FreeBSD 9.0-RELEASE Release Notes

2.2.1 Boot Loader Changes

The default boot loader menu has been updated.[r222417]

[ia64] The loader(8) loader now supports PBVM (Pre-Boot Virtual Memory). This allows linking the kernel at a fixed virtual address without having to make any assumptions about the physical memory layout. The PBVM also allows fine control of the address where the kernel and its modules are to be loaded.[r219541]

2.2.2 Hardware Support

[powerpc] FreeBSD/powerpc now supports Sony Playstation 3 using the OtherOS feature available on firmwares 3.15 and earlier.[r217044]

A new loader(8) tunable machdep.disable_tsc has been added. Setting this to a non-zero value disables use of TSC (Time Stamp Counter) by turning off boot-time CPU frequency calibration, DELAY(9) with TSC, and using TSC as a CPU ticker. Another new loader(8) tunable machdep.disable_tsc_calibration allows to skip the TSC frequency calibration only. This is useful when one wants to use the nominal frequency of the chip in Intel processors, for example.[r219473, r220577]

[amd64, i386] The FreeBSD usb(4) subsystem now supports USB 3.0 by default.[r223098]

The FreeBSD usb(4) subsystem now supports USB packet filter. This allows to capture packets which go through each USB host controller. The implementation is almost based on bpf(4) code. The userland program usbdump(8) has been added.[r215649]

wpi0: could not map mbuf (error 12)
wpi0: wpi_rx_intr: bus_dmamap_load failed, error 12

2.2.3 Network Protocols

ipfw(8) now supports IPv6 in the fwd action.[r225044]

ipfw(8) now supports the call and return actions. Upon the call number action, the current rule number is saved in the internal stack and ruleset processing continues with the first rule numbered number or higher. The return action takes the rule number saved to internal stack by the latest call action and returns ruleset processing to the first rule with number greater than that saved number.[r223666]

FreeBSD's ipsec(4) support now uses half of the hash size as the authenticator hash size in Hashed Message Authentication Mode (HMAC-SHA-256, HMAC-SHA-384, and HMAC-SHA-512) as described in RFC 4868. This was a fixed 96-bit length in prior releases because the implementation was based on an old Internet draft draft-ietf-ipsec-ciph-sha-256-00. Note that this means 9.0-RELEASE and later are no longer interoperable with the older FreeBSD releases.[r218794]

For Infiniband support, OFED (OpenFabrics Enterprise Distribution) version 1.5.3 has been imported into the base system. Note that this is not built nor installed by default. To build and install them, specify WITH_OFED=yes in /etc/src.conf and rebuild the base system as described in the FreeBSD Handbook.[r219820]

The FreeBSD TCP/IP network stack now supports IPv4 prefixes with /31 as described in RFC 3021, “Using 31-Bit Prefixes on IPv4 Point-to-Point Links”.[r226572]

A bug in the FreeBSD TCP/IP network stack has been fixed. Source address selection could not be performed when multicast options were present but without an interface being specified.[r217169]

A bug in the IPV6_PKTINFO option used in sendmsg(2) has been fixed. The IPV6_USE_MIN_MTU state set by setsockopt(2) was ignored.[r225682]

The FreeBSD TCP/IP network stack now supports the mod_cc(9) pluggable congestion control framework. This allows TCP congestion control algorithms to be implemented as dynamically loadable kernel modules. The following kernel modules are available as of 9.0-RELEASE: cc_chd(4) for the CAIA-Hamilton-Delay algorithm, cc_cubic(4) for the CUBIC algorithm, cc_hd(4) for the Hamilton-Delay algorithm, cc_htcp(4) for the H-TCP algorithm, cc_newreno(4) for the NewReno algorithm, and cc_vegas(4) for the Vegas algorithm. The default algorithm can be set by a new sysctl(8) variable net.inet.tcp.cc.algorithm. The value must be set to one of the names listed by net.inet.tcp.cc.available, and newreno is the default set at boot time. For more detail, see the mod_cc(4) and mod_cc(9) manual pages.[r216109, r216114, r216115, r218152, r218153, r218155]

An h_ertt(4) (Enhanced Round Trip Time) khelp(9) module has been added. This module allows per-connection, low noise estimates of the instantaneous RTT in the TCP/IP network stack with a robust implementation even in the face of delayed acknowledgments and/or TSO (TCP Segmentation Offload) being in use for a connection.[r217806]

A new tcp(4) socket option TCP_CONGESTION has been added. This allows to select or query the congestion control algorithm that the TCP/IP network stack will use for connections on the socket.[r218912]

The ng_ipfw(4) netgraph(4) node now supports IPv6.[r225586]

The ng_one2many(4) netgraph(4) node now supports the XMIT_FAILOVER transmit algorithm. This makes packets deliver out of the first active many hook.[r219127]

The ng_netflow(4) netgraph(4) node now supports NetFlow version 9. A new export9 hook has been added for NetFlow v9 data. Note that data export can be done simultaneously in both version 5 and version 9.[r219183]

2.2.4 Disks and Storage

The ada(4) driver now supports write cache control. A new sysctl(8) variable kern.cam.ada.write_cache determines whether the write cache of ada(4) devices is enabled or not. Setting to 1 enables and 0 disables the write cache, and -1 leaves the device default behavior. sysctl(8) variables kern.cam.ada.N.write_cache can override the configuration in a per-device basis (the default value is -1, which means to use the global setting). Note that the value can be changed at runtime, but it takes effect only after a device reset.[r220412]

The arcmsr(4) driver has been updated to version 1.20.00.22.[r224905]

The cam(4) subsystem now supports the descriptor format sense data of the SPC-3 (SCSI Primary Commands 3) specification.[r226067]

The geom_map(4) GEOM class has been added. This allows to generate multiple geom providers based on a hard-coded layout of a device with no explicit partition table such as embedded flash storage. For more information, see the geom_map(4) manual page.[r220559]

The gpart(8) GEOM class now supports the following aliases for the MBR and EBR schemes: fat32, ebr, linux-data, linux-raid, and linux-swap.[r218014]

The gpart(8) GEOM class now supports bios-boot GUID for the GPT scheme which is used in GRUB 2 loader.[r218014]

The graid(8) GEOM class has been added. This is a replacement of the ataraid(4) driver supporting various BIOS-based software RAID.[r219974]

The sysctl(8) variable kern.geom.confxml now contains information about disk identification in an <ident> tag and disk model strings in a <descr> tag.[r219056]

The md(4) memory-backed pseudo disk device driver now supports a sysctl(8) variable vm.md_malloc_wait to specify whether a malloc-backed disk will use M_WAITOK or M_NOWAIT for malloc(9) calls. The M_WAITOK setting can prevent memory allocation failure under high load. If it is set to 0, a malloc-backed disk uses M_NOWAIT for memory allocation. The default value is 0.[r216793]

A bug in the mmc(4) driver that could cause device detection to fail has been fixed.[r216941, r217509]

The mxge(4) driver has been updated.[r223958]

A tws(4) driver for 3ware 9750 SATA+SAS 6Gb/s RAID controllers has been added.[r226115]

2.2.5 File Systems

The FreeBSD Fast File System now supports softupdates journaling. It introduces a intent log into a softupdates-enabled file system which eliminates the need for background fsck(8) even on unclean shutdown. This can be enabled in a per-filesystem basis by using the -j flag of the newfs(8) utility or the -j enable option of the tunefs(8) utility. Note that the 9.0-RELEASE installer automatically enables softupdates journaling for newly-created UFS file systems.[r207141, r218726]

The FreeBSD Fast File System now supports the TRIM command when freeing data blocks. A new flag -t in the newfs(8) and tunefs(8) utilities sets the TRIM-enable flag for a file system. The TRIM-enable flag makes the file system send a delete request to the underlying device for each freed block. The TRIM command is specified as a Data Set Management Command in the ATA8-ACS2 standard to carry the information related to deleted data blocks to a device, especially for a SSD (Solid-State Drive) for optimization.[r216796]

A new flag -E has been added to the newfs(8) and fsck_ffs(8) utilities. This clears unallocated blocks, notifying the underlying device that they are not used and that their contents may be discarded. This is useful in fsck_ffs(8) for file systems which have been mounted on systems without TRIM support, or with TRIM support disabled, as well as filesystems which have been copied from one device to another.[r221233]

The FreeBSD NFS subsystem has been updated. The new implementation supports NFS version 4 in addition to 2 and 3. The kernel options for the NFS server and client are changed from NFSSERVER and NFSCLIENT to NFSD and NFSCL. sysctl(8) variables which start with vfs.nfssrv. have been renamed to vfs.nfsd.. The NFS server now supports vfs.nfsd.server_max_nfsvers and vfs.nfsd.server_min_nfsvers sysctl(8) variables to specify the maximum and the minimum NFS version number which the server accepts. The default value is set to 3 and 2, respectively.[r221124]

To enable NFSv4, the following variables are needed on the server side in rc.conf(5):

nfsv_server_enable="YES"
nfsv4_server_enable="YES"
nfsuserd_enable="YES"

and the following line is needed in /etc/exports:

V4: /

For more information about NFSv4 and its configuration, see the nfsv4(4) and exports(5) manual pages.

The FreeBSD NFS subsystem now supports a nocto mount option. This disables the close-to-open cache coherency check at open time. This option may improve performance for read-only mounts, but should only be used only if the data on the server changes rarely. The mount_nfs(8) utility now also supports this flag keyword.[r221436]

A loader(8) tunable vfs.typenumhash has been added and set to 1 by default. This enables to use a hash calculation on the file system identification number internally used in the kernel. This fixes the “Stale NFS file handle” error on NFS clients when upgrading or rebuilding the kernel on the NFS server due to unexpected change of these identification number values.[r225537]

The FreeBSD ZFS subsystem has been updated to the SPA (Storage Pool Allocator, also known as zpool) version 28. It now supports data deduplication, triple parity RAIDZ (raidz3), snapshot holds, log device removal, zfs diff, zpool split, zpool import -F, and read-only zpool import.[r219089]

3.2.1 Update of dialog

The dialog library is used in FreeBSD's new installer and the FreeBSD Ports Collection to display a dialog window and allow users to select various options. Note that it is updated in 9.0-RELEASE and there are several differences in key operations which might confuse users who are familiar with releases prior to 9.0-RELEASE. For example, pushing the enter key in a checklist window will no longer check an item. The new version consistently uses space bar for selecting an item and the enter key for OK/Cancel selection.

3.2.2 Partition Metadata Integrity Check

FreeBSD now checks the integrity of partition metadata when a partition table is found on a disk though the GEOM PART subsystem. This detection is automatically performed when a disk device is ready. The GEOM PART class in the kernel verifies all generic partition parameters obtained from the disk metadata, and if some inconsistency is detected, the partition table will be rejected with the following diagnostic message:

GEOM_PART: Integrity check failed

This integrity check is enabled by default. On a system prior to 9.0-RELEASE, the inconsistencies were silently ignored. Therefore, there is a possibility that this prevents a system from booting after upgrading it to 9.0-RELEASE. More specifically, the kernel cannot mount the system partition at boot time in some cases.

If this happens, a loader(8) tunable kern.geom.part.check_integrity can be used as a workaround. Enter the following lines in the loader(8) prompt at boot time:

set kern.geom.part.check_integrity="0"
boot

These commands temporarily disable the integrity check. If it was the cause of the boot failure, the FreeBSD kernel should detect the partitions as the prior release did, after entering the commands. This configuration can be added into /boot/loader.conf as follows:

kern.geom.part.check_integrity="0"

To check inconsistent metadata after booting on the system, use the gpart(8) utility on the system. A corrupted entry will be displayed like the following:

% gpart show
=>        63  1953525104  mirror/gm0  MBR  (931G) [CORRUPT]
          63  1953525105           1  freebsd  [active]  (931G)

For more information, see the gpart(8) manual page.

3.2.3 ATA/SATA subsystem now cam(4)-based

In 9.0-RELEASE, the FreeBSD ATA/SATA disk subsystem has been replaced with a new cam(4)-based implementation. cam(4) stands for Common Access Method, which is an implementation of an API set originally for SCSI-2 and standardized as "SCSI-2 Common Access Method Transport and SCSI Interface Module". FreeBSD has used the cam(4) subsystem to handle SCSI devices since 3.X.

Although the new cam(4)-based ATA/SATA subsystem provides various functionality which the old ata(4) did not have, it also has some incompatibilities:

• An ATA/SATA disk is now recognized as a device node with a name ada0 instead of ad0. Currently, a symbolic link /dev/ad0 is automatically generated for /dev/ada0 to keep backward compatibility. This symbolic link generation can be controlled by a kern.cam.ada.legacy_aliases (enabled by default). You might want to update /etc/fstab and/or consider using volume labels (see glabel(8) for more details) for specifying each file system to be mounted.

• The atacontrol(8) utility cannot be used for cam(4)-based devices. The camcontrol(8) utility is a replacement.

• ataraid(4) software RAID is now supported by the graid(8) GEOM class. It generates a device node with a name /dev/raid/r0 if you previously had /dev/ar0. Note that this is not enabled by default. To enable it, enter the following line in the loader(8) prompt:

  set geom_raid_load="YES"
  boot
  

  or add the following line to /boot/loader.conf:

  geom_raid_load="YES"
  

  and reboot the system. A symbolic link like /dev/ar0 will NOT be generated for /dev/raid/r0. Therefore, if your system used /dev/ar0 as the root partition, mounting local file systems will fail because it is renamed to /dev/raid/r0. You need to update /etc/fstab manually in that case.

• The burncd(8) utility does not work with cam(4)-based devices. Use the cdrecord(1) utility in sysutils/cdrtools instead.

3.2.4 Network Configuration Changes in /etc/rc.conf

Although variables in rc.conf(5) are basically compatible with earlier releases, ones related to network configuration are changed because of reorganization of the rc(8) scripts.

• An address configuration now always needs an address family keyword. For example, the following line

  ifconfig_em0="192.168.2.1 netmask 255.255.255.0"
  

  should be

  ifconfig_em0="inet 192.168.2.1 netmask 255.255.255.0"
  

  Although the old convention is still supported in the existing variables for backward compatibility, some new variables do not support it.

• The ifconfig_IF_alias0 variable now requires an address family keyword to support non-IPv4 address families. For instance,

  ifconfig_em0_alias0="192.168.2.10 netmask 255.255.255.255"
  

  should be

  ifconfig_em0_alias0="inet 192.168.2.10 netmask 255.255.255.255"
  

  Different address families can coexist like the following:

  ifconfig_em0_alias0="inet 192.168.2.10 netmask 255.255.255.255"
  ifconfig_em0_alias1="inet6 2001:db8:1::1 prefixlen 64"
  

  Note that IPv6 alias configurations in ifconfig_IF_aliasN will be ignored when no ifconfig_IF_ipv6 variable is defined because it determines whether IPv6 functionality is enabled on that interface or not (this variable will be explained later).

• All alias and static routing configurations through rc.conf(5) variables will be deactivated when invoking rc(8) scripts or the service(8) command with the stop keyword.

  # service netif stop em0
  

  stops the interface em0.

  # service routing stop
  

  deactivates all static route configurations.

  Releases prior to FreeBSD 9.0-RELEASE did not support this functionality properly for non-IPv4 protocols.

• IPv6 configuration handling has been changed in the following way. Before in-depth explanations, here is a before-and-after example. What was previously:

  ifconfig_em0="192.168.2.1 netmask 255.255.255.0"
  ifconfig_em0_alias0="192.168.2.2 netmask 255.255.255.255"
  
  ipv6_enable="YES"
  ipv6_ifconfig_em0="2001:db8:1::1 prefixlen 64"
  ipv6_ifconfig_em0_alias0="2001:db8:2::1 prefixlen 64"
  # em1 uses SLAAC for IPv6 address configuration
  

  should be in 9.0-RELEASE:

  ifconfig_em0="inet 192.168.2.1 netmask 255.255.255.0"
  ifconfig_em0_ipv6="inet6 2001:db8:1::1 prefixlen 64 accept_rtadv"
  ifconfig_em0_alias0="inet 192.168.2.2 netmask 255.255.255.255"
  ifconfig_em0_alias1="inet6 2001:db8:2::1 prefixlen 64"
  
  ifconfig_em1_ipv6="inet6 accept_rtadv"
  

  More specific explanations of the changes are as follows:

  • The ipv6_enable variable is deprecated. IPv6 functionality on the system is enabled by default. No IPv6 communication will happen if you configure no IPv6 address.

    9.0-RELEASE now supports intermediate configurations between a host and a router IPv6 node. The ipv6_enable variable assumed that the system was a host node when ipv6_gateway_enable was set to NO (default), and a router node if not. A host node always accepted ICMPv6 Router Advertise messages, and a router did not.

    In 9.0-RELEASE, this model is still applied but on a per-interface basis, not a system-wide basis. Specifically, if an interface has an ACCEPT_RTADV flag, RA messages will be accepted on that interface for SLAAC (StateLess Address AutoConfiguration) regardless of whether the packet forwarding is enabled or not.

    In addition to them, a per-interface flag NO_RADR and a sysctl(8) variable net.inet6.ip6.rfc6204w3 have been added. This controls whether default router list information via RA messages on an RA-accepting interface should be ignored or not. In an IPv6 router model, it is not supposed to accept RA messages as an information source for the default router list. Because of that, FreeBSD 9.0-RELEASE ignores the default router list part when IPv6 packet forwarding is enabled, even if the interface has an ACCEPT_RTADV flag. However, this can make for a difficult situation when the system has to work as a CPE (Customer Premises Equipment) which needs RA messages from the upstream network for network configuration and acts as a router for the LAN simultaneously. For more information about this kind of configuration, see RFC 6204.

    To support this kind of configuration, the ipv6_cpe_wanif variable in rc.conf(5) can be used.

    ipv6_gateway_enable="YES"
    ipv6_cpe_wanif="em0"
    

    means the em0 interface accepts RA messages and the default router information in them, and the other interfaces ignore the default router information part even when the ACCEPT_RTADV flag is set on them.

    ipv6_cpe_wanif handling internally sets the net.inet6.ip6.rfc6204w3 and the net.inet6.ip6.no_radr sysctl(8) variables to 1. Note that both are set to 0 by default. When the former is set to 1, FreeBSD accepts the default router list even when IPv6 packet forwarding is enabled. Note that a system administrator needs to set a NO_RADR flag on the other RA-accepting interfaces, if any, to prevent it from accepting unexpected default router information. The latter variable means the NO_RADR flag is automatically set on them.

    If ipv6_enable="YES" is defined in FreeBSD 9.0-RELEASE, it sets ipv6_activate_all_interfaces="YES" in /etc/rc.conf and the inet6 accept_rtadv ifconfig(8) option on all network interfaces. Note that this is only for backward compatibility. The ipv6_enable should not be used in FreeBSD 9.0-RELEASE.

  • The ipv6_ifconfig_IF variable is renamed to ifconfig_IF_ipv6. This variable controls whether IPv6 functionality should be enabled on that interface or not. If ifconfig_IF_ipv6, is not set, there is no IPv6 functionality on the interface IF.

    Note that the ifconfig_IF_ipv6 variable always needs the address family keyword inet6. If you need an automatic link-local address only, the following line is enough:

    ifconfig_em0_ipv6="inet6 auto_linklocal"
    

    If you need full-blown IPv6 functionality on all interfaces like prior releases with ipv6_enable="YES", including ones with no ifconfig_IF_ipv6 line, you might want to use the ipv6_activate_all_interfaces variable as explained later.

    If ipv6_ifconfig_IF="..." is defined in FreeBSD 9.0-RELEASE, it means ifconfig_IF_ipv6="inet6 ...". Note that this is only for backward compatibility. The inet6 address family keyword is required for ifconfig_IF_ipv6, but was NOT required for ipv6_ifconfig_IF. The ipv6_ifconfig_IF variables should not be used in 9.0-RELEASE.

  • An interface with no corresponding ifconfig_IF_ipv6 variable is marked with an IFDISABLED flag by devd(8) daemon. This flag means IPv6 communication is disabled on that interface. This can also be found in output of ifconfig(8):

    % ifconfig em0
    em0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
            options=9b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM>
            ether xx:xx:xx:xx:xx:xx
            inet 192.168.2.1 netmask 0xffffff00 broadcast 192.168.2.255
            nd6 options=3<PERFORMNUD,IFDISABLED,ACCEPT_RTADV>
            media: Ethernet autoselect (1000baseT <full-duplex>)
            status: active
    

    To enable IPv6 functionality, this flag should be removed first. There are several ways to do so. Adding an IPv6 address automatically removes this flag. It is possible to remove this flag explicitly by using the following command:

    # ifconfig em0 inet6 -ifdisabled
    

    Note that defining an ifconfig_IF_ipv6 is the most reasonable way to activate IPv6 functionality on that interface. This IFDISABLED flag is to prevent unintended IPv6 communications in an IPv4-only environment even when the interface has an IPv6 link-local address. If you need full-blown IPv6 functionality on all interfaces, you might want to use the ipv6_activate_all_interfaces variable as explained later.

  • The sysctl(8) variable net.inet6.ip6.accept_rtadv has been changed. It was a system-wide configuration knob which controlled whether the system accepts ICMPv6 Router Advertisement messages or not. In FreeBSD 9.0-RELEASE, this knob is converted into a per-interface inet6 accept_rtadv ifconfig(8) option. Although the sysctl(8) variable is available still in FreeBSD 9.0-RELEASE, it now controls whether the per-interface option is set by default or not. The default value is 0 (not accept the RA messages).

  • The sysctl(8) variable net.inet6.ip6.auto_linklocal has been changed. It was a system-wide configuration knob which controlled whether an IPv6 link-local address was generated on a network interface when it became up. In FreeBSD 9.0-RELEASE, this knob is converted into a per-interface inet6 auto_linklocal ifconfig(8) option. Although the sysctl(8) variable is still available in FreeBSD 9.0-RELEASE, it now controls whether the per-interface option is set by default or not. The default value is 1 (generate a link-local automatically).

  • The functionality of ipv6_ifconfig_IF_alias0 is integrated into ifconfig_IF_alias0. Note that address family keywords are always required:

    ifconfig_em0_alias0="inet 192.168.2.10 netmask 255.255.255.255"
    ifconfig_em0_alias1="inet6 2001:db8:1::1 prefixlen 64
    

    Although ipv6_ifconfig_IF_aliasN is still usable in FreeBSD 9.0-RELEASE, it is only for backward compatibility.

  • A new ipv6_activate_all_interfaces variable has been added. If this variable is set to YES, the IFDISABLED option will not be added even if ifconfig_IF_ipv6 variables are not defined. This can prevent IFDISABLED on dynamically-added interfaces such as ppp(4), tap(4), and ng_iface(4) where defining ifconfig_IF_ipv6 in advance is difficult.

3.2.5 Openresolv and /etc/resolv.conf

The resolvconf(8) utility has been added and it now handles updating the resolv.conf(5) file. Direct modifications to /etc/resolv.conf can be overwritten by network configuration utilities such as dhclient(8) and rtsold(8).

3.2.6 Disk Partition Management Utilities

In earlier releases various utilities were available to manage disk partition information. They are deprecated in favor of the gpart(8) utility. Specifically, the fdisk(8), disklabel(8) bsdlabel(8), and sunlabel(8) utilities are no longer supported actively though these are still available for backward compatibility.