1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 http://www.freebsd.org/doc/en_US.ISO8859-1/books/handbook/makeworld.html
11 Items affecting the ports and packages system can be found in
12 /usr/ports/UPDATING. Please read that file before running portupgrade.
15 A new compression method (lz4) has been merged to. Please refer to
16 zpool-features(7) for more information.
18 Please refer to the "ZFS notes" section of this file for information
19 on upgrading boot ZFS pools.
22 The VFS KBI was changed with the merge of several nullfs
23 optimizations and fixes. All filesystem modules must be
27 With the addition of auditdistd(8), a new auditdistd user is now
28 depended on during installworld. "mergemaster -p" can be used to add
29 the user prior to installworld, as documented in the handbook.
35 A new version of ZFS (pool version 5000) has been merged to 9-STABLE.
36 Starting with this version the old system of ZFS pool versioning
37 is superseded by "feature flags". This concept enables forward
38 compatibility against certain future changes in functionality of ZFS
39 pools. The first two read-only compatible "feature flags" for ZFS
40 pools are "com.delphix:async_destroy" and "com.delphix:empty_bpobj".
41 For more information read the new zpool-features(7) manual page.
42 Please refer to the "ZFS notes" section of this file for information
43 on upgrading boot ZFS pools.
46 The commit introducing bsd.compiler.mk breaks the traditional
47 building of kernels before this point. Add -m ${SRC}/share/mk
48 (for the right value of SRC) to your command lines to work
49 around; update your useland to a point after this; or use the
50 buildkernel/installkernel top-level targets. See also 20120829.
53 The IPFIREWALL_FORWARD kernel option has been removed. Its
54 functionality now turned on by default.
57 The random(4) support for the VIA hardware random number
58 generator (`PADLOCK') is no longer enabled unconditionally.
59 Add the PADLOCK_RNG option in the custom kernel config if
60 needed. The GENERIC kernels on i386 and amd64 do include the
61 option, so the change only affects the custom kernel
65 The amd64 kernel now uses xsetbv, xrstor instructions. To compile with
66 the traditional method, you must update your system with an installworld
67 before the kernel will build. The documented make buildkernel/installkernel
68 interfaces (coupled with fresh make kernel-toolchain) continue to work.
71 The sparc64 ZFS loader has been changed to no longer try to auto-
72 detect ZFS providers based on diskN aliases but now requires these
73 to be explicitly listed in the OFW boot-device environment variable.
76 Now unix domain sockets behave "as expected" on nullfs(5). Previously
77 nullfs(5) did not pass through all behaviours to the underlying layer,
78 as a result if we bound to a socket on the lower layer we could connect
79 only to the lower path; if we bound to the upper layer we could connect
80 only to the upper path. The new behavior is one can connect to both the
81 lower and the upper paths regardless what layer path one binds to.
84 The acpi_wmi(4) status device /dev/wmistat has been renamed to
88 A new VOP_ADVISE() was added to support posix_fadvise(2). All
89 filesystem modules must be recompiled.
92 The interface of the VOP_VPTOCNP(9) changed, now the returned
93 vnode shall be referenced, previously it was required to be
94 only held. All in-tree filesystems are converted.
100 The broken amd(4) driver has been replaced with esp(4) in the amd64,
101 i386 and pc98 GENERIC kernel configuration files.
104 This commit modifies vfs_register() so that it uses a hash
105 calculation to set vfc_typenum, which is enabled by default.
106 The first time a system is booted after this change, the
107 vfc_typenum values will change for all file systems. The
108 main effect of this is a change to the NFS server file handles
109 for file systems that use vfc_typenum in their fsid, such as ZFS.
110 It will, however, prevent vfc_typenum from changing when file
111 systems are loaded in a different order for subsequent reboots.
112 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
113 until you are ready to remount all NFS clients after a reboot.
116 Bump the shared library version numbers for libraries that
117 do not use symbol versioning, have changed the ABI compared
118 to stable/8 and which shared library version was not bumped.
119 Done as part of 9.0-RELEASE cycle.
122 During the merge of Capsicum features, the fget(9) KPI was modified.
123 This may require the rebuilding of out-of-tree device drivers --
124 issues have been reported specifically with the nVidia device driver.
125 __FreeBSD_version is bumped to 900041.
127 Also, there is a period between 20110811 and 20110814 where the
128 special devices /dev/{stdin,stdout,stderr} did not work correctly.
129 Building world from a kernel during that window may not work.
132 The packet filter (pf) code has been updated to OpenBSD 4.5.
133 You need to update userland tools to be in sync with kernel.
134 This update breaks backward compatibility with earlier pfsync(4)
135 versions. Care must be taken when updating redundant firewall setups.
138 The following sysctls and tunables are retired on x86 platforms:
140 machdep.hlt_logical_cpus
141 The following sysctl is retired:
142 machdep.hyperthreading_allowed
143 The sysctls were supposed to provide a way to dynamically offline and
144 online selected CPUs on x86 platforms, but the implementation has not
145 been reliable especially with SCHED_ULE scheduler.
146 machdep.hyperthreading_allowed tunable is still available to ignore
147 hyperthreading CPUs at OS level.
148 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
149 where X is an APIC ID of a CPU. Be advised, though, that disabling
150 CPUs in non-uniform fashion will result in non-uniform topology and
151 may lead to sub-optimal system performance with SCHED_ULE, which is
155 cpumask_t type is retired and cpuset_t is used in order to describe
159 Changes to ifconfig(8) for dynamic address family detection mandate
160 that you are running a kernel of 20110525 or later. Make sure to
161 follow the update procedure to boot a new kernel before installing
165 Support for sun4v architecture is officially dropped
168 Several KPI breaking changes have been committed to the mii(4) layer,
169 the PHY drivers and consequently some Ethernet drivers using mii(4).
170 This means that miibus.ko and the modules of the affected Ethernet
171 drivers need to be recompiled.
173 Note to kernel developers: Given that the OUI bit reversion problem
174 was fixed as part of these changes all mii(4) commits related to OUIs,
175 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
176 handling, no longer can be merged verbatim to stable/8 and previous
180 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
181 into their kernel configurations along with 'device pci'.
184 The default NFS client is now the new NFS client, so fstype "newnfs"
185 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
186 Although mounts via fstype "nfs" will usually work without userland
187 changes, it is recommended that the mount(8) and mount_nfs(8)
188 commands be rebuilt from sources and that a link to mount_nfs called
189 mount_oldnfs be created. The new client is compiled into the
190 kernel with "options NFSCL" and this is needed for diskless root
191 file systems. The GENERIC kernel configs have been changed to use
192 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
193 To use the regular/old client, you can "mount -t oldnfs ...". For
194 a diskless root file system, you must also include a line like:
196 vfs.root.mountfrom="oldnfs:"
198 in the boot/loader.conf on the root fs on the NFS server to make
199 a diskless root fs use the old client.
202 The GENERIC kernels for all architectures now default to the new
203 CAM-based ATA stack. It means that all legacy ATA drivers were
204 removed and replaced by respective CAM drivers. If you are using
205 ATA device names in /etc/fstab or other places, make sure to update
206 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
207 where 'Y's are the sequential numbers starting from zero for each type
208 in order of detection, unless configured otherwise with tunables,
209 see cam(4)). There will be symbolic links created in /dev/ to map
210 old adX devices to the respective adaY. They should provide basic
211 compatibility for file systems mounting in most cases, but they do
212 not support old user-level APIs and do not have respective providers
213 in GEOM. Consider using updated management tools with new device names.
215 It is possible to load devices ahci, ata, siis and mvs as modules,
216 but option ATA_CAM should remain in kernel configuration to make ata
217 module work as CAM driver supporting legacy ATA controllers. Device ata
218 still can be used in modular fashion (atacore + ...). Modules atadisk
219 and atapi* are not used and won't affect operation in ATA_CAM mode.
220 Note that to use CAM-based ATA kernel should include CAM devices
221 scbus, pass, da (or explicitly ada), cd and optionally others. All of
222 them are parts of the cam module.
224 ataraid(4) functionality is now supported by the RAID GEOM class.
225 To use it you can load geom_raid kernel module and use graid(8) tool
226 for management. Instead of /dev/arX device names, use /dev/raid/rX.
228 No kernel config options or code have been removed, so if a problem
229 arises, please report it and optionally revert to the old ATA stack.
230 In order to do it you can remove from the kernel config:
235 , and instead add back:
236 device atadisk # ATA disk drives
237 device ataraid # ATA RAID drives
238 device atapicd # ATAPI CDROM drives
239 device atapifd # ATAPI floppy drives
240 device atapist # ATAPI tape drives
243 The default NFS server has been changed to the new server, which
244 was referred to as the experimental server. If you need to switch
245 back to the old NFS server, you must now put the "-o" option on
246 both the mountd and nfsd commands. This can be done using the
247 mountd_flags and nfs_server_flags rc.conf variables until an
248 update to the rc scripts is committed, which is coming soon.
251 The GNU Objective-C runtime library (libobjc), and other Objective-C
252 related components have been removed from the base system. If you
253 require an Objective-C library, please use one of the available ports.
256 ath(4) has been split into bus- and device- modules. if_ath contains
257 the HAL, the TX rate control and the network device code. if_ath_pci
258 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
259 contains the AHB glue. Users need to load both if_ath_pci and if_ath
260 in order to use ath on everything else.
262 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
263 need to load if_ath and if_ath_pci for ath(4) operation.
266 As part of the replacement of sysinstall, the process of building
267 release media has changed significantly. For details, please re-read
268 release(7), which has been updated to reflect the new build process.
271 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
272 is the last available version under GPLv2. It brings a number of new
273 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
274 4.1 and SSE 4.2), better support for powerpc64, a number of new
275 directives, and lots of other small improvements. See the ChangeLog
276 file in contrib/binutils for the full details.
279 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
280 compliant, and will now use half of hash for authentication.
281 This will break interoperability with all stacks (including all
282 actual FreeBSD versions) who implement
283 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
285 The only workaround with such peers is to use another HMAC
286 algorithm for IPsec ("phase 2") authentication.
289 Remove the uio_yield prototype and symbol. This function has
290 been misnamed since it was introduced and should not be
291 globally exposed with this name. The equivalent functionality
292 is now available using kern_yield(curthread->td_user_pri).
293 The function remains undocumented.
296 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
297 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
298 for scalar sysctls is defined but disabled. Code that needs
299 UQUAD to pass the type checking that must compile on older
300 systems where the define is not present can check against
301 __FreeBSD_version >= 900030.
303 The system dialog(1) has been replaced with a new version previously
304 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
305 with the previous version, but the libdialog associated with it has
306 a largely incompatible API. As such, the original version of libdialog
307 will be kept temporarily as libodialog, until its base system consumers
308 are replaced or updated. Bump __FreeBSD_version to 900030.
311 If you are trying to run make universe on a -stable system, and you get
312 the following warning:
313 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
314 unknown. config(8) likely too old."
315 or something similar to it, then you must upgrade your -stable system
316 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
317 stable/8) or build the config from the latest stable/8 branch and
318 install it on your system.
320 Prior to this date, building a current universe on 8-stable system from
321 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
322 error in the first kernel build phase. A new config on those old
323 systems will fix that problem for older versions of -current.
326 The TCP stack has been modified to allow Khelp modules to interact with
327 it via helper hook points and store per-connection data in the TCP
328 control block. Bump __FreeBSD_version to 900029. User space tools that
329 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
333 Generic IEEE 802.3 annex 31B full duplex flow control support has been
334 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
335 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
336 to take advantage of it instead of using custom implementations. This
337 means that these drivers now no longer unconditionally advertise
338 support for flow control but only do so if flow control is a selected
339 media option. This was implemented in the generic support that way in
340 order to allow flow control to be switched on and off via ifconfig(8)
341 with the PHY specific default to typically off in order to protect
342 from unwanted effects. Consequently, if you used flow control with
343 one of the above mentioned drivers you now need to explicitly enable
345 ifconfig bge0 media auto mediaopt flowcontrol
347 Along with the above mentioned changes generic support for setting
348 1000baseT master mode also has been added and brgphy(4), ciphy(4),
349 e1000phy(4) as well as ip1000phy(4) have been converted to take
350 advantage of it. This means that these drivers now no longer take the
351 link0 parameter for selecting master mode but the master media option
352 has to be used instead, for example like in the following:
353 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
355 Selection of master mode now is also available with all other PHY
356 drivers supporting 1000baseT.
359 The TCP stack has received a significant update to add support for
360 modularised congestion control and generally improve the clarity of
361 congestion control decisions. Bump __FreeBSD_version to 900025. User
362 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
363 sockstat) need to be recompiled.
366 The man(1) utility has been replaced by a new version that no longer
367 uses /etc/manpath.config. Please consult man.conf(5) for how to
368 migrate local entries to the new format.
371 The copyright strings printed by login(1) and sshd(8) at the time of a
372 new connection have been removed to follow other operating systems and
376 A workaround for a fixed ld bug has been removed in kernel code,
377 so make sure that your system ld is built from sources after
378 revision 210245 from 2010-07-19 (r211583 if building head kernel
379 on stable/8, r211584 for stable/7; both from 2010-08-21).
380 A symptom of incorrect ld version is different addresses for
381 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
384 The $ipv6_prefer variable in rc.conf(5) has been split into
385 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
387 The $ip6addrctl_policy is a variable to choose a pre-defined
388 address selection policy set by ip6addrctl(8). A value
389 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
392 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
393 flag (see an entry of 20090926) is set on an interface with no
394 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
395 security reason. If you want IPv6 link-local address on all
396 interfaces by default, set this to "YES".
398 The old ipv6_prefer="YES" is equivalent to
399 ipv6_activate_all_interfaces="YES" and
400 ip6addrctl_policy="ipv6_prefer".
403 DTrace has grown support for userland tracing. Due to this, DTrace is
404 now i386 and amd64 only.
405 dtruss(1) is now installed by default on those systems and a new
406 kernel module is needed for userland tracing: fasttrap.
407 No changes to your kernel config file are necessary to enable
408 userland tracing, but you might consider adding 'STRIP=' and
409 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
410 to have informative userland stack traces in DTrace (ustack).
413 The acpi_aiboost(4) driver has been removed in favor of the new
414 aibs(4) driver. You should update your kernel configuration file.
417 BSD grep has been imported to the base system and it is built by
418 default. It is completely BSD licensed, highly GNU-compatible, uses
419 less memory than its GNU counterpart and has a small codebase.
420 However, it is slower than its GNU counterpart, which is mostly
421 noticeable for larger searches, for smaller ones it is measurable
422 but not significant. The reason is complex, the most important factor
423 is that we lack a modern and efficient regex library and GNU
424 overcomes this by optimizing the searches internally. Future work
425 on improving the regex performance is planned, for the meantime,
426 users that need better performance, can build GNU grep instead by
427 setting the WITH_GNU_GREP knob.
430 Due to the import of powerpc64 support, all existing powerpc kernel
431 configuration files must be updated with a machine directive like this:
432 machine powerpc powerpc
434 In addition, an updated config(8) is required to build powerpc kernels
438 A new version of ZFS (version 15) has been merged to -HEAD.
439 This version uses a python library for the following subcommands:
440 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
441 For full functionality of these commands the following port must
442 be installed: sysutils/py-zfs
445 'vm_page's are now hashed by physical address to an array of mutexes.
446 Currently this is only used to serialize access to hold_count. Over
447 time the page queue mutex will be peeled away. This changes the size
448 of pmap on every architecture. And requires all callers of vm_page_hold
449 and vm_page_unhold to be updated.
452 WITH_CTF can now be specified in src.conf (not recommended, there
453 are some problems with static executables), make.conf (would also
454 affect ports which do not use GNU make and do not override the
455 compile targets) or in the kernel config (via "makeoptions
457 When WITH_CTF was specified there before this was silently ignored,
458 so make sure that WITH_CTF is not used in places which could lead
459 to unwanted behavior.
462 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
463 to allow 32-bit compatibility on non-x86 platforms. All kernel
464 configurations on amd64 and ia64 platforms using these options must
465 be modified accordingly.
468 The utmp user accounting database has been replaced with utmpx,
469 the user accounting interface standardized by POSIX.
470 Unfortunately the semantics of utmp and utmpx don't match,
471 making it practically impossible to support both interfaces.
472 The user accounting database is used by tools like finger(1),
473 last(1), talk(1), w(1) and ac(8).
475 All applications in the base system use utmpx. This means only
476 local binaries (e.g. from the ports tree) may still use these
477 utmp database files. These applications must be rebuilt to make
480 After the system has been upgraded, it is safe to remove the old
481 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
482 assuming their contents is of no importance anymore. Old wtmp
483 databases can only be used by last(1) and ac(8) after they have
484 been converted to the new format using wtmpcvt(1).
487 Introduce the kernel thread "deadlock resolver" (which can be enabled
488 via the DEADLKRES option, see NOTES for more details) and the
489 sleepq_type() function for sleepqueues.
492 The rc.firewall and rc.firewall6 were unified, and
493 rc.firewall6 and rc.d/ip6fw were removed.
494 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
495 variables are obsoleted. Instead, the following new rc
496 variables are added to rc.d/ipfw:
498 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
499 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
500 firewall_simple_onet_ipv6, firewall_trusted_ipv6
502 The meanings correspond to the relevant IPv4 variables.
508 The default terminal emulation for syscons(4) has been changed
509 from cons25 to xterm on all platforms except pc98. This means
510 that the /etc/ttys file needs to be updated to ensure correct
511 operation of applications on the console.
513 The terminal emulation style can be toggled per window by using
514 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
515 options can be used to change the compile-time default back to
518 To prevent graphical artifacts, make sure the TERM environment
519 variable is set to match the terminal emulation that is being
520 performed by syscons(4).
523 The layout of the structure ieee80211req_scan_result has changed.
524 Applications that require wireless scan results (e.g. ifconfig(8))
525 from net80211 need to be recompiled.
527 Applications such as wpa_supplicant(8) may require a full world
528 build without using NO_CLEAN in order to get synchronized with the
532 The iwn(4) driver has been updated to support the 5000 and 5150 series.
533 There's one kernel module for each firmware. Adding "device iwnfw"
534 to the kernel configuration file means including all three firmware
535 images inside the kernel. If you want to include just the one for
536 your wireless card, use the devices iwn4965fw, iwn5000fw or
540 The rc.d/network_ipv6, IPv6 configuration script has been integrated
541 into rc.d/netif. The changes are the following:
543 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
544 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
545 Note that both variables need the "inet6" keyword at the head.
547 Do not set $ipv6_network_interfaces manually if you do not
548 understand what you are doing. It is not needed in most cases.
550 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
553 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
554 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
556 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
557 all configured interfaces have "inet6 accept_rtadv" in the
558 $ifconfig_IF_ipv6. These are for backward compatibility.
560 3. A new variable $ipv6_prefer has been added. If NO, IPv6
561 functionality of interfaces with no corresponding
562 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
563 and the default address selection policy of ip6addrctl(8)
564 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
565 Note that if you want to configure IPv6 functionality on the
566 disabled interfaces after boot, first you need to clear the flag by
567 using ifconfig(8) like:
569 ifconfig em0 inet6 -ifdisabled
571 If YES, the default address selection policy is set as
574 The default value of $ipv6_prefer is NO.
576 4. If your system need to receive Router Advertisement messages,
577 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
578 scripts automatically invoke rtsol(8) when the interface becomes
579 UP. The Router Advertisement messages are used for SLAAC
580 (State-Less Address AutoConfiguration).
583 802.11s D3.03 support was committed. This is incompatible with the
584 previous code, which was based on D3.0.
587 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
588 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
589 control whether accepting Router Advertisement messages or not.
590 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
591 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
592 The ifconfig(8) utility now supports these flags.
595 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
596 mount(8) and -a option for df(1) to see them.
599 The old tunable hw.bus.devctl_disable has been superseded by
600 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
601 replaced by hw.bus.devctl_queue=0. The default for this new tunable
605 Remove the option STOP_NMI. The default action is now to use NMI only
606 for KDB via the newly introduced function stop_cpus_hard() and
607 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
610 The stable/8 branch created in subversion. This corresponds to the
611 RELENG_8 branch in CVS.
614 Bump the shared library version numbers for all libraries that do not
615 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
616 __FreeBSD_version to 800105.
619 Due to changes in the implementation of virtual network stack support,
620 all network-related kernel modules must be recompiled. As this change
621 breaks the ABI, bump __FreeBSD_version to 800104.
624 The TOE interface to the TCP syncache has been modified to remove
625 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
626 The cxgb driver is the only TOE consumer affected by this change, and
627 needs to be recompiled along with the kernel. As this change breaks
628 the ABI, bump __FreeBSD_version to 800103.
631 Padding has been added to struct tcpcb, sackhint and tcpstat in
632 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
633 maintaining the ABI. However, this change breaks the ABI, so bump
634 __FreeBSD_version to 800102. User space tools that rely on the size of
635 any of these structs (e.g. sockstat) need to be recompiled.
638 The NFS_LEGACYRPC option has been removed along with the old kernel
639 RPC implementation that this option selected. Kernel configurations
640 may need to be adjusted.
643 The network interface device nodes at /dev/net/<interface> have been
644 removed. All ioctl operations can be performed the normal way using
645 routing sockets. The kqueue functionality can generally be replaced
646 with routing sockets.
649 The documentation from the FreeBSD Documentation Project (Handbook,
650 FAQ, etc.) is now installed via packages by sysinstall(8) and under
651 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
654 The ABI of various structures related to the SYSV IPC API have been
655 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
656 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
660 Layout of struct vnet has changed as routing related variables were
661 moved to their own Vimage module. Modules need to be recompiled. Bump
662 __FreeBSD_version to 800099.
665 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
666 respectively. As long as no more than 16 groups per process are used,
667 no changes should be visible. When more than 16 groups are used, old
668 binaries may fail if they call getgroups() or getgrouplist() with
669 statically sized storage. Recompiling will work around this, but
670 applications should be modified to use dynamically allocated storage
671 for group arrays as POSIX.1-2008 does not cap an implementation's
672 number of supported groups at NGROUPS_MAX+1 as previous versions did.
674 NFS and portalfs mounts may also be affected as the list of groups is
675 truncated to 16. Users of NFS who use more than 16 groups, should
676 take care that negative group permissions are not used on the exported
677 file systems as they will not be reliable unless a GSSAPI based
678 authentication method is used.
681 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
682 option compiles in the support for adaptive spinning for lockmgrs
683 which want to enable it. The lockinit() function now accepts the flag
684 LK_ADAPTIVE in order to make the lock object subject to adaptive
685 spinning when both held in write and read mode.
688 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
689 changed. User applications that use this ioctl need to be rebuilt.
692 The layout of struct thread has changed. Kernel and modules need to
696 The layout of structs ifnet, domain, protosw and vnet_net has changed.
697 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
700 window(1) has been removed from the base system. It can now be
701 installed from ports. The port is called misc/window.
704 The way we are storing and accessing `routing table' entries has
705 changed. Programs reading the FIB, like netstat, need to be
709 A new netisr implementation has been added for FreeBSD 8. Network
710 file system modules, such as igmp, ipdivert, and others, should be
712 Bump __FreeBSD_version to 800096.
715 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
719 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
720 Bump __FreeBSD_version to 800094.
723 Add mnt_xflag field to 'struct mount'. File system modules need to be
725 Bump __FreeBSD_version to 800093.
728 The compiling option ADAPTIVE_SX has been retired while it has been
729 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
730 The KPI for sx_init_flags() changes as accepting flags:
731 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
732 been introduced in order to handle the reversed logic.
733 Bump __FreeBSD_version to 800092.
736 Add support for hierarchical jails. Remove global securelevel.
737 Bump __FreeBSD_version to 800091.
740 The layout of struct vnet_net has changed, therefore modules
742 Bump __FreeBSD_version to 800090.
745 The newly imported zic(8) produces a new format in the output. Please
746 run tzsetup(8) to install the newly created data to /etc/localtime.
749 The sysctl tree for the usb stack has renamed from hw.usb2.* to
750 hw.usb.* and is now consistent again with previous releases.
753 802.11 monitor mode support was revised and driver api's were changed.
754 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
755 of DLT_IEEE802_11. No user-visible data structures were changed but
756 applications that use DLT_IEEE802_11 may require changes.
757 Bump __FreeBSD_version to 800088.
760 The layout of the following structs has changed: sysctl_oid,
761 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
762 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
763 panics may be experienced. World rebuild is required for
764 correctly checking networking state from userland.
765 Bump __FreeBSD_version to 800085.
768 MLDv2 and Source-Specific Multicast (SSM) have been merged
769 to the IPv6 stack. VIMAGE hooks are in but not yet used.
770 The implementation of SSM within FreeBSD's IPv6 stack closely
771 follows the IPv4 implementation.
773 For kernel developers:
775 * The most important changes are that the ip6_output() and
776 ip6_input() paths no longer take the IN6_MULTI_LOCK,
777 and this lock has been downgraded to a non-recursive mutex.
779 * As with the changes to the IPv4 stack to support SSM, filtering
780 of inbound multicast traffic must now be performed by transport
781 protocols within the IPv6 stack. This does not apply to TCP and
782 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
784 * The KPIs used by IPv6 multicast are similar to those used by
785 the IPv4 stack, with the following differences:
786 * im6o_mc_filter() is analogous to imo_multicast_filter().
787 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
788 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
789 * IN6_LOOKUP_MULTI() has been deprecated and removed.
790 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
791 for MLDv1 have an additional 'timer' argument which is used to
792 jitter the initial membership report for the solicited-node
793 multicast membership on-link.
794 * This is not strictly needed for MLDv2, which already jitters
795 its report transmissions. However, the 'timer' argument is
796 preserved in case MLDv1 is active on the interface.
798 * The KAME linked-list based IPv6 membership implementation has
799 been refactored to use a vector similar to that used by the IPv4
801 Code which maintains a list of its own multicast memberships
802 internally, e.g. carp, has been updated to reflect the new
805 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
806 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
807 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
808 implementation constraint which needs to be addressed in HEAD.
810 For application developers:
812 * The changes are broadly similar to those made for the IPv4
815 * The use of IPv4 and IPv6 multicast socket options on the same
816 socket, using mapped addresses, HAS NOT been tested or supported.
818 * There are a number of issues with the implementation of various
819 IPv6 multicast APIs which need to be resolved in the API surface
820 before the implementation is fully compatible with KAME userland
821 use, and these are mostly to do with interface index treatment.
823 * The literature available discusses the use of either the delta / ASM
824 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
825 using setsourcefilter(3)/getsourcefilter(3). For more information
826 please refer to RFC 3768, 'Socket Interface Extensions for
827 Multicast Source Filters'.
829 * Applications which use the published RFC 3678 APIs should be fine.
831 For systems administrators:
833 * The mtest(8) utility has been refactored to support IPv6, in
834 addition to IPv4. Interface addresses are no longer accepted
835 as arguments, their names must be used instead. The utility
836 will map the interface name to its first IPv4 address as
837 returned by getifaddrs(3).
839 * The ifmcstat(8) utility has also been updated to print the MLDv2
840 endpoint state and source filter lists via sysctl(3).
842 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
843 loopback of IPv6 multicast datagrams by default; it defaults to 1
844 to preserve the existing behaviour. Disabling multicast loopback is
845 recommended for optimal system performance.
847 * The IPv6 MROUTING code has been changed to examine this sysctl
848 instead of attempting to perform a group lookup before looping
849 back forwarded datagrams.
851 Bump __FreeBSD_version to 800084.
854 Implement low-level Bluetooth HCI API.
855 Bump __FreeBSD_version to 800083.
858 The layout of struct malloc_type, used by modules to register new
859 memory allocation types, has changed. Most modules will need to
860 be rebuilt or panics may be experienced.
861 Bump __FreeBSD_version to 800081.
864 Anticipate overflowing inp_flags - add inp_flags2.
865 This changes most offsets in inpcb, so checking v4 connection
866 state will require a world rebuild.
867 Bump __FreeBSD_version to 800080.
870 Add an llentry to struct route and struct route_in6. Modules
871 embedding a struct route will need to be recompiled.
872 Bump __FreeBSD_version to 800079.
875 The size of rt_metrics_lite and by extension rtentry has changed.
876 Networking administration apps will need to be recompiled.
877 The route command now supports show as an alias for get, weighting
878 of routes, sticky and nostick flags to alter the behavior of stateful
880 Bump __FreeBSD_version to 800078.
883 Do not use Giant for kbdmux(4) locking. This is wrong and
884 apparently causing more problems than it solves. This will
885 re-open the issue where interrupt handlers may race with
886 kbdmux(4) in polling mode. Typical symptoms include (but
887 not limited to) duplicated and/or missing characters when
888 low level console functions (such as gets) are used while
889 interrupts are enabled (for example geli password prompt,
890 mountroot prompt etc.). Disabling kbdmux(4) may help.
893 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
894 kernel modules referencing any of the above need to be recompiled.
895 Bump __FreeBSD_version to 800075.
898 GEOM_PART has become the default partition slicer for storage devices,
899 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
900 introduces some changes:
902 MSDOS/EBR: the devices created from MSDOS extended partition entries
903 (EBR) can be named differently than with GEOM_MBR and are now symlinks
904 to devices with offset-based names. fstabs may need to be modified.
906 BSD: the "geometry does not match label" warning is harmless in most
907 cases but it points to problems in file system misalignment with
908 disk geometry. The "c" partition is now implicit, covers the whole
909 top-level drive and cannot be (mis)used by users.
911 General: Kernel dumps are now not allowed to be written to devices
912 whose partition types indicate they are meant to be used for file
913 systems (or, in case of MSDOS partitions, as something else than
916 Most of these changes date approximately from 200812.
919 The uscanner(4) driver has been removed from the kernel. This follows
920 Linux removing theirs in 2.6 and making libusb the default interface
924 The multicast forwarding code has been cleaned up. netstat(1)
925 only relies on KVM now for printing bandwidth upcall meters.
926 The IPv4 and IPv6 modules are split into ip_mroute_mod and
927 ip6_mroute_mod respectively. The config(5) options for statically
928 compiling this code remain the same, i.e. 'options MROUTING'.
931 Support for the IFF_NEEDSGIANT network interface flag has been
932 removed, which means that non-MPSAFE network device drivers are no
933 longer supported. In particular, if_ar, if_sr, and network device
934 drivers from the old (legacy) USB stack can no longer be built or
938 POSIX.1 Native Language Support (NLS) has been enabled in libc and
939 a bunch of new language catalog files have also been added.
940 This means that some common libc messages are now localized and
941 they depend on the LC_MESSAGES environmental variable.
944 The k8temp(4) driver has been renamed to amdtemp(4) since
945 support for Family 10 and Family 11 CPU families was added.
948 IGMPv3 and Source-Specific Multicast (SSM) have been merged
949 to the IPv4 stack. VIMAGE hooks are in but not yet used.
951 For kernel developers, the most important changes are that the
952 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
953 and this lock has been downgraded to a non-recursive mutex.
955 Transport protocols (UDP, Raw IP) are now responsible for filtering
956 inbound multicast traffic according to group membership and source
957 filters. The imo_multicast_filter() KPI exists for this purpose.
958 Transports which do not use multicast (SCTP, TCP) already reject
959 multicast by default. Forwarding and receive performance may improve
960 as a mutex acquisition is no longer needed in the ip_input()
961 low-level input path. in_addmulti() and in_delmulti() are shimmed
962 to new KPIs which exist to support SSM in-kernel.
964 For application developers, it is recommended that loopback of
965 multicast datagrams be disabled for best performance, as this
966 will still cause the lock to be taken for each looped-back
967 datagram transmission. The net.inet.ip.mcast.loop sysctl may
968 be tuned to 0 to disable loopback by default; it defaults to 1
969 to preserve the existing behaviour.
971 For systems administrators, to obtain best performance with
972 multicast reception and multiple groups, it is always recommended
973 that a card with a suitably precise hash filter is used. Hash
974 collisions will still result in the lock being taken within the
975 transport protocol input path to check group membership.
977 If deploying FreeBSD in an environment with IGMP snooping switches,
978 it is recommended that the net.inet.igmp.sendlocal sysctl remain
979 enabled; this forces 224.0.0.0/24 group membership to be announced
982 The size of 'struct igmpstat' has changed; netstat needs to be
983 recompiled to reflect this.
984 Bump __FreeBSD_version to 800070.
987 libusb20.so.1 is now installed as libusb.so.1 and the ports system
988 updated to use it. This requires a buildworld/installworld in order to
989 update the library and dependencies (usbconfig, etc). Its advisable to
990 rebuild all ports which uses libusb. More specific directions are given
991 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
992 libusb are no longer required and can be removed.
995 A workaround is committed to allow the creation of System V shared
996 memory segment of size > 2 GB on the 64-bit architectures.
997 Due to a limitation of the existing ABI, the shm_segsz member
998 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
999 wrong for large segments. Note that limits must be explicitly
1000 raised to allow such segments to be created.
1003 The layout of struct ifnet has changed, requiring a rebuild of all
1004 network device driver modules.
1007 The /dev handling for the new USB stack has changed, a
1008 buildworld/installworld is required for libusb20.
1011 The new USB2 stack has now been permanently moved in and all kernel and
1012 module names reverted to their previous values (eg, usb, ehci, ohci,
1013 ums, ...). The old usb stack can be compiled in by prefixing the name
1014 with the letter 'o', the old usb modules have been removed.
1015 Updating entry 20090216 for xorg and 20090215 for libmap may still
1019 The rc.conf(5) option if_up_delay has been renamed to
1020 defaultroute_delay to better reflect its purpose. If you have
1021 customized this setting in /etc/rc.conf you need to update it to
1025 xorg 7.4 wants to configure its input devices via hald which does not
1026 yet work with USB2. If the keyboard/mouse does not work in xorg then
1028 Option "AllowEmptyInput" "off"
1029 to your ServerLayout section. This will cause X to use the configured
1030 kbd and mouse sections from your xorg.conf.
1033 The GENERIC kernels for all architectures now default to the new USB2
1034 stack. No kernel config options or code have been removed so if a
1035 problem arises please report it and optionally revert to the old USB
1036 stack. If you are loading USB kernel modules or have a custom kernel
1037 that includes GENERIC then ensure that usb names are also changed over,
1038 eg uftdi -> usb2_serial_ftdi.
1040 Older programs linked against the ports libusb 0.1 need to be
1041 redirected to the new stack's libusb20. /etc/libmap.conf can
1043 # Map old usb library to new one for usb2 stack
1044 libusb-0.1.so.8 libusb20.so.1
1047 All USB ethernet devices now attach as interfaces under the name ueN
1048 (eg. ue0). This is to provide a predictable name as vendors often
1049 change usb chipsets in a product without notice.
1052 The ichsmb(4) driver has been changed to require SMBus slave
1053 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1054 All of the other SMBus controller drivers require left-justified
1055 slave addresses, so this change makes all the drivers provide the
1059 INET6 statistics (struct ip6stat) was updated.
1060 netstat(1) needs to be recompiled.
1063 NTFS has been removed from GENERIC kernel on amd64 to match
1064 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1065 will load ntfs.ko module automatically when NTFS support is
1066 actually needed, unless ntfs.ko is not installed or security
1067 level prohibits loading kernel modules. If either is the case,
1068 "options NTFS" has to be added into kernel config.
1071 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1072 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1073 800061. User space tools that rely on the size of struct tcpcb in
1074 tcp_var.h (e.g. sockstat) need to be recompiled.
1077 ng_tty(4) module updated to match the new TTY subsystem.
1078 Due to API change, user-level applications must be updated.
1079 New API support added to mpd5 CVS and expected to be present
1080 in next mpd5.3 release.
1083 With __FreeBSD_version 800060 the makefs tool is part of
1084 the base system (it was a port).
1087 The afdata and ifnet locks have been changed from mutexes to
1088 rwlocks, network modules will need to be re-compiled.
1091 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1092 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1093 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1094 architecture and 40 bytes on 64-bit architecture. The userland
1095 applications "arp" and "ndp" have been updated accordingly.
1096 The output from "netstat -r" shows only routing entries and
1097 none of the L2 information.
1100 __FreeBSD_version 800057 marks the switchover from the
1101 binary ath hal to source code. Users must add the line:
1103 options AH_SUPPORT_AR5416
1105 to their kernel config files when specifying:
1109 The ath_hal module no longer exists; the code is now compiled
1110 together with the driver in the ath module. It is now
1111 possible to tailor chip support (i.e. reduce the set of chips
1112 and thereby the code size); consult ath_hal(4) for details.
1115 __FreeBSD_version 800054 adds memory barriers to
1116 <machine/atomic.h>, new interfaces to ifnet to facilitate
1117 multiple hardware transmit queues for cards that support
1118 them, and a lock-less ring-buffer implementation to
1119 enable drivers to more efficiently manage queueing of
1123 A new version of ZFS (version 13) has been merged to -HEAD.
1124 This version has zpool attribute "listsnapshots" off by
1125 default, which means "zfs list" does not show snapshots,
1126 and is the same as Solaris behavior.
1129 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1132 The uhci, ohci, ehci and slhci USB Host controller drivers have
1133 been put into separate modules. If you load the usb module
1134 separately through loader.conf you will need to load the
1135 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1136 controller add the following to loader.conf:
1142 The ABI used by the PMC toolset has changed. Please keep
1143 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1147 atapci kernel module now includes only generic PCI ATA
1148 driver. AHCI driver moved to ataahci kernel module.
1149 All vendor-specific code moved into separate kernel modules:
1150 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1151 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1152 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1153 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1156 The TTY subsystem of the kernel has been replaced by a new
1157 implementation, which provides better scalability and an
1158 improved driver model. Most common drivers have been migrated to
1159 the new TTY subsystem, while others have not. The following
1160 drivers have not yet been ported to the new TTY layer:
1163 cy, digi, rc, rp, sio
1169 ng_h4, ng_tty, ppp, sl, snp
1171 Adding these drivers to your kernel configuration file shall
1172 cause compilation to fail.
1175 ntpd has been upgraded to 4.2.4p5.
1178 OpenSSH has been upgraded to 5.1p1.
1180 For many years, FreeBSD's version of OpenSSH preferred DSA
1181 over RSA for host and user authentication keys. With this
1182 upgrade, we've switched to the vendor's default of RSA over
1183 DSA. This may cause upgraded clients to warn about unknown
1184 host keys even for previously known hosts. Users should
1185 follow the usual procedure for verifying host keys before
1186 accepting the RSA key.
1188 This can be circumvented by setting the "HostKeyAlgorithms"
1189 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1192 Please note that the sequence of keys offered for
1193 authentication has been changed as well. You may want to
1194 specify IdentityFile in a different order to revert this
1198 The sio(4) driver has been removed from the i386 and amd64
1199 kernel configuration files. This means uart(4) is now the
1200 default serial port driver on those platforms as well.
1202 To prevent collisions with the sio(4) driver, the uart(4) driver
1203 uses different names for its device nodes. This means the
1204 onboard serial port will now most likely be called "ttyu0"
1205 instead of "ttyd0". You may need to reconfigure applications to
1206 use the new device names.
1208 When using the serial port as a boot console, be sure to update
1209 /boot/device.hints and /etc/ttys before booting the new kernel.
1210 If you forget to do so, you can still manually specify the hints
1211 at the loader prompt:
1213 set hint.uart.0.at="isa"
1214 set hint.uart.0.port="0x3F8"
1215 set hint.uart.0.flags="0x10"
1216 set hint.uart.0.irq="4"
1220 The gpt(8) utility has been removed. Use gpart(8) to partition
1224 The version that Linuxulator emulates was changed from 2.4.2
1225 to 2.6.16. If you experience any problems with Linux binaries
1226 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1227 if it fixes the problem contact emulation mailing list.
1230 ISDN4BSD (I4B) was removed from the src tree. You may need to
1231 update a your kernel configuration and remove relevant entries.
1234 I have checked in code to support multiple routing tables.
1235 See the man pages setfib(1) and setfib(2).
1236 This is a hopefully backwards compatible version,
1237 but to make use of it you need to compile your kernel
1238 with options ROUTETABLES=2 (or more up to 16).
1241 The 802.11 wireless support was redone to enable multi-bss
1242 operation on devices that are capable. The underlying device
1243 is no longer used directly but instead wlanX devices are
1244 cloned with ifconfig. This requires changes to rc.conf files.
1245 For example, change:
1246 ifconfig_ath0="WPA DHCP"
1249 ifconfig_wlan0="WPA DHCP"
1250 see rc.conf(5) for more details. In addition, mergemaster of
1251 /etc/rc.d is highly recommended. Simultaneous update of userland
1252 and kernel wouldn't hurt either.
1254 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1255 modules were merged into the base wlan module. All references
1256 to these modules (e.g. in kernel config files) must be removed.
1259 psm(4) has gained write(2) support in native operation level.
1260 Arbitrary commands can be written to /dev/psm%d and status can
1261 be read back from it. Therefore, an application is responsible
1262 for status validation and error recovery. It is a no-op in
1263 other operation levels.
1266 Support for KSE threading has been removed from the kernel. To
1267 run legacy applications linked against KSE libmap.conf may
1268 be used. The following libmap.conf may be used to ensure
1269 compatibility with any prior release:
1271 libpthread.so.1 libthr.so.1
1272 libpthread.so.2 libthr.so.2
1273 libkse.so.3 libthr.so.3
1276 The layout of struct vmspace has changed. This affects libkvm
1277 and any executables that link against libkvm and use the
1278 kvm_getprocs() function. In particular, but not exclusively,
1279 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1280 The effects are minimal, but it's advisable to upgrade world
1284 The latest em driver no longer has support in it for the
1285 82575 adapter, this is now moved to the igb driver. The
1286 split was done to make new features that are incompatible
1287 with older hardware easier to do.
1290 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1291 likewise the kernel option is now GEOM_LINUX_LVM.
1294 The default NFS mount mode has changed from UDP to TCP for
1295 increased reliability. If you rely on (insecurely) NFS
1296 mounting across a firewall you may need to update your
1300 Belatedly note the addition of m_collapse for compacting
1304 The fts(3) structures have been changed to use adequate
1305 integer types for their members and so to be able to cope
1306 with huge file trees. The old fts(3) ABI is preserved
1307 through symbol versioning in libc, so third-party binaries
1308 using fts(3) should still work, although they will not take
1309 advantage of the extended types. At the same time, some
1310 third-party software might fail to build after this change
1311 due to unportable assumptions made in its source code about
1312 fts(3) structure members. Such software should be fixed
1313 by its vendor or, in the worst case, in the ports tree.
1314 FreeBSD_version 800015 marks this change for the unlikely
1315 case that a portable fix is impossible.
1318 To upgrade to -current after this date, you must be running
1319 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1320 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1323 The ADAPTIVE_GIANT kernel option has been retired because its
1324 functionality is the default now.
1327 The AT keyboard emulation of sunkbd(4) has been turned on
1328 by default. In order to make the special symbols of the Sun
1329 keyboards driven by sunkbd(4) work under X these now have
1330 to be configured the same way as Sun USB keyboards driven
1331 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1333 Option "XkbLayout" "us"
1334 Option "XkbRules" "xorg"
1335 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1338 It has been decided that it is desirable to provide ABI
1339 backwards compatibility to the FreeBSD 4/5/6 versions of the
1340 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1341 broken with the introduction of PCI domain support (see the
1342 20070930 entry). Unfortunately, this required the ABI of
1343 PCIOCGETCONF to be broken again in order to be able to
1344 provide backwards compatibility to the old version of that
1345 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1346 again. As for prominent ports this affects neither pciutils
1347 nor xorg-server this time, the hal port needs to be rebuilt
1351 The misnamed kthread_create() and friends have been renamed
1352 to kproc_create() etc. Many of the callers already
1353 used kproc_start()..
1354 I will return kthread_create() and friends in a while
1355 with implementations that actually create threads, not procs.
1356 Renaming corresponds with version 800002.
1365 Avoid using make -j when upgrading. While generally safe, there are
1366 sometimes problems using -j to upgrade. If your upgrade fails with
1367 -j, please try again without -j. From time to time in the past there
1368 have been problems using -j with buildworld and/or installworld. This
1369 is especially true when upgrading between "distant" versions (eg one
1370 that cross a major release boundary or several minor releases, or when
1371 several months have passed on the -current branch).
1373 Sometimes, obscure build problems are the result of environment
1374 poisoning. This can happen because the make utility reads its
1375 environment when searching for values for global variables. To run
1376 your build attempts in an "environmental clean room", prefix all make
1377 commands with 'env -i '. See the env(1) manual page for more details.
1379 When upgrading from one major version to another it is generally best
1380 to upgrade to the latest code in the currently installed branch first,
1381 then do an upgrade to the new branch. This is the best-tested upgrade
1382 path, and has the highest probability of being successful. Please try
1383 this approach before reporting problems with a major version upgrade.
1387 When upgrading the boot ZFS pool to a new version, always follow
1390 1.) recompile and reinstall the ZFS boot loader and boot block
1391 (this is part of "make buildworld" and "make installworld")
1393 2.) update the ZFS boot block on your boot drive
1395 The following example updates the ZFS boot block on the first
1396 partition (freebsd-boot) of a GPT partitioned drive ad0:
1397 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1399 Non-boot pools do not need these updates.
1403 If you are updating from a prior version of FreeBSD (even one just
1404 a few days old), you should follow this procedure. It is the most
1405 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1407 make kernel-toolchain
1408 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1409 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1411 To test a kernel once
1412 ---------------------
1413 If you just want to boot a kernel once (because you are not sure
1414 if it works, or if you want to boot a known bad kernel to provide
1415 debugging information) run
1416 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1417 nextboot -k testkernel
1419 To just build a kernel when you know that it won't mess you up
1420 --------------------------------------------------------------
1421 This assumes you are already running a CURRENT system. Replace
1422 ${arch} with the architecture of your machine (e.g. "i386",
1423 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1425 cd src/sys/${arch}/conf
1426 config KERNEL_NAME_HERE
1427 cd ../compile/KERNEL_NAME_HERE
1432 If this fails, go to the "To build a kernel" section.
1434 To rebuild everything and install it on the current system.
1435 -----------------------------------------------------------
1436 # Note: sometimes if you are running current you gotta do more than
1437 # is listed here if you are upgrading from a really old current.
1439 <make sure you have good level 0 dumps>
1441 make kernel KERNCONF=YOUR_KERNEL_HERE
1443 <reboot in single user> [3]
1451 To cross-install current onto a separate partition
1452 --------------------------------------------------
1453 # In this approach we use a separate partition to hold
1454 # current's root, 'usr', and 'var' directories. A partition
1455 # holding "/", "/usr" and "/var" should be about 2GB in
1458 <make sure you have good level 0 dumps>
1461 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1462 <maybe newfs current's root partition>
1463 <mount current's root partition on directory ${CURRENT_ROOT}>
1464 make installworld DESTDIR=${CURRENT_ROOT}
1465 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1466 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1467 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1468 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1469 <reboot into current>
1470 <do a "native" rebuild/install as described in the previous section>
1471 <maybe install compatibility libraries from ports/misc/compat*>
1475 To upgrade in-place from stable to current
1476 ----------------------------------------------
1477 <make sure you have good level 0 dumps>
1479 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1481 <reboot in single user> [3]
1488 Make sure that you've read the UPDATING file to understand the
1489 tweaks to various things you need. At this point in the life
1490 cycle of current, things change often and you are on your own
1491 to cope. The defaults can also change, so please read ALL of
1492 the UPDATING entries.
1494 Also, if you are tracking -current, you must be subscribed to
1495 freebsd-current@freebsd.org. Make sure that before you update
1496 your sources that you have read and understood all the recent
1497 messages there. If in doubt, please track -stable which has
1498 much fewer pitfalls.
1500 [1] If you have third party modules, such as vmware, you
1501 should disable them at this point so they don't crash your
1504 [3] From the bootblocks, boot -s, and then do
1509 adjkerntz -i # if CMOS is wall time
1510 Also, when doing a major release upgrade, it is required that
1511 you boot into single user mode to do the installworld.
1513 [4] Note: This step is non-optional. Failure to do this step
1514 can result in a significant reduction in the functionality of the
1515 system. Attempting to do it by hand is not recommended and those
1516 that pursue this avenue should read this file carefully, as well
1517 as the archives of freebsd-current and freebsd-hackers mailing lists
1518 for potential gotchas. The -U option is also useful to consider.
1519 See mergemaster(8) for more information.
1521 [5] Usually this step is a noop. However, from time to time
1522 you may need to do this if you get unknown user in the following
1523 step. It never hurts to do it all the time. You may need to
1524 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1525 install) after the buildworld before this step if you last updated
1526 from current before 20020224 or from -stable before 20020408.
1528 [6] This only deletes old files and directories. Old libraries
1529 can be deleted by "make delete-old-libs", but you have to make
1530 sure that no program is using those libraries anymore.
1532 [8] In order to have a kernel that can run the 4.x binaries needed to
1533 do an installworld, you must include the COMPAT_FREEBSD4 option in
1534 your kernel. Failure to do so may leave you with a system that is
1535 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1536 required to run the 5.x binaries on more recent kernels. And so on
1537 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1539 Make sure that you merge any new devices from GENERIC since the
1540 last time you updated your kernel config file.
1542 [9] When checking out sources, you must include the -P flag to have
1543 cvs prune empty directories.
1545 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1546 "?=" instead of the "=" assignment operator, so that buildworld can
1547 override the CPUTYPE if it needs to.
1549 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1550 not on the command line, or in /etc/make.conf. buildworld will
1551 warn if it is improperly defined.
1554 This file contains a list, in reverse chronological order, of major
1555 breakages in tracking -current. It is not guaranteed to be a complete
1556 list of such breakages, and only contains entries since October 10, 2007.
1557 If you need to see UPDATING entries from before that date, you will need
1558 to fetch an UPDATING file from an older FreeBSD release.
1560 Copyright information:
1562 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1564 Redistribution, publication, translation and use, with or without
1565 modification, in full or in part, in any form or format of this
1566 document are permitted without further permission from the author.
1568 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1569 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1570 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1571 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1572 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1573 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1574 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1575 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1576 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1577 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1578 POSSIBILITY OF SUCH DAMAGE.
1580 Contact Warner Losh if you have any questions about your use of