Update packet filter (pf) code to OpenBSD 4.5.

[FreeBSD/FreeBSD.git] / UPDATING

Updating Information for FreeBSD current users

This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
See end of file for further details.  For commonly done items, please see the
COMMON ITEMS: section later in the file.  These instructions assume that you
basically know what you are doing.  If not, then please consult the FreeBSD
handbook.

Items affecting the ports and packages system can be found in
/usr/ports/UPDATING.  Please read that file before running portupgrade.

NOTE TO PEOPLE WHO THINK THAT FreeBSD 9.x IS SLOW:
        FreeBSD 9.x has many debugging features turned on, in both the kernel
        and userland.  These features attempt to detect incorrect use of
        system primitives, and encourage loud failure through extra sanity
        checking and fail stop semantics.  They also substantially impact
        system performance.  If you want to do performance measurement,
        benchmarking, and optimization, you'll want to turn them off.  This
        includes various WITNESS- related kernel options, INVARIANTS, malloc
        debugging flags in userland, and various verbose features in the
        kernel.  Many developers choose to disable these features on build
        machines to maximize performance.  (To disable malloc debugging, run
        ln -s aj /etc/malloc.conf.)

20110628:
        The packet filter (pf) code has been updated to OpenBSD 4.5.
        You need to update userland tools to be in sync with kernel.

20110608:
        The following sysctls and tunables are retired on x86 platforms:
                machdep.hlt_cpus
                machdep.hlt_logical_cpus
        The following sysctl is retired:
                machdep.hyperthreading_allowed
        The sysctls were supposed to provide a way to dynamically offline and
        online selected CPUs on x86 platforms, but the implementation has not
        been reliable especially with SCHED_ULE scheduler.
        machdep.hyperthreading_allowed tunable is still available to ignore
        hyperthreading CPUs at OS level.
        Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
        where X is an APIC ID of a CPU.  Be advised, though, that disabling
        CPUs in non-uniform fashion will result in non-uniform topology and
        may lead to sub-optimal system performance with SCHED_ULE, which is
        a default scheduler.

20110607:
        cpumask_t type is retired and cpuset_t is used in order to describe
        a mask of CPUs.

20110531:
        Changes to ifconfig(8) for dynamic address family detection mandate
        that you are running a kernel of 20110525 or later.  Make sure to
        follow the update procedure to boot a new kernel before installing
        world.

20110513:
        Support for sun4v architecture is officially dropped

20110503:
        Several KPI breaking changes have been committed to the mii(4) layer,
        the PHY drivers and consequently some Ethernet drivers using mii(4).
        This means that miibus.ko and the modules of the affected Ethernet
        drivers need to be recompiled.

        Note to kernel developers: Given that the OUI bit reversion problem
        was fixed as part of these changes all mii(4) commits related to OUIs,
        i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
        handling, no longer can be merged verbatim to stable/8 and previous
        branches.

20110430:
        Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
        into their kernel configurations along with 'device pci'.

20110427:
        The default NFS client is now the new NFS client, so fstype "newnfs"
        is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
        Although mounts via fstype "nfs" will usually work without userland
        changes, it is recommended that the mount(8) and mount_nfs(8)
        commands be rebuilt from sources and that a link to mount_nfs called
        mount_oldnfs be created. The new client is compiled into the
        kernel with "options NFSCL" and this is needed for diskless root
        file systems. The GENERIC kernel configs have been changed to use
        NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
        To use the regular/old client, you can "mount -t oldnfs ...". For
        a diskless root file system, you must also include a line like:
        
        vfs.root.mountfrom="oldnfs:"

        in the boot/loader.conf on the root fs on the NFS server to make
        a diskless root fs use the old client.

20110424:
        The GENERIC kernels for all architectures now default to the new
        CAM-based ATA stack. It means that all legacy ATA drivers were
        removed and replaced by respective CAM drivers. If you are using
        ATA device names in /etc/fstab or other places, make sure to update
        them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
        where 'Y's are the sequential numbers starting from zero for each type
        in order of detection, unless configured otherwise with tunables,
        see cam(4)). There will be symbolic links created in /dev/ to map
        old adX devices to the respective adaY. They should provide basic
        compatibility for file systems mounting in most cases, but they do
        not support old user-level APIs and do not have respective providers
        in GEOM. Consider using updated management tools with new device names.

        It is possible to load devices ahci, ata, siis and mvs as modules,
        but option ATA_CAM should remain in kernel configuration to make ata
        module work as CAM driver supporting legacy ATA controllers. Device ata
        still can be used in modular fashion (atacore + ...). Modules atadisk
        and atapi* are not used and won't affect operation in ATA_CAM mode.
        Note that to use CAM-based ATA kernel should include CAM devices
        scbus, pass, da (or explicitly ada), cd and optionally others. All of
        them are parts of the cam module.

        ataraid(4) functionality is now supported by the RAID GEOM class.
        To use it you can load geom_raid kernel module and use graid(8) tool
        for management. Instead of /dev/arX device names, use /dev/raid/rX.

        No kernel config options or code have been removed, so if a problem
        arises, please report it and optionally revert to the old ATA stack.
        In order to do it you can remove from the kernel config:
            options        ATA_CAM
            device         ahci
            device         mvs
            device         siis
        , and instead add back:
            device         atadisk         # ATA disk drives
            device         ataraid         # ATA RAID drives
            device         atapicd         # ATAPI CDROM drives
            device         atapifd         # ATAPI floppy drives
            device         atapist         # ATAPI tape drives

20110423:
        The default NFS server has been changed to the new server, which
        was referred to as the experimental server. If you need to switch
        back to the old NFS server, you must now put the "-o" option on
        both the mountd and nfsd commands. This can be done using the
        mountd_flags and nfs_server_flags rc.conf variables until an
        update to the rc scripts is committed, which is coming soon.

20110418:
        The GNU Objective-C runtime library (libobjc), and other Objective-C
        related components have been removed from the base system.  If you
        require an Objective-C library, please use one of the available ports.

20110331:
        ath(4) has been split into bus- and device- modules. if_ath contains
        the HAL, the TX rate control and the network device code. if_ath_pci
        contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
        contains the AHB glue. Users need to load both if_ath_pci and if_ath
        in order to use ath on everything else.

        TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
        need to load if_ath and if_ath_pci for ath(4) operation.

20110314:
        As part of the replacement of sysinstall, the process of building
        release media has changed significantly. For details, please re-read
        release(7), which has been updated to reflect the new build process.

20110218:
        GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD.  This
        is the last available version under GPLv2.  It brings a number of new
        features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
        4.1 and SSE 4.2), better support for powerpc64, a number of new
        directives, and lots of other small improvements.  See the ChangeLog
        file in contrib/binutils for the full details.

20110218:
        IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
        compliant, and will now use half of hash for authentication.
        This will break interoperability with all stacks (including all
        actual FreeBSD versions) who implement
        draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
        authentication).
        The only workaround with such peers is to use another HMAC
        algorithm for IPsec ("phase 2") authentication.

20110207:
        Remove the uio_yield prototype and symbol.  This function has
        been misnamed since it was introduced and should not be
        globally exposed with this name.  The equivalent functionality
        is now available using kern_yield(curthread->td_user_pri).
        The function remains undocumented.

20110112:
        A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
        symmetric with the existing SYSCTL_[ADD_]QUAD.  Type checking
        for scalar sysctls is defined but disabled.  Code that needs
        UQUAD to pass the type checking that must compile on older
        systems where the define is not present can check against
        __FreeBSD_version >= 900030.

        The system dialog(1) has been replaced with a new version previously
        in ports as devel/cdialog. dialog(1) is mostly command-line compatible
        with the previous version, but the libdialog associated with it has
        a largely incompatible API. As such, the original version of libdialog
        will be kept temporarily as libodialog, until its base system consumers
        are replaced or updated. Bump __FreeBSD_version to 900030.

20110103:
        If you are trying to run make universe on a -stable system, and you get
        the following warning:
        "Makefile", line 356: "Target architecture for i386/conf/GENERIC 
        unknown.  config(8) likely too old."
        or something similar to it, then you must upgrade your -stable system
        to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
        stable/8) or build the config from the latest stable/8 branch and
        install it on your system.

        Prior to this date, building a current universe on 8-stable system from
        between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
        error in the first kernel build phase.  A new config on those old 
        systems will fix that problem for older versions of -current.

20101228:
        The TCP stack has been modified to allow Khelp modules to interact with
        it via helper hook points and store per-connection data in the TCP
        control block. Bump __FreeBSD_version to 900029. User space tools that
        rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
        be recompiled.

20101114:
        Generic IEEE 802.3 annex 31B full duplex flow control support has been
        added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
        with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
        to take advantage of it instead of using custom implementations.  This
        means that these drivers now no longer unconditionally advertise
        support for flow control but only do so if flow control is a selected
        media option.  This was implemented in the generic support that way in
        order to allow flow control to be switched on and off via ifconfig(8)
        with the PHY specific default to typically off in order to protect
        from unwanted effects.  Consequently, if you used flow control with
        one of the above mentioned drivers you now need to explicitly enable
        it, for example via:
                ifconfig bge0 media auto mediaopt flowcontrol

        Along with the above mentioned changes generic support for setting
        1000baseT master mode also has been added and brgphy(4), ciphy(4),
        e1000phy(4) as well as ip1000phy(4) have been converted to take
        advantage of it.  This means that these drivers now no longer take the
        link0 parameter for selecting master mode but the master media option
        has to be used instead, for example like in the following:
                ifconfig bge0 media 1000baseT mediaopt full-duplex,master

        Selection of master mode now is also available with all other PHY
        drivers supporting 1000baseT.

20101111:
        The TCP stack has received a significant update to add support for
        modularised congestion control and generally improve the clarity of
        congestion control decisions. Bump __FreeBSD_version to 900025. User
        space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
        sockstat) need to be recompiled.

20101002:
        The man(1) utility has been replaced by a new version that no longer
        uses /etc/manpath.config. Please consult man.conf(5) for how to
        migrate local entries to the new format.

20100928:
        The copyright strings printed by login(1) and sshd(8) at the time of a
        new connection have been removed to follow other operating systems and
        upstream sshd.

20100915:
        A workaround for a fixed ld bug has been removed in kernel code,
        so make sure that your system ld is built from sources after
        revision 210245 from 2010-07-19 (r211583 if building head kernel
        on stable/8, r211584 for stable/7; both from 2010-08-21).
        A symptom of incorrect ld version is different addresses for
        set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.

20100913:
        The $ipv6_prefer variable in rc.conf(5) has been split into
        $ip6addrctl_policy and $ipv6_activate_all_interfaces.

        The $ip6addrctl_policy is a variable to choose a pre-defined
        address selection policy set by ip6addrctl(8).  A value
        "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified.  The
        default is "AUTO".

        The $ipv6_activate_all_interfaces specifies whether IFDISABLED
        flag (see an entry of 20090926) is set on an interface with no
        corresponding $ifconfig_IF_ipv6 line.  The default is "NO" for
        security reason.  If you want IPv6 link-local address on all
        interfaces by default, set this to "YES".

        The old ipv6_prefer="YES" is equivalent to
        ipv6_activate_all_interfaces="YES" and
        ip6addrctl_policy="ipv6_prefer".

20100913:
        DTrace has grown support for userland tracing. Due to this, DTrace is
        now i386 and amd64 only.
        dtruss(1) is now installed by default on those systems and a new
        kernel module is needed for userland tracing: fasttrap.
        No changes to your kernel config file are necessary to enable
        userland tracing, but you might consider adding 'STRIP=' and
        'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
        to have informative userland stack traces in DTrace (ustack).

20100725:
        The acpi_aiboost(4) driver has been removed in favor of the new
        aibs(4) driver. You should update your kernel configuration file.

20100722:
        BSD grep has been imported to the base system and it is built by
        default.  It is completely BSD licensed, highly GNU-compatible, uses
        less memory than its GNU counterpart and has a small codebase.
        However, it is slower than its GNU counterpart, which is mostly
        noticeable for larger searches, for smaller ones it is measurable
        but not significant.  The reason is complex, the most important factor
        is that we lack a modern and efficient regex library and GNU
        overcomes this by optimizing the searches internally.  Future work
        on improving the regex performance is planned, for the meantime,
        users that need better performance, can build GNU grep instead by
        setting the WITH_GNU_GREP knob.

20100713:
        Due to the import of powerpc64 support, all existing powerpc kernel
        configuration files must be updated with a machine directive like this:
            machine powerpc powerpc

        In addition, an updated config(8) is required to build powerpc kernels
        after this change.

20100713:
        A new version of ZFS (version 15) has been merged to -HEAD.
        This version uses a python library for the following subcommands:
        zfs allow, zfs unallow, zfs groupspace, zfs userspace.
        For full functionality of these commands the following port must
        be installed: sysutils/py-zfs

20100429:
        'vm_page's are now hashed by physical address to an array of mutexes.
        Currently this is only used to serialize access to hold_count. Over 
        time the page queue mutex will be peeled away. This changes the size
        of pmap on every architecture. And requires all callers of vm_page_hold
        and vm_page_unhold to be updated. 
 
20100402:
        WITH_CTF can now be specified in src.conf (not recommended, there
        are some problems with static executables), make.conf (would also
        affect ports which do not use GNU make and do not override the
        compile targets) or in the kernel config (via "makeoptions
        WITH_CTF=yes").
        When WITH_CTF was specified there before this was silently ignored,
        so make sure that WITH_CTF is not used in places which could lead
        to unwanted behavior.

20100311:
        The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
        to allow 32-bit compatibility on non-x86 platforms. All kernel
        configurations on amd64 and ia64 platforms using these options must
        be modified accordingly.

20100113:
        The utmp user accounting database has been replaced with utmpx,
        the user accounting interface standardized by POSIX.
        Unfortunately the semantics of utmp and utmpx don't match,
        making it practically impossible to support both interfaces.
        The user accounting database is used by tools like finger(1),
        last(1), talk(1), w(1) and ac(8).

        All applications in the base system use utmpx.  This means only
        local binaries (e.g. from the ports tree) may still use these
        utmp database files.  These applications must be rebuilt to make
        use of utmpx.

        After the system has been upgraded, it is safe to remove the old
        log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
        assuming their contents is of no importance anymore.  Old wtmp
        databases can only be used by last(1) and ac(8) after they have
        been converted to the new format using wtmpcvt(1).

20100108:
        Introduce the kernel thread "deadlock resolver" (which can be enabled
        via the DEADLKRES option, see NOTES for more details) and the
        sleepq_type() function for sleepqueues.

20091202:
        The rc.firewall and rc.firewall6 were unified, and
        rc.firewall6 and rc.d/ip6fw were removed.
        According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
        variables are obsoleted.  Instead, the following new rc
        variables are added to rc.d/ipfw:

                firewall_client_net_ipv6, firewall_simple_iif_ipv6,
                firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
                firewall_simple_onet_ipv6, firewall_trusted_ipv6

        The meanings correspond to the relevant IPv4 variables.

20091125:
        8.0-RELEASE.

20091113:
        The default terminal emulation for syscons(4) has been changed
        from cons25 to xterm on all platforms except pc98.  This means
        that the /etc/ttys file needs to be updated to ensure correct
        operation of applications on the console.

        The terminal emulation style can be toggled per window by using
        vidcontrol(1)'s -T flag.  The TEKEN_CONS25 kernel configuration
        options can be used to change the compile-time default back to
        cons25.

        To prevent graphical artifacts, make sure the TERM environment
        variable is set to match the terminal emulation that is being
        performed by syscons(4).

20091109:
        The layout of the structure ieee80211req_scan_result has changed.
        Applications that require wireless scan results (e.g. ifconfig(8))
        from net80211 need to be recompiled.

        Applications such as wpa_supplicant(8) may require a full world
        build without using NO_CLEAN in order to get synchronized with the
        new structure.

20091025:
        The iwn(4) driver has been updated to support the 5000 and 5150 series.
        There's one kernel module for each firmware. Adding "device iwnfw"
        to the kernel configuration file means including all three firmware
        images inside the kernel. If you want to include just the one for
        your wireless card, use the the devices iwn4965fw, iwn5000fw or
        iwn5150fw.

20090926:
        The rc.d/network_ipv6, IPv6 configuration script has been integrated
        into rc.d/netif.  The changes are the following:

        1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
           for IPv4.  For aliases, $ifconfig_IF_aliasN should be used.
           Note that both variables need the "inet6" keyword at the head.

           Do not set $ipv6_network_interfaces manually if you do not
           understand what you are doing.  It is not needed in most cases. 

           $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
           they are obsolete.

        2. $ipv6_enable is obsolete.  Use $ipv6_prefer and
           "inet6 accept_rtadv" keyword in ifconfig(8) instead.

           If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
           all configured interfaces have "inet6 accept_rtadv" in the
           $ifconfig_IF_ipv6.  These are for backward compatibility.

        3. A new variable $ipv6_prefer has been added.  If NO, IPv6
           functionality of interfaces with no corresponding
           $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
           and the default address selection policy of ip6addrctl(8) 
           is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
           Note that if you want to configure IPv6 functionality on the
           disabled interfaces after boot, first you need to clear the flag by
           using ifconfig(8) like:

                ifconfig em0 inet6 -ifdisabled

           If YES, the default address selection policy is set as
           IPv6-preferred.

           The default value of $ipv6_prefer is NO.

        4. If your system need to receive Router Advertisement messages,
           define "inet6 accept_rtadv" in $ifconfig_IF_ipv6.  The rc(8)
           scripts automatically invoke rtsol(8) when the interface becomes
           UP.  The Router Advertisement messages are used for SLAAC
           (State-Less Address AutoConfiguration).

20090922:
        802.11s D3.03 support was committed. This is incompatible with the
        previous code, which was based on D3.0.

20090912:
        A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
        of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
        control whether accepting Router Advertisement messages or not.
        Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
        a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
        The ifconfig(8) utility now supports these flags.

20090910:
        ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
        mount(8) and -a option for df(1) to see them.

20090825:
        The old tunable hw.bus.devctl_disable has been superseded by
        hw.bus.devctl_queue.  hw.bus.devctl_disable=1 in loader.conf should be
        replaced by hw.bus.devctl_queue=0.  The default for this new tunable
        is 1000.

20090813:
        Remove the option STOP_NMI.  The default action is now to use NMI only
        for KDB via the newly introduced function stop_cpus_hard() and
        maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.

20090803:
        The stable/8 branch created in subversion.  This corresponds to the
        RELENG_8 branch in CVS.

20090719:
        Bump the shared library version numbers for all libraries that do not
        use symbol versioning as part of the 8.0-RELEASE cycle.  Bump
        __FreeBSD_version to 800105.

20090714:
        Due to changes in the implementation of virtual network stack support,
        all network-related kernel modules must be recompiled.  As this change
        breaks the ABI, bump __FreeBSD_version to 800104.

20090713:
        The TOE interface to the TCP syncache has been modified to remove
        struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
        The cxgb driver is the only TOE consumer affected by this change, and
        needs to be recompiled along with the kernel. As this change breaks
        the ABI, bump __FreeBSD_version to 800103.

20090712: 
        Padding has been added to struct tcpcb, sackhint and tcpstat in
        <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
        maintaining the ABI. However, this change breaks the ABI, so bump
        __FreeBSD_version to 800102. User space tools that rely on the size of
        any of these structs (e.g. sockstat) need to be recompiled.

20090630:
        The NFS_LEGACYRPC option has been removed along with the old kernel
        RPC implementation that this option selected. Kernel configurations
        may need to be adjusted.

20090629:
        The network interface device nodes at /dev/net/<interface> have been
        removed.  All ioctl operations can be performed the normal way using
        routing sockets.  The kqueue functionality can generally be replaced
        with routing sockets.

20090628:
        The documentation from the FreeBSD Documentation Project (Handbook,
        FAQ, etc.) is now installed via packages by sysinstall(8) and under
        the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.

20090624:
        The ABI of various structures related to the SYSV IPC API have been
        changed.  As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
        options now all require COMPAT_FREEBSD7.  Bump __FreeBSD_version to
        800100.

20090622:
        Layout of struct vnet has changed as routing related variables were
        moved to their own Vimage module. Modules need to be recompiled.  Bump
        __FreeBSD_version to 800099.

20090619:
        NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
        respectively.  As long as no more than 16 groups per process are used,
        no changes should be visible.  When more than 16 groups are used, old
        binaries may fail if they call getgroups() or getgrouplist() with
        statically sized storage.  Recompiling will work around this, but
        applications should be modified to use dynamically allocated storage
        for group arrays as POSIX.1-2008 does not cap an implementation's
        number of supported groups at NGROUPS_MAX+1 as previous versions did.

        NFS and portalfs mounts may also be affected as the list of groups is
        truncated to 16.  Users of NFS who use more than 16 groups, should
        take care that negative group permissions are not used on the exported
        file systems as they will not be reliable unless a GSSAPI based
        authentication method is used.

20090616: 
        The compiling option ADAPTIVE_LOCKMGRS has been introduced.  This
        option compiles in the support for adaptive spinning for lockmgrs
        which want to enable it.  The lockinit() function now accepts the flag
        LK_ADAPTIVE in order to make the lock object subject to adaptive
        spinning when both held in write and read mode.

20090613:
        The layout of the structure returned by IEEE80211_IOC_STA_INFO has
        changed.  User applications that use this ioctl need to be rebuilt.

20090611:
        The layout of struct thread has changed.  Kernel and modules need to
        be rebuilt.

20090608:
        The layout of structs ifnet, domain, protosw and vnet_net has changed.
        Kernel modules need to be rebuilt.  Bump __FreeBSD_version to 800097.

20090602:
        window(1) has been removed from the base system. It can now be
        installed from ports. The port is called misc/window.

20090601:
        The way we are storing and accessing `routing table' entries has
        changed. Programs reading the FIB, like netstat, need to be
        re-compiled.

20090601:
        A new netisr implementation has been added for FreeBSD 8.  Network
        file system modules, such as igmp, ipdivert, and others, should be
        rebuilt.
        Bump __FreeBSD_version to 800096.

20090530:
        Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
        more valid.

20090530:
        Add VOP_ACCESSX(9).  File system modules need to be rebuilt.
        Bump __FreeBSD_version to 800094.

20090529:
        Add mnt_xflag field to 'struct mount'.  File system modules need to be
        rebuilt.
        Bump __FreeBSD_version to 800093.

20090528:
        The compiling option ADAPTIVE_SX has been retired while it has been
        introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
        The KPI for sx_init_flags() changes as accepting flags:
        SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
        been introduced in order to handle the reversed logic.
        Bump __FreeBSD_version to 800092.

20090527:
        Add support for hierarchical jails.  Remove global securelevel.
        Bump __FreeBSD_version to 800091.

20090523:
        The layout of struct vnet_net has changed, therefore modules
        need to be rebuilt.
        Bump __FreeBSD_version to 800090.

20090523:
        The newly imported zic(8) produces a new format in the output. Please
        run tzsetup(8) to install the newly created data to /etc/localtime.

20090520:
        The sysctl tree for the usb stack has renamed from hw.usb2.* to
        hw.usb.* and is now consistent again with previous releases.

20090520:
        802.11 monitor mode support was revised and driver api's were changed.
        Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
        of DLT_IEEE802_11.  No user-visible data structures were changed but
        applications that use DLT_IEEE802_11 may require changes.
        Bump __FreeBSD_version to 800088.

20090430:
        The layout of the following structs has changed: sysctl_oid,
        socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
        vnet_inet6 and vnet_ipfw.  Most modules need to be rebuild or
        panics may be experienced.  World rebuild is required for
        correctly checking networking state from userland.
        Bump __FreeBSD_version to 800085.

20090429:
        MLDv2 and Source-Specific Multicast (SSM) have been merged
        to the IPv6 stack. VIMAGE hooks are in but not yet used.
        The implementation of SSM within FreeBSD's IPv6 stack closely
        follows the IPv4 implementation.

        For kernel developers:

        * The most important changes are that the ip6_output() and
          ip6_input() paths no longer take the IN6_MULTI_LOCK,
          and this lock has been downgraded to a non-recursive mutex.

        * As with the changes to the IPv4 stack to support SSM, filtering
          of inbound multicast traffic must now be performed by transport
          protocols within the IPv6 stack. This does not apply to TCP and
          SCTP, however, it does apply to UDP in IPv6 and raw IPv6.

        * The KPIs used by IPv6 multicast are similar to those used by
          the IPv4 stack, with the following differences:
           * im6o_mc_filter() is analogous to imo_multicast_filter().
           * The legacy KAME entry points in6_joingroup and in6_leavegroup()
             are shimmed to in6_mc_join() and in6_mc_leave() respectively.
           * IN6_LOOKUP_MULTI() has been deprecated and removed.
           * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
             for MLDv1 have an additional 'timer' argument which is used to
             jitter the initial membership report for the solicited-node
             multicast membership on-link.
           * This is not strictly needed for MLDv2, which already jitters
             its report transmissions.  However, the 'timer' argument is
             preserved in case MLDv1 is active on the interface.

        * The KAME linked-list based IPv6 membership implementation has
          been refactored to use a vector similar to that used by the IPv4
          stack.
          Code which maintains a list of its own multicast memberships
          internally, e.g. carp, has been updated to reflect the new
          semantics.

        * There is a known Lock Order Reversal (LOR) due to in6_setscope()
          acquiring the IF_AFDATA_LOCK and being called within ip6_output().
          Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
          implementation constraint which needs to be addressed in HEAD.

        For application developers:

        * The changes are broadly similar to those made for the IPv4
          stack.

        * The use of IPv4 and IPv6 multicast socket options on the same
          socket, using mapped addresses, HAS NOT been tested or supported.

        * There are a number of issues with the implementation of various
          IPv6 multicast APIs which need to be resolved in the API surface
          before the implementation is fully compatible with KAME userland
          use, and these are mostly to do with interface index treatment.

        * The literature available discusses the use of either the delta / ASM
          API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
          using setsourcefilter(3)/getsourcefilter(3). For more information
          please refer to RFC 3768, 'Socket Interface Extensions for
          Multicast Source Filters'.

        * Applications which use the published RFC 3678 APIs should be fine.

        For systems administrators:

        * The mtest(8) utility has been refactored to support IPv6, in
          addition to IPv4. Interface addresses are no longer accepted
          as arguments, their names must be used instead. The utility
          will map the interface name to its first IPv4 address as
          returned by getifaddrs(3).

        * The ifmcstat(8) utility has also been updated to print the MLDv2
          endpoint state and source filter lists via sysctl(3).

        * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
          loopback of IPv6 multicast datagrams by default; it defaults to 1
          to preserve the existing behaviour. Disabling multicast loopback is
          recommended for optimal system performance.

        * The IPv6 MROUTING code has been changed to examine this sysctl
          instead of attempting to perform a group lookup before looping
          back forwarded datagrams.

        Bump __FreeBSD_version to 800084.

20090422:
        Implement low-level Bluetooth HCI API.
        Bump __FreeBSD_version to 800083.

20090419:
        The layout of struct malloc_type, used by modules to register new
        memory allocation types, has changed.  Most modules will need to
        be rebuilt or panics may be experienced.
        Bump __FreeBSD_version to 800081.

20090415:
        Anticipate overflowing inp_flags - add inp_flags2.
        This changes most offsets in inpcb, so checking v4 connection
        state will require a world rebuild.
        Bump __FreeBSD_version to 800080.

20090415:
        Add an llentry to struct route and struct route_in6. Modules
        embedding a struct route will need to be recompiled.
        Bump __FreeBSD_version to 800079.

20090414:
        The size of rt_metrics_lite and by extension rtentry has changed.
        Networking administration apps will need to be recompiled.
        The route command now supports show as an alias for get, weighting
        of routes, sticky and nostick flags to alter the behavior of stateful
        load balancing.
        Bump __FreeBSD_version to 800078.

20090408:
        Do not use Giant for kbdmux(4) locking. This is wrong and
        apparently causing more problems than it solves. This will
        re-open the issue where interrupt handlers may race with
        kbdmux(4) in polling mode. Typical symptoms include (but
        not limited to) duplicated and/or missing characters when
        low level console functions (such as gets) are used while
        interrupts are enabled (for example geli password prompt,
        mountroot prompt etc.). Disabling kbdmux(4) may help.

20090407:
        The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
        kernel modules referencing any of the above need to be recompiled.
        Bump __FreeBSD_version to 800075.

20090320:
        GEOM_PART has become the default partition slicer for storage devices,
        replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
        introduces some changes:

        MSDOS/EBR: the devices created from MSDOS extended partition entries
        (EBR) can be named differently than with GEOM_MBR and are now symlinks
        to devices with offset-based names. fstabs may need to be modified.

        BSD: the "geometry does not match label" warning is harmless in most
        cases but it points to problems in file system misalignment with
        disk geometry. The "c" partition is now implicit, covers the whole
        top-level drive and cannot be (mis)used by users.

        General: Kernel dumps are now not allowed to be written to devices
        whose partition types indicate they are meant to be used for file
        systems (or, in case of MSDOS partitions, as something else than
        the "386BSD" type).

        Most of these changes date approximately from 200812.

20090319:
        The uscanner(4) driver has been removed from the kernel. This follows
        Linux removing theirs in 2.6 and making libusb the default interface
        (supported by sane).

20090319:
        The multicast forwarding code has been cleaned up. netstat(1)
        only relies on KVM now for printing bandwidth upcall meters.
        The IPv4 and IPv6 modules are split into ip_mroute_mod and
        ip6_mroute_mod respectively. The config(5) options for statically
        compiling this code remain the same, i.e. 'options MROUTING'.

20090315:
        Support for the IFF_NEEDSGIANT network interface flag has been
        removed, which means that non-MPSAFE network device drivers are no
        longer supported.  In particular, if_ar, if_sr, and network device
        drivers from the old (legacy) USB stack can no longer be built or
        used.

20090313:
        POSIX.1 Native Language Support (NLS) has been enabled in libc and
        a bunch of new language catalog files have also been added.
        This means that some common libc messages are now localized and
        they depend on the LC_MESSAGES environmental variable.

20090313:
        The k8temp(4) driver has been renamed to amdtemp(4) since
        support for Family 10 and Family 11 CPU families was added.

20090309:
        IGMPv3 and Source-Specific Multicast (SSM) have been merged
        to the IPv4 stack. VIMAGE hooks are in but not yet used.

        For kernel developers, the most important changes are that the
        ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
        and this lock has been downgraded to a non-recursive mutex.

        Transport protocols (UDP, Raw IP) are now responsible for filtering
        inbound multicast traffic according to group membership and source
        filters. The imo_multicast_filter() KPI exists for this purpose.
        Transports which do not use multicast (SCTP, TCP) already reject
        multicast by default. Forwarding and receive performance may improve
        as a mutex acquisition is no longer needed in the ip_input()
        low-level input path.  in_addmulti() and in_delmulti() are shimmed
        to new KPIs which exist to support SSM in-kernel.

        For application developers, it is recommended that loopback of
        multicast datagrams be disabled for best performance, as this
        will still cause the lock to be taken for each looped-back
        datagram transmission. The net.inet.ip.mcast.loop sysctl may
        be tuned to 0 to disable loopback by default; it defaults to 1
        to preserve the existing behaviour.

        For systems administrators, to obtain best performance with
        multicast reception and multiple groups, it is always recommended
        that a card with a suitably precise hash filter is used. Hash
        collisions will still result in the lock being taken within the
        transport protocol input path to check group membership.

        If deploying FreeBSD in an environment with IGMP snooping switches,
        it is recommended that the net.inet.igmp.sendlocal sysctl remain
        enabled; this forces 224.0.0.0/24 group membership to be announced
        via IGMP.

        The size of 'struct igmpstat' has changed; netstat needs to be
        recompiled to reflect this.
        Bump __FreeBSD_version to 800070.

20090309:
        libusb20.so.1 is now installed as libusb.so.1 and the ports system
        updated to use it. This requires a buildworld/installworld in order to
        update the library and dependencies (usbconfig, etc). Its advisable to
        rebuild all ports which uses libusb. More specific directions are given
        in the ports collection UPDATING file. Any /etc/libmap.conf entries for
        libusb are no longer required and can be removed.

20090302:
        A workaround is committed to allow the creation of System V shared
        memory segment of size > 2 GB on the 64-bit architectures.
        Due to a limitation of the existing ABI, the shm_segsz member
        of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
        wrong for large segments. Note that limits must be explicitly
        raised to allow such segments to be created.

20090301:
        The layout of struct ifnet has changed, requiring a rebuild of all
        network device driver modules.

20090227:
        The /dev handling for the new USB stack has changed, a
        buildworld/installworld is required for libusb20.

20090223:
        The new USB2 stack has now been permanently moved in and all kernel and
        module names reverted to their previous values (eg, usb, ehci, ohci,
        ums, ...).  The old usb stack can be compiled in by prefixing the name
        with the letter 'o', the old usb modules have been removed.
        Updating entry 20090216 for xorg and 20090215 for libmap may still
        apply.

20090217:
        The rc.conf(5) option if_up_delay has been renamed to
        defaultroute_delay to better reflect its purpose. If you have
        customized this setting in /etc/rc.conf you need to update it to
        use the new name.

20090216:
        xorg 7.4 wants to configure its input devices via hald which does not
        yet work with USB2. If the keyboard/mouse does not work in xorg then
        add
                Option "AllowEmptyInput" "off"
        to your ServerLayout section.  This will cause X to use the configured
        kbd and mouse sections from your xorg.conf.

20090215:
        The GENERIC kernels for all architectures now default to the new USB2
        stack. No kernel config options or code have been removed so if a
        problem arises please report it and optionally revert to the old USB
        stack. If you are loading USB kernel modules or have a custom kernel
        that includes GENERIC then ensure that usb names are also changed over,
        eg uftdi -> usb2_serial_ftdi.

        Older programs linked against the ports libusb 0.1 need to be
        redirected to the new stack's libusb20.  /etc/libmap.conf can
        be used for this:
                # Map old usb library to new one for usb2 stack
                libusb-0.1.so.8 libusb20.so.1

20090209:
        All USB ethernet devices now attach as interfaces under the name ueN
        (eg. ue0). This is to provide a predictable name as vendors often
        change usb chipsets in a product without notice.

20090203:
        The ichsmb(4) driver has been changed to require SMBus slave
        addresses be left-justified (xxxxxxx0b) rather than right-justified.
        All of the other SMBus controller drivers require left-justified
        slave addresses, so this change makes all the drivers provide the
        same interface.

20090201:
        INET6 statistics (struct ip6stat) was updated.
        netstat(1) needs to be recompiled.

20090119:
        NTFS has been removed from GENERIC kernel on amd64 to match
        GENERIC on i386. Should not cause any issues since mount_ntfs(8)
        will load ntfs.ko module automatically when NTFS support is
        actually needed, unless ntfs.ko is not installed or security
        level prohibits loading kernel modules. If either is the case,
        "options NTFS" has to be added into kernel config.

20090115:
        TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
        New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
        800061. User space tools that rely on the size of struct tcpcb in
        tcp_var.h (e.g. sockstat) need to be recompiled.

20081225:
        ng_tty(4) module updated to match the new TTY subsystem.
        Due to API change, user-level applications must be updated.
        New API support added to mpd5 CVS and expected to be present
        in next mpd5.3 release.

20081219:
        With __FreeBSD_version 800060 the makefs tool is part of
        the base system (it was a port).

20081216:
        The afdata and ifnet locks have been changed from mutexes to
        rwlocks, network modules will need to be re-compiled.

20081214:
        __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
        RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
        The new code reduced struct rtentry{} by 16 bytes on 32-bit
        architecture and 40 bytes on 64-bit architecture. The userland
        applications "arp" and "ndp" have been updated accordingly.
        The output from "netstat -r" shows only routing entries and
        none of the L2 information.

20081130:
        __FreeBSD_version 800057 marks the switchover from the
        binary ath hal to source code. Users must add the line:

        options AH_SUPPORT_AR5416

        to their kernel config files when specifying:

        device  ath_hal

        The ath_hal module no longer exists; the code is now compiled
        together with the driver in the ath module.  It is now
        possible to tailor chip support (i.e. reduce the set of chips
        and thereby the code size); consult ath_hal(4) for details.

20081121:
        __FreeBSD_version 800054 adds memory barriers to
        <machine/atomic.h>, new interfaces to ifnet to facilitate
        multiple hardware transmit queues for cards that support
        them, and a lock-less ring-buffer implementation to
        enable drivers to more efficiently manage queueing of
        packets.

20081117:
        A new version of ZFS (version 13) has been merged to -HEAD.
        This version has zpool attribute "listsnapshots" off by
        default, which means "zfs list" does not show snapshots,
        and is the same as Solaris behavior.

20081028:
        dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.

20081009:
        The uhci, ohci, ehci and slhci USB Host controller drivers have
        been put into separate modules. If you load the usb module
        separately through loader.conf you will need to load the
        appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
        controller add the following to loader.conf:

                uhci_load="YES"
                ehci_load="YES"

20081009:
        The ABI used by the PMC toolset has changed.  Please keep
        userland (libpmc(3)) and the kernel module (hwpmc(4)) in
        sync.

20081009:
        atapci kernel module now includes only generic PCI ATA
        driver. AHCI driver moved to ataahci kernel module.
        All vendor-specific code moved into separate kernel modules:
        ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
        atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
        atamarvell, atamicron, atanational, atanetcell, atanvidia,
        atapromise, ataserverworks, atasiliconimage, atasis, atavia

20080820:
        The TTY subsystem of the kernel has been replaced by a new
        implementation, which provides better scalability and an
        improved driver model. Most common drivers have been migrated to
        the new TTY subsystem, while others have not. The following
        drivers have not yet been ported to the new TTY layer:

        PCI/ISA:
                cy, digi, rc, rp, sio

        USB:
                ubser, ucycom

        Line disciplines:
                ng_h4, ng_tty, ppp, sl, snp

        Adding these drivers to your kernel configuration file shall
        cause compilation to fail.

20080818:
        ntpd has been upgraded to 4.2.4p5.

20080801:
        OpenSSH has been upgraded to 5.1p1.

        For many years, FreeBSD's version of OpenSSH preferred DSA
        over RSA for host and user authentication keys.  With this
        upgrade, we've switched to the vendor's default of RSA over
        DSA.  This may cause upgraded clients to warn about unknown
        host keys even for previously known hosts.  Users should
        follow the usual procedure for verifying host keys before
        accepting the RSA key.

        This can be circumvented by setting the "HostKeyAlgorithms"
        option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
        command line.

        Please note that the sequence of keys offered for
        authentication has been changed as well.  You may want to
        specify IdentityFile in a different order to revert this
        behavior.

20080713:
        The sio(4) driver has been removed from the i386 and amd64
        kernel configuration files. This means uart(4) is now the
        default serial port driver on those platforms as well.

        To prevent collisions with the sio(4) driver, the uart(4) driver
        uses different names for its device nodes. This means the
        onboard serial port will now most likely be called "ttyu0"
        instead of "ttyd0". You may need to reconfigure applications to
        use the new device names.

        When using the serial port as a boot console, be sure to update
        /boot/device.hints and /etc/ttys before booting the new kernel.
        If you forget to do so, you can still manually specify the hints
        at the loader prompt:

                set hint.uart.0.at="isa"
                set hint.uart.0.port="0x3F8"
                set hint.uart.0.flags="0x10"
                set hint.uart.0.irq="4"
                boot -s

20080609:
        The gpt(8) utility has been removed. Use gpart(8) to partition
        disks instead.

20080603:
        The version that Linuxulator emulates was changed from 2.4.2
        to 2.6.16. If you experience any problems with Linux binaries
        please try to set sysctl compat.linux.osrelease to 2.4.2 and
        if it fixes the problem contact emulation mailing list.

20080525:
        ISDN4BSD (I4B) was removed from the src tree. You may need to
        update a your kernel configuration and remove relevant entries.

20080509:
        I have checked in code to support multiple routing tables.
        See the man pages setfib(1) and setfib(2).
        This is a hopefully backwards compatible version,
        but to make use of it you need to compile your kernel
        with options ROUTETABLES=2 (or more up to 16).

20080420:
        The 802.11 wireless support was redone to enable multi-bss
        operation on devices that are capable.  The underlying device
        is no longer used directly but instead wlanX devices are
        cloned with ifconfig.  This requires changes to rc.conf files.
        For example, change:
                ifconfig_ath0="WPA DHCP"
        to
                wlans_ath0=wlan0
                ifconfig_wlan0="WPA DHCP"
        see rc.conf(5) for more details.  In addition, mergemaster of
        /etc/rc.d is highly recommended.  Simultaneous update of userland
        and kernel wouldn't hurt either.

        As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
        modules were merged into the base wlan module.  All references
        to these modules (e.g. in kernel config files) must be removed.

20080408:
        psm(4) has gained write(2) support in native operation level.
        Arbitrary commands can be written to /dev/psm%d and status can
        be read back from it.  Therefore, an application is responsible
        for status validation and error recovery.  It is a no-op in
        other operation levels.

20080312:
        Support for KSE threading has been removed from the kernel.  To
        run legacy applications linked against KSE libmap.conf may
        be used.  The following libmap.conf may be used to ensure
        compatibility with any prior release:

        libpthread.so.1 libthr.so.1
        libpthread.so.2 libthr.so.2
        libkse.so.3 libthr.so.3

20080301:
        The layout of struct vmspace has changed. This affects libkvm
        and any executables that link against libkvm and use the
        kvm_getprocs() function. In particular, but not exclusively,
        it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
        The effects are minimal, but it's advisable to upgrade world
        nonetheless.

20080229:
        The latest em driver no longer has support in it for the
        82575 adapter, this is now moved to the igb driver. The
        split was done to make new features that are incompatible
        with older hardware easier to do.

20080220:
        The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
        likewise the kernel option is now GEOM_LINUX_LVM.

20080211:
        The default NFS mount mode has changed from UDP to TCP for
        increased reliability.  If you rely on (insecurely) NFS
        mounting across a firewall you may need to update your
        firewall rules.

20080208:
        Belatedly note the addition of m_collapse for compacting
        mbuf chains.

20080126:
        The fts(3) structures have been changed to use adequate
        integer types for their members and so to be able to cope
        with huge file trees.  The old fts(3) ABI is preserved
        through symbol versioning in libc, so third-party binaries
        using fts(3) should still work, although they will not take
        advantage of the extended types.  At the same time, some
        third-party software might fail to build after this change
        due to unportable assumptions made in its source code about
        fts(3) structure members.  Such software should be fixed
        by its vendor or, in the worst case, in the ports tree.
        FreeBSD_version 800015 marks this change for the unlikely
        case that a portable fix is impossible.

20080123:
        To upgrade to -current after this date, you must be running
        FreeBSD not older than 6.0-RELEASE.  Upgrading to -current
        from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.

20071128:
        The ADAPTIVE_GIANT kernel option has been retired because its
        functionality is the default now.

20071118:
        The AT keyboard emulation of sunkbd(4) has been turned on
        by default. In order to make the special symbols of the Sun
        keyboards driven by sunkbd(4) work under X these now have
        to be configured the same way as Sun USB keyboards driven
        by ukbd(4) (which also does AT keyboard emulation), f.e.:

        Option  "XkbLayout" "us"
        Option  "XkbRules" "xorg"
        Option  "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"

20071024:
        It has been decided that it is desirable to provide ABI
        backwards compatibility to the FreeBSD 4/5/6 versions of the
        PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
        broken with the introduction of PCI domain support (see the
        20070930 entry). Unfortunately, this required the ABI of
        PCIOCGETCONF to be broken again in order to be able to
        provide backwards compatibility to the old version of that
        IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
        again. As for prominent ports this affects neither pciutils
        nor xorg-server this time, the hal port needs to be rebuilt
        however.

20071020:
        The misnamed kthread_create() and friends have been renamed
        to kproc_create() etc. Many of the callers already
        used kproc_start()..
        I will return kthread_create() and friends in a while
        with implementations that actually create threads, not procs.
        Renaming corresponds with version 800002.

20071010:
        RELENG_7 branched.

COMMON ITEMS:

        General Notes
        -------------
        Avoid using make -j when upgrading.  While generally safe, there are
        sometimes problems using -j to upgrade.  If your upgrade fails with
        -j, please try again without -j.  From time to time in the past there
        have been problems using -j with buildworld and/or installworld.  This
        is especially true when upgrading between "distant" versions (eg one
        that cross a major release boundary or several minor releases, or when
        several months have passed on the -current branch).

        Sometimes, obscure build problems are the result of environment
        poisoning.  This can happen because the make utility reads its
        environment when searching for values for global variables.  To run
        your build attempts in an "environmental clean room", prefix all make
        commands with 'env -i '.  See the env(1) manual page for more details.

        When upgrading from one major version to another it is generally best
        to upgrade to the latest code in the currently installed branch first,
        then do an upgrade to the new branch. This is the best-tested upgrade
        path, and has the highest probability of being successful.  Please try
        this approach before reporting problems with a major version upgrade.

        ZFS notes
        ---------
        When upgrading the boot ZFS pool to a new version, always follow
        these two steps:

        1.) recompile and reinstall the ZFS boot loader and boot block
        (this is part of "make buildworld" and "make installworld")

        2.) update the ZFS boot block on your boot drive

        The following example updates the ZFS boot block on the first
        partition (freebsd-boot) of a GPT partitioned drive ad0:
        "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"

        Non-boot pools do not need these updates.

        To build a kernel
        -----------------
        If you are updating from a prior version of FreeBSD (even one just
        a few days old), you should follow this procedure.  It is the most
        failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,

        make kernel-toolchain
        make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
        make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE

        To test a kernel once
        ---------------------
        If you just want to boot a kernel once (because you are not sure
        if it works, or if you want to boot a known bad kernel to provide
        debugging information) run
        make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
        nextboot -k testkernel

        To just build a kernel when you know that it won't mess you up
        --------------------------------------------------------------
        This assumes you are already running a CURRENT system.  Replace
        ${arch} with the architecture of your machine (e.g. "i386",
        "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).

        cd src/sys/${arch}/conf
        config KERNEL_NAME_HERE
        cd ../compile/KERNEL_NAME_HERE
        make depend
        make
        make install

        If this fails, go to the "To build a kernel" section.

        To rebuild everything and install it on the current system.
        -----------------------------------------------------------
        # Note: sometimes if you are running current you gotta do more than
        # is listed here if you are upgrading from a really old current.

        <make sure you have good level 0 dumps>
        make buildworld
        make kernel KERNCONF=YOUR_KERNEL_HERE
                                                        [1]
        <reboot in single user>                         [3]
        mergemaster -p                                  [5]
        make installworld
        mergemaster -i                                  [4]
        make delete-old                                 [6]
        <reboot>


        To cross-install current onto a separate partition
        --------------------------------------------------
        # In this approach we use a separate partition to hold
        # current's root, 'usr', and 'var' directories.   A partition
        # holding "/", "/usr" and "/var" should be about 2GB in
        # size.

        <make sure you have good level 0 dumps>
        <boot into -stable>
        make buildworld
        make buildkernel KERNCONF=YOUR_KERNEL_HERE
        <maybe newfs current's root partition>
        <mount current's root partition on directory ${CURRENT_ROOT}>
        make installworld DESTDIR=${CURRENT_ROOT}
        make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
        make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
        cp /etc/fstab ${CURRENT_ROOT}/etc/fstab                    # if newfs'd
        <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
        <reboot into current>
        <do a "native" rebuild/install as described in the previous section>
        <maybe install compatibility libraries from ports/misc/compat*>
        <reboot>


        To upgrade in-place from 8.x-stable to current
        ----------------------------------------------
        <make sure you have good level 0 dumps>
        make buildworld                                 [9]
        make kernel KERNCONF=YOUR_KERNEL_HERE           [8]
                                                        [1]
        <reboot in single user>                         [3]
        mergemaster -p                                  [5]
        make installworld
        mergemaster -i                                  [4]
        make delete-old                                 [6]
        <reboot>

        Make sure that you've read the UPDATING file to understand the
        tweaks to various things you need.  At this point in the life
        cycle of current, things change often and you are on your own
        to cope.  The defaults can also change, so please read ALL of
        the UPDATING entries.

        Also, if you are tracking -current, you must be subscribed to
        freebsd-current@freebsd.org.  Make sure that before you update
        your sources that you have read and understood all the recent
        messages there.  If in doubt, please track -stable which has
        much fewer pitfalls.

        [1] If you have third party modules, such as vmware, you
        should disable them at this point so they don't crash your
        system on reboot.

        [3] From the bootblocks, boot -s, and then do
                fsck -p
                mount -u /
                mount -a
                cd src
                adjkerntz -i            # if CMOS is wall time
        Also, when doing a major release upgrade, it is required that
        you boot into single user mode to do the installworld.

        [4] Note: This step is non-optional.  Failure to do this step
        can result in a significant reduction in the functionality of the
        system.  Attempting to do it by hand is not recommended and those
        that pursue this avenue should read this file carefully, as well
        as the archives of freebsd-current and freebsd-hackers mailing lists
        for potential gotchas.  The -U option is also useful to consider.
        See mergemaster(8) for more information.

        [5] Usually this step is a noop.  However, from time to time
        you may need to do this if you get unknown user in the following
        step.  It never hurts to do it all the time.  You may need to
        install a new mergemaster (cd src/usr.sbin/mergemaster && make
        install) after the buildworld before this step if you last updated
        from current before 20020224 or from -stable before 20020408.

        [6] This only deletes old files and directories. Old libraries
        can be deleted by "make delete-old-libs", but you have to make
        sure that no program is using those libraries anymore.

        [8] In order to have a kernel that can run the 4.x binaries needed to
        do an installworld, you must include the COMPAT_FREEBSD4 option in
        your kernel.  Failure to do so may leave you with a system that is
        hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
        required to run the 5.x binaries on more recent kernels.  And so on
        for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.

        Make sure that you merge any new devices from GENERIC since the
        last time you updated your kernel config file.

        [9] When checking out sources, you must include the -P flag to have
        cvs prune empty directories.

        If CPUTYPE is defined in your /etc/make.conf, make sure to use the
        "?=" instead of the "=" assignment operator, so that buildworld can
        override the CPUTYPE if it needs to.

        MAKEOBJDIRPREFIX must be defined in an environment variable, and
        not on the command line, or in /etc/make.conf.  buildworld will
        warn if it is improperly defined.
FORMAT:

This file contains a list, in reverse chronological order, of major
breakages in tracking -current.  Not all things will be listed here,
and it only starts on October 16, 2004.  Updating files can found in
previous releases if your system is older than this.

Copyright information:

Copyright 1998-2009 M. Warner Losh.  All Rights Reserved.

Redistribution, publication, translation and use, with or without
modification, in full or in part, in any form or format of this
document are permitted without further permission from the author.

THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED.  IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.

Contact Warner Losh if you have any questions about your use of
this document.

$FreeBSD$