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 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
12 NOTE TO PEOPLE WHO THINK THAT FreeBSD 9.x IS SLOW:
13 FreeBSD 9.x has many debugging features turned on, in both the kernel
14 and userland. These features attempt to detect incorrect use of
15 system primitives, and encourage loud failure through extra sanity
16 checking and fail stop semantics. They also substantially impact
17 system performance. If you want to do performance measurement,
18 benchmarking, and optimization, you'll want to turn them off. This
19 includes various WITNESS- related kernel options, INVARIANTS, malloc
20 debugging flags in userland, and various verbose features in the
21 kernel. Many developers choose to disable these features on build
22 machines to maximize performance. (To disable malloc debugging, run
23 ln -s aj /etc/malloc.conf.)
26 During the merge of Capsicum features, the fget(9) KPI was modified.
27 This may require the rebuilding of out-of-tree device drivers --
28 issues have been reported specifically with the nVidia device driver.
29 __FreeBSD_version is bumped to 900041.
31 Also, there is a period between 20110811 and 20110814 where the
32 special devices /dev/{stdin,stdout,stderr} did not work correctly.
33 Building world from a kernel during that window may not work.
36 The packet filter (pf) code has been updated to OpenBSD 4.5.
37 You need to update userland tools to be in sync with kernel.
38 This update breaks backward compatibility with earlier pfsync(4)
39 versions. Care must be taken when updating redundant firewall setups.
42 The following sysctls and tunables are retired on x86 platforms:
44 machdep.hlt_logical_cpus
45 The following sysctl is retired:
46 machdep.hyperthreading_allowed
47 The sysctls were supposed to provide a way to dynamically offline and
48 online selected CPUs on x86 platforms, but the implementation has not
49 been reliable especially with SCHED_ULE scheduler.
50 machdep.hyperthreading_allowed tunable is still available to ignore
51 hyperthreading CPUs at OS level.
52 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
53 where X is an APIC ID of a CPU. Be advised, though, that disabling
54 CPUs in non-uniform fashion will result in non-uniform topology and
55 may lead to sub-optimal system performance with SCHED_ULE, which is
59 cpumask_t type is retired and cpuset_t is used in order to describe
63 Changes to ifconfig(8) for dynamic address family detection mandate
64 that you are running a kernel of 20110525 or later. Make sure to
65 follow the update procedure to boot a new kernel before installing
69 Support for sun4v architecture is officially dropped
72 Several KPI breaking changes have been committed to the mii(4) layer,
73 the PHY drivers and consequently some Ethernet drivers using mii(4).
74 This means that miibus.ko and the modules of the affected Ethernet
75 drivers need to be recompiled.
77 Note to kernel developers: Given that the OUI bit reversion problem
78 was fixed as part of these changes all mii(4) commits related to OUIs,
79 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
80 handling, no longer can be merged verbatim to stable/8 and previous
84 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
85 into their kernel configurations along with 'device pci'.
88 The default NFS client is now the new NFS client, so fstype "newnfs"
89 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
90 Although mounts via fstype "nfs" will usually work without userland
91 changes, it is recommended that the mount(8) and mount_nfs(8)
92 commands be rebuilt from sources and that a link to mount_nfs called
93 mount_oldnfs be created. The new client is compiled into the
94 kernel with "options NFSCL" and this is needed for diskless root
95 file systems. The GENERIC kernel configs have been changed to use
96 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
97 To use the regular/old client, you can "mount -t oldnfs ...". For
98 a diskless root file system, you must also include a line like:
100 vfs.root.mountfrom="oldnfs:"
102 in the boot/loader.conf on the root fs on the NFS server to make
103 a diskless root fs use the old client.
106 The GENERIC kernels for all architectures now default to the new
107 CAM-based ATA stack. It means that all legacy ATA drivers were
108 removed and replaced by respective CAM drivers. If you are using
109 ATA device names in /etc/fstab or other places, make sure to update
110 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
111 where 'Y's are the sequential numbers starting from zero for each type
112 in order of detection, unless configured otherwise with tunables,
113 see cam(4)). There will be symbolic links created in /dev/ to map
114 old adX devices to the respective adaY. They should provide basic
115 compatibility for file systems mounting in most cases, but they do
116 not support old user-level APIs and do not have respective providers
117 in GEOM. Consider using updated management tools with new device names.
119 It is possible to load devices ahci, ata, siis and mvs as modules,
120 but option ATA_CAM should remain in kernel configuration to make ata
121 module work as CAM driver supporting legacy ATA controllers. Device ata
122 still can be used in modular fashion (atacore + ...). Modules atadisk
123 and atapi* are not used and won't affect operation in ATA_CAM mode.
124 Note that to use CAM-based ATA kernel should include CAM devices
125 scbus, pass, da (or explicitly ada), cd and optionally others. All of
126 them are parts of the cam module.
128 ataraid(4) functionality is now supported by the RAID GEOM class.
129 To use it you can load geom_raid kernel module and use graid(8) tool
130 for management. Instead of /dev/arX device names, use /dev/raid/rX.
132 No kernel config options or code have been removed, so if a problem
133 arises, please report it and optionally revert to the old ATA stack.
134 In order to do it you can remove from the kernel config:
139 , and instead add back:
140 device atadisk # ATA disk drives
141 device ataraid # ATA RAID drives
142 device atapicd # ATAPI CDROM drives
143 device atapifd # ATAPI floppy drives
144 device atapist # ATAPI tape drives
147 The default NFS server has been changed to the new server, which
148 was referred to as the experimental server. If you need to switch
149 back to the old NFS server, you must now put the "-o" option on
150 both the mountd and nfsd commands. This can be done using the
151 mountd_flags and nfs_server_flags rc.conf variables until an
152 update to the rc scripts is committed, which is coming soon.
155 The GNU Objective-C runtime library (libobjc), and other Objective-C
156 related components have been removed from the base system. If you
157 require an Objective-C library, please use one of the available ports.
160 ath(4) has been split into bus- and device- modules. if_ath contains
161 the HAL, the TX rate control and the network device code. if_ath_pci
162 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
163 contains the AHB glue. Users need to load both if_ath_pci and if_ath
164 in order to use ath on everything else.
166 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
167 need to load if_ath and if_ath_pci for ath(4) operation.
170 As part of the replacement of sysinstall, the process of building
171 release media has changed significantly. For details, please re-read
172 release(7), which has been updated to reflect the new build process.
175 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
176 is the last available version under GPLv2. It brings a number of new
177 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
178 4.1 and SSE 4.2), better support for powerpc64, a number of new
179 directives, and lots of other small improvements. See the ChangeLog
180 file in contrib/binutils for the full details.
183 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
184 compliant, and will now use half of hash for authentication.
185 This will break interoperability with all stacks (including all
186 actual FreeBSD versions) who implement
187 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
189 The only workaround with such peers is to use another HMAC
190 algorithm for IPsec ("phase 2") authentication.
193 Remove the uio_yield prototype and symbol. This function has
194 been misnamed since it was introduced and should not be
195 globally exposed with this name. The equivalent functionality
196 is now available using kern_yield(curthread->td_user_pri).
197 The function remains undocumented.
200 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
201 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
202 for scalar sysctls is defined but disabled. Code that needs
203 UQUAD to pass the type checking that must compile on older
204 systems where the define is not present can check against
205 __FreeBSD_version >= 900030.
207 The system dialog(1) has been replaced with a new version previously
208 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
209 with the previous version, but the libdialog associated with it has
210 a largely incompatible API. As such, the original version of libdialog
211 will be kept temporarily as libodialog, until its base system consumers
212 are replaced or updated. Bump __FreeBSD_version to 900030.
215 If you are trying to run make universe on a -stable system, and you get
216 the following warning:
217 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
218 unknown. config(8) likely too old."
219 or something similar to it, then you must upgrade your -stable system
220 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
221 stable/8) or build the config from the latest stable/8 branch and
222 install it on your system.
224 Prior to this date, building a current universe on 8-stable system from
225 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
226 error in the first kernel build phase. A new config on those old
227 systems will fix that problem for older versions of -current.
230 The TCP stack has been modified to allow Khelp modules to interact with
231 it via helper hook points and store per-connection data in the TCP
232 control block. Bump __FreeBSD_version to 900029. User space tools that
233 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
237 Generic IEEE 802.3 annex 31B full duplex flow control support has been
238 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
239 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
240 to take advantage of it instead of using custom implementations. This
241 means that these drivers now no longer unconditionally advertise
242 support for flow control but only do so if flow control is a selected
243 media option. This was implemented in the generic support that way in
244 order to allow flow control to be switched on and off via ifconfig(8)
245 with the PHY specific default to typically off in order to protect
246 from unwanted effects. Consequently, if you used flow control with
247 one of the above mentioned drivers you now need to explicitly enable
249 ifconfig bge0 media auto mediaopt flowcontrol
251 Along with the above mentioned changes generic support for setting
252 1000baseT master mode also has been added and brgphy(4), ciphy(4),
253 e1000phy(4) as well as ip1000phy(4) have been converted to take
254 advantage of it. This means that these drivers now no longer take the
255 link0 parameter for selecting master mode but the master media option
256 has to be used instead, for example like in the following:
257 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
259 Selection of master mode now is also available with all other PHY
260 drivers supporting 1000baseT.
263 The TCP stack has received a significant update to add support for
264 modularised congestion control and generally improve the clarity of
265 congestion control decisions. Bump __FreeBSD_version to 900025. User
266 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
267 sockstat) need to be recompiled.
270 The man(1) utility has been replaced by a new version that no longer
271 uses /etc/manpath.config. Please consult man.conf(5) for how to
272 migrate local entries to the new format.
275 The copyright strings printed by login(1) and sshd(8) at the time of a
276 new connection have been removed to follow other operating systems and
280 A workaround for a fixed ld bug has been removed in kernel code,
281 so make sure that your system ld is built from sources after
282 revision 210245 from 2010-07-19 (r211583 if building head kernel
283 on stable/8, r211584 for stable/7; both from 2010-08-21).
284 A symptom of incorrect ld version is different addresses for
285 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
288 The $ipv6_prefer variable in rc.conf(5) has been split into
289 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
291 The $ip6addrctl_policy is a variable to choose a pre-defined
292 address selection policy set by ip6addrctl(8). A value
293 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
296 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
297 flag (see an entry of 20090926) is set on an interface with no
298 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
299 security reason. If you want IPv6 link-local address on all
300 interfaces by default, set this to "YES".
302 The old ipv6_prefer="YES" is equivalent to
303 ipv6_activate_all_interfaces="YES" and
304 ip6addrctl_policy="ipv6_prefer".
307 DTrace has grown support for userland tracing. Due to this, DTrace is
308 now i386 and amd64 only.
309 dtruss(1) is now installed by default on those systems and a new
310 kernel module is needed for userland tracing: fasttrap.
311 No changes to your kernel config file are necessary to enable
312 userland tracing, but you might consider adding 'STRIP=' and
313 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
314 to have informative userland stack traces in DTrace (ustack).
317 The acpi_aiboost(4) driver has been removed in favor of the new
318 aibs(4) driver. You should update your kernel configuration file.
321 BSD grep has been imported to the base system and it is built by
322 default. It is completely BSD licensed, highly GNU-compatible, uses
323 less memory than its GNU counterpart and has a small codebase.
324 However, it is slower than its GNU counterpart, which is mostly
325 noticeable for larger searches, for smaller ones it is measurable
326 but not significant. The reason is complex, the most important factor
327 is that we lack a modern and efficient regex library and GNU
328 overcomes this by optimizing the searches internally. Future work
329 on improving the regex performance is planned, for the meantime,
330 users that need better performance, can build GNU grep instead by
331 setting the WITH_GNU_GREP knob.
334 Due to the import of powerpc64 support, all existing powerpc kernel
335 configuration files must be updated with a machine directive like this:
336 machine powerpc powerpc
338 In addition, an updated config(8) is required to build powerpc kernels
342 A new version of ZFS (version 15) has been merged to -HEAD.
343 This version uses a python library for the following subcommands:
344 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
345 For full functionality of these commands the following port must
346 be installed: sysutils/py-zfs
349 'vm_page's are now hashed by physical address to an array of mutexes.
350 Currently this is only used to serialize access to hold_count. Over
351 time the page queue mutex will be peeled away. This changes the size
352 of pmap on every architecture. And requires all callers of vm_page_hold
353 and vm_page_unhold to be updated.
356 WITH_CTF can now be specified in src.conf (not recommended, there
357 are some problems with static executables), make.conf (would also
358 affect ports which do not use GNU make and do not override the
359 compile targets) or in the kernel config (via "makeoptions
361 When WITH_CTF was specified there before this was silently ignored,
362 so make sure that WITH_CTF is not used in places which could lead
363 to unwanted behavior.
366 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
367 to allow 32-bit compatibility on non-x86 platforms. All kernel
368 configurations on amd64 and ia64 platforms using these options must
369 be modified accordingly.
372 The utmp user accounting database has been replaced with utmpx,
373 the user accounting interface standardized by POSIX.
374 Unfortunately the semantics of utmp and utmpx don't match,
375 making it practically impossible to support both interfaces.
376 The user accounting database is used by tools like finger(1),
377 last(1), talk(1), w(1) and ac(8).
379 All applications in the base system use utmpx. This means only
380 local binaries (e.g. from the ports tree) may still use these
381 utmp database files. These applications must be rebuilt to make
384 After the system has been upgraded, it is safe to remove the old
385 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
386 assuming their contents is of no importance anymore. Old wtmp
387 databases can only be used by last(1) and ac(8) after they have
388 been converted to the new format using wtmpcvt(1).
391 Introduce the kernel thread "deadlock resolver" (which can be enabled
392 via the DEADLKRES option, see NOTES for more details) and the
393 sleepq_type() function for sleepqueues.
396 The rc.firewall and rc.firewall6 were unified, and
397 rc.firewall6 and rc.d/ip6fw were removed.
398 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
399 variables are obsoleted. Instead, the following new rc
400 variables are added to rc.d/ipfw:
402 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
403 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
404 firewall_simple_onet_ipv6, firewall_trusted_ipv6
406 The meanings correspond to the relevant IPv4 variables.
412 The default terminal emulation for syscons(4) has been changed
413 from cons25 to xterm on all platforms except pc98. This means
414 that the /etc/ttys file needs to be updated to ensure correct
415 operation of applications on the console.
417 The terminal emulation style can be toggled per window by using
418 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
419 options can be used to change the compile-time default back to
422 To prevent graphical artifacts, make sure the TERM environment
423 variable is set to match the terminal emulation that is being
424 performed by syscons(4).
427 The layout of the structure ieee80211req_scan_result has changed.
428 Applications that require wireless scan results (e.g. ifconfig(8))
429 from net80211 need to be recompiled.
431 Applications such as wpa_supplicant(8) may require a full world
432 build without using NO_CLEAN in order to get synchronized with the
436 The iwn(4) driver has been updated to support the 5000 and 5150 series.
437 There's one kernel module for each firmware. Adding "device iwnfw"
438 to the kernel configuration file means including all three firmware
439 images inside the kernel. If you want to include just the one for
440 your wireless card, use the the devices iwn4965fw, iwn5000fw or
444 The rc.d/network_ipv6, IPv6 configuration script has been integrated
445 into rc.d/netif. The changes are the following:
447 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
448 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
449 Note that both variables need the "inet6" keyword at the head.
451 Do not set $ipv6_network_interfaces manually if you do not
452 understand what you are doing. It is not needed in most cases.
454 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
457 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
458 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
460 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
461 all configured interfaces have "inet6 accept_rtadv" in the
462 $ifconfig_IF_ipv6. These are for backward compatibility.
464 3. A new variable $ipv6_prefer has been added. If NO, IPv6
465 functionality of interfaces with no corresponding
466 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
467 and the default address selection policy of ip6addrctl(8)
468 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
469 Note that if you want to configure IPv6 functionality on the
470 disabled interfaces after boot, first you need to clear the flag by
471 using ifconfig(8) like:
473 ifconfig em0 inet6 -ifdisabled
475 If YES, the default address selection policy is set as
478 The default value of $ipv6_prefer is NO.
480 4. If your system need to receive Router Advertisement messages,
481 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
482 scripts automatically invoke rtsol(8) when the interface becomes
483 UP. The Router Advertisement messages are used for SLAAC
484 (State-Less Address AutoConfiguration).
487 802.11s D3.03 support was committed. This is incompatible with the
488 previous code, which was based on D3.0.
491 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
492 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
493 control whether accepting Router Advertisement messages or not.
494 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
495 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
496 The ifconfig(8) utility now supports these flags.
499 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
500 mount(8) and -a option for df(1) to see them.
503 The old tunable hw.bus.devctl_disable has been superseded by
504 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
505 replaced by hw.bus.devctl_queue=0. The default for this new tunable
509 Remove the option STOP_NMI. The default action is now to use NMI only
510 for KDB via the newly introduced function stop_cpus_hard() and
511 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
514 The stable/8 branch created in subversion. This corresponds to the
515 RELENG_8 branch in CVS.
518 Bump the shared library version numbers for all libraries that do not
519 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
520 __FreeBSD_version to 800105.
523 Due to changes in the implementation of virtual network stack support,
524 all network-related kernel modules must be recompiled. As this change
525 breaks the ABI, bump __FreeBSD_version to 800104.
528 The TOE interface to the TCP syncache has been modified to remove
529 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
530 The cxgb driver is the only TOE consumer affected by this change, and
531 needs to be recompiled along with the kernel. As this change breaks
532 the ABI, bump __FreeBSD_version to 800103.
535 Padding has been added to struct tcpcb, sackhint and tcpstat in
536 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
537 maintaining the ABI. However, this change breaks the ABI, so bump
538 __FreeBSD_version to 800102. User space tools that rely on the size of
539 any of these structs (e.g. sockstat) need to be recompiled.
542 The NFS_LEGACYRPC option has been removed along with the old kernel
543 RPC implementation that this option selected. Kernel configurations
544 may need to be adjusted.
547 The network interface device nodes at /dev/net/<interface> have been
548 removed. All ioctl operations can be performed the normal way using
549 routing sockets. The kqueue functionality can generally be replaced
550 with routing sockets.
553 The documentation from the FreeBSD Documentation Project (Handbook,
554 FAQ, etc.) is now installed via packages by sysinstall(8) and under
555 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
558 The ABI of various structures related to the SYSV IPC API have been
559 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
560 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
564 Layout of struct vnet has changed as routing related variables were
565 moved to their own Vimage module. Modules need to be recompiled. Bump
566 __FreeBSD_version to 800099.
569 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
570 respectively. As long as no more than 16 groups per process are used,
571 no changes should be visible. When more than 16 groups are used, old
572 binaries may fail if they call getgroups() or getgrouplist() with
573 statically sized storage. Recompiling will work around this, but
574 applications should be modified to use dynamically allocated storage
575 for group arrays as POSIX.1-2008 does not cap an implementation's
576 number of supported groups at NGROUPS_MAX+1 as previous versions did.
578 NFS and portalfs mounts may also be affected as the list of groups is
579 truncated to 16. Users of NFS who use more than 16 groups, should
580 take care that negative group permissions are not used on the exported
581 file systems as they will not be reliable unless a GSSAPI based
582 authentication method is used.
585 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
586 option compiles in the support for adaptive spinning for lockmgrs
587 which want to enable it. The lockinit() function now accepts the flag
588 LK_ADAPTIVE in order to make the lock object subject to adaptive
589 spinning when both held in write and read mode.
592 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
593 changed. User applications that use this ioctl need to be rebuilt.
596 The layout of struct thread has changed. Kernel and modules need to
600 The layout of structs ifnet, domain, protosw and vnet_net has changed.
601 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
604 window(1) has been removed from the base system. It can now be
605 installed from ports. The port is called misc/window.
608 The way we are storing and accessing `routing table' entries has
609 changed. Programs reading the FIB, like netstat, need to be
613 A new netisr implementation has been added for FreeBSD 8. Network
614 file system modules, such as igmp, ipdivert, and others, should be
616 Bump __FreeBSD_version to 800096.
619 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
623 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
624 Bump __FreeBSD_version to 800094.
627 Add mnt_xflag field to 'struct mount'. File system modules need to be
629 Bump __FreeBSD_version to 800093.
632 The compiling option ADAPTIVE_SX has been retired while it has been
633 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
634 The KPI for sx_init_flags() changes as accepting flags:
635 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
636 been introduced in order to handle the reversed logic.
637 Bump __FreeBSD_version to 800092.
640 Add support for hierarchical jails. Remove global securelevel.
641 Bump __FreeBSD_version to 800091.
644 The layout of struct vnet_net has changed, therefore modules
646 Bump __FreeBSD_version to 800090.
649 The newly imported zic(8) produces a new format in the output. Please
650 run tzsetup(8) to install the newly created data to /etc/localtime.
653 The sysctl tree for the usb stack has renamed from hw.usb2.* to
654 hw.usb.* and is now consistent again with previous releases.
657 802.11 monitor mode support was revised and driver api's were changed.
658 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
659 of DLT_IEEE802_11. No user-visible data structures were changed but
660 applications that use DLT_IEEE802_11 may require changes.
661 Bump __FreeBSD_version to 800088.
664 The layout of the following structs has changed: sysctl_oid,
665 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
666 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
667 panics may be experienced. World rebuild is required for
668 correctly checking networking state from userland.
669 Bump __FreeBSD_version to 800085.
672 MLDv2 and Source-Specific Multicast (SSM) have been merged
673 to the IPv6 stack. VIMAGE hooks are in but not yet used.
674 The implementation of SSM within FreeBSD's IPv6 stack closely
675 follows the IPv4 implementation.
677 For kernel developers:
679 * The most important changes are that the ip6_output() and
680 ip6_input() paths no longer take the IN6_MULTI_LOCK,
681 and this lock has been downgraded to a non-recursive mutex.
683 * As with the changes to the IPv4 stack to support SSM, filtering
684 of inbound multicast traffic must now be performed by transport
685 protocols within the IPv6 stack. This does not apply to TCP and
686 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
688 * The KPIs used by IPv6 multicast are similar to those used by
689 the IPv4 stack, with the following differences:
690 * im6o_mc_filter() is analogous to imo_multicast_filter().
691 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
692 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
693 * IN6_LOOKUP_MULTI() has been deprecated and removed.
694 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
695 for MLDv1 have an additional 'timer' argument which is used to
696 jitter the initial membership report for the solicited-node
697 multicast membership on-link.
698 * This is not strictly needed for MLDv2, which already jitters
699 its report transmissions. However, the 'timer' argument is
700 preserved in case MLDv1 is active on the interface.
702 * The KAME linked-list based IPv6 membership implementation has
703 been refactored to use a vector similar to that used by the IPv4
705 Code which maintains a list of its own multicast memberships
706 internally, e.g. carp, has been updated to reflect the new
709 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
710 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
711 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
712 implementation constraint which needs to be addressed in HEAD.
714 For application developers:
716 * The changes are broadly similar to those made for the IPv4
719 * The use of IPv4 and IPv6 multicast socket options on the same
720 socket, using mapped addresses, HAS NOT been tested or supported.
722 * There are a number of issues with the implementation of various
723 IPv6 multicast APIs which need to be resolved in the API surface
724 before the implementation is fully compatible with KAME userland
725 use, and these are mostly to do with interface index treatment.
727 * The literature available discusses the use of either the delta / ASM
728 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
729 using setsourcefilter(3)/getsourcefilter(3). For more information
730 please refer to RFC 3768, 'Socket Interface Extensions for
731 Multicast Source Filters'.
733 * Applications which use the published RFC 3678 APIs should be fine.
735 For systems administrators:
737 * The mtest(8) utility has been refactored to support IPv6, in
738 addition to IPv4. Interface addresses are no longer accepted
739 as arguments, their names must be used instead. The utility
740 will map the interface name to its first IPv4 address as
741 returned by getifaddrs(3).
743 * The ifmcstat(8) utility has also been updated to print the MLDv2
744 endpoint state and source filter lists via sysctl(3).
746 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
747 loopback of IPv6 multicast datagrams by default; it defaults to 1
748 to preserve the existing behaviour. Disabling multicast loopback is
749 recommended for optimal system performance.
751 * The IPv6 MROUTING code has been changed to examine this sysctl
752 instead of attempting to perform a group lookup before looping
753 back forwarded datagrams.
755 Bump __FreeBSD_version to 800084.
758 Implement low-level Bluetooth HCI API.
759 Bump __FreeBSD_version to 800083.
762 The layout of struct malloc_type, used by modules to register new
763 memory allocation types, has changed. Most modules will need to
764 be rebuilt or panics may be experienced.
765 Bump __FreeBSD_version to 800081.
768 Anticipate overflowing inp_flags - add inp_flags2.
769 This changes most offsets in inpcb, so checking v4 connection
770 state will require a world rebuild.
771 Bump __FreeBSD_version to 800080.
774 Add an llentry to struct route and struct route_in6. Modules
775 embedding a struct route will need to be recompiled.
776 Bump __FreeBSD_version to 800079.
779 The size of rt_metrics_lite and by extension rtentry has changed.
780 Networking administration apps will need to be recompiled.
781 The route command now supports show as an alias for get, weighting
782 of routes, sticky and nostick flags to alter the behavior of stateful
784 Bump __FreeBSD_version to 800078.
787 Do not use Giant for kbdmux(4) locking. This is wrong and
788 apparently causing more problems than it solves. This will
789 re-open the issue where interrupt handlers may race with
790 kbdmux(4) in polling mode. Typical symptoms include (but
791 not limited to) duplicated and/or missing characters when
792 low level console functions (such as gets) are used while
793 interrupts are enabled (for example geli password prompt,
794 mountroot prompt etc.). Disabling kbdmux(4) may help.
797 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
798 kernel modules referencing any of the above need to be recompiled.
799 Bump __FreeBSD_version to 800075.
802 GEOM_PART has become the default partition slicer for storage devices,
803 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
804 introduces some changes:
806 MSDOS/EBR: the devices created from MSDOS extended partition entries
807 (EBR) can be named differently than with GEOM_MBR and are now symlinks
808 to devices with offset-based names. fstabs may need to be modified.
810 BSD: the "geometry does not match label" warning is harmless in most
811 cases but it points to problems in file system misalignment with
812 disk geometry. The "c" partition is now implicit, covers the whole
813 top-level drive and cannot be (mis)used by users.
815 General: Kernel dumps are now not allowed to be written to devices
816 whose partition types indicate they are meant to be used for file
817 systems (or, in case of MSDOS partitions, as something else than
820 Most of these changes date approximately from 200812.
823 The uscanner(4) driver has been removed from the kernel. This follows
824 Linux removing theirs in 2.6 and making libusb the default interface
828 The multicast forwarding code has been cleaned up. netstat(1)
829 only relies on KVM now for printing bandwidth upcall meters.
830 The IPv4 and IPv6 modules are split into ip_mroute_mod and
831 ip6_mroute_mod respectively. The config(5) options for statically
832 compiling this code remain the same, i.e. 'options MROUTING'.
835 Support for the IFF_NEEDSGIANT network interface flag has been
836 removed, which means that non-MPSAFE network device drivers are no
837 longer supported. In particular, if_ar, if_sr, and network device
838 drivers from the old (legacy) USB stack can no longer be built or
842 POSIX.1 Native Language Support (NLS) has been enabled in libc and
843 a bunch of new language catalog files have also been added.
844 This means that some common libc messages are now localized and
845 they depend on the LC_MESSAGES environmental variable.
848 The k8temp(4) driver has been renamed to amdtemp(4) since
849 support for Family 10 and Family 11 CPU families was added.
852 IGMPv3 and Source-Specific Multicast (SSM) have been merged
853 to the IPv4 stack. VIMAGE hooks are in but not yet used.
855 For kernel developers, the most important changes are that the
856 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
857 and this lock has been downgraded to a non-recursive mutex.
859 Transport protocols (UDP, Raw IP) are now responsible for filtering
860 inbound multicast traffic according to group membership and source
861 filters. The imo_multicast_filter() KPI exists for this purpose.
862 Transports which do not use multicast (SCTP, TCP) already reject
863 multicast by default. Forwarding and receive performance may improve
864 as a mutex acquisition is no longer needed in the ip_input()
865 low-level input path. in_addmulti() and in_delmulti() are shimmed
866 to new KPIs which exist to support SSM in-kernel.
868 For application developers, it is recommended that loopback of
869 multicast datagrams be disabled for best performance, as this
870 will still cause the lock to be taken for each looped-back
871 datagram transmission. The net.inet.ip.mcast.loop sysctl may
872 be tuned to 0 to disable loopback by default; it defaults to 1
873 to preserve the existing behaviour.
875 For systems administrators, to obtain best performance with
876 multicast reception and multiple groups, it is always recommended
877 that a card with a suitably precise hash filter is used. Hash
878 collisions will still result in the lock being taken within the
879 transport protocol input path to check group membership.
881 If deploying FreeBSD in an environment with IGMP snooping switches,
882 it is recommended that the net.inet.igmp.sendlocal sysctl remain
883 enabled; this forces 224.0.0.0/24 group membership to be announced
886 The size of 'struct igmpstat' has changed; netstat needs to be
887 recompiled to reflect this.
888 Bump __FreeBSD_version to 800070.
891 libusb20.so.1 is now installed as libusb.so.1 and the ports system
892 updated to use it. This requires a buildworld/installworld in order to
893 update the library and dependencies (usbconfig, etc). Its advisable to
894 rebuild all ports which uses libusb. More specific directions are given
895 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
896 libusb are no longer required and can be removed.
899 A workaround is committed to allow the creation of System V shared
900 memory segment of size > 2 GB on the 64-bit architectures.
901 Due to a limitation of the existing ABI, the shm_segsz member
902 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
903 wrong for large segments. Note that limits must be explicitly
904 raised to allow such segments to be created.
907 The layout of struct ifnet has changed, requiring a rebuild of all
908 network device driver modules.
911 The /dev handling for the new USB stack has changed, a
912 buildworld/installworld is required for libusb20.
915 The new USB2 stack has now been permanently moved in and all kernel and
916 module names reverted to their previous values (eg, usb, ehci, ohci,
917 ums, ...). The old usb stack can be compiled in by prefixing the name
918 with the letter 'o', the old usb modules have been removed.
919 Updating entry 20090216 for xorg and 20090215 for libmap may still
923 The rc.conf(5) option if_up_delay has been renamed to
924 defaultroute_delay to better reflect its purpose. If you have
925 customized this setting in /etc/rc.conf you need to update it to
929 xorg 7.4 wants to configure its input devices via hald which does not
930 yet work with USB2. If the keyboard/mouse does not work in xorg then
932 Option "AllowEmptyInput" "off"
933 to your ServerLayout section. This will cause X to use the configured
934 kbd and mouse sections from your xorg.conf.
937 The GENERIC kernels for all architectures now default to the new USB2
938 stack. No kernel config options or code have been removed so if a
939 problem arises please report it and optionally revert to the old USB
940 stack. If you are loading USB kernel modules or have a custom kernel
941 that includes GENERIC then ensure that usb names are also changed over,
942 eg uftdi -> usb2_serial_ftdi.
944 Older programs linked against the ports libusb 0.1 need to be
945 redirected to the new stack's libusb20. /etc/libmap.conf can
947 # Map old usb library to new one for usb2 stack
948 libusb-0.1.so.8 libusb20.so.1
951 All USB ethernet devices now attach as interfaces under the name ueN
952 (eg. ue0). This is to provide a predictable name as vendors often
953 change usb chipsets in a product without notice.
956 The ichsmb(4) driver has been changed to require SMBus slave
957 addresses be left-justified (xxxxxxx0b) rather than right-justified.
958 All of the other SMBus controller drivers require left-justified
959 slave addresses, so this change makes all the drivers provide the
963 INET6 statistics (struct ip6stat) was updated.
964 netstat(1) needs to be recompiled.
967 NTFS has been removed from GENERIC kernel on amd64 to match
968 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
969 will load ntfs.ko module automatically when NTFS support is
970 actually needed, unless ntfs.ko is not installed or security
971 level prohibits loading kernel modules. If either is the case,
972 "options NTFS" has to be added into kernel config.
975 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
976 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
977 800061. User space tools that rely on the size of struct tcpcb in
978 tcp_var.h (e.g. sockstat) need to be recompiled.
981 ng_tty(4) module updated to match the new TTY subsystem.
982 Due to API change, user-level applications must be updated.
983 New API support added to mpd5 CVS and expected to be present
984 in next mpd5.3 release.
987 With __FreeBSD_version 800060 the makefs tool is part of
988 the base system (it was a port).
991 The afdata and ifnet locks have been changed from mutexes to
992 rwlocks, network modules will need to be re-compiled.
995 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
996 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
997 The new code reduced struct rtentry{} by 16 bytes on 32-bit
998 architecture and 40 bytes on 64-bit architecture. The userland
999 applications "arp" and "ndp" have been updated accordingly.
1000 The output from "netstat -r" shows only routing entries and
1001 none of the L2 information.
1004 __FreeBSD_version 800057 marks the switchover from the
1005 binary ath hal to source code. Users must add the line:
1007 options AH_SUPPORT_AR5416
1009 to their kernel config files when specifying:
1013 The ath_hal module no longer exists; the code is now compiled
1014 together with the driver in the ath module. It is now
1015 possible to tailor chip support (i.e. reduce the set of chips
1016 and thereby the code size); consult ath_hal(4) for details.
1019 __FreeBSD_version 800054 adds memory barriers to
1020 <machine/atomic.h>, new interfaces to ifnet to facilitate
1021 multiple hardware transmit queues for cards that support
1022 them, and a lock-less ring-buffer implementation to
1023 enable drivers to more efficiently manage queueing of
1027 A new version of ZFS (version 13) has been merged to -HEAD.
1028 This version has zpool attribute "listsnapshots" off by
1029 default, which means "zfs list" does not show snapshots,
1030 and is the same as Solaris behavior.
1033 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1036 The uhci, ohci, ehci and slhci USB Host controller drivers have
1037 been put into separate modules. If you load the usb module
1038 separately through loader.conf you will need to load the
1039 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1040 controller add the following to loader.conf:
1046 The ABI used by the PMC toolset has changed. Please keep
1047 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1051 atapci kernel module now includes only generic PCI ATA
1052 driver. AHCI driver moved to ataahci kernel module.
1053 All vendor-specific code moved into separate kernel modules:
1054 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1055 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1056 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1057 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1060 The TTY subsystem of the kernel has been replaced by a new
1061 implementation, which provides better scalability and an
1062 improved driver model. Most common drivers have been migrated to
1063 the new TTY subsystem, while others have not. The following
1064 drivers have not yet been ported to the new TTY layer:
1067 cy, digi, rc, rp, sio
1073 ng_h4, ng_tty, ppp, sl, snp
1075 Adding these drivers to your kernel configuration file shall
1076 cause compilation to fail.
1079 ntpd has been upgraded to 4.2.4p5.
1082 OpenSSH has been upgraded to 5.1p1.
1084 For many years, FreeBSD's version of OpenSSH preferred DSA
1085 over RSA for host and user authentication keys. With this
1086 upgrade, we've switched to the vendor's default of RSA over
1087 DSA. This may cause upgraded clients to warn about unknown
1088 host keys even for previously known hosts. Users should
1089 follow the usual procedure for verifying host keys before
1090 accepting the RSA key.
1092 This can be circumvented by setting the "HostKeyAlgorithms"
1093 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1096 Please note that the sequence of keys offered for
1097 authentication has been changed as well. You may want to
1098 specify IdentityFile in a different order to revert this
1102 The sio(4) driver has been removed from the i386 and amd64
1103 kernel configuration files. This means uart(4) is now the
1104 default serial port driver on those platforms as well.
1106 To prevent collisions with the sio(4) driver, the uart(4) driver
1107 uses different names for its device nodes. This means the
1108 onboard serial port will now most likely be called "ttyu0"
1109 instead of "ttyd0". You may need to reconfigure applications to
1110 use the new device names.
1112 When using the serial port as a boot console, be sure to update
1113 /boot/device.hints and /etc/ttys before booting the new kernel.
1114 If you forget to do so, you can still manually specify the hints
1115 at the loader prompt:
1117 set hint.uart.0.at="isa"
1118 set hint.uart.0.port="0x3F8"
1119 set hint.uart.0.flags="0x10"
1120 set hint.uart.0.irq="4"
1124 The gpt(8) utility has been removed. Use gpart(8) to partition
1128 The version that Linuxulator emulates was changed from 2.4.2
1129 to 2.6.16. If you experience any problems with Linux binaries
1130 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1131 if it fixes the problem contact emulation mailing list.
1134 ISDN4BSD (I4B) was removed from the src tree. You may need to
1135 update a your kernel configuration and remove relevant entries.
1138 I have checked in code to support multiple routing tables.
1139 See the man pages setfib(1) and setfib(2).
1140 This is a hopefully backwards compatible version,
1141 but to make use of it you need to compile your kernel
1142 with options ROUTETABLES=2 (or more up to 16).
1145 The 802.11 wireless support was redone to enable multi-bss
1146 operation on devices that are capable. The underlying device
1147 is no longer used directly but instead wlanX devices are
1148 cloned with ifconfig. This requires changes to rc.conf files.
1149 For example, change:
1150 ifconfig_ath0="WPA DHCP"
1153 ifconfig_wlan0="WPA DHCP"
1154 see rc.conf(5) for more details. In addition, mergemaster of
1155 /etc/rc.d is highly recommended. Simultaneous update of userland
1156 and kernel wouldn't hurt either.
1158 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1159 modules were merged into the base wlan module. All references
1160 to these modules (e.g. in kernel config files) must be removed.
1163 psm(4) has gained write(2) support in native operation level.
1164 Arbitrary commands can be written to /dev/psm%d and status can
1165 be read back from it. Therefore, an application is responsible
1166 for status validation and error recovery. It is a no-op in
1167 other operation levels.
1170 Support for KSE threading has been removed from the kernel. To
1171 run legacy applications linked against KSE libmap.conf may
1172 be used. The following libmap.conf may be used to ensure
1173 compatibility with any prior release:
1175 libpthread.so.1 libthr.so.1
1176 libpthread.so.2 libthr.so.2
1177 libkse.so.3 libthr.so.3
1180 The layout of struct vmspace has changed. This affects libkvm
1181 and any executables that link against libkvm and use the
1182 kvm_getprocs() function. In particular, but not exclusively,
1183 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1184 The effects are minimal, but it's advisable to upgrade world
1188 The latest em driver no longer has support in it for the
1189 82575 adapter, this is now moved to the igb driver. The
1190 split was done to make new features that are incompatible
1191 with older hardware easier to do.
1194 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1195 likewise the kernel option is now GEOM_LINUX_LVM.
1198 The default NFS mount mode has changed from UDP to TCP for
1199 increased reliability. If you rely on (insecurely) NFS
1200 mounting across a firewall you may need to update your
1204 Belatedly note the addition of m_collapse for compacting
1208 The fts(3) structures have been changed to use adequate
1209 integer types for their members and so to be able to cope
1210 with huge file trees. The old fts(3) ABI is preserved
1211 through symbol versioning in libc, so third-party binaries
1212 using fts(3) should still work, although they will not take
1213 advantage of the extended types. At the same time, some
1214 third-party software might fail to build after this change
1215 due to unportable assumptions made in its source code about
1216 fts(3) structure members. Such software should be fixed
1217 by its vendor or, in the worst case, in the ports tree.
1218 FreeBSD_version 800015 marks this change for the unlikely
1219 case that a portable fix is impossible.
1222 To upgrade to -current after this date, you must be running
1223 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1224 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1227 The ADAPTIVE_GIANT kernel option has been retired because its
1228 functionality is the default now.
1231 The AT keyboard emulation of sunkbd(4) has been turned on
1232 by default. In order to make the special symbols of the Sun
1233 keyboards driven by sunkbd(4) work under X these now have
1234 to be configured the same way as Sun USB keyboards driven
1235 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1237 Option "XkbLayout" "us"
1238 Option "XkbRules" "xorg"
1239 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1242 It has been decided that it is desirable to provide ABI
1243 backwards compatibility to the FreeBSD 4/5/6 versions of the
1244 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1245 broken with the introduction of PCI domain support (see the
1246 20070930 entry). Unfortunately, this required the ABI of
1247 PCIOCGETCONF to be broken again in order to be able to
1248 provide backwards compatibility to the old version of that
1249 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1250 again. As for prominent ports this affects neither pciutils
1251 nor xorg-server this time, the hal port needs to be rebuilt
1255 The misnamed kthread_create() and friends have been renamed
1256 to kproc_create() etc. Many of the callers already
1257 used kproc_start()..
1258 I will return kthread_create() and friends in a while
1259 with implementations that actually create threads, not procs.
1260 Renaming corresponds with version 800002.
1269 Avoid using make -j when upgrading. While generally safe, there are
1270 sometimes problems using -j to upgrade. If your upgrade fails with
1271 -j, please try again without -j. From time to time in the past there
1272 have been problems using -j with buildworld and/or installworld. This
1273 is especially true when upgrading between "distant" versions (eg one
1274 that cross a major release boundary or several minor releases, or when
1275 several months have passed on the -current branch).
1277 Sometimes, obscure build problems are the result of environment
1278 poisoning. This can happen because the make utility reads its
1279 environment when searching for values for global variables. To run
1280 your build attempts in an "environmental clean room", prefix all make
1281 commands with 'env -i '. See the env(1) manual page for more details.
1283 When upgrading from one major version to another it is generally best
1284 to upgrade to the latest code in the currently installed branch first,
1285 then do an upgrade to the new branch. This is the best-tested upgrade
1286 path, and has the highest probability of being successful. Please try
1287 this approach before reporting problems with a major version upgrade.
1291 When upgrading the boot ZFS pool to a new version, always follow
1294 1.) recompile and reinstall the ZFS boot loader and boot block
1295 (this is part of "make buildworld" and "make installworld")
1297 2.) update the ZFS boot block on your boot drive
1299 The following example updates the ZFS boot block on the first
1300 partition (freebsd-boot) of a GPT partitioned drive ad0:
1301 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1303 Non-boot pools do not need these updates.
1307 If you are updating from a prior version of FreeBSD (even one just
1308 a few days old), you should follow this procedure. It is the most
1309 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1311 make kernel-toolchain
1312 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1313 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1315 To test a kernel once
1316 ---------------------
1317 If you just want to boot a kernel once (because you are not sure
1318 if it works, or if you want to boot a known bad kernel to provide
1319 debugging information) run
1320 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1321 nextboot -k testkernel
1323 To just build a kernel when you know that it won't mess you up
1324 --------------------------------------------------------------
1325 This assumes you are already running a CURRENT system. Replace
1326 ${arch} with the architecture of your machine (e.g. "i386",
1327 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1329 cd src/sys/${arch}/conf
1330 config KERNEL_NAME_HERE
1331 cd ../compile/KERNEL_NAME_HERE
1336 If this fails, go to the "To build a kernel" section.
1338 To rebuild everything and install it on the current system.
1339 -----------------------------------------------------------
1340 # Note: sometimes if you are running current you gotta do more than
1341 # is listed here if you are upgrading from a really old current.
1343 <make sure you have good level 0 dumps>
1345 make kernel KERNCONF=YOUR_KERNEL_HERE
1347 <reboot in single user> [3]
1355 To cross-install current onto a separate partition
1356 --------------------------------------------------
1357 # In this approach we use a separate partition to hold
1358 # current's root, 'usr', and 'var' directories. A partition
1359 # holding "/", "/usr" and "/var" should be about 2GB in
1362 <make sure you have good level 0 dumps>
1365 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1366 <maybe newfs current's root partition>
1367 <mount current's root partition on directory ${CURRENT_ROOT}>
1368 make installworld DESTDIR=${CURRENT_ROOT}
1369 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1370 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1371 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1372 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1373 <reboot into current>
1374 <do a "native" rebuild/install as described in the previous section>
1375 <maybe install compatibility libraries from ports/misc/compat*>
1379 To upgrade in-place from 8.x-stable to current
1380 ----------------------------------------------
1381 <make sure you have good level 0 dumps>
1383 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1385 <reboot in single user> [3]
1392 Make sure that you've read the UPDATING file to understand the
1393 tweaks to various things you need. At this point in the life
1394 cycle of current, things change often and you are on your own
1395 to cope. The defaults can also change, so please read ALL of
1396 the UPDATING entries.
1398 Also, if you are tracking -current, you must be subscribed to
1399 freebsd-current@freebsd.org. Make sure that before you update
1400 your sources that you have read and understood all the recent
1401 messages there. If in doubt, please track -stable which has
1402 much fewer pitfalls.
1404 [1] If you have third party modules, such as vmware, you
1405 should disable them at this point so they don't crash your
1408 [3] From the bootblocks, boot -s, and then do
1413 adjkerntz -i # if CMOS is wall time
1414 Also, when doing a major release upgrade, it is required that
1415 you boot into single user mode to do the installworld.
1417 [4] Note: This step is non-optional. Failure to do this step
1418 can result in a significant reduction in the functionality of the
1419 system. Attempting to do it by hand is not recommended and those
1420 that pursue this avenue should read this file carefully, as well
1421 as the archives of freebsd-current and freebsd-hackers mailing lists
1422 for potential gotchas. The -U option is also useful to consider.
1423 See mergemaster(8) for more information.
1425 [5] Usually this step is a noop. However, from time to time
1426 you may need to do this if you get unknown user in the following
1427 step. It never hurts to do it all the time. You may need to
1428 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1429 install) after the buildworld before this step if you last updated
1430 from current before 20020224 or from -stable before 20020408.
1432 [6] This only deletes old files and directories. Old libraries
1433 can be deleted by "make delete-old-libs", but you have to make
1434 sure that no program is using those libraries anymore.
1436 [8] In order to have a kernel that can run the 4.x binaries needed to
1437 do an installworld, you must include the COMPAT_FREEBSD4 option in
1438 your kernel. Failure to do so may leave you with a system that is
1439 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1440 required to run the 5.x binaries on more recent kernels. And so on
1441 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1443 Make sure that you merge any new devices from GENERIC since the
1444 last time you updated your kernel config file.
1446 [9] When checking out sources, you must include the -P flag to have
1447 cvs prune empty directories.
1449 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1450 "?=" instead of the "=" assignment operator, so that buildworld can
1451 override the CPUTYPE if it needs to.
1453 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1454 not on the command line, or in /etc/make.conf. buildworld will
1455 warn if it is improperly defined.
1458 This file contains a list, in reverse chronological order, of major
1459 breakages in tracking -current. Not all things will be listed here,
1460 and it only starts on October 16, 2004. Updating files can found in
1461 previous releases if your system is older than this.
1463 Copyright information:
1465 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1467 Redistribution, publication, translation and use, with or without
1468 modification, in full or in part, in any form or format of this
1469 document are permitted without further permission from the author.
1471 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1472 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1473 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1474 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1475 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1476 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1477 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1478 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1479 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1480 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1481 POSSIBILITY OF SUCH DAMAGE.
1483 Contact Warner Losh if you have any questions about your use of