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.
13 This commit modifies vfs_register() so that it uses a hash
14 calculation to set vfc_typenum, which is enabled by default.
15 The first time a system is booted after this change, the
16 vfc_typenum values will change for all file systems. The
17 main effect of this is a change to the NFS server file handles
18 for file systems that use vfc_typenum in their fsid, such as ZFS.
19 It will, however, prevent vfc_typenum from changing when file
20 systems are loaded in a different order for subsequent reboots.
21 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
22 until you are ready to remount all NFS clients after a reboot.
25 Bump the shared library version numbers for libraries that
26 do not use symbol versioning, have changed the ABI compared
27 to stable/8 and which shared library version was not bumped.
28 Done as part of 9.0-RELEASE cycle.
31 During the merge of Capsicum features, the fget(9) KPI was modified.
32 This may require the rebuilding of out-of-tree device drivers --
33 issues have been reported specifically with the nVidia device driver.
34 __FreeBSD_version is bumped to 900041.
36 Also, there is a period between 20110811 and 20110814 where the
37 special devices /dev/{stdin,stdout,stderr} did not work correctly.
38 Building world from a kernel during that window may not work.
41 The packet filter (pf) code has been updated to OpenBSD 4.5.
42 You need to update userland tools to be in sync with kernel.
43 This update breaks backward compatibility with earlier pfsync(4)
44 versions. Care must be taken when updating redundant firewall setups.
47 The following sysctls and tunables are retired on x86 platforms:
49 machdep.hlt_logical_cpus
50 The following sysctl is retired:
51 machdep.hyperthreading_allowed
52 The sysctls were supposed to provide a way to dynamically offline and
53 online selected CPUs on x86 platforms, but the implementation has not
54 been reliable especially with SCHED_ULE scheduler.
55 machdep.hyperthreading_allowed tunable is still available to ignore
56 hyperthreading CPUs at OS level.
57 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
58 where X is an APIC ID of a CPU. Be advised, though, that disabling
59 CPUs in non-uniform fashion will result in non-uniform topology and
60 may lead to sub-optimal system performance with SCHED_ULE, which is
64 cpumask_t type is retired and cpuset_t is used in order to describe
68 Changes to ifconfig(8) for dynamic address family detection mandate
69 that you are running a kernel of 20110525 or later. Make sure to
70 follow the update procedure to boot a new kernel before installing
74 Support for sun4v architecture is officially dropped
77 Several KPI breaking changes have been committed to the mii(4) layer,
78 the PHY drivers and consequently some Ethernet drivers using mii(4).
79 This means that miibus.ko and the modules of the affected Ethernet
80 drivers need to be recompiled.
82 Note to kernel developers: Given that the OUI bit reversion problem
83 was fixed as part of these changes all mii(4) commits related to OUIs,
84 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
85 handling, no longer can be merged verbatim to stable/8 and previous
89 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
90 into their kernel configurations along with 'device pci'.
93 The default NFS client is now the new NFS client, so fstype "newnfs"
94 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
95 Although mounts via fstype "nfs" will usually work without userland
96 changes, it is recommended that the mount(8) and mount_nfs(8)
97 commands be rebuilt from sources and that a link to mount_nfs called
98 mount_oldnfs be created. The new client is compiled into the
99 kernel with "options NFSCL" and this is needed for diskless root
100 file systems. The GENERIC kernel configs have been changed to use
101 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
102 To use the regular/old client, you can "mount -t oldnfs ...". For
103 a diskless root file system, you must also include a line like:
105 vfs.root.mountfrom="oldnfs:"
107 in the boot/loader.conf on the root fs on the NFS server to make
108 a diskless root fs use the old client.
111 The GENERIC kernels for all architectures now default to the new
112 CAM-based ATA stack. It means that all legacy ATA drivers were
113 removed and replaced by respective CAM drivers. If you are using
114 ATA device names in /etc/fstab or other places, make sure to update
115 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
116 where 'Y's are the sequential numbers starting from zero for each type
117 in order of detection, unless configured otherwise with tunables,
118 see cam(4)). There will be symbolic links created in /dev/ to map
119 old adX devices to the respective adaY. They should provide basic
120 compatibility for file systems mounting in most cases, but they do
121 not support old user-level APIs and do not have respective providers
122 in GEOM. Consider using updated management tools with new device names.
124 It is possible to load devices ahci, ata, siis and mvs as modules,
125 but option ATA_CAM should remain in kernel configuration to make ata
126 module work as CAM driver supporting legacy ATA controllers. Device ata
127 still can be used in modular fashion (atacore + ...). Modules atadisk
128 and atapi* are not used and won't affect operation in ATA_CAM mode.
129 Note that to use CAM-based ATA kernel should include CAM devices
130 scbus, pass, da (or explicitly ada), cd and optionally others. All of
131 them are parts of the cam module.
133 ataraid(4) functionality is now supported by the RAID GEOM class.
134 To use it you can load geom_raid kernel module and use graid(8) tool
135 for management. Instead of /dev/arX device names, use /dev/raid/rX.
137 No kernel config options or code have been removed, so if a problem
138 arises, please report it and optionally revert to the old ATA stack.
139 In order to do it you can remove from the kernel config:
144 , and instead add back:
145 device atadisk # ATA disk drives
146 device ataraid # ATA RAID drives
147 device atapicd # ATAPI CDROM drives
148 device atapifd # ATAPI floppy drives
149 device atapist # ATAPI tape drives
152 The default NFS server has been changed to the new server, which
153 was referred to as the experimental server. If you need to switch
154 back to the old NFS server, you must now put the "-o" option on
155 both the mountd and nfsd commands. This can be done using the
156 mountd_flags and nfs_server_flags rc.conf variables until an
157 update to the rc scripts is committed, which is coming soon.
160 The GNU Objective-C runtime library (libobjc), and other Objective-C
161 related components have been removed from the base system. If you
162 require an Objective-C library, please use one of the available ports.
165 ath(4) has been split into bus- and device- modules. if_ath contains
166 the HAL, the TX rate control and the network device code. if_ath_pci
167 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
168 contains the AHB glue. Users need to load both if_ath_pci and if_ath
169 in order to use ath on everything else.
171 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
172 need to load if_ath and if_ath_pci for ath(4) operation.
175 As part of the replacement of sysinstall, the process of building
176 release media has changed significantly. For details, please re-read
177 release(7), which has been updated to reflect the new build process.
180 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
181 is the last available version under GPLv2. It brings a number of new
182 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
183 4.1 and SSE 4.2), better support for powerpc64, a number of new
184 directives, and lots of other small improvements. See the ChangeLog
185 file in contrib/binutils for the full details.
188 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
189 compliant, and will now use half of hash for authentication.
190 This will break interoperability with all stacks (including all
191 actual FreeBSD versions) who implement
192 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
194 The only workaround with such peers is to use another HMAC
195 algorithm for IPsec ("phase 2") authentication.
198 Remove the uio_yield prototype and symbol. This function has
199 been misnamed since it was introduced and should not be
200 globally exposed with this name. The equivalent functionality
201 is now available using kern_yield(curthread->td_user_pri).
202 The function remains undocumented.
205 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
206 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
207 for scalar sysctls is defined but disabled. Code that needs
208 UQUAD to pass the type checking that must compile on older
209 systems where the define is not present can check against
210 __FreeBSD_version >= 900030.
212 The system dialog(1) has been replaced with a new version previously
213 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
214 with the previous version, but the libdialog associated with it has
215 a largely incompatible API. As such, the original version of libdialog
216 will be kept temporarily as libodialog, until its base system consumers
217 are replaced or updated. Bump __FreeBSD_version to 900030.
220 If you are trying to run make universe on a -stable system, and you get
221 the following warning:
222 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
223 unknown. config(8) likely too old."
224 or something similar to it, then you must upgrade your -stable system
225 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
226 stable/8) or build the config from the latest stable/8 branch and
227 install it on your system.
229 Prior to this date, building a current universe on 8-stable system from
230 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
231 error in the first kernel build phase. A new config on those old
232 systems will fix that problem for older versions of -current.
235 The TCP stack has been modified to allow Khelp modules to interact with
236 it via helper hook points and store per-connection data in the TCP
237 control block. Bump __FreeBSD_version to 900029. User space tools that
238 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
242 Generic IEEE 802.3 annex 31B full duplex flow control support has been
243 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
244 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
245 to take advantage of it instead of using custom implementations. This
246 means that these drivers now no longer unconditionally advertise
247 support for flow control but only do so if flow control is a selected
248 media option. This was implemented in the generic support that way in
249 order to allow flow control to be switched on and off via ifconfig(8)
250 with the PHY specific default to typically off in order to protect
251 from unwanted effects. Consequently, if you used flow control with
252 one of the above mentioned drivers you now need to explicitly enable
254 ifconfig bge0 media auto mediaopt flowcontrol
256 Along with the above mentioned changes generic support for setting
257 1000baseT master mode also has been added and brgphy(4), ciphy(4),
258 e1000phy(4) as well as ip1000phy(4) have been converted to take
259 advantage of it. This means that these drivers now no longer take the
260 link0 parameter for selecting master mode but the master media option
261 has to be used instead, for example like in the following:
262 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
264 Selection of master mode now is also available with all other PHY
265 drivers supporting 1000baseT.
268 The TCP stack has received a significant update to add support for
269 modularised congestion control and generally improve the clarity of
270 congestion control decisions. Bump __FreeBSD_version to 900025. User
271 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
272 sockstat) need to be recompiled.
275 The man(1) utility has been replaced by a new version that no longer
276 uses /etc/manpath.config. Please consult man.conf(5) for how to
277 migrate local entries to the new format.
280 The copyright strings printed by login(1) and sshd(8) at the time of a
281 new connection have been removed to follow other operating systems and
285 A workaround for a fixed ld bug has been removed in kernel code,
286 so make sure that your system ld is built from sources after
287 revision 210245 from 2010-07-19 (r211583 if building head kernel
288 on stable/8, r211584 for stable/7; both from 2010-08-21).
289 A symptom of incorrect ld version is different addresses for
290 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
293 The $ipv6_prefer variable in rc.conf(5) has been split into
294 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
296 The $ip6addrctl_policy is a variable to choose a pre-defined
297 address selection policy set by ip6addrctl(8). A value
298 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
301 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
302 flag (see an entry of 20090926) is set on an interface with no
303 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
304 security reason. If you want IPv6 link-local address on all
305 interfaces by default, set this to "YES".
307 The old ipv6_prefer="YES" is equivalent to
308 ipv6_activate_all_interfaces="YES" and
309 ip6addrctl_policy="ipv6_prefer".
312 DTrace has grown support for userland tracing. Due to this, DTrace is
313 now i386 and amd64 only.
314 dtruss(1) is now installed by default on those systems and a new
315 kernel module is needed for userland tracing: fasttrap.
316 No changes to your kernel config file are necessary to enable
317 userland tracing, but you might consider adding 'STRIP=' and
318 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
319 to have informative userland stack traces in DTrace (ustack).
322 The acpi_aiboost(4) driver has been removed in favor of the new
323 aibs(4) driver. You should update your kernel configuration file.
326 BSD grep has been imported to the base system and it is built by
327 default. It is completely BSD licensed, highly GNU-compatible, uses
328 less memory than its GNU counterpart and has a small codebase.
329 However, it is slower than its GNU counterpart, which is mostly
330 noticeable for larger searches, for smaller ones it is measurable
331 but not significant. The reason is complex, the most important factor
332 is that we lack a modern and efficient regex library and GNU
333 overcomes this by optimizing the searches internally. Future work
334 on improving the regex performance is planned, for the meantime,
335 users that need better performance, can build GNU grep instead by
336 setting the WITH_GNU_GREP knob.
339 Due to the import of powerpc64 support, all existing powerpc kernel
340 configuration files must be updated with a machine directive like this:
341 machine powerpc powerpc
343 In addition, an updated config(8) is required to build powerpc kernels
347 A new version of ZFS (version 15) has been merged to -HEAD.
348 This version uses a python library for the following subcommands:
349 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
350 For full functionality of these commands the following port must
351 be installed: sysutils/py-zfs
354 'vm_page's are now hashed by physical address to an array of mutexes.
355 Currently this is only used to serialize access to hold_count. Over
356 time the page queue mutex will be peeled away. This changes the size
357 of pmap on every architecture. And requires all callers of vm_page_hold
358 and vm_page_unhold to be updated.
361 WITH_CTF can now be specified in src.conf (not recommended, there
362 are some problems with static executables), make.conf (would also
363 affect ports which do not use GNU make and do not override the
364 compile targets) or in the kernel config (via "makeoptions
366 When WITH_CTF was specified there before this was silently ignored,
367 so make sure that WITH_CTF is not used in places which could lead
368 to unwanted behavior.
371 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
372 to allow 32-bit compatibility on non-x86 platforms. All kernel
373 configurations on amd64 and ia64 platforms using these options must
374 be modified accordingly.
377 The utmp user accounting database has been replaced with utmpx,
378 the user accounting interface standardized by POSIX.
379 Unfortunately the semantics of utmp and utmpx don't match,
380 making it practically impossible to support both interfaces.
381 The user accounting database is used by tools like finger(1),
382 last(1), talk(1), w(1) and ac(8).
384 All applications in the base system use utmpx. This means only
385 local binaries (e.g. from the ports tree) may still use these
386 utmp database files. These applications must be rebuilt to make
389 After the system has been upgraded, it is safe to remove the old
390 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
391 assuming their contents is of no importance anymore. Old wtmp
392 databases can only be used by last(1) and ac(8) after they have
393 been converted to the new format using wtmpcvt(1).
396 Introduce the kernel thread "deadlock resolver" (which can be enabled
397 via the DEADLKRES option, see NOTES for more details) and the
398 sleepq_type() function for sleepqueues.
401 The rc.firewall and rc.firewall6 were unified, and
402 rc.firewall6 and rc.d/ip6fw were removed.
403 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
404 variables are obsoleted. Instead, the following new rc
405 variables are added to rc.d/ipfw:
407 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
408 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
409 firewall_simple_onet_ipv6, firewall_trusted_ipv6
411 The meanings correspond to the relevant IPv4 variables.
417 The default terminal emulation for syscons(4) has been changed
418 from cons25 to xterm on all platforms except pc98. This means
419 that the /etc/ttys file needs to be updated to ensure correct
420 operation of applications on the console.
422 The terminal emulation style can be toggled per window by using
423 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
424 options can be used to change the compile-time default back to
427 To prevent graphical artifacts, make sure the TERM environment
428 variable is set to match the terminal emulation that is being
429 performed by syscons(4).
432 The layout of the structure ieee80211req_scan_result has changed.
433 Applications that require wireless scan results (e.g. ifconfig(8))
434 from net80211 need to be recompiled.
436 Applications such as wpa_supplicant(8) may require a full world
437 build without using NO_CLEAN in order to get synchronized with the
441 The iwn(4) driver has been updated to support the 5000 and 5150 series.
442 There's one kernel module for each firmware. Adding "device iwnfw"
443 to the kernel configuration file means including all three firmware
444 images inside the kernel. If you want to include just the one for
445 your wireless card, use the the devices iwn4965fw, iwn5000fw or
449 The rc.d/network_ipv6, IPv6 configuration script has been integrated
450 into rc.d/netif. The changes are the following:
452 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
453 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
454 Note that both variables need the "inet6" keyword at the head.
456 Do not set $ipv6_network_interfaces manually if you do not
457 understand what you are doing. It is not needed in most cases.
459 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
462 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
463 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
465 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
466 all configured interfaces have "inet6 accept_rtadv" in the
467 $ifconfig_IF_ipv6. These are for backward compatibility.
469 3. A new variable $ipv6_prefer has been added. If NO, IPv6
470 functionality of interfaces with no corresponding
471 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
472 and the default address selection policy of ip6addrctl(8)
473 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
474 Note that if you want to configure IPv6 functionality on the
475 disabled interfaces after boot, first you need to clear the flag by
476 using ifconfig(8) like:
478 ifconfig em0 inet6 -ifdisabled
480 If YES, the default address selection policy is set as
483 The default value of $ipv6_prefer is NO.
485 4. If your system need to receive Router Advertisement messages,
486 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
487 scripts automatically invoke rtsol(8) when the interface becomes
488 UP. The Router Advertisement messages are used for SLAAC
489 (State-Less Address AutoConfiguration).
492 802.11s D3.03 support was committed. This is incompatible with the
493 previous code, which was based on D3.0.
496 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
497 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
498 control whether accepting Router Advertisement messages or not.
499 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
500 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
501 The ifconfig(8) utility now supports these flags.
504 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
505 mount(8) and -a option for df(1) to see them.
508 The old tunable hw.bus.devctl_disable has been superseded by
509 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
510 replaced by hw.bus.devctl_queue=0. The default for this new tunable
514 Remove the option STOP_NMI. The default action is now to use NMI only
515 for KDB via the newly introduced function stop_cpus_hard() and
516 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
519 The stable/8 branch created in subversion. This corresponds to the
520 RELENG_8 branch in CVS.
523 Bump the shared library version numbers for all libraries that do not
524 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
525 __FreeBSD_version to 800105.
528 Due to changes in the implementation of virtual network stack support,
529 all network-related kernel modules must be recompiled. As this change
530 breaks the ABI, bump __FreeBSD_version to 800104.
533 The TOE interface to the TCP syncache has been modified to remove
534 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
535 The cxgb driver is the only TOE consumer affected by this change, and
536 needs to be recompiled along with the kernel. As this change breaks
537 the ABI, bump __FreeBSD_version to 800103.
540 Padding has been added to struct tcpcb, sackhint and tcpstat in
541 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
542 maintaining the ABI. However, this change breaks the ABI, so bump
543 __FreeBSD_version to 800102. User space tools that rely on the size of
544 any of these structs (e.g. sockstat) need to be recompiled.
547 The NFS_LEGACYRPC option has been removed along with the old kernel
548 RPC implementation that this option selected. Kernel configurations
549 may need to be adjusted.
552 The network interface device nodes at /dev/net/<interface> have been
553 removed. All ioctl operations can be performed the normal way using
554 routing sockets. The kqueue functionality can generally be replaced
555 with routing sockets.
558 The documentation from the FreeBSD Documentation Project (Handbook,
559 FAQ, etc.) is now installed via packages by sysinstall(8) and under
560 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
563 The ABI of various structures related to the SYSV IPC API have been
564 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
565 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
569 Layout of struct vnet has changed as routing related variables were
570 moved to their own Vimage module. Modules need to be recompiled. Bump
571 __FreeBSD_version to 800099.
574 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
575 respectively. As long as no more than 16 groups per process are used,
576 no changes should be visible. When more than 16 groups are used, old
577 binaries may fail if they call getgroups() or getgrouplist() with
578 statically sized storage. Recompiling will work around this, but
579 applications should be modified to use dynamically allocated storage
580 for group arrays as POSIX.1-2008 does not cap an implementation's
581 number of supported groups at NGROUPS_MAX+1 as previous versions did.
583 NFS and portalfs mounts may also be affected as the list of groups is
584 truncated to 16. Users of NFS who use more than 16 groups, should
585 take care that negative group permissions are not used on the exported
586 file systems as they will not be reliable unless a GSSAPI based
587 authentication method is used.
590 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
591 option compiles in the support for adaptive spinning for lockmgrs
592 which want to enable it. The lockinit() function now accepts the flag
593 LK_ADAPTIVE in order to make the lock object subject to adaptive
594 spinning when both held in write and read mode.
597 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
598 changed. User applications that use this ioctl need to be rebuilt.
601 The layout of struct thread has changed. Kernel and modules need to
605 The layout of structs ifnet, domain, protosw and vnet_net has changed.
606 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
609 window(1) has been removed from the base system. It can now be
610 installed from ports. The port is called misc/window.
613 The way we are storing and accessing `routing table' entries has
614 changed. Programs reading the FIB, like netstat, need to be
618 A new netisr implementation has been added for FreeBSD 8. Network
619 file system modules, such as igmp, ipdivert, and others, should be
621 Bump __FreeBSD_version to 800096.
624 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
628 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
629 Bump __FreeBSD_version to 800094.
632 Add mnt_xflag field to 'struct mount'. File system modules need to be
634 Bump __FreeBSD_version to 800093.
637 The compiling option ADAPTIVE_SX has been retired while it has been
638 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
639 The KPI for sx_init_flags() changes as accepting flags:
640 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
641 been introduced in order to handle the reversed logic.
642 Bump __FreeBSD_version to 800092.
645 Add support for hierarchical jails. Remove global securelevel.
646 Bump __FreeBSD_version to 800091.
649 The layout of struct vnet_net has changed, therefore modules
651 Bump __FreeBSD_version to 800090.
654 The newly imported zic(8) produces a new format in the output. Please
655 run tzsetup(8) to install the newly created data to /etc/localtime.
658 The sysctl tree for the usb stack has renamed from hw.usb2.* to
659 hw.usb.* and is now consistent again with previous releases.
662 802.11 monitor mode support was revised and driver api's were changed.
663 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
664 of DLT_IEEE802_11. No user-visible data structures were changed but
665 applications that use DLT_IEEE802_11 may require changes.
666 Bump __FreeBSD_version to 800088.
669 The layout of the following structs has changed: sysctl_oid,
670 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
671 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
672 panics may be experienced. World rebuild is required for
673 correctly checking networking state from userland.
674 Bump __FreeBSD_version to 800085.
677 MLDv2 and Source-Specific Multicast (SSM) have been merged
678 to the IPv6 stack. VIMAGE hooks are in but not yet used.
679 The implementation of SSM within FreeBSD's IPv6 stack closely
680 follows the IPv4 implementation.
682 For kernel developers:
684 * The most important changes are that the ip6_output() and
685 ip6_input() paths no longer take the IN6_MULTI_LOCK,
686 and this lock has been downgraded to a non-recursive mutex.
688 * As with the changes to the IPv4 stack to support SSM, filtering
689 of inbound multicast traffic must now be performed by transport
690 protocols within the IPv6 stack. This does not apply to TCP and
691 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
693 * The KPIs used by IPv6 multicast are similar to those used by
694 the IPv4 stack, with the following differences:
695 * im6o_mc_filter() is analogous to imo_multicast_filter().
696 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
697 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
698 * IN6_LOOKUP_MULTI() has been deprecated and removed.
699 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
700 for MLDv1 have an additional 'timer' argument which is used to
701 jitter the initial membership report for the solicited-node
702 multicast membership on-link.
703 * This is not strictly needed for MLDv2, which already jitters
704 its report transmissions. However, the 'timer' argument is
705 preserved in case MLDv1 is active on the interface.
707 * The KAME linked-list based IPv6 membership implementation has
708 been refactored to use a vector similar to that used by the IPv4
710 Code which maintains a list of its own multicast memberships
711 internally, e.g. carp, has been updated to reflect the new
714 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
715 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
716 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
717 implementation constraint which needs to be addressed in HEAD.
719 For application developers:
721 * The changes are broadly similar to those made for the IPv4
724 * The use of IPv4 and IPv6 multicast socket options on the same
725 socket, using mapped addresses, HAS NOT been tested or supported.
727 * There are a number of issues with the implementation of various
728 IPv6 multicast APIs which need to be resolved in the API surface
729 before the implementation is fully compatible with KAME userland
730 use, and these are mostly to do with interface index treatment.
732 * The literature available discusses the use of either the delta / ASM
733 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
734 using setsourcefilter(3)/getsourcefilter(3). For more information
735 please refer to RFC 3768, 'Socket Interface Extensions for
736 Multicast Source Filters'.
738 * Applications which use the published RFC 3678 APIs should be fine.
740 For systems administrators:
742 * The mtest(8) utility has been refactored to support IPv6, in
743 addition to IPv4. Interface addresses are no longer accepted
744 as arguments, their names must be used instead. The utility
745 will map the interface name to its first IPv4 address as
746 returned by getifaddrs(3).
748 * The ifmcstat(8) utility has also been updated to print the MLDv2
749 endpoint state and source filter lists via sysctl(3).
751 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
752 loopback of IPv6 multicast datagrams by default; it defaults to 1
753 to preserve the existing behaviour. Disabling multicast loopback is
754 recommended for optimal system performance.
756 * The IPv6 MROUTING code has been changed to examine this sysctl
757 instead of attempting to perform a group lookup before looping
758 back forwarded datagrams.
760 Bump __FreeBSD_version to 800084.
763 Implement low-level Bluetooth HCI API.
764 Bump __FreeBSD_version to 800083.
767 The layout of struct malloc_type, used by modules to register new
768 memory allocation types, has changed. Most modules will need to
769 be rebuilt or panics may be experienced.
770 Bump __FreeBSD_version to 800081.
773 Anticipate overflowing inp_flags - add inp_flags2.
774 This changes most offsets in inpcb, so checking v4 connection
775 state will require a world rebuild.
776 Bump __FreeBSD_version to 800080.
779 Add an llentry to struct route and struct route_in6. Modules
780 embedding a struct route will need to be recompiled.
781 Bump __FreeBSD_version to 800079.
784 The size of rt_metrics_lite and by extension rtentry has changed.
785 Networking administration apps will need to be recompiled.
786 The route command now supports show as an alias for get, weighting
787 of routes, sticky and nostick flags to alter the behavior of stateful
789 Bump __FreeBSD_version to 800078.
792 Do not use Giant for kbdmux(4) locking. This is wrong and
793 apparently causing more problems than it solves. This will
794 re-open the issue where interrupt handlers may race with
795 kbdmux(4) in polling mode. Typical symptoms include (but
796 not limited to) duplicated and/or missing characters when
797 low level console functions (such as gets) are used while
798 interrupts are enabled (for example geli password prompt,
799 mountroot prompt etc.). Disabling kbdmux(4) may help.
802 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
803 kernel modules referencing any of the above need to be recompiled.
804 Bump __FreeBSD_version to 800075.
807 GEOM_PART has become the default partition slicer for storage devices,
808 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
809 introduces some changes:
811 MSDOS/EBR: the devices created from MSDOS extended partition entries
812 (EBR) can be named differently than with GEOM_MBR and are now symlinks
813 to devices with offset-based names. fstabs may need to be modified.
815 BSD: the "geometry does not match label" warning is harmless in most
816 cases but it points to problems in file system misalignment with
817 disk geometry. The "c" partition is now implicit, covers the whole
818 top-level drive and cannot be (mis)used by users.
820 General: Kernel dumps are now not allowed to be written to devices
821 whose partition types indicate they are meant to be used for file
822 systems (or, in case of MSDOS partitions, as something else than
825 Most of these changes date approximately from 200812.
828 The uscanner(4) driver has been removed from the kernel. This follows
829 Linux removing theirs in 2.6 and making libusb the default interface
833 The multicast forwarding code has been cleaned up. netstat(1)
834 only relies on KVM now for printing bandwidth upcall meters.
835 The IPv4 and IPv6 modules are split into ip_mroute_mod and
836 ip6_mroute_mod respectively. The config(5) options for statically
837 compiling this code remain the same, i.e. 'options MROUTING'.
840 Support for the IFF_NEEDSGIANT network interface flag has been
841 removed, which means that non-MPSAFE network device drivers are no
842 longer supported. In particular, if_ar, if_sr, and network device
843 drivers from the old (legacy) USB stack can no longer be built or
847 POSIX.1 Native Language Support (NLS) has been enabled in libc and
848 a bunch of new language catalog files have also been added.
849 This means that some common libc messages are now localized and
850 they depend on the LC_MESSAGES environmental variable.
853 The k8temp(4) driver has been renamed to amdtemp(4) since
854 support for Family 10 and Family 11 CPU families was added.
857 IGMPv3 and Source-Specific Multicast (SSM) have been merged
858 to the IPv4 stack. VIMAGE hooks are in but not yet used.
860 For kernel developers, the most important changes are that the
861 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
862 and this lock has been downgraded to a non-recursive mutex.
864 Transport protocols (UDP, Raw IP) are now responsible for filtering
865 inbound multicast traffic according to group membership and source
866 filters. The imo_multicast_filter() KPI exists for this purpose.
867 Transports which do not use multicast (SCTP, TCP) already reject
868 multicast by default. Forwarding and receive performance may improve
869 as a mutex acquisition is no longer needed in the ip_input()
870 low-level input path. in_addmulti() and in_delmulti() are shimmed
871 to new KPIs which exist to support SSM in-kernel.
873 For application developers, it is recommended that loopback of
874 multicast datagrams be disabled for best performance, as this
875 will still cause the lock to be taken for each looped-back
876 datagram transmission. The net.inet.ip.mcast.loop sysctl may
877 be tuned to 0 to disable loopback by default; it defaults to 1
878 to preserve the existing behaviour.
880 For systems administrators, to obtain best performance with
881 multicast reception and multiple groups, it is always recommended
882 that a card with a suitably precise hash filter is used. Hash
883 collisions will still result in the lock being taken within the
884 transport protocol input path to check group membership.
886 If deploying FreeBSD in an environment with IGMP snooping switches,
887 it is recommended that the net.inet.igmp.sendlocal sysctl remain
888 enabled; this forces 224.0.0.0/24 group membership to be announced
891 The size of 'struct igmpstat' has changed; netstat needs to be
892 recompiled to reflect this.
893 Bump __FreeBSD_version to 800070.
896 libusb20.so.1 is now installed as libusb.so.1 and the ports system
897 updated to use it. This requires a buildworld/installworld in order to
898 update the library and dependencies (usbconfig, etc). Its advisable to
899 rebuild all ports which uses libusb. More specific directions are given
900 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
901 libusb are no longer required and can be removed.
904 A workaround is committed to allow the creation of System V shared
905 memory segment of size > 2 GB on the 64-bit architectures.
906 Due to a limitation of the existing ABI, the shm_segsz member
907 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
908 wrong for large segments. Note that limits must be explicitly
909 raised to allow such segments to be created.
912 The layout of struct ifnet has changed, requiring a rebuild of all
913 network device driver modules.
916 The /dev handling for the new USB stack has changed, a
917 buildworld/installworld is required for libusb20.
920 The new USB2 stack has now been permanently moved in and all kernel and
921 module names reverted to their previous values (eg, usb, ehci, ohci,
922 ums, ...). The old usb stack can be compiled in by prefixing the name
923 with the letter 'o', the old usb modules have been removed.
924 Updating entry 20090216 for xorg and 20090215 for libmap may still
928 The rc.conf(5) option if_up_delay has been renamed to
929 defaultroute_delay to better reflect its purpose. If you have
930 customized this setting in /etc/rc.conf you need to update it to
934 xorg 7.4 wants to configure its input devices via hald which does not
935 yet work with USB2. If the keyboard/mouse does not work in xorg then
937 Option "AllowEmptyInput" "off"
938 to your ServerLayout section. This will cause X to use the configured
939 kbd and mouse sections from your xorg.conf.
942 The GENERIC kernels for all architectures now default to the new USB2
943 stack. No kernel config options or code have been removed so if a
944 problem arises please report it and optionally revert to the old USB
945 stack. If you are loading USB kernel modules or have a custom kernel
946 that includes GENERIC then ensure that usb names are also changed over,
947 eg uftdi -> usb2_serial_ftdi.
949 Older programs linked against the ports libusb 0.1 need to be
950 redirected to the new stack's libusb20. /etc/libmap.conf can
952 # Map old usb library to new one for usb2 stack
953 libusb-0.1.so.8 libusb20.so.1
956 All USB ethernet devices now attach as interfaces under the name ueN
957 (eg. ue0). This is to provide a predictable name as vendors often
958 change usb chipsets in a product without notice.
961 The ichsmb(4) driver has been changed to require SMBus slave
962 addresses be left-justified (xxxxxxx0b) rather than right-justified.
963 All of the other SMBus controller drivers require left-justified
964 slave addresses, so this change makes all the drivers provide the
968 INET6 statistics (struct ip6stat) was updated.
969 netstat(1) needs to be recompiled.
972 NTFS has been removed from GENERIC kernel on amd64 to match
973 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
974 will load ntfs.ko module automatically when NTFS support is
975 actually needed, unless ntfs.ko is not installed or security
976 level prohibits loading kernel modules. If either is the case,
977 "options NTFS" has to be added into kernel config.
980 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
981 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
982 800061. User space tools that rely on the size of struct tcpcb in
983 tcp_var.h (e.g. sockstat) need to be recompiled.
986 ng_tty(4) module updated to match the new TTY subsystem.
987 Due to API change, user-level applications must be updated.
988 New API support added to mpd5 CVS and expected to be present
989 in next mpd5.3 release.
992 With __FreeBSD_version 800060 the makefs tool is part of
993 the base system (it was a port).
996 The afdata and ifnet locks have been changed from mutexes to
997 rwlocks, network modules will need to be re-compiled.
1000 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1001 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1002 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1003 architecture and 40 bytes on 64-bit architecture. The userland
1004 applications "arp" and "ndp" have been updated accordingly.
1005 The output from "netstat -r" shows only routing entries and
1006 none of the L2 information.
1009 __FreeBSD_version 800057 marks the switchover from the
1010 binary ath hal to source code. Users must add the line:
1012 options AH_SUPPORT_AR5416
1014 to their kernel config files when specifying:
1018 The ath_hal module no longer exists; the code is now compiled
1019 together with the driver in the ath module. It is now
1020 possible to tailor chip support (i.e. reduce the set of chips
1021 and thereby the code size); consult ath_hal(4) for details.
1024 __FreeBSD_version 800054 adds memory barriers to
1025 <machine/atomic.h>, new interfaces to ifnet to facilitate
1026 multiple hardware transmit queues for cards that support
1027 them, and a lock-less ring-buffer implementation to
1028 enable drivers to more efficiently manage queueing of
1032 A new version of ZFS (version 13) has been merged to -HEAD.
1033 This version has zpool attribute "listsnapshots" off by
1034 default, which means "zfs list" does not show snapshots,
1035 and is the same as Solaris behavior.
1038 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1041 The uhci, ohci, ehci and slhci USB Host controller drivers have
1042 been put into separate modules. If you load the usb module
1043 separately through loader.conf you will need to load the
1044 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1045 controller add the following to loader.conf:
1051 The ABI used by the PMC toolset has changed. Please keep
1052 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1056 atapci kernel module now includes only generic PCI ATA
1057 driver. AHCI driver moved to ataahci kernel module.
1058 All vendor-specific code moved into separate kernel modules:
1059 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1060 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1061 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1062 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1065 The TTY subsystem of the kernel has been replaced by a new
1066 implementation, which provides better scalability and an
1067 improved driver model. Most common drivers have been migrated to
1068 the new TTY subsystem, while others have not. The following
1069 drivers have not yet been ported to the new TTY layer:
1072 cy, digi, rc, rp, sio
1078 ng_h4, ng_tty, ppp, sl, snp
1080 Adding these drivers to your kernel configuration file shall
1081 cause compilation to fail.
1084 ntpd has been upgraded to 4.2.4p5.
1087 OpenSSH has been upgraded to 5.1p1.
1089 For many years, FreeBSD's version of OpenSSH preferred DSA
1090 over RSA for host and user authentication keys. With this
1091 upgrade, we've switched to the vendor's default of RSA over
1092 DSA. This may cause upgraded clients to warn about unknown
1093 host keys even for previously known hosts. Users should
1094 follow the usual procedure for verifying host keys before
1095 accepting the RSA key.
1097 This can be circumvented by setting the "HostKeyAlgorithms"
1098 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1101 Please note that the sequence of keys offered for
1102 authentication has been changed as well. You may want to
1103 specify IdentityFile in a different order to revert this
1107 The sio(4) driver has been removed from the i386 and amd64
1108 kernel configuration files. This means uart(4) is now the
1109 default serial port driver on those platforms as well.
1111 To prevent collisions with the sio(4) driver, the uart(4) driver
1112 uses different names for its device nodes. This means the
1113 onboard serial port will now most likely be called "ttyu0"
1114 instead of "ttyd0". You may need to reconfigure applications to
1115 use the new device names.
1117 When using the serial port as a boot console, be sure to update
1118 /boot/device.hints and /etc/ttys before booting the new kernel.
1119 If you forget to do so, you can still manually specify the hints
1120 at the loader prompt:
1122 set hint.uart.0.at="isa"
1123 set hint.uart.0.port="0x3F8"
1124 set hint.uart.0.flags="0x10"
1125 set hint.uart.0.irq="4"
1129 The gpt(8) utility has been removed. Use gpart(8) to partition
1133 The version that Linuxulator emulates was changed from 2.4.2
1134 to 2.6.16. If you experience any problems with Linux binaries
1135 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1136 if it fixes the problem contact emulation mailing list.
1139 ISDN4BSD (I4B) was removed from the src tree. You may need to
1140 update a your kernel configuration and remove relevant entries.
1143 I have checked in code to support multiple routing tables.
1144 See the man pages setfib(1) and setfib(2).
1145 This is a hopefully backwards compatible version,
1146 but to make use of it you need to compile your kernel
1147 with options ROUTETABLES=2 (or more up to 16).
1150 The 802.11 wireless support was redone to enable multi-bss
1151 operation on devices that are capable. The underlying device
1152 is no longer used directly but instead wlanX devices are
1153 cloned with ifconfig. This requires changes to rc.conf files.
1154 For example, change:
1155 ifconfig_ath0="WPA DHCP"
1158 ifconfig_wlan0="WPA DHCP"
1159 see rc.conf(5) for more details. In addition, mergemaster of
1160 /etc/rc.d is highly recommended. Simultaneous update of userland
1161 and kernel wouldn't hurt either.
1163 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1164 modules were merged into the base wlan module. All references
1165 to these modules (e.g. in kernel config files) must be removed.
1168 psm(4) has gained write(2) support in native operation level.
1169 Arbitrary commands can be written to /dev/psm%d and status can
1170 be read back from it. Therefore, an application is responsible
1171 for status validation and error recovery. It is a no-op in
1172 other operation levels.
1175 Support for KSE threading has been removed from the kernel. To
1176 run legacy applications linked against KSE libmap.conf may
1177 be used. The following libmap.conf may be used to ensure
1178 compatibility with any prior release:
1180 libpthread.so.1 libthr.so.1
1181 libpthread.so.2 libthr.so.2
1182 libkse.so.3 libthr.so.3
1185 The layout of struct vmspace has changed. This affects libkvm
1186 and any executables that link against libkvm and use the
1187 kvm_getprocs() function. In particular, but not exclusively,
1188 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1189 The effects are minimal, but it's advisable to upgrade world
1193 The latest em driver no longer has support in it for the
1194 82575 adapter, this is now moved to the igb driver. The
1195 split was done to make new features that are incompatible
1196 with older hardware easier to do.
1199 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1200 likewise the kernel option is now GEOM_LINUX_LVM.
1203 The default NFS mount mode has changed from UDP to TCP for
1204 increased reliability. If you rely on (insecurely) NFS
1205 mounting across a firewall you may need to update your
1209 Belatedly note the addition of m_collapse for compacting
1213 The fts(3) structures have been changed to use adequate
1214 integer types for their members and so to be able to cope
1215 with huge file trees. The old fts(3) ABI is preserved
1216 through symbol versioning in libc, so third-party binaries
1217 using fts(3) should still work, although they will not take
1218 advantage of the extended types. At the same time, some
1219 third-party software might fail to build after this change
1220 due to unportable assumptions made in its source code about
1221 fts(3) structure members. Such software should be fixed
1222 by its vendor or, in the worst case, in the ports tree.
1223 FreeBSD_version 800015 marks this change for the unlikely
1224 case that a portable fix is impossible.
1227 To upgrade to -current after this date, you must be running
1228 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1229 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1232 The ADAPTIVE_GIANT kernel option has been retired because its
1233 functionality is the default now.
1236 The AT keyboard emulation of sunkbd(4) has been turned on
1237 by default. In order to make the special symbols of the Sun
1238 keyboards driven by sunkbd(4) work under X these now have
1239 to be configured the same way as Sun USB keyboards driven
1240 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1242 Option "XkbLayout" "us"
1243 Option "XkbRules" "xorg"
1244 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1247 It has been decided that it is desirable to provide ABI
1248 backwards compatibility to the FreeBSD 4/5/6 versions of the
1249 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1250 broken with the introduction of PCI domain support (see the
1251 20070930 entry). Unfortunately, this required the ABI of
1252 PCIOCGETCONF to be broken again in order to be able to
1253 provide backwards compatibility to the old version of that
1254 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1255 again. As for prominent ports this affects neither pciutils
1256 nor xorg-server this time, the hal port needs to be rebuilt
1260 The misnamed kthread_create() and friends have been renamed
1261 to kproc_create() etc. Many of the callers already
1262 used kproc_start()..
1263 I will return kthread_create() and friends in a while
1264 with implementations that actually create threads, not procs.
1265 Renaming corresponds with version 800002.
1274 Avoid using make -j when upgrading. While generally safe, there are
1275 sometimes problems using -j to upgrade. If your upgrade fails with
1276 -j, please try again without -j. From time to time in the past there
1277 have been problems using -j with buildworld and/or installworld. This
1278 is especially true when upgrading between "distant" versions (eg one
1279 that cross a major release boundary or several minor releases, or when
1280 several months have passed on the -current branch).
1282 Sometimes, obscure build problems are the result of environment
1283 poisoning. This can happen because the make utility reads its
1284 environment when searching for values for global variables. To run
1285 your build attempts in an "environmental clean room", prefix all make
1286 commands with 'env -i '. See the env(1) manual page for more details.
1288 When upgrading from one major version to another it is generally best
1289 to upgrade to the latest code in the currently installed branch first,
1290 then do an upgrade to the new branch. This is the best-tested upgrade
1291 path, and has the highest probability of being successful. Please try
1292 this approach before reporting problems with a major version upgrade.
1296 When upgrading the boot ZFS pool to a new version, always follow
1299 1.) recompile and reinstall the ZFS boot loader and boot block
1300 (this is part of "make buildworld" and "make installworld")
1302 2.) update the ZFS boot block on your boot drive
1304 The following example updates the ZFS boot block on the first
1305 partition (freebsd-boot) of a GPT partitioned drive ad0:
1306 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1308 Non-boot pools do not need these updates.
1312 If you are updating from a prior version of FreeBSD (even one just
1313 a few days old), you should follow this procedure. It is the most
1314 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1316 make kernel-toolchain
1317 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1318 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1320 To test a kernel once
1321 ---------------------
1322 If you just want to boot a kernel once (because you are not sure
1323 if it works, or if you want to boot a known bad kernel to provide
1324 debugging information) run
1325 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1326 nextboot -k testkernel
1328 To just build a kernel when you know that it won't mess you up
1329 --------------------------------------------------------------
1330 This assumes you are already running a CURRENT system. Replace
1331 ${arch} with the architecture of your machine (e.g. "i386",
1332 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1334 cd src/sys/${arch}/conf
1335 config KERNEL_NAME_HERE
1336 cd ../compile/KERNEL_NAME_HERE
1341 If this fails, go to the "To build a kernel" section.
1343 To rebuild everything and install it on the current system.
1344 -----------------------------------------------------------
1345 # Note: sometimes if you are running current you gotta do more than
1346 # is listed here if you are upgrading from a really old current.
1348 <make sure you have good level 0 dumps>
1350 make kernel KERNCONF=YOUR_KERNEL_HERE
1352 <reboot in single user> [3]
1360 To cross-install current onto a separate partition
1361 --------------------------------------------------
1362 # In this approach we use a separate partition to hold
1363 # current's root, 'usr', and 'var' directories. A partition
1364 # holding "/", "/usr" and "/var" should be about 2GB in
1367 <make sure you have good level 0 dumps>
1370 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1371 <maybe newfs current's root partition>
1372 <mount current's root partition on directory ${CURRENT_ROOT}>
1373 make installworld DESTDIR=${CURRENT_ROOT}
1374 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1375 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1376 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1377 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1378 <reboot into current>
1379 <do a "native" rebuild/install as described in the previous section>
1380 <maybe install compatibility libraries from ports/misc/compat*>
1384 To upgrade in-place from 8.x-stable to current
1385 ----------------------------------------------
1386 <make sure you have good level 0 dumps>
1388 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1390 <reboot in single user> [3]
1397 Make sure that you've read the UPDATING file to understand the
1398 tweaks to various things you need. At this point in the life
1399 cycle of current, things change often and you are on your own
1400 to cope. The defaults can also change, so please read ALL of
1401 the UPDATING entries.
1403 Also, if you are tracking -current, you must be subscribed to
1404 freebsd-current@freebsd.org. Make sure that before you update
1405 your sources that you have read and understood all the recent
1406 messages there. If in doubt, please track -stable which has
1407 much fewer pitfalls.
1409 [1] If you have third party modules, such as vmware, you
1410 should disable them at this point so they don't crash your
1413 [3] From the bootblocks, boot -s, and then do
1418 adjkerntz -i # if CMOS is wall time
1419 Also, when doing a major release upgrade, it is required that
1420 you boot into single user mode to do the installworld.
1422 [4] Note: This step is non-optional. Failure to do this step
1423 can result in a significant reduction in the functionality of the
1424 system. Attempting to do it by hand is not recommended and those
1425 that pursue this avenue should read this file carefully, as well
1426 as the archives of freebsd-current and freebsd-hackers mailing lists
1427 for potential gotchas. The -U option is also useful to consider.
1428 See mergemaster(8) for more information.
1430 [5] Usually this step is a noop. However, from time to time
1431 you may need to do this if you get unknown user in the following
1432 step. It never hurts to do it all the time. You may need to
1433 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1434 install) after the buildworld before this step if you last updated
1435 from current before 20020224 or from -stable before 20020408.
1437 [6] This only deletes old files and directories. Old libraries
1438 can be deleted by "make delete-old-libs", but you have to make
1439 sure that no program is using those libraries anymore.
1441 [8] In order to have a kernel that can run the 4.x binaries needed to
1442 do an installworld, you must include the COMPAT_FREEBSD4 option in
1443 your kernel. Failure to do so may leave you with a system that is
1444 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1445 required to run the 5.x binaries on more recent kernels. And so on
1446 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1448 Make sure that you merge any new devices from GENERIC since the
1449 last time you updated your kernel config file.
1451 [9] When checking out sources, you must include the -P flag to have
1452 cvs prune empty directories.
1454 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1455 "?=" instead of the "=" assignment operator, so that buildworld can
1456 override the CPUTYPE if it needs to.
1458 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1459 not on the command line, or in /etc/make.conf. buildworld will
1460 warn if it is improperly defined.
1463 This file contains a list, in reverse chronological order, of major
1464 breakages in tracking -current. It is not guaranteed to be a complete
1465 list of such breakages, and only contains entries since October 10, 2007.
1466 If you need to see UPDATING entries from before that date, you will need
1467 to fetch an UPDATING file from an older FreeBSD release.
1469 Copyright information:
1471 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1473 Redistribution, publication, translation and use, with or without
1474 modification, in full or in part, in any form or format of this
1475 document are permitted without further permission from the author.
1477 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1478 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1479 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1480 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1481 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1482 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1483 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1484 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1485 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1486 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1487 POSSIBILITY OF SUCH DAMAGE.
1489 Contact Warner Losh if you have any questions about your use of