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 With the addition of auditdistd(8), a new auditdistd user is now
16 depended on during installworld. "mergemaster -p" can be used to add
17 the user prior to installworld, as documented in the handbook.
23 A new version of ZFS (pool version 5000) has been merged to 9-STABLE.
24 Starting with this version the old system of ZFS pool versioning
25 is superseded by "feature flags". This concept enables forward
26 compatibility against certain future changes in functionality of ZFS
27 pools. The first two read-only compatible "feature flags" for ZFS
28 pools are "com.delphix:async_destroy" and "com.delphix:empty_bpobj".
29 For more information read the new zpool-features(7) manual page.
30 Please refer to the "ZFS notes" section of this file for information
31 on upgrading boot ZFS pools.
34 The IPFIREWALL_FORWARD kernel option has been removed. Its
35 functionality now turned on by default.
38 The random(4) support for the VIA hardware random number
39 generator (`PADLOCK') is no longer enabled unconditionally.
40 Add the PADLOCK_RNG option in the custom kernel config if
41 needed. The GENERIC kernels on i386 and amd64 do include the
42 option, so the change only affects the custom kernel
46 The sparc64 ZFS loader has been changed to no longer try to auto-
47 detect ZFS providers based on diskN aliases but now requires these
48 to be explicitly listed in the OFW boot-device environment variable.
51 Now unix domain sockets behave "as expected" on nullfs(5). Previously
52 nullfs(5) did not pass through all behaviours to the underlying layer,
53 as a result if we bound to a socket on the lower layer we could connect
54 only to the lower path; if we bound to the upper layer we could connect
55 only to the upper path. The new behavior is one can connect to both the
56 lower and the upper paths regardless what layer path one binds to.
59 The acpi_wmi(4) status device /dev/wmistat has been renamed to
63 A new VOP_ADVISE() was added to support posix_fadvise(2). All
64 filesystem modules must be recompiled.
67 The interface of the VOP_VPTOCNP(9) changed, now the returned
68 vnode shall be referenced, previously it was required to be
69 only held. All in-tree filesystems are converted.
75 The broken amd(4) driver has been replaced with esp(4) in the amd64,
76 i386 and pc98 GENERIC kernel configuration files.
79 This commit modifies vfs_register() so that it uses a hash
80 calculation to set vfc_typenum, which is enabled by default.
81 The first time a system is booted after this change, the
82 vfc_typenum values will change for all file systems. The
83 main effect of this is a change to the NFS server file handles
84 for file systems that use vfc_typenum in their fsid, such as ZFS.
85 It will, however, prevent vfc_typenum from changing when file
86 systems are loaded in a different order for subsequent reboots.
87 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
88 until you are ready to remount all NFS clients after a reboot.
91 Bump the shared library version numbers for libraries that
92 do not use symbol versioning, have changed the ABI compared
93 to stable/8 and which shared library version was not bumped.
94 Done as part of 9.0-RELEASE cycle.
97 During the merge of Capsicum features, the fget(9) KPI was modified.
98 This may require the rebuilding of out-of-tree device drivers --
99 issues have been reported specifically with the nVidia device driver.
100 __FreeBSD_version is bumped to 900041.
102 Also, there is a period between 20110811 and 20110814 where the
103 special devices /dev/{stdin,stdout,stderr} did not work correctly.
104 Building world from a kernel during that window may not work.
107 The packet filter (pf) code has been updated to OpenBSD 4.5.
108 You need to update userland tools to be in sync with kernel.
109 This update breaks backward compatibility with earlier pfsync(4)
110 versions. Care must be taken when updating redundant firewall setups.
113 The following sysctls and tunables are retired on x86 platforms:
115 machdep.hlt_logical_cpus
116 The following sysctl is retired:
117 machdep.hyperthreading_allowed
118 The sysctls were supposed to provide a way to dynamically offline and
119 online selected CPUs on x86 platforms, but the implementation has not
120 been reliable especially with SCHED_ULE scheduler.
121 machdep.hyperthreading_allowed tunable is still available to ignore
122 hyperthreading CPUs at OS level.
123 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
124 where X is an APIC ID of a CPU. Be advised, though, that disabling
125 CPUs in non-uniform fashion will result in non-uniform topology and
126 may lead to sub-optimal system performance with SCHED_ULE, which is
130 cpumask_t type is retired and cpuset_t is used in order to describe
134 Changes to ifconfig(8) for dynamic address family detection mandate
135 that you are running a kernel of 20110525 or later. Make sure to
136 follow the update procedure to boot a new kernel before installing
140 Support for sun4v architecture is officially dropped
143 Several KPI breaking changes have been committed to the mii(4) layer,
144 the PHY drivers and consequently some Ethernet drivers using mii(4).
145 This means that miibus.ko and the modules of the affected Ethernet
146 drivers need to be recompiled.
148 Note to kernel developers: Given that the OUI bit reversion problem
149 was fixed as part of these changes all mii(4) commits related to OUIs,
150 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
151 handling, no longer can be merged verbatim to stable/8 and previous
155 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
156 into their kernel configurations along with 'device pci'.
159 The default NFS client is now the new NFS client, so fstype "newnfs"
160 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
161 Although mounts via fstype "nfs" will usually work without userland
162 changes, it is recommended that the mount(8) and mount_nfs(8)
163 commands be rebuilt from sources and that a link to mount_nfs called
164 mount_oldnfs be created. The new client is compiled into the
165 kernel with "options NFSCL" and this is needed for diskless root
166 file systems. The GENERIC kernel configs have been changed to use
167 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
168 To use the regular/old client, you can "mount -t oldnfs ...". For
169 a diskless root file system, you must also include a line like:
171 vfs.root.mountfrom="oldnfs:"
173 in the boot/loader.conf on the root fs on the NFS server to make
174 a diskless root fs use the old client.
177 The GENERIC kernels for all architectures now default to the new
178 CAM-based ATA stack. It means that all legacy ATA drivers were
179 removed and replaced by respective CAM drivers. If you are using
180 ATA device names in /etc/fstab or other places, make sure to update
181 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
182 where 'Y's are the sequential numbers starting from zero for each type
183 in order of detection, unless configured otherwise with tunables,
184 see cam(4)). There will be symbolic links created in /dev/ to map
185 old adX devices to the respective adaY. They should provide basic
186 compatibility for file systems mounting in most cases, but they do
187 not support old user-level APIs and do not have respective providers
188 in GEOM. Consider using updated management tools with new device names.
190 It is possible to load devices ahci, ata, siis and mvs as modules,
191 but option ATA_CAM should remain in kernel configuration to make ata
192 module work as CAM driver supporting legacy ATA controllers. Device ata
193 still can be used in modular fashion (atacore + ...). Modules atadisk
194 and atapi* are not used and won't affect operation in ATA_CAM mode.
195 Note that to use CAM-based ATA kernel should include CAM devices
196 scbus, pass, da (or explicitly ada), cd and optionally others. All of
197 them are parts of the cam module.
199 ataraid(4) functionality is now supported by the RAID GEOM class.
200 To use it you can load geom_raid kernel module and use graid(8) tool
201 for management. Instead of /dev/arX device names, use /dev/raid/rX.
203 No kernel config options or code have been removed, so if a problem
204 arises, please report it and optionally revert to the old ATA stack.
205 In order to do it you can remove from the kernel config:
210 , and instead add back:
211 device atadisk # ATA disk drives
212 device ataraid # ATA RAID drives
213 device atapicd # ATAPI CDROM drives
214 device atapifd # ATAPI floppy drives
215 device atapist # ATAPI tape drives
218 The default NFS server has been changed to the new server, which
219 was referred to as the experimental server. If you need to switch
220 back to the old NFS server, you must now put the "-o" option on
221 both the mountd and nfsd commands. This can be done using the
222 mountd_flags and nfs_server_flags rc.conf variables until an
223 update to the rc scripts is committed, which is coming soon.
226 The GNU Objective-C runtime library (libobjc), and other Objective-C
227 related components have been removed from the base system. If you
228 require an Objective-C library, please use one of the available ports.
231 ath(4) has been split into bus- and device- modules. if_ath contains
232 the HAL, the TX rate control and the network device code. if_ath_pci
233 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
234 contains the AHB glue. Users need to load both if_ath_pci and if_ath
235 in order to use ath on everything else.
237 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
238 need to load if_ath and if_ath_pci for ath(4) operation.
241 As part of the replacement of sysinstall, the process of building
242 release media has changed significantly. For details, please re-read
243 release(7), which has been updated to reflect the new build process.
246 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
247 is the last available version under GPLv2. It brings a number of new
248 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
249 4.1 and SSE 4.2), better support for powerpc64, a number of new
250 directives, and lots of other small improvements. See the ChangeLog
251 file in contrib/binutils for the full details.
254 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
255 compliant, and will now use half of hash for authentication.
256 This will break interoperability with all stacks (including all
257 actual FreeBSD versions) who implement
258 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
260 The only workaround with such peers is to use another HMAC
261 algorithm for IPsec ("phase 2") authentication.
264 Remove the uio_yield prototype and symbol. This function has
265 been misnamed since it was introduced and should not be
266 globally exposed with this name. The equivalent functionality
267 is now available using kern_yield(curthread->td_user_pri).
268 The function remains undocumented.
271 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
272 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
273 for scalar sysctls is defined but disabled. Code that needs
274 UQUAD to pass the type checking that must compile on older
275 systems where the define is not present can check against
276 __FreeBSD_version >= 900030.
278 The system dialog(1) has been replaced with a new version previously
279 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
280 with the previous version, but the libdialog associated with it has
281 a largely incompatible API. As such, the original version of libdialog
282 will be kept temporarily as libodialog, until its base system consumers
283 are replaced or updated. Bump __FreeBSD_version to 900030.
286 If you are trying to run make universe on a -stable system, and you get
287 the following warning:
288 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
289 unknown. config(8) likely too old."
290 or something similar to it, then you must upgrade your -stable system
291 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
292 stable/8) or build the config from the latest stable/8 branch and
293 install it on your system.
295 Prior to this date, building a current universe on 8-stable system from
296 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
297 error in the first kernel build phase. A new config on those old
298 systems will fix that problem for older versions of -current.
301 The TCP stack has been modified to allow Khelp modules to interact with
302 it via helper hook points and store per-connection data in the TCP
303 control block. Bump __FreeBSD_version to 900029. User space tools that
304 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
308 Generic IEEE 802.3 annex 31B full duplex flow control support has been
309 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
310 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
311 to take advantage of it instead of using custom implementations. This
312 means that these drivers now no longer unconditionally advertise
313 support for flow control but only do so if flow control is a selected
314 media option. This was implemented in the generic support that way in
315 order to allow flow control to be switched on and off via ifconfig(8)
316 with the PHY specific default to typically off in order to protect
317 from unwanted effects. Consequently, if you used flow control with
318 one of the above mentioned drivers you now need to explicitly enable
320 ifconfig bge0 media auto mediaopt flowcontrol
322 Along with the above mentioned changes generic support for setting
323 1000baseT master mode also has been added and brgphy(4), ciphy(4),
324 e1000phy(4) as well as ip1000phy(4) have been converted to take
325 advantage of it. This means that these drivers now no longer take the
326 link0 parameter for selecting master mode but the master media option
327 has to be used instead, for example like in the following:
328 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
330 Selection of master mode now is also available with all other PHY
331 drivers supporting 1000baseT.
334 The TCP stack has received a significant update to add support for
335 modularised congestion control and generally improve the clarity of
336 congestion control decisions. Bump __FreeBSD_version to 900025. User
337 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
338 sockstat) need to be recompiled.
341 The man(1) utility has been replaced by a new version that no longer
342 uses /etc/manpath.config. Please consult man.conf(5) for how to
343 migrate local entries to the new format.
346 The copyright strings printed by login(1) and sshd(8) at the time of a
347 new connection have been removed to follow other operating systems and
351 A workaround for a fixed ld bug has been removed in kernel code,
352 so make sure that your system ld is built from sources after
353 revision 210245 from 2010-07-19 (r211583 if building head kernel
354 on stable/8, r211584 for stable/7; both from 2010-08-21).
355 A symptom of incorrect ld version is different addresses for
356 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
359 The $ipv6_prefer variable in rc.conf(5) has been split into
360 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
362 The $ip6addrctl_policy is a variable to choose a pre-defined
363 address selection policy set by ip6addrctl(8). A value
364 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
367 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
368 flag (see an entry of 20090926) is set on an interface with no
369 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
370 security reason. If you want IPv6 link-local address on all
371 interfaces by default, set this to "YES".
373 The old ipv6_prefer="YES" is equivalent to
374 ipv6_activate_all_interfaces="YES" and
375 ip6addrctl_policy="ipv6_prefer".
378 DTrace has grown support for userland tracing. Due to this, DTrace is
379 now i386 and amd64 only.
380 dtruss(1) is now installed by default on those systems and a new
381 kernel module is needed for userland tracing: fasttrap.
382 No changes to your kernel config file are necessary to enable
383 userland tracing, but you might consider adding 'STRIP=' and
384 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
385 to have informative userland stack traces in DTrace (ustack).
388 The acpi_aiboost(4) driver has been removed in favor of the new
389 aibs(4) driver. You should update your kernel configuration file.
392 BSD grep has been imported to the base system and it is built by
393 default. It is completely BSD licensed, highly GNU-compatible, uses
394 less memory than its GNU counterpart and has a small codebase.
395 However, it is slower than its GNU counterpart, which is mostly
396 noticeable for larger searches, for smaller ones it is measurable
397 but not significant. The reason is complex, the most important factor
398 is that we lack a modern and efficient regex library and GNU
399 overcomes this by optimizing the searches internally. Future work
400 on improving the regex performance is planned, for the meantime,
401 users that need better performance, can build GNU grep instead by
402 setting the WITH_GNU_GREP knob.
405 Due to the import of powerpc64 support, all existing powerpc kernel
406 configuration files must be updated with a machine directive like this:
407 machine powerpc powerpc
409 In addition, an updated config(8) is required to build powerpc kernels
413 A new version of ZFS (version 15) has been merged to -HEAD.
414 This version uses a python library for the following subcommands:
415 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
416 For full functionality of these commands the following port must
417 be installed: sysutils/py-zfs
420 'vm_page's are now hashed by physical address to an array of mutexes.
421 Currently this is only used to serialize access to hold_count. Over
422 time the page queue mutex will be peeled away. This changes the size
423 of pmap on every architecture. And requires all callers of vm_page_hold
424 and vm_page_unhold to be updated.
427 WITH_CTF can now be specified in src.conf (not recommended, there
428 are some problems with static executables), make.conf (would also
429 affect ports which do not use GNU make and do not override the
430 compile targets) or in the kernel config (via "makeoptions
432 When WITH_CTF was specified there before this was silently ignored,
433 so make sure that WITH_CTF is not used in places which could lead
434 to unwanted behavior.
437 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
438 to allow 32-bit compatibility on non-x86 platforms. All kernel
439 configurations on amd64 and ia64 platforms using these options must
440 be modified accordingly.
443 The utmp user accounting database has been replaced with utmpx,
444 the user accounting interface standardized by POSIX.
445 Unfortunately the semantics of utmp and utmpx don't match,
446 making it practically impossible to support both interfaces.
447 The user accounting database is used by tools like finger(1),
448 last(1), talk(1), w(1) and ac(8).
450 All applications in the base system use utmpx. This means only
451 local binaries (e.g. from the ports tree) may still use these
452 utmp database files. These applications must be rebuilt to make
455 After the system has been upgraded, it is safe to remove the old
456 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
457 assuming their contents is of no importance anymore. Old wtmp
458 databases can only be used by last(1) and ac(8) after they have
459 been converted to the new format using wtmpcvt(1).
462 Introduce the kernel thread "deadlock resolver" (which can be enabled
463 via the DEADLKRES option, see NOTES for more details) and the
464 sleepq_type() function for sleepqueues.
467 The rc.firewall and rc.firewall6 were unified, and
468 rc.firewall6 and rc.d/ip6fw were removed.
469 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
470 variables are obsoleted. Instead, the following new rc
471 variables are added to rc.d/ipfw:
473 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
474 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
475 firewall_simple_onet_ipv6, firewall_trusted_ipv6
477 The meanings correspond to the relevant IPv4 variables.
483 The default terminal emulation for syscons(4) has been changed
484 from cons25 to xterm on all platforms except pc98. This means
485 that the /etc/ttys file needs to be updated to ensure correct
486 operation of applications on the console.
488 The terminal emulation style can be toggled per window by using
489 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
490 options can be used to change the compile-time default back to
493 To prevent graphical artifacts, make sure the TERM environment
494 variable is set to match the terminal emulation that is being
495 performed by syscons(4).
498 The layout of the structure ieee80211req_scan_result has changed.
499 Applications that require wireless scan results (e.g. ifconfig(8))
500 from net80211 need to be recompiled.
502 Applications such as wpa_supplicant(8) may require a full world
503 build without using NO_CLEAN in order to get synchronized with the
507 The iwn(4) driver has been updated to support the 5000 and 5150 series.
508 There's one kernel module for each firmware. Adding "device iwnfw"
509 to the kernel configuration file means including all three firmware
510 images inside the kernel. If you want to include just the one for
511 your wireless card, use the devices iwn4965fw, iwn5000fw or
515 The rc.d/network_ipv6, IPv6 configuration script has been integrated
516 into rc.d/netif. The changes are the following:
518 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
519 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
520 Note that both variables need the "inet6" keyword at the head.
522 Do not set $ipv6_network_interfaces manually if you do not
523 understand what you are doing. It is not needed in most cases.
525 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
528 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
529 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
531 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
532 all configured interfaces have "inet6 accept_rtadv" in the
533 $ifconfig_IF_ipv6. These are for backward compatibility.
535 3. A new variable $ipv6_prefer has been added. If NO, IPv6
536 functionality of interfaces with no corresponding
537 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
538 and the default address selection policy of ip6addrctl(8)
539 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
540 Note that if you want to configure IPv6 functionality on the
541 disabled interfaces after boot, first you need to clear the flag by
542 using ifconfig(8) like:
544 ifconfig em0 inet6 -ifdisabled
546 If YES, the default address selection policy is set as
549 The default value of $ipv6_prefer is NO.
551 4. If your system need to receive Router Advertisement messages,
552 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
553 scripts automatically invoke rtsol(8) when the interface becomes
554 UP. The Router Advertisement messages are used for SLAAC
555 (State-Less Address AutoConfiguration).
558 802.11s D3.03 support was committed. This is incompatible with the
559 previous code, which was based on D3.0.
562 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
563 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
564 control whether accepting Router Advertisement messages or not.
565 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
566 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
567 The ifconfig(8) utility now supports these flags.
570 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
571 mount(8) and -a option for df(1) to see them.
574 The old tunable hw.bus.devctl_disable has been superseded by
575 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
576 replaced by hw.bus.devctl_queue=0. The default for this new tunable
580 Remove the option STOP_NMI. The default action is now to use NMI only
581 for KDB via the newly introduced function stop_cpus_hard() and
582 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
585 The stable/8 branch created in subversion. This corresponds to the
586 RELENG_8 branch in CVS.
589 Bump the shared library version numbers for all libraries that do not
590 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
591 __FreeBSD_version to 800105.
594 Due to changes in the implementation of virtual network stack support,
595 all network-related kernel modules must be recompiled. As this change
596 breaks the ABI, bump __FreeBSD_version to 800104.
599 The TOE interface to the TCP syncache has been modified to remove
600 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
601 The cxgb driver is the only TOE consumer affected by this change, and
602 needs to be recompiled along with the kernel. As this change breaks
603 the ABI, bump __FreeBSD_version to 800103.
606 Padding has been added to struct tcpcb, sackhint and tcpstat in
607 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
608 maintaining the ABI. However, this change breaks the ABI, so bump
609 __FreeBSD_version to 800102. User space tools that rely on the size of
610 any of these structs (e.g. sockstat) need to be recompiled.
613 The NFS_LEGACYRPC option has been removed along with the old kernel
614 RPC implementation that this option selected. Kernel configurations
615 may need to be adjusted.
618 The network interface device nodes at /dev/net/<interface> have been
619 removed. All ioctl operations can be performed the normal way using
620 routing sockets. The kqueue functionality can generally be replaced
621 with routing sockets.
624 The documentation from the FreeBSD Documentation Project (Handbook,
625 FAQ, etc.) is now installed via packages by sysinstall(8) and under
626 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
629 The ABI of various structures related to the SYSV IPC API have been
630 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
631 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
635 Layout of struct vnet has changed as routing related variables were
636 moved to their own Vimage module. Modules need to be recompiled. Bump
637 __FreeBSD_version to 800099.
640 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
641 respectively. As long as no more than 16 groups per process are used,
642 no changes should be visible. When more than 16 groups are used, old
643 binaries may fail if they call getgroups() or getgrouplist() with
644 statically sized storage. Recompiling will work around this, but
645 applications should be modified to use dynamically allocated storage
646 for group arrays as POSIX.1-2008 does not cap an implementation's
647 number of supported groups at NGROUPS_MAX+1 as previous versions did.
649 NFS and portalfs mounts may also be affected as the list of groups is
650 truncated to 16. Users of NFS who use more than 16 groups, should
651 take care that negative group permissions are not used on the exported
652 file systems as they will not be reliable unless a GSSAPI based
653 authentication method is used.
656 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
657 option compiles in the support for adaptive spinning for lockmgrs
658 which want to enable it. The lockinit() function now accepts the flag
659 LK_ADAPTIVE in order to make the lock object subject to adaptive
660 spinning when both held in write and read mode.
663 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
664 changed. User applications that use this ioctl need to be rebuilt.
667 The layout of struct thread has changed. Kernel and modules need to
671 The layout of structs ifnet, domain, protosw and vnet_net has changed.
672 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
675 window(1) has been removed from the base system. It can now be
676 installed from ports. The port is called misc/window.
679 The way we are storing and accessing `routing table' entries has
680 changed. Programs reading the FIB, like netstat, need to be
684 A new netisr implementation has been added for FreeBSD 8. Network
685 file system modules, such as igmp, ipdivert, and others, should be
687 Bump __FreeBSD_version to 800096.
690 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
694 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
695 Bump __FreeBSD_version to 800094.
698 Add mnt_xflag field to 'struct mount'. File system modules need to be
700 Bump __FreeBSD_version to 800093.
703 The compiling option ADAPTIVE_SX has been retired while it has been
704 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
705 The KPI for sx_init_flags() changes as accepting flags:
706 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
707 been introduced in order to handle the reversed logic.
708 Bump __FreeBSD_version to 800092.
711 Add support for hierarchical jails. Remove global securelevel.
712 Bump __FreeBSD_version to 800091.
715 The layout of struct vnet_net has changed, therefore modules
717 Bump __FreeBSD_version to 800090.
720 The newly imported zic(8) produces a new format in the output. Please
721 run tzsetup(8) to install the newly created data to /etc/localtime.
724 The sysctl tree for the usb stack has renamed from hw.usb2.* to
725 hw.usb.* and is now consistent again with previous releases.
728 802.11 monitor mode support was revised and driver api's were changed.
729 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
730 of DLT_IEEE802_11. No user-visible data structures were changed but
731 applications that use DLT_IEEE802_11 may require changes.
732 Bump __FreeBSD_version to 800088.
735 The layout of the following structs has changed: sysctl_oid,
736 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
737 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
738 panics may be experienced. World rebuild is required for
739 correctly checking networking state from userland.
740 Bump __FreeBSD_version to 800085.
743 MLDv2 and Source-Specific Multicast (SSM) have been merged
744 to the IPv6 stack. VIMAGE hooks are in but not yet used.
745 The implementation of SSM within FreeBSD's IPv6 stack closely
746 follows the IPv4 implementation.
748 For kernel developers:
750 * The most important changes are that the ip6_output() and
751 ip6_input() paths no longer take the IN6_MULTI_LOCK,
752 and this lock has been downgraded to a non-recursive mutex.
754 * As with the changes to the IPv4 stack to support SSM, filtering
755 of inbound multicast traffic must now be performed by transport
756 protocols within the IPv6 stack. This does not apply to TCP and
757 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
759 * The KPIs used by IPv6 multicast are similar to those used by
760 the IPv4 stack, with the following differences:
761 * im6o_mc_filter() is analogous to imo_multicast_filter().
762 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
763 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
764 * IN6_LOOKUP_MULTI() has been deprecated and removed.
765 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
766 for MLDv1 have an additional 'timer' argument which is used to
767 jitter the initial membership report for the solicited-node
768 multicast membership on-link.
769 * This is not strictly needed for MLDv2, which already jitters
770 its report transmissions. However, the 'timer' argument is
771 preserved in case MLDv1 is active on the interface.
773 * The KAME linked-list based IPv6 membership implementation has
774 been refactored to use a vector similar to that used by the IPv4
776 Code which maintains a list of its own multicast memberships
777 internally, e.g. carp, has been updated to reflect the new
780 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
781 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
782 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
783 implementation constraint which needs to be addressed in HEAD.
785 For application developers:
787 * The changes are broadly similar to those made for the IPv4
790 * The use of IPv4 and IPv6 multicast socket options on the same
791 socket, using mapped addresses, HAS NOT been tested or supported.
793 * There are a number of issues with the implementation of various
794 IPv6 multicast APIs which need to be resolved in the API surface
795 before the implementation is fully compatible with KAME userland
796 use, and these are mostly to do with interface index treatment.
798 * The literature available discusses the use of either the delta / ASM
799 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
800 using setsourcefilter(3)/getsourcefilter(3). For more information
801 please refer to RFC 3768, 'Socket Interface Extensions for
802 Multicast Source Filters'.
804 * Applications which use the published RFC 3678 APIs should be fine.
806 For systems administrators:
808 * The mtest(8) utility has been refactored to support IPv6, in
809 addition to IPv4. Interface addresses are no longer accepted
810 as arguments, their names must be used instead. The utility
811 will map the interface name to its first IPv4 address as
812 returned by getifaddrs(3).
814 * The ifmcstat(8) utility has also been updated to print the MLDv2
815 endpoint state and source filter lists via sysctl(3).
817 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
818 loopback of IPv6 multicast datagrams by default; it defaults to 1
819 to preserve the existing behaviour. Disabling multicast loopback is
820 recommended for optimal system performance.
822 * The IPv6 MROUTING code has been changed to examine this sysctl
823 instead of attempting to perform a group lookup before looping
824 back forwarded datagrams.
826 Bump __FreeBSD_version to 800084.
829 Implement low-level Bluetooth HCI API.
830 Bump __FreeBSD_version to 800083.
833 The layout of struct malloc_type, used by modules to register new
834 memory allocation types, has changed. Most modules will need to
835 be rebuilt or panics may be experienced.
836 Bump __FreeBSD_version to 800081.
839 Anticipate overflowing inp_flags - add inp_flags2.
840 This changes most offsets in inpcb, so checking v4 connection
841 state will require a world rebuild.
842 Bump __FreeBSD_version to 800080.
845 Add an llentry to struct route and struct route_in6. Modules
846 embedding a struct route will need to be recompiled.
847 Bump __FreeBSD_version to 800079.
850 The size of rt_metrics_lite and by extension rtentry has changed.
851 Networking administration apps will need to be recompiled.
852 The route command now supports show as an alias for get, weighting
853 of routes, sticky and nostick flags to alter the behavior of stateful
855 Bump __FreeBSD_version to 800078.
858 Do not use Giant for kbdmux(4) locking. This is wrong and
859 apparently causing more problems than it solves. This will
860 re-open the issue where interrupt handlers may race with
861 kbdmux(4) in polling mode. Typical symptoms include (but
862 not limited to) duplicated and/or missing characters when
863 low level console functions (such as gets) are used while
864 interrupts are enabled (for example geli password prompt,
865 mountroot prompt etc.). Disabling kbdmux(4) may help.
868 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
869 kernel modules referencing any of the above need to be recompiled.
870 Bump __FreeBSD_version to 800075.
873 GEOM_PART has become the default partition slicer for storage devices,
874 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
875 introduces some changes:
877 MSDOS/EBR: the devices created from MSDOS extended partition entries
878 (EBR) can be named differently than with GEOM_MBR and are now symlinks
879 to devices with offset-based names. fstabs may need to be modified.
881 BSD: the "geometry does not match label" warning is harmless in most
882 cases but it points to problems in file system misalignment with
883 disk geometry. The "c" partition is now implicit, covers the whole
884 top-level drive and cannot be (mis)used by users.
886 General: Kernel dumps are now not allowed to be written to devices
887 whose partition types indicate they are meant to be used for file
888 systems (or, in case of MSDOS partitions, as something else than
891 Most of these changes date approximately from 200812.
894 The uscanner(4) driver has been removed from the kernel. This follows
895 Linux removing theirs in 2.6 and making libusb the default interface
899 The multicast forwarding code has been cleaned up. netstat(1)
900 only relies on KVM now for printing bandwidth upcall meters.
901 The IPv4 and IPv6 modules are split into ip_mroute_mod and
902 ip6_mroute_mod respectively. The config(5) options for statically
903 compiling this code remain the same, i.e. 'options MROUTING'.
906 Support for the IFF_NEEDSGIANT network interface flag has been
907 removed, which means that non-MPSAFE network device drivers are no
908 longer supported. In particular, if_ar, if_sr, and network device
909 drivers from the old (legacy) USB stack can no longer be built or
913 POSIX.1 Native Language Support (NLS) has been enabled in libc and
914 a bunch of new language catalog files have also been added.
915 This means that some common libc messages are now localized and
916 they depend on the LC_MESSAGES environmental variable.
919 The k8temp(4) driver has been renamed to amdtemp(4) since
920 support for Family 10 and Family 11 CPU families was added.
923 IGMPv3 and Source-Specific Multicast (SSM) have been merged
924 to the IPv4 stack. VIMAGE hooks are in but not yet used.
926 For kernel developers, the most important changes are that the
927 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
928 and this lock has been downgraded to a non-recursive mutex.
930 Transport protocols (UDP, Raw IP) are now responsible for filtering
931 inbound multicast traffic according to group membership and source
932 filters. The imo_multicast_filter() KPI exists for this purpose.
933 Transports which do not use multicast (SCTP, TCP) already reject
934 multicast by default. Forwarding and receive performance may improve
935 as a mutex acquisition is no longer needed in the ip_input()
936 low-level input path. in_addmulti() and in_delmulti() are shimmed
937 to new KPIs which exist to support SSM in-kernel.
939 For application developers, it is recommended that loopback of
940 multicast datagrams be disabled for best performance, as this
941 will still cause the lock to be taken for each looped-back
942 datagram transmission. The net.inet.ip.mcast.loop sysctl may
943 be tuned to 0 to disable loopback by default; it defaults to 1
944 to preserve the existing behaviour.
946 For systems administrators, to obtain best performance with
947 multicast reception and multiple groups, it is always recommended
948 that a card with a suitably precise hash filter is used. Hash
949 collisions will still result in the lock being taken within the
950 transport protocol input path to check group membership.
952 If deploying FreeBSD in an environment with IGMP snooping switches,
953 it is recommended that the net.inet.igmp.sendlocal sysctl remain
954 enabled; this forces 224.0.0.0/24 group membership to be announced
957 The size of 'struct igmpstat' has changed; netstat needs to be
958 recompiled to reflect this.
959 Bump __FreeBSD_version to 800070.
962 libusb20.so.1 is now installed as libusb.so.1 and the ports system
963 updated to use it. This requires a buildworld/installworld in order to
964 update the library and dependencies (usbconfig, etc). Its advisable to
965 rebuild all ports which uses libusb. More specific directions are given
966 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
967 libusb are no longer required and can be removed.
970 A workaround is committed to allow the creation of System V shared
971 memory segment of size > 2 GB on the 64-bit architectures.
972 Due to a limitation of the existing ABI, the shm_segsz member
973 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
974 wrong for large segments. Note that limits must be explicitly
975 raised to allow such segments to be created.
978 The layout of struct ifnet has changed, requiring a rebuild of all
979 network device driver modules.
982 The /dev handling for the new USB stack has changed, a
983 buildworld/installworld is required for libusb20.
986 The new USB2 stack has now been permanently moved in and all kernel and
987 module names reverted to their previous values (eg, usb, ehci, ohci,
988 ums, ...). The old usb stack can be compiled in by prefixing the name
989 with the letter 'o', the old usb modules have been removed.
990 Updating entry 20090216 for xorg and 20090215 for libmap may still
994 The rc.conf(5) option if_up_delay has been renamed to
995 defaultroute_delay to better reflect its purpose. If you have
996 customized this setting in /etc/rc.conf you need to update it to
1000 xorg 7.4 wants to configure its input devices via hald which does not
1001 yet work with USB2. If the keyboard/mouse does not work in xorg then
1003 Option "AllowEmptyInput" "off"
1004 to your ServerLayout section. This will cause X to use the configured
1005 kbd and mouse sections from your xorg.conf.
1008 The GENERIC kernels for all architectures now default to the new USB2
1009 stack. No kernel config options or code have been removed so if a
1010 problem arises please report it and optionally revert to the old USB
1011 stack. If you are loading USB kernel modules or have a custom kernel
1012 that includes GENERIC then ensure that usb names are also changed over,
1013 eg uftdi -> usb2_serial_ftdi.
1015 Older programs linked against the ports libusb 0.1 need to be
1016 redirected to the new stack's libusb20. /etc/libmap.conf can
1018 # Map old usb library to new one for usb2 stack
1019 libusb-0.1.so.8 libusb20.so.1
1022 All USB ethernet devices now attach as interfaces under the name ueN
1023 (eg. ue0). This is to provide a predictable name as vendors often
1024 change usb chipsets in a product without notice.
1027 The ichsmb(4) driver has been changed to require SMBus slave
1028 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1029 All of the other SMBus controller drivers require left-justified
1030 slave addresses, so this change makes all the drivers provide the
1034 INET6 statistics (struct ip6stat) was updated.
1035 netstat(1) needs to be recompiled.
1038 NTFS has been removed from GENERIC kernel on amd64 to match
1039 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1040 will load ntfs.ko module automatically when NTFS support is
1041 actually needed, unless ntfs.ko is not installed or security
1042 level prohibits loading kernel modules. If either is the case,
1043 "options NTFS" has to be added into kernel config.
1046 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1047 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1048 800061. User space tools that rely on the size of struct tcpcb in
1049 tcp_var.h (e.g. sockstat) need to be recompiled.
1052 ng_tty(4) module updated to match the new TTY subsystem.
1053 Due to API change, user-level applications must be updated.
1054 New API support added to mpd5 CVS and expected to be present
1055 in next mpd5.3 release.
1058 With __FreeBSD_version 800060 the makefs tool is part of
1059 the base system (it was a port).
1062 The afdata and ifnet locks have been changed from mutexes to
1063 rwlocks, network modules will need to be re-compiled.
1066 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1067 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1068 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1069 architecture and 40 bytes on 64-bit architecture. The userland
1070 applications "arp" and "ndp" have been updated accordingly.
1071 The output from "netstat -r" shows only routing entries and
1072 none of the L2 information.
1075 __FreeBSD_version 800057 marks the switchover from the
1076 binary ath hal to source code. Users must add the line:
1078 options AH_SUPPORT_AR5416
1080 to their kernel config files when specifying:
1084 The ath_hal module no longer exists; the code is now compiled
1085 together with the driver in the ath module. It is now
1086 possible to tailor chip support (i.e. reduce the set of chips
1087 and thereby the code size); consult ath_hal(4) for details.
1090 __FreeBSD_version 800054 adds memory barriers to
1091 <machine/atomic.h>, new interfaces to ifnet to facilitate
1092 multiple hardware transmit queues for cards that support
1093 them, and a lock-less ring-buffer implementation to
1094 enable drivers to more efficiently manage queueing of
1098 A new version of ZFS (version 13) has been merged to -HEAD.
1099 This version has zpool attribute "listsnapshots" off by
1100 default, which means "zfs list" does not show snapshots,
1101 and is the same as Solaris behavior.
1104 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1107 The uhci, ohci, ehci and slhci USB Host controller drivers have
1108 been put into separate modules. If you load the usb module
1109 separately through loader.conf you will need to load the
1110 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1111 controller add the following to loader.conf:
1117 The ABI used by the PMC toolset has changed. Please keep
1118 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1122 atapci kernel module now includes only generic PCI ATA
1123 driver. AHCI driver moved to ataahci kernel module.
1124 All vendor-specific code moved into separate kernel modules:
1125 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1126 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1127 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1128 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1131 The TTY subsystem of the kernel has been replaced by a new
1132 implementation, which provides better scalability and an
1133 improved driver model. Most common drivers have been migrated to
1134 the new TTY subsystem, while others have not. The following
1135 drivers have not yet been ported to the new TTY layer:
1138 cy, digi, rc, rp, sio
1144 ng_h4, ng_tty, ppp, sl, snp
1146 Adding these drivers to your kernel configuration file shall
1147 cause compilation to fail.
1150 ntpd has been upgraded to 4.2.4p5.
1153 OpenSSH has been upgraded to 5.1p1.
1155 For many years, FreeBSD's version of OpenSSH preferred DSA
1156 over RSA for host and user authentication keys. With this
1157 upgrade, we've switched to the vendor's default of RSA over
1158 DSA. This may cause upgraded clients to warn about unknown
1159 host keys even for previously known hosts. Users should
1160 follow the usual procedure for verifying host keys before
1161 accepting the RSA key.
1163 This can be circumvented by setting the "HostKeyAlgorithms"
1164 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1167 Please note that the sequence of keys offered for
1168 authentication has been changed as well. You may want to
1169 specify IdentityFile in a different order to revert this
1173 The sio(4) driver has been removed from the i386 and amd64
1174 kernel configuration files. This means uart(4) is now the
1175 default serial port driver on those platforms as well.
1177 To prevent collisions with the sio(4) driver, the uart(4) driver
1178 uses different names for its device nodes. This means the
1179 onboard serial port will now most likely be called "ttyu0"
1180 instead of "ttyd0". You may need to reconfigure applications to
1181 use the new device names.
1183 When using the serial port as a boot console, be sure to update
1184 /boot/device.hints and /etc/ttys before booting the new kernel.
1185 If you forget to do so, you can still manually specify the hints
1186 at the loader prompt:
1188 set hint.uart.0.at="isa"
1189 set hint.uart.0.port="0x3F8"
1190 set hint.uart.0.flags="0x10"
1191 set hint.uart.0.irq="4"
1195 The gpt(8) utility has been removed. Use gpart(8) to partition
1199 The version that Linuxulator emulates was changed from 2.4.2
1200 to 2.6.16. If you experience any problems with Linux binaries
1201 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1202 if it fixes the problem contact emulation mailing list.
1205 ISDN4BSD (I4B) was removed from the src tree. You may need to
1206 update a your kernel configuration and remove relevant entries.
1209 I have checked in code to support multiple routing tables.
1210 See the man pages setfib(1) and setfib(2).
1211 This is a hopefully backwards compatible version,
1212 but to make use of it you need to compile your kernel
1213 with options ROUTETABLES=2 (or more up to 16).
1216 The 802.11 wireless support was redone to enable multi-bss
1217 operation on devices that are capable. The underlying device
1218 is no longer used directly but instead wlanX devices are
1219 cloned with ifconfig. This requires changes to rc.conf files.
1220 For example, change:
1221 ifconfig_ath0="WPA DHCP"
1224 ifconfig_wlan0="WPA DHCP"
1225 see rc.conf(5) for more details. In addition, mergemaster of
1226 /etc/rc.d is highly recommended. Simultaneous update of userland
1227 and kernel wouldn't hurt either.
1229 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1230 modules were merged into the base wlan module. All references
1231 to these modules (e.g. in kernel config files) must be removed.
1234 psm(4) has gained write(2) support in native operation level.
1235 Arbitrary commands can be written to /dev/psm%d and status can
1236 be read back from it. Therefore, an application is responsible
1237 for status validation and error recovery. It is a no-op in
1238 other operation levels.
1241 Support for KSE threading has been removed from the kernel. To
1242 run legacy applications linked against KSE libmap.conf may
1243 be used. The following libmap.conf may be used to ensure
1244 compatibility with any prior release:
1246 libpthread.so.1 libthr.so.1
1247 libpthread.so.2 libthr.so.2
1248 libkse.so.3 libthr.so.3
1251 The layout of struct vmspace has changed. This affects libkvm
1252 and any executables that link against libkvm and use the
1253 kvm_getprocs() function. In particular, but not exclusively,
1254 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1255 The effects are minimal, but it's advisable to upgrade world
1259 The latest em driver no longer has support in it for the
1260 82575 adapter, this is now moved to the igb driver. The
1261 split was done to make new features that are incompatible
1262 with older hardware easier to do.
1265 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1266 likewise the kernel option is now GEOM_LINUX_LVM.
1269 The default NFS mount mode has changed from UDP to TCP for
1270 increased reliability. If you rely on (insecurely) NFS
1271 mounting across a firewall you may need to update your
1275 Belatedly note the addition of m_collapse for compacting
1279 The fts(3) structures have been changed to use adequate
1280 integer types for their members and so to be able to cope
1281 with huge file trees. The old fts(3) ABI is preserved
1282 through symbol versioning in libc, so third-party binaries
1283 using fts(3) should still work, although they will not take
1284 advantage of the extended types. At the same time, some
1285 third-party software might fail to build after this change
1286 due to unportable assumptions made in its source code about
1287 fts(3) structure members. Such software should be fixed
1288 by its vendor or, in the worst case, in the ports tree.
1289 FreeBSD_version 800015 marks this change for the unlikely
1290 case that a portable fix is impossible.
1293 To upgrade to -current after this date, you must be running
1294 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1295 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1298 The ADAPTIVE_GIANT kernel option has been retired because its
1299 functionality is the default now.
1302 The AT keyboard emulation of sunkbd(4) has been turned on
1303 by default. In order to make the special symbols of the Sun
1304 keyboards driven by sunkbd(4) work under X these now have
1305 to be configured the same way as Sun USB keyboards driven
1306 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1308 Option "XkbLayout" "us"
1309 Option "XkbRules" "xorg"
1310 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1313 It has been decided that it is desirable to provide ABI
1314 backwards compatibility to the FreeBSD 4/5/6 versions of the
1315 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1316 broken with the introduction of PCI domain support (see the
1317 20070930 entry). Unfortunately, this required the ABI of
1318 PCIOCGETCONF to be broken again in order to be able to
1319 provide backwards compatibility to the old version of that
1320 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1321 again. As for prominent ports this affects neither pciutils
1322 nor xorg-server this time, the hal port needs to be rebuilt
1326 The misnamed kthread_create() and friends have been renamed
1327 to kproc_create() etc. Many of the callers already
1328 used kproc_start()..
1329 I will return kthread_create() and friends in a while
1330 with implementations that actually create threads, not procs.
1331 Renaming corresponds with version 800002.
1340 Avoid using make -j when upgrading. While generally safe, there are
1341 sometimes problems using -j to upgrade. If your upgrade fails with
1342 -j, please try again without -j. From time to time in the past there
1343 have been problems using -j with buildworld and/or installworld. This
1344 is especially true when upgrading between "distant" versions (eg one
1345 that cross a major release boundary or several minor releases, or when
1346 several months have passed on the -current branch).
1348 Sometimes, obscure build problems are the result of environment
1349 poisoning. This can happen because the make utility reads its
1350 environment when searching for values for global variables. To run
1351 your build attempts in an "environmental clean room", prefix all make
1352 commands with 'env -i '. See the env(1) manual page for more details.
1354 When upgrading from one major version to another it is generally best
1355 to upgrade to the latest code in the currently installed branch first,
1356 then do an upgrade to the new branch. This is the best-tested upgrade
1357 path, and has the highest probability of being successful. Please try
1358 this approach before reporting problems with a major version upgrade.
1362 When upgrading the boot ZFS pool to a new version, always follow
1365 1.) recompile and reinstall the ZFS boot loader and boot block
1366 (this is part of "make buildworld" and "make installworld")
1368 2.) update the ZFS boot block on your boot drive
1370 The following example updates the ZFS boot block on the first
1371 partition (freebsd-boot) of a GPT partitioned drive ad0:
1372 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1374 Non-boot pools do not need these updates.
1378 If you are updating from a prior version of FreeBSD (even one just
1379 a few days old), you should follow this procedure. It is the most
1380 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1382 make kernel-toolchain
1383 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1384 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1386 To test a kernel once
1387 ---------------------
1388 If you just want to boot a kernel once (because you are not sure
1389 if it works, or if you want to boot a known bad kernel to provide
1390 debugging information) run
1391 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1392 nextboot -k testkernel
1394 To just build a kernel when you know that it won't mess you up
1395 --------------------------------------------------------------
1396 This assumes you are already running a CURRENT system. Replace
1397 ${arch} with the architecture of your machine (e.g. "i386",
1398 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1400 cd src/sys/${arch}/conf
1401 config KERNEL_NAME_HERE
1402 cd ../compile/KERNEL_NAME_HERE
1407 If this fails, go to the "To build a kernel" section.
1409 To rebuild everything and install it on the current system.
1410 -----------------------------------------------------------
1411 # Note: sometimes if you are running current you gotta do more than
1412 # is listed here if you are upgrading from a really old current.
1414 <make sure you have good level 0 dumps>
1416 make kernel KERNCONF=YOUR_KERNEL_HERE
1418 <reboot in single user> [3]
1426 To cross-install current onto a separate partition
1427 --------------------------------------------------
1428 # In this approach we use a separate partition to hold
1429 # current's root, 'usr', and 'var' directories. A partition
1430 # holding "/", "/usr" and "/var" should be about 2GB in
1433 <make sure you have good level 0 dumps>
1436 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1437 <maybe newfs current's root partition>
1438 <mount current's root partition on directory ${CURRENT_ROOT}>
1439 make installworld DESTDIR=${CURRENT_ROOT}
1440 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1441 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1442 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1443 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1444 <reboot into current>
1445 <do a "native" rebuild/install as described in the previous section>
1446 <maybe install compatibility libraries from ports/misc/compat*>
1450 To upgrade in-place from stable to current
1451 ----------------------------------------------
1452 <make sure you have good level 0 dumps>
1454 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1456 <reboot in single user> [3]
1463 Make sure that you've read the UPDATING file to understand the
1464 tweaks to various things you need. At this point in the life
1465 cycle of current, things change often and you are on your own
1466 to cope. The defaults can also change, so please read ALL of
1467 the UPDATING entries.
1469 Also, if you are tracking -current, you must be subscribed to
1470 freebsd-current@freebsd.org. Make sure that before you update
1471 your sources that you have read and understood all the recent
1472 messages there. If in doubt, please track -stable which has
1473 much fewer pitfalls.
1475 [1] If you have third party modules, such as vmware, you
1476 should disable them at this point so they don't crash your
1479 [3] From the bootblocks, boot -s, and then do
1484 adjkerntz -i # if CMOS is wall time
1485 Also, when doing a major release upgrade, it is required that
1486 you boot into single user mode to do the installworld.
1488 [4] Note: This step is non-optional. Failure to do this step
1489 can result in a significant reduction in the functionality of the
1490 system. Attempting to do it by hand is not recommended and those
1491 that pursue this avenue should read this file carefully, as well
1492 as the archives of freebsd-current and freebsd-hackers mailing lists
1493 for potential gotchas. The -U option is also useful to consider.
1494 See mergemaster(8) for more information.
1496 [5] Usually this step is a noop. However, from time to time
1497 you may need to do this if you get unknown user in the following
1498 step. It never hurts to do it all the time. You may need to
1499 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1500 install) after the buildworld before this step if you last updated
1501 from current before 20020224 or from -stable before 20020408.
1503 [6] This only deletes old files and directories. Old libraries
1504 can be deleted by "make delete-old-libs", but you have to make
1505 sure that no program is using those libraries anymore.
1507 [8] In order to have a kernel that can run the 4.x binaries needed to
1508 do an installworld, you must include the COMPAT_FREEBSD4 option in
1509 your kernel. Failure to do so may leave you with a system that is
1510 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1511 required to run the 5.x binaries on more recent kernels. And so on
1512 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1514 Make sure that you merge any new devices from GENERIC since the
1515 last time you updated your kernel config file.
1517 [9] When checking out sources, you must include the -P flag to have
1518 cvs prune empty directories.
1520 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1521 "?=" instead of the "=" assignment operator, so that buildworld can
1522 override the CPUTYPE if it needs to.
1524 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1525 not on the command line, or in /etc/make.conf. buildworld will
1526 warn if it is improperly defined.
1529 This file contains a list, in reverse chronological order, of major
1530 breakages in tracking -current. It is not guaranteed to be a complete
1531 list of such breakages, and only contains entries since October 10, 2007.
1532 If you need to see UPDATING entries from before that date, you will need
1533 to fetch an UPDATING file from an older FreeBSD release.
1535 Copyright information:
1537 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1539 Redistribution, publication, translation and use, with or without
1540 modification, in full or in part, in any form or format of this
1541 document are permitted without further permission from the author.
1543 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1544 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1545 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1546 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1547 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1548 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1549 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1550 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1551 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1552 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1553 POSSIBILITY OF SUCH DAMAGE.
1555 Contact Warner Losh if you have any questions about your use of