added support for PPPOE links. The code has been fully contributed
    by Vasiliy Ponomarev. A big thank goes to him.
  + Added a 'sql_startup_delay' configuration key to allow more
    plugin instances that need to write to the DB, to flush their
    data at same intervals but in different times to avoid locking
    stalls or DB overkills.
  + Improved handling of syslog connections. SIGHUP signal, used to
    reopen a connection with syslog (eg. for log rotation purposes),
    now is supported in all plugins. 
  + A simple LRU (Last Recently Used) cache has been added to the
    in-memory table plugin. The cache gives great benefits (exploiting
    some kind of locality in communication flows) when the table gets
    large (and chain in buckets become long and expensive to traverse).
  + Down-up of listening interface are now handled properly. Such an
    event traps a reopening of connection with libpcap. [EXPERIMENTAL]
  + Some work has been done (mostly via directives to preprocessor)
    in order to get pmacct compiled under Solaris. [HIGLY EXPERIMENTAL,
    translates: don't assume it works but, please, try it out and some
    kind of feedback would be appreciated] 
  ! Plugins have been better structured; plugin hooking has been
    simplified and re-documented; configuration parser has been
    strongly improved.
  ! Fixed a bug in 'configure' script; when supplying custom paths to
    MySQL libraries an erroneous library filename was searched for.
    (thanks to Wim Kerkhoff) 
  

0.6.0p3 -- 09-Feb-2004
  ! Fixed an issue concerning promiscuous mode; it was
    erroneously defaulting to 'false' under certain
    conditions. (Thanks to Royston Boot for signalling the
    problem)

0.6.0p2 -- 05-Feb-2004
  ! Fixed pmacct daemon in-memory table plugin unstability,
    noticed under sustained loads. (A thank for signalling
    the problem goes to Martin Pot) 
  ! Minor code rewritings for better optimizazion done in
    both in-memory table plugin and pmacct client.

0.6.0p1 -- 28-Jan-2004
  ! Fixed a bug in in-memory table plugin that was causing
    incorrect memorization of statistics. (Many thanks for
    promptly signalling it go to Martin Pot)
  ! Fixed a bug in pmacct client, used to gather stats from
    in-memory table. Under high loads and certain conditions
    the client was returning SEGV due to a realloc() issue.
    (Thanks to Martin Pot)

0.6.0  -- 27-Jan-2004
  + PMACCT OPENS TO POSTGRESQL: fully featured PostgreSQL
    plugin has been added; it's transaction based and
    already supports "recovery mode" both via logfile and
    backup DB actions. pmpgplay is the new tool that allows
    to play logfiles written in recovery mode by the plugin
    into a PostgreSQL DB. See CONFIG-KEYS and EXAMPLES for
    further informations. (Again, many thanks to Wim Kerkoff) 
  + Added new "recovery mode" action to MySQL plugin: write
    data to a backup DB if primary DB fails. DB table/user/
    password need to be the same as in the primary DB. The
    action could be enabled via "sql_backup_host" config
    key. 
  + Added a "sql_data" configuration optinion; a "frontend"
    value means to write human readable (strings) data; a
    "backend" value means to write integers in network byte
    order. Currently, this option is supported only into the
    new PostgreSQL plugin. See CONFIG-KEYS and README.pgsql
    for further informations.
  + Added support for simple password authentication in 
    client/server query mechanism for in-memory table
    statistics. It's available via "imt_passwd" config key. 
  + Added a "-t" commandline switch to pmmyplay; it runs
    the tool in a test only mode; useful to check header
    infos or logfile integrity.
  ! Fixed an ugly bug that made impossible MAC accounting 
    over certain links. Was affected only version 0.5.4. 
  ! Many code and structure cleanups.

0.5.4  -- 18-Dec-2003
  + Added a commandline and configuration switch to use 
    or not promiscuous mode for traffic capturing; useful 
    to avoid waste of resources if running over a router. 
  + Introduced a "recovery mode" concept for MySQL plugin:
    if DB fails an action is taken; currently is possible
    to write data to a logfile. More failover solutions to
    come in next releases. Thanks also to Wim Kerkhoff. 
  + Added a new "pmmyplay" tool. Allows to play logfiles
    previously written by a MySQL plugin in recovery mode.
    Check EXAMPLES for hints; see INTERNALS for further
    details about recovery mode and pmmyplay. 
  + Added syslog logging and debugging. Thanks for long
    brainstormings to Wim Kerkhoff.
  + Added chance to write PID of pmacctd core process to
    a specified file; it could help in automating tasks
    that need to send signals to pmacctd (eg. to rotate
    logfiles and reopen syslog connection). Take a look
    to SIGNALS file for further informations.
  + support for 802.11 Wireless links. [EXPERIMENTAL]  
  + support for linux cooked device links (DLT_LINUX_SLL).
    pcap library >= 0.6.x is needed. A big thank goes to
    KP Kirchdoerfer.
  ! Simplified client/server query mechanism; avoided all
    string comparison stuff. 
  ! Large parts of in-memory table plugin code has been
    revised to achieve better efficiency and optimization of
    available resources.

0.5.3  -- 20-Nov-2003
  ! pmacctd core has been optimized and a new loop-callback
    scheme driven by pcap library has been introduced; I/O
    multiplexing is avoided.
  ! In MySQL plugin, refresh of entries in the DB has been
    switched from a signal-driven approach to a lazy timeslot
    based one. If using historical recording, taking care
    to the choosen values, this greatly alleviates cache 
    aliasing.
  ! In MySQL plugin, modulo function (for insertion of data in
    the direct mapped cache) has been changed: crc32 algorithm
    has been adopted. Experimental tests shown the reduction of
    cache aliasing to about 0.45%. 
  ! The whole MySQL plugin has been inspected for performance
    bottlenecks resulted by the addition of new features in
    last releases.
  ! Fixed a bug in link layer handlers. 

0.5.2  -- 03-Nov-2003
  + "sql_history" configuration key syntax has been changed to
    support history recording at fixed times with mins, hrs and
    days granularity. A little of date arithmetics has been
    introduced (merely multiplicative factors, eg. to ease 95th
    percentile operations).
  + Added "sql_history_roundoff" configuration key to round off
    time of first timeslot. This little care gives cleaner time
    results and inductively affects all subsequent slots.
  + Achieved more precise calculations via timestamps added to
    the cache structure to avoid data counted during the current
    timeslot and not already fed in the DB to be accounted in next
    slot. 
  ! Monthly historical aggregation is no more available.
  ! Fixed portability issues posed by vsnprintf() in MySQL
    plugin. Now the plugin compiles smoothly under Tru64 Unix.

0.5.1  -- 01-Oct-2003
  + due to the proliferation of command-line options, the
    support for a configuration file has been added. All
    commandline switches until version 0.5.0 will be supported
    in the future.
    New configurable options (eg. log to a remote SQL server)
    will be only supported via configuration file. See 
    CONFIG-KEYS file for available configuration keys.
  + added support for historical recording of counters in the
    MySQL database. Available granularities of aggregation are
    hourly, daily or monthly (eg. counters are separated hour
    by hour, daily of monthly for each record). Timestamps of
    last INSERT and UPDATE have been added over each record.
    (thanks to Wim Kerkhoff for his strong collaboration)
  + support for IP header options.
  + support for PPP links. [EXPERIMENTAL]  
  ! Fixed a MySQL plugin direct-mapped cache issue: the cache
    now traps INSERT queries when an UPDATE fails due to any
    asyncronous table manipulation event (eg. external scripts,
    table truncation, etc.).
  ! MySQL plugin has been strongly revised and optimized; added
    options to save data to a remote sql server and to customize
    username, password and table; added MySQL locking stuff. 
    (another big thank to Wim Kerkhoff).
  ! various code cleanups.

0.5.0  -- 22-Jul-2003
  + static aggregation directives (src_host, dst_host, ..)
    are now superseded by primitives that can be stacked
    together to form complex aggregation methods. 
    The commandline syntax of the client program has been
    consequently changed to support these new features. 
  + two new primitives have been added: source MAC address
    and destination MAC address. 
  + support for 802.1Q (VLANs) tagged packets (thanks to
    Rich Gade). 
  + support for FDDI links. [EXPERIMENTAL]
  ! the core pmacctd loop (that gathers packets off the
    wire and feeds data to plugins) has been revised and
    strongly optimized. 
  ! the main loop of MySQL plugin has been optimized with
    the introduction of adaptive selection queries during 
    the update process.
  ! fixed a memory allocation issue (that caused a SIGSEGV,
    under certain circustances) in pmacct client: now the
    upper bound of dss is checked for large data retrieval.

0.4.2  -- 20-Jun-2003
  + limited support for transport protocols (currently
    only tcp and udp): aggregation of statistics for
    source or destination port. 
  + optimized query mechanism for in-memory table; solved
    few generalization issues that will enable (in future
    versions) to support complex queries. 
  + added "-t" pmacctd commandline switch to specify a
    custom database table. 
  ! fixed realloc() issue in pmacct client (thanks to
    Arjen Nienhuis).
  ! fixed an issue regarding mysql headers in the configure
    script. 

0.4.1  -- 08-May-2003
  ! missing break in a case statement that led pmacctd
    to misbehaviours; a cleaner approach to global vars
    (thanks to Peter Payne). 
  ! fixed an issue with getopt() and external vars. Now
    pmacct has reported to compile without problems on
    FreeBSD 4.x (thanks to Kirill Ponomarew).
  ! missing conditional statement to check the runtime
    execution of compiled plugins in exec_plugins()

0.4.0  -- 02-May-2003
  + switched to a plugin architecture: plugins need to
    be activated at configure time to be compiled and
    then used via "-P" command-line switch in pmacctd.
    See PLUGINS for more details.
  + added first plugin: Mysql driver. It uses a Mysql
    database as backend to store statistics other than
    in-memory table. See sql/ directory for scripts for
    creation of db needed to store data.
  + added the choice to collect statistics for traffic
    flows in addition to src|dst|sum aggregation via
    the "-c flows" command-line switch in pmacctd.
  + major code cleanups.
  + mostly rewritten configure script; switched back to
    autoconf 2.1.

0.3.4  -- 24-Mar-2003
  + accounting of IP traffic for source, destination
    and aggregation of both. Introduced -c switch to
    pmacctd (thanks to Martynas Bieliauskas).
  + added daemonization of pmacctd process via -D
    command line switch
  + added buffering via pcap_open_live() timeout handling
    on those architectures where it is supported.
  + It compiles and works fine over FreeBSD 5.x;
    solved some pcap library issues.  
  + added customization of pipe for client/server
    communication via -p command line switch both in
    pmacct and pmacctd
 
0.3.3  -- 19-Mar-2003
  + introduced synchronous I/O multiplexing
  + support for -m 0 pmacctd switch, in-memory table
    can grow undefinitely. 
  + revised memory pool descriptors table structure
  ! introduced realloc() in pmacct to support really
    large in-memory table transfers; solved additional
    alignment problems. 
  ! solved compatibility issues with libpcap 0.4 
  ! solved nasty problem with -i pmacctd switch
  ! solved various memory code bugs and open issues

0.3.2  -- 13-Mar-2003
  + support for pcap library filters 
  ! minor bugfixes

0.3.1  -- 12-Mar-2003
  + documentation stuff: updated TODO and added INTERNALS
  + revised query mechanism to server process, added a
    standard header to find command and optional values
    carried in query buffer.
  + added -s commandline switch to customize the size of
    each memory pool; see INTERNLS for more informations
  ! stability tests and fixes 
  ! configure script enhancements

0.3.0  -- 11-Mar-2003
  ! not public release
  + increased efficiency through allocation of memory pools 
    instead of sparse malloc() calls when inserting new
    elements in in-memory table.
  + added -m commandline switch to pmacctd to set the number
    of available memory pools; the size of each memory pool is
    the number of buckets, chosen with -b commandline option,
    see INTERNALS for more informations.
  + switched client program to getopt() to acquire commandline
    inputs.
  + new -m commandline  option in client program to acquire
    statistics of a specified IP address in a format useful for
    acquisition by MRTG program; see examples directory for a
    sample mrtg configuration.
  ! major bugfixes
  ! minor code cleanups

0.2.4  -- 07-Mar-2003
  + portability: Tru64 5.x 
  ! configure script fixes
  ! minor bugfixes

0.2.3  -- 05-Mar-2003
  + first public release
  ! portability fixes
  ! minor bugfixes


0.2.2  -- 04-Mar-2003
  + minor code cleanups
  + added autoconf, automake stuff


0.2.1  -- 03-Mar-2003
  + fork()ing when handling queries  
  + signal handling
  + command-line options using getopt() 
  + usage instructions
  ! major bugfixes


0.2.0  -- 01-Mar-2003
  + dynamic allocation of in-memory table
  + query (client/server) mechanism 
  + added a Makefile
  ! major bugfixes


0.1.0  -- late Feb, 2003
  + Initial release