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 ON IA64:
13 For ia64 the INVARIANTS and INVARIANT_SUPPORT kernel options
14 were left in the GENERIC kernel because the kernel does not
15 work properly without them.
18 This commit modifies vfs_register() so that it uses a hash
19 calculation to set vfc_typenum, which is enabled by default.
20 The first time a system is booted after this change, the
21 vfc_typenum values will change for all file systems. The
22 main effect of this is a change to the NFS server file handles
23 for file systems that use vfc_typenum in their fsid, such as ZFS.
24 It will, however, prevent vfc_typenum from changing when file
25 systems are loaded in a different order for subsequent reboots.
26 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
27 until you are ready to remount all NFS clients after a reboot.
30 Bump the shared library version numbers for libraries that
31 do not use symbol versioning, have changed the ABI compared
32 to stable/8 and which shared library version was not bumped.
33 Done as part of 9.0-RELEASE cycle.
36 During the merge of Capsicum features, the fget(9) KPI was modified.
37 This may require the rebuilding of out-of-tree device drivers --
38 issues have been reported specifically with the nVidia device driver.
39 __FreeBSD_version is bumped to 900041.
41 Also, there is a period between 20110811 and 20110814 where the
42 special devices /dev/{stdin,stdout,stderr} did not work correctly.
43 Building world from a kernel during that window may not work.
46 The packet filter (pf) code has been updated to OpenBSD 4.5.
47 You need to update userland tools to be in sync with kernel.
48 This update breaks backward compatibility with earlier pfsync(4)
49 versions. Care must be taken when updating redundant firewall setups.
52 The following sysctls and tunables are retired on x86 platforms:
54 machdep.hlt_logical_cpus
55 The following sysctl is retired:
56 machdep.hyperthreading_allowed
57 The sysctls were supposed to provide a way to dynamically offline and
58 online selected CPUs on x86 platforms, but the implementation has not
59 been reliable especially with SCHED_ULE scheduler.
60 machdep.hyperthreading_allowed tunable is still available to ignore
61 hyperthreading CPUs at OS level.
62 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
63 where X is an APIC ID of a CPU. Be advised, though, that disabling
64 CPUs in non-uniform fashion will result in non-uniform topology and
65 may lead to sub-optimal system performance with SCHED_ULE, which is
69 cpumask_t type is retired and cpuset_t is used in order to describe
73 Changes to ifconfig(8) for dynamic address family detection mandate
74 that you are running a kernel of 20110525 or later. Make sure to
75 follow the update procedure to boot a new kernel before installing
79 Support for sun4v architecture is officially dropped
82 Several KPI breaking changes have been committed to the mii(4) layer,
83 the PHY drivers and consequently some Ethernet drivers using mii(4).
84 This means that miibus.ko and the modules of the affected Ethernet
85 drivers need to be recompiled.
87 Note to kernel developers: Given that the OUI bit reversion problem
88 was fixed as part of these changes all mii(4) commits related to OUIs,
89 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
90 handling, no longer can be merged verbatim to stable/8 and previous
94 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
95 into their kernel configurations along with 'device pci'.
98 The default NFS client is now the new NFS client, so fstype "newnfs"
99 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
100 Although mounts via fstype "nfs" will usually work without userland
101 changes, it is recommended that the mount(8) and mount_nfs(8)
102 commands be rebuilt from sources and that a link to mount_nfs called
103 mount_oldnfs be created. The new client is compiled into the
104 kernel with "options NFSCL" and this is needed for diskless root
105 file systems. The GENERIC kernel configs have been changed to use
106 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
107 To use the regular/old client, you can "mount -t oldnfs ...". For
108 a diskless root file system, you must also include a line like:
110 vfs.root.mountfrom="oldnfs:"
112 in the boot/loader.conf on the root fs on the NFS server to make
113 a diskless root fs use the old client.
116 The GENERIC kernels for all architectures now default to the new
117 CAM-based ATA stack. It means that all legacy ATA drivers were
118 removed and replaced by respective CAM drivers. If you are using
119 ATA device names in /etc/fstab or other places, make sure to update
120 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
121 where 'Y's are the sequential numbers starting from zero for each type
122 in order of detection, unless configured otherwise with tunables,
123 see cam(4)). There will be symbolic links created in /dev/ to map
124 old adX devices to the respective adaY. They should provide basic
125 compatibility for file systems mounting in most cases, but they do
126 not support old user-level APIs and do not have respective providers
127 in GEOM. Consider using updated management tools with new device names.
129 It is possible to load devices ahci, ata, siis and mvs as modules,
130 but option ATA_CAM should remain in kernel configuration to make ata
131 module work as CAM driver supporting legacy ATA controllers. Device ata
132 still can be used in modular fashion (atacore + ...). Modules atadisk
133 and atapi* are not used and won't affect operation in ATA_CAM mode.
134 Note that to use CAM-based ATA kernel should include CAM devices
135 scbus, pass, da (or explicitly ada), cd and optionally others. All of
136 them are parts of the cam module.
138 ataraid(4) functionality is now supported by the RAID GEOM class.
139 To use it you can load geom_raid kernel module and use graid(8) tool
140 for management. Instead of /dev/arX device names, use /dev/raid/rX.
142 No kernel config options or code have been removed, so if a problem
143 arises, please report it and optionally revert to the old ATA stack.
144 In order to do it you can remove from the kernel config:
149 , and instead add back:
150 device atadisk # ATA disk drives
151 device ataraid # ATA RAID drives
152 device atapicd # ATAPI CDROM drives
153 device atapifd # ATAPI floppy drives
154 device atapist # ATAPI tape drives
157 The default NFS server has been changed to the new server, which
158 was referred to as the experimental server. If you need to switch
159 back to the old NFS server, you must now put the "-o" option on
160 both the mountd and nfsd commands. This can be done using the
161 mountd_flags and nfs_server_flags rc.conf variables until an
162 update to the rc scripts is committed, which is coming soon.
165 The GNU Objective-C runtime library (libobjc), and other Objective-C
166 related components have been removed from the base system. If you
167 require an Objective-C library, please use one of the available ports.
170 ath(4) has been split into bus- and device- modules. if_ath contains
171 the HAL, the TX rate control and the network device code. if_ath_pci
172 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
173 contains the AHB glue. Users need to load both if_ath_pci and if_ath
174 in order to use ath on everything else.
176 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
177 need to load if_ath and if_ath_pci for ath(4) operation.
180 As part of the replacement of sysinstall, the process of building
181 release media has changed significantly. For details, please re-read
182 release(7), which has been updated to reflect the new build process.
185 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
186 is the last available version under GPLv2. It brings a number of new
187 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
188 4.1 and SSE 4.2), better support for powerpc64, a number of new
189 directives, and lots of other small improvements. See the ChangeLog
190 file in contrib/binutils for the full details.
193 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
194 compliant, and will now use half of hash for authentication.
195 This will break interoperability with all stacks (including all
196 actual FreeBSD versions) who implement
197 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
199 The only workaround with such peers is to use another HMAC
200 algorithm for IPsec ("phase 2") authentication.
203 Remove the uio_yield prototype and symbol. This function has
204 been misnamed since it was introduced and should not be
205 globally exposed with this name. The equivalent functionality
206 is now available using kern_yield(curthread->td_user_pri).
207 The function remains undocumented.
210 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
211 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
212 for scalar sysctls is defined but disabled. Code that needs
213 UQUAD to pass the type checking that must compile on older
214 systems where the define is not present can check against
215 __FreeBSD_version >= 900030.
217 The system dialog(1) has been replaced with a new version previously
218 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
219 with the previous version, but the libdialog associated with it has
220 a largely incompatible API. As such, the original version of libdialog
221 will be kept temporarily as libodialog, until its base system consumers
222 are replaced or updated. Bump __FreeBSD_version to 900030.
225 If you are trying to run make universe on a -stable system, and you get
226 the following warning:
227 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
228 unknown. config(8) likely too old."
229 or something similar to it, then you must upgrade your -stable system
230 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
231 stable/8) or build the config from the latest stable/8 branch and
232 install it on your system.
234 Prior to this date, building a current universe on 8-stable system from
235 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
236 error in the first kernel build phase. A new config on those old
237 systems will fix that problem for older versions of -current.
240 The TCP stack has been modified to allow Khelp modules to interact with
241 it via helper hook points and store per-connection data in the TCP
242 control block. Bump __FreeBSD_version to 900029. User space tools that
243 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
247 Generic IEEE 802.3 annex 31B full duplex flow control support has been
248 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
249 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
250 to take advantage of it instead of using custom implementations. This
251 means that these drivers now no longer unconditionally advertise
252 support for flow control but only do so if flow control is a selected
253 media option. This was implemented in the generic support that way in
254 order to allow flow control to be switched on and off via ifconfig(8)
255 with the PHY specific default to typically off in order to protect
256 from unwanted effects. Consequently, if you used flow control with
257 one of the above mentioned drivers you now need to explicitly enable
259 ifconfig bge0 media auto mediaopt flowcontrol
261 Along with the above mentioned changes generic support for setting
262 1000baseT master mode also has been added and brgphy(4), ciphy(4),
263 e1000phy(4) as well as ip1000phy(4) have been converted to take
264 advantage of it. This means that these drivers now no longer take the
265 link0 parameter for selecting master mode but the master media option
266 has to be used instead, for example like in the following:
267 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
269 Selection of master mode now is also available with all other PHY
270 drivers supporting 1000baseT.
273 The TCP stack has received a significant update to add support for
274 modularised congestion control and generally improve the clarity of
275 congestion control decisions. Bump __FreeBSD_version to 900025. User
276 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
277 sockstat) need to be recompiled.
280 The man(1) utility has been replaced by a new version that no longer
281 uses /etc/manpath.config. Please consult man.conf(5) for how to
282 migrate local entries to the new format.
285 The copyright strings printed by login(1) and sshd(8) at the time of a
286 new connection have been removed to follow other operating systems and
290 A workaround for a fixed ld bug has been removed in kernel code,
291 so make sure that your system ld is built from sources after
292 revision 210245 from 2010-07-19 (r211583 if building head kernel
293 on stable/8, r211584 for stable/7; both from 2010-08-21).
294 A symptom of incorrect ld version is different addresses for
295 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
298 The $ipv6_prefer variable in rc.conf(5) has been split into
299 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
301 The $ip6addrctl_policy is a variable to choose a pre-defined
302 address selection policy set by ip6addrctl(8). A value
303 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
306 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
307 flag (see an entry of 20090926) is set on an interface with no
308 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
309 security reason. If you want IPv6 link-local address on all
310 interfaces by default, set this to "YES".
312 The old ipv6_prefer="YES" is equivalent to
313 ipv6_activate_all_interfaces="YES" and
314 ip6addrctl_policy="ipv6_prefer".
317 DTrace has grown support for userland tracing. Due to this, DTrace is
318 now i386 and amd64 only.
319 dtruss(1) is now installed by default on those systems and a new
320 kernel module is needed for userland tracing: fasttrap.
321 No changes to your kernel config file are necessary to enable
322 userland tracing, but you might consider adding 'STRIP=' and
323 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
324 to have informative userland stack traces in DTrace (ustack).
327 The acpi_aiboost(4) driver has been removed in favor of the new
328 aibs(4) driver. You should update your kernel configuration file.
331 BSD grep has been imported to the base system and it is built by
332 default. It is completely BSD licensed, highly GNU-compatible, uses
333 less memory than its GNU counterpart and has a small codebase.
334 However, it is slower than its GNU counterpart, which is mostly
335 noticeable for larger searches, for smaller ones it is measurable
336 but not significant. The reason is complex, the most important factor
337 is that we lack a modern and efficient regex library and GNU
338 overcomes this by optimizing the searches internally. Future work
339 on improving the regex performance is planned, for the meantime,
340 users that need better performance, can build GNU grep instead by
341 setting the WITH_GNU_GREP knob.
344 Due to the import of powerpc64 support, all existing powerpc kernel
345 configuration files must be updated with a machine directive like this:
346 machine powerpc powerpc
348 In addition, an updated config(8) is required to build powerpc kernels
352 A new version of ZFS (version 15) has been merged to -HEAD.
353 This version uses a python library for the following subcommands:
354 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
355 For full functionality of these commands the following port must
356 be installed: sysutils/py-zfs
359 'vm_page's are now hashed by physical address to an array of mutexes.
360 Currently this is only used to serialize access to hold_count. Over
361 time the page queue mutex will be peeled away. This changes the size
362 of pmap on every architecture. And requires all callers of vm_page_hold
363 and vm_page_unhold to be updated.
366 WITH_CTF can now be specified in src.conf (not recommended, there
367 are some problems with static executables), make.conf (would also
368 affect ports which do not use GNU make and do not override the
369 compile targets) or in the kernel config (via "makeoptions
371 When WITH_CTF was specified there before this was silently ignored,
372 so make sure that WITH_CTF is not used in places which could lead
373 to unwanted behavior.
376 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
377 to allow 32-bit compatibility on non-x86 platforms. All kernel
378 configurations on amd64 and ia64 platforms using these options must
379 be modified accordingly.
382 The utmp user accounting database has been replaced with utmpx,
383 the user accounting interface standardized by POSIX.
384 Unfortunately the semantics of utmp and utmpx don't match,
385 making it practically impossible to support both interfaces.
386 The user accounting database is used by tools like finger(1),
387 last(1), talk(1), w(1) and ac(8).
389 All applications in the base system use utmpx. This means only
390 local binaries (e.g. from the ports tree) may still use these
391 utmp database files. These applications must be rebuilt to make
394 After the system has been upgraded, it is safe to remove the old
395 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
396 assuming their contents is of no importance anymore. Old wtmp
397 databases can only be used by last(1) and ac(8) after they have
398 been converted to the new format using wtmpcvt(1).
401 Introduce the kernel thread "deadlock resolver" (which can be enabled
402 via the DEADLKRES option, see NOTES for more details) and the
403 sleepq_type() function for sleepqueues.
406 The rc.firewall and rc.firewall6 were unified, and
407 rc.firewall6 and rc.d/ip6fw were removed.
408 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
409 variables are obsoleted. Instead, the following new rc
410 variables are added to rc.d/ipfw:
412 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
413 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
414 firewall_simple_onet_ipv6, firewall_trusted_ipv6
416 The meanings correspond to the relevant IPv4 variables.
422 The default terminal emulation for syscons(4) has been changed
423 from cons25 to xterm on all platforms except pc98. This means
424 that the /etc/ttys file needs to be updated to ensure correct
425 operation of applications on the console.
427 The terminal emulation style can be toggled per window by using
428 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
429 options can be used to change the compile-time default back to
432 To prevent graphical artifacts, make sure the TERM environment
433 variable is set to match the terminal emulation that is being
434 performed by syscons(4).
437 The layout of the structure ieee80211req_scan_result has changed.
438 Applications that require wireless scan results (e.g. ifconfig(8))
439 from net80211 need to be recompiled.
441 Applications such as wpa_supplicant(8) may require a full world
442 build without using NO_CLEAN in order to get synchronized with the
446 The iwn(4) driver has been updated to support the 5000 and 5150 series.
447 There's one kernel module for each firmware. Adding "device iwnfw"
448 to the kernel configuration file means including all three firmware
449 images inside the kernel. If you want to include just the one for
450 your wireless card, use the the devices iwn4965fw, iwn5000fw or
454 The rc.d/network_ipv6, IPv6 configuration script has been integrated
455 into rc.d/netif. The changes are the following:
457 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
458 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
459 Note that both variables need the "inet6" keyword at the head.
461 Do not set $ipv6_network_interfaces manually if you do not
462 understand what you are doing. It is not needed in most cases.
464 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
467 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
468 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
470 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
471 all configured interfaces have "inet6 accept_rtadv" in the
472 $ifconfig_IF_ipv6. These are for backward compatibility.
474 3. A new variable $ipv6_prefer has been added. If NO, IPv6
475 functionality of interfaces with no corresponding
476 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
477 and the default address selection policy of ip6addrctl(8)
478 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
479 Note that if you want to configure IPv6 functionality on the
480 disabled interfaces after boot, first you need to clear the flag by
481 using ifconfig(8) like:
483 ifconfig em0 inet6 -ifdisabled
485 If YES, the default address selection policy is set as
488 The default value of $ipv6_prefer is NO.
490 4. If your system need to receive Router Advertisement messages,
491 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
492 scripts automatically invoke rtsol(8) when the interface becomes
493 UP. The Router Advertisement messages are used for SLAAC
494 (State-Less Address AutoConfiguration).
497 802.11s D3.03 support was committed. This is incompatible with the
498 previous code, which was based on D3.0.
501 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
502 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
503 control whether accepting Router Advertisement messages or not.
504 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
505 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
506 The ifconfig(8) utility now supports these flags.
509 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
510 mount(8) and -a option for df(1) to see them.
513 The old tunable hw.bus.devctl_disable has been superseded by
514 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
515 replaced by hw.bus.devctl_queue=0. The default for this new tunable
519 Remove the option STOP_NMI. The default action is now to use NMI only
520 for KDB via the newly introduced function stop_cpus_hard() and
521 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
524 The stable/8 branch created in subversion. This corresponds to the
525 RELENG_8 branch in CVS.
528 Bump the shared library version numbers for all libraries that do not
529 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
530 __FreeBSD_version to 800105.
533 Due to changes in the implementation of virtual network stack support,
534 all network-related kernel modules must be recompiled. As this change
535 breaks the ABI, bump __FreeBSD_version to 800104.
538 The TOE interface to the TCP syncache has been modified to remove
539 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
540 The cxgb driver is the only TOE consumer affected by this change, and
541 needs to be recompiled along with the kernel. As this change breaks
542 the ABI, bump __FreeBSD_version to 800103.
545 Padding has been added to struct tcpcb, sackhint and tcpstat in
546 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
547 maintaining the ABI. However, this change breaks the ABI, so bump
548 __FreeBSD_version to 800102. User space tools that rely on the size of
549 any of these structs (e.g. sockstat) need to be recompiled.
552 The NFS_LEGACYRPC option has been removed along with the old kernel
553 RPC implementation that this option selected. Kernel configurations
554 may need to be adjusted.
557 The network interface device nodes at /dev/net/<interface> have been
558 removed. All ioctl operations can be performed the normal way using
559 routing sockets. The kqueue functionality can generally be replaced
560 with routing sockets.
563 The documentation from the FreeBSD Documentation Project (Handbook,
564 FAQ, etc.) is now installed via packages by sysinstall(8) and under
565 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
568 The ABI of various structures related to the SYSV IPC API have been
569 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
570 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
574 Layout of struct vnet has changed as routing related variables were
575 moved to their own Vimage module. Modules need to be recompiled. Bump
576 __FreeBSD_version to 800099.
579 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
580 respectively. As long as no more than 16 groups per process are used,
581 no changes should be visible. When more than 16 groups are used, old
582 binaries may fail if they call getgroups() or getgrouplist() with
583 statically sized storage. Recompiling will work around this, but
584 applications should be modified to use dynamically allocated storage
585 for group arrays as POSIX.1-2008 does not cap an implementation's
586 number of supported groups at NGROUPS_MAX+1 as previous versions did.
588 NFS and portalfs mounts may also be affected as the list of groups is
589 truncated to 16. Users of NFS who use more than 16 groups, should
590 take care that negative group permissions are not used on the exported
591 file systems as they will not be reliable unless a GSSAPI based
592 authentication method is used.
595 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
596 option compiles in the support for adaptive spinning for lockmgrs
597 which want to enable it. The lockinit() function now accepts the flag
598 LK_ADAPTIVE in order to make the lock object subject to adaptive
599 spinning when both held in write and read mode.
602 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
603 changed. User applications that use this ioctl need to be rebuilt.
606 The layout of struct thread has changed. Kernel and modules need to
610 The layout of structs ifnet, domain, protosw and vnet_net has changed.
611 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
614 window(1) has been removed from the base system. It can now be
615 installed from ports. The port is called misc/window.
618 The way we are storing and accessing `routing table' entries has
619 changed. Programs reading the FIB, like netstat, need to be
623 A new netisr implementation has been added for FreeBSD 8. Network
624 file system modules, such as igmp, ipdivert, and others, should be
626 Bump __FreeBSD_version to 800096.
629 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
633 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
634 Bump __FreeBSD_version to 800094.
637 Add mnt_xflag field to 'struct mount'. File system modules need to be
639 Bump __FreeBSD_version to 800093.
642 The compiling option ADAPTIVE_SX has been retired while it has been
643 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
644 The KPI for sx_init_flags() changes as accepting flags:
645 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
646 been introduced in order to handle the reversed logic.
647 Bump __FreeBSD_version to 800092.
650 Add support for hierarchical jails. Remove global securelevel.
651 Bump __FreeBSD_version to 800091.
654 The layout of struct vnet_net has changed, therefore modules
656 Bump __FreeBSD_version to 800090.
659 The newly imported zic(8) produces a new format in the output. Please
660 run tzsetup(8) to install the newly created data to /etc/localtime.
663 The sysctl tree for the usb stack has renamed from hw.usb2.* to
664 hw.usb.* and is now consistent again with previous releases.
667 802.11 monitor mode support was revised and driver api's were changed.
668 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
669 of DLT_IEEE802_11. No user-visible data structures were changed but
670 applications that use DLT_IEEE802_11 may require changes.
671 Bump __FreeBSD_version to 800088.
674 The layout of the following structs has changed: sysctl_oid,
675 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
676 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
677 panics may be experienced. World rebuild is required for
678 correctly checking networking state from userland.
679 Bump __FreeBSD_version to 800085.
682 MLDv2 and Source-Specific Multicast (SSM) have been merged
683 to the IPv6 stack. VIMAGE hooks are in but not yet used.
684 The implementation of SSM within FreeBSD's IPv6 stack closely
685 follows the IPv4 implementation.
687 For kernel developers:
689 * The most important changes are that the ip6_output() and
690 ip6_input() paths no longer take the IN6_MULTI_LOCK,
691 and this lock has been downgraded to a non-recursive mutex.
693 * As with the changes to the IPv4 stack to support SSM, filtering
694 of inbound multicast traffic must now be performed by transport
695 protocols within the IPv6 stack. This does not apply to TCP and
696 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
698 * The KPIs used by IPv6 multicast are similar to those used by
699 the IPv4 stack, with the following differences:
700 * im6o_mc_filter() is analogous to imo_multicast_filter().
701 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
702 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
703 * IN6_LOOKUP_MULTI() has been deprecated and removed.
704 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
705 for MLDv1 have an additional 'timer' argument which is used to
706 jitter the initial membership report for the solicited-node
707 multicast membership on-link.
708 * This is not strictly needed for MLDv2, which already jitters
709 its report transmissions. However, the 'timer' argument is
710 preserved in case MLDv1 is active on the interface.
712 * The KAME linked-list based IPv6 membership implementation has
713 been refactored to use a vector similar to that used by the IPv4
715 Code which maintains a list of its own multicast memberships
716 internally, e.g. carp, has been updated to reflect the new
719 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
720 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
721 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
722 implementation constraint which needs to be addressed in HEAD.
724 For application developers:
726 * The changes are broadly similar to those made for the IPv4
729 * The use of IPv4 and IPv6 multicast socket options on the same
730 socket, using mapped addresses, HAS NOT been tested or supported.
732 * There are a number of issues with the implementation of various
733 IPv6 multicast APIs which need to be resolved in the API surface
734 before the implementation is fully compatible with KAME userland
735 use, and these are mostly to do with interface index treatment.
737 * The literature available discusses the use of either the delta / ASM
738 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
739 using setsourcefilter(3)/getsourcefilter(3). For more information
740 please refer to RFC 3768, 'Socket Interface Extensions for
741 Multicast Source Filters'.
743 * Applications which use the published RFC 3678 APIs should be fine.
745 For systems administrators:
747 * The mtest(8) utility has been refactored to support IPv6, in
748 addition to IPv4. Interface addresses are no longer accepted
749 as arguments, their names must be used instead. The utility
750 will map the interface name to its first IPv4 address as
751 returned by getifaddrs(3).
753 * The ifmcstat(8) utility has also been updated to print the MLDv2
754 endpoint state and source filter lists via sysctl(3).
756 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
757 loopback of IPv6 multicast datagrams by default; it defaults to 1
758 to preserve the existing behaviour. Disabling multicast loopback is
759 recommended for optimal system performance.
761 * The IPv6 MROUTING code has been changed to examine this sysctl
762 instead of attempting to perform a group lookup before looping
763 back forwarded datagrams.
765 Bump __FreeBSD_version to 800084.
768 Implement low-level Bluetooth HCI API.
769 Bump __FreeBSD_version to 800083.
772 The layout of struct malloc_type, used by modules to register new
773 memory allocation types, has changed. Most modules will need to
774 be rebuilt or panics may be experienced.
775 Bump __FreeBSD_version to 800081.
778 Anticipate overflowing inp_flags - add inp_flags2.
779 This changes most offsets in inpcb, so checking v4 connection
780 state will require a world rebuild.
781 Bump __FreeBSD_version to 800080.
784 Add an llentry to struct route and struct route_in6. Modules
785 embedding a struct route will need to be recompiled.
786 Bump __FreeBSD_version to 800079.
789 The size of rt_metrics_lite and by extension rtentry has changed.
790 Networking administration apps will need to be recompiled.
791 The route command now supports show as an alias for get, weighting
792 of routes, sticky and nostick flags to alter the behavior of stateful
794 Bump __FreeBSD_version to 800078.
797 Do not use Giant for kbdmux(4) locking. This is wrong and
798 apparently causing more problems than it solves. This will
799 re-open the issue where interrupt handlers may race with
800 kbdmux(4) in polling mode. Typical symptoms include (but
801 not limited to) duplicated and/or missing characters when
802 low level console functions (such as gets) are used while
803 interrupts are enabled (for example geli password prompt,
804 mountroot prompt etc.). Disabling kbdmux(4) may help.
807 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
808 kernel modules referencing any of the above need to be recompiled.
809 Bump __FreeBSD_version to 800075.
812 GEOM_PART has become the default partition slicer for storage devices,
813 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
814 introduces some changes:
816 MSDOS/EBR: the devices created from MSDOS extended partition entries
817 (EBR) can be named differently than with GEOM_MBR and are now symlinks
818 to devices with offset-based names. fstabs may need to be modified.
820 BSD: the "geometry does not match label" warning is harmless in most
821 cases but it points to problems in file system misalignment with
822 disk geometry. The "c" partition is now implicit, covers the whole
823 top-level drive and cannot be (mis)used by users.
825 General: Kernel dumps are now not allowed to be written to devices
826 whose partition types indicate they are meant to be used for file
827 systems (or, in case of MSDOS partitions, as something else than
830 Most of these changes date approximately from 200812.
833 The uscanner(4) driver has been removed from the kernel. This follows
834 Linux removing theirs in 2.6 and making libusb the default interface
838 The multicast forwarding code has been cleaned up. netstat(1)
839 only relies on KVM now for printing bandwidth upcall meters.
840 The IPv4 and IPv6 modules are split into ip_mroute_mod and
841 ip6_mroute_mod respectively. The config(5) options for statically
842 compiling this code remain the same, i.e. 'options MROUTING'.
845 Support for the IFF_NEEDSGIANT network interface flag has been
846 removed, which means that non-MPSAFE network device drivers are no
847 longer supported. In particular, if_ar, if_sr, and network device
848 drivers from the old (legacy) USB stack can no longer be built or
852 POSIX.1 Native Language Support (NLS) has been enabled in libc and
853 a bunch of new language catalog files have also been added.
854 This means that some common libc messages are now localized and
855 they depend on the LC_MESSAGES environmental variable.
858 The k8temp(4) driver has been renamed to amdtemp(4) since
859 support for Family 10 and Family 11 CPU families was added.
862 IGMPv3 and Source-Specific Multicast (SSM) have been merged
863 to the IPv4 stack. VIMAGE hooks are in but not yet used.
865 For kernel developers, the most important changes are that the
866 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
867 and this lock has been downgraded to a non-recursive mutex.
869 Transport protocols (UDP, Raw IP) are now responsible for filtering
870 inbound multicast traffic according to group membership and source
871 filters. The imo_multicast_filter() KPI exists for this purpose.
872 Transports which do not use multicast (SCTP, TCP) already reject
873 multicast by default. Forwarding and receive performance may improve
874 as a mutex acquisition is no longer needed in the ip_input()
875 low-level input path. in_addmulti() and in_delmulti() are shimmed
876 to new KPIs which exist to support SSM in-kernel.
878 For application developers, it is recommended that loopback of
879 multicast datagrams be disabled for best performance, as this
880 will still cause the lock to be taken for each looped-back
881 datagram transmission. The net.inet.ip.mcast.loop sysctl may
882 be tuned to 0 to disable loopback by default; it defaults to 1
883 to preserve the existing behaviour.
885 For systems administrators, to obtain best performance with
886 multicast reception and multiple groups, it is always recommended
887 that a card with a suitably precise hash filter is used. Hash
888 collisions will still result in the lock being taken within the
889 transport protocol input path to check group membership.
891 If deploying FreeBSD in an environment with IGMP snooping switches,
892 it is recommended that the net.inet.igmp.sendlocal sysctl remain
893 enabled; this forces 224.0.0.0/24 group membership to be announced
896 The size of 'struct igmpstat' has changed; netstat needs to be
897 recompiled to reflect this.
898 Bump __FreeBSD_version to 800070.
901 libusb20.so.1 is now installed as libusb.so.1 and the ports system
902 updated to use it. This requires a buildworld/installworld in order to
903 update the library and dependencies (usbconfig, etc). Its advisable to
904 rebuild all ports which uses libusb. More specific directions are given
905 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
906 libusb are no longer required and can be removed.
909 A workaround is committed to allow the creation of System V shared
910 memory segment of size > 2 GB on the 64-bit architectures.
911 Due to a limitation of the existing ABI, the shm_segsz member
912 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
913 wrong for large segments. Note that limits must be explicitly
914 raised to allow such segments to be created.
917 The layout of struct ifnet has changed, requiring a rebuild of all
918 network device driver modules.
921 The /dev handling for the new USB stack has changed, a
922 buildworld/installworld is required for libusb20.
925 The new USB2 stack has now been permanently moved in and all kernel and
926 module names reverted to their previous values (eg, usb, ehci, ohci,
927 ums, ...). The old usb stack can be compiled in by prefixing the name
928 with the letter 'o', the old usb modules have been removed.
929 Updating entry 20090216 for xorg and 20090215 for libmap may still
933 The rc.conf(5) option if_up_delay has been renamed to
934 defaultroute_delay to better reflect its purpose. If you have
935 customized this setting in /etc/rc.conf you need to update it to
939 xorg 7.4 wants to configure its input devices via hald which does not
940 yet work with USB2. If the keyboard/mouse does not work in xorg then
942 Option "AllowEmptyInput" "off"
943 to your ServerLayout section. This will cause X to use the configured
944 kbd and mouse sections from your xorg.conf.
947 The GENERIC kernels for all architectures now default to the new USB2
948 stack. No kernel config options or code have been removed so if a
949 problem arises please report it and optionally revert to the old USB
950 stack. If you are loading USB kernel modules or have a custom kernel
951 that includes GENERIC then ensure that usb names are also changed over,
952 eg uftdi -> usb2_serial_ftdi.
954 Older programs linked against the ports libusb 0.1 need to be
955 redirected to the new stack's libusb20. /etc/libmap.conf can
957 # Map old usb library to new one for usb2 stack
958 libusb-0.1.so.8 libusb20.so.1
961 All USB ethernet devices now attach as interfaces under the name ueN
962 (eg. ue0). This is to provide a predictable name as vendors often
963 change usb chipsets in a product without notice.
966 The ichsmb(4) driver has been changed to require SMBus slave
967 addresses be left-justified (xxxxxxx0b) rather than right-justified.
968 All of the other SMBus controller drivers require left-justified
969 slave addresses, so this change makes all the drivers provide the
973 INET6 statistics (struct ip6stat) was updated.
974 netstat(1) needs to be recompiled.
977 NTFS has been removed from GENERIC kernel on amd64 to match
978 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
979 will load ntfs.ko module automatically when NTFS support is
980 actually needed, unless ntfs.ko is not installed or security
981 level prohibits loading kernel modules. If either is the case,
982 "options NTFS" has to be added into kernel config.
985 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
986 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
987 800061. User space tools that rely on the size of struct tcpcb in
988 tcp_var.h (e.g. sockstat) need to be recompiled.
991 ng_tty(4) module updated to match the new TTY subsystem.
992 Due to API change, user-level applications must be updated.
993 New API support added to mpd5 CVS and expected to be present
994 in next mpd5.3 release.
997 With __FreeBSD_version 800060 the makefs tool is part of
998 the base system (it was a port).
1001 The afdata and ifnet locks have been changed from mutexes to
1002 rwlocks, network modules will need to be re-compiled.
1005 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1006 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1007 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1008 architecture and 40 bytes on 64-bit architecture. The userland
1009 applications "arp" and "ndp" have been updated accordingly.
1010 The output from "netstat -r" shows only routing entries and
1011 none of the L2 information.
1014 __FreeBSD_version 800057 marks the switchover from the
1015 binary ath hal to source code. Users must add the line:
1017 options AH_SUPPORT_AR5416
1019 to their kernel config files when specifying:
1023 The ath_hal module no longer exists; the code is now compiled
1024 together with the driver in the ath module. It is now
1025 possible to tailor chip support (i.e. reduce the set of chips
1026 and thereby the code size); consult ath_hal(4) for details.
1029 __FreeBSD_version 800054 adds memory barriers to
1030 <machine/atomic.h>, new interfaces to ifnet to facilitate
1031 multiple hardware transmit queues for cards that support
1032 them, and a lock-less ring-buffer implementation to
1033 enable drivers to more efficiently manage queueing of
1037 A new version of ZFS (version 13) has been merged to -HEAD.
1038 This version has zpool attribute "listsnapshots" off by
1039 default, which means "zfs list" does not show snapshots,
1040 and is the same as Solaris behavior.
1043 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1046 The uhci, ohci, ehci and slhci USB Host controller drivers have
1047 been put into separate modules. If you load the usb module
1048 separately through loader.conf you will need to load the
1049 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1050 controller add the following to loader.conf:
1056 The ABI used by the PMC toolset has changed. Please keep
1057 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1061 atapci kernel module now includes only generic PCI ATA
1062 driver. AHCI driver moved to ataahci kernel module.
1063 All vendor-specific code moved into separate kernel modules:
1064 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1065 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1066 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1067 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1070 The TTY subsystem of the kernel has been replaced by a new
1071 implementation, which provides better scalability and an
1072 improved driver model. Most common drivers have been migrated to
1073 the new TTY subsystem, while others have not. The following
1074 drivers have not yet been ported to the new TTY layer:
1077 cy, digi, rc, rp, sio
1083 ng_h4, ng_tty, ppp, sl, snp
1085 Adding these drivers to your kernel configuration file shall
1086 cause compilation to fail.
1089 ntpd has been upgraded to 4.2.4p5.
1092 OpenSSH has been upgraded to 5.1p1.
1094 For many years, FreeBSD's version of OpenSSH preferred DSA
1095 over RSA for host and user authentication keys. With this
1096 upgrade, we've switched to the vendor's default of RSA over
1097 DSA. This may cause upgraded clients to warn about unknown
1098 host keys even for previously known hosts. Users should
1099 follow the usual procedure for verifying host keys before
1100 accepting the RSA key.
1102 This can be circumvented by setting the "HostKeyAlgorithms"
1103 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1106 Please note that the sequence of keys offered for
1107 authentication has been changed as well. You may want to
1108 specify IdentityFile in a different order to revert this
1112 The sio(4) driver has been removed from the i386 and amd64
1113 kernel configuration files. This means uart(4) is now the
1114 default serial port driver on those platforms as well.
1116 To prevent collisions with the sio(4) driver, the uart(4) driver
1117 uses different names for its device nodes. This means the
1118 onboard serial port will now most likely be called "ttyu0"
1119 instead of "ttyd0". You may need to reconfigure applications to
1120 use the new device names.
1122 When using the serial port as a boot console, be sure to update
1123 /boot/device.hints and /etc/ttys before booting the new kernel.
1124 If you forget to do so, you can still manually specify the hints
1125 at the loader prompt:
1127 set hint.uart.0.at="isa"
1128 set hint.uart.0.port="0x3F8"
1129 set hint.uart.0.flags="0x10"
1130 set hint.uart.0.irq="4"
1134 The gpt(8) utility has been removed. Use gpart(8) to partition
1138 The version that Linuxulator emulates was changed from 2.4.2
1139 to 2.6.16. If you experience any problems with Linux binaries
1140 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1141 if it fixes the problem contact emulation mailing list.
1144 ISDN4BSD (I4B) was removed from the src tree. You may need to
1145 update a your kernel configuration and remove relevant entries.
1148 I have checked in code to support multiple routing tables.
1149 See the man pages setfib(1) and setfib(2).
1150 This is a hopefully backwards compatible version,
1151 but to make use of it you need to compile your kernel
1152 with options ROUTETABLES=2 (or more up to 16).
1155 The 802.11 wireless support was redone to enable multi-bss
1156 operation on devices that are capable. The underlying device
1157 is no longer used directly but instead wlanX devices are
1158 cloned with ifconfig. This requires changes to rc.conf files.
1159 For example, change:
1160 ifconfig_ath0="WPA DHCP"
1163 ifconfig_wlan0="WPA DHCP"
1164 see rc.conf(5) for more details. In addition, mergemaster of
1165 /etc/rc.d is highly recommended. Simultaneous update of userland
1166 and kernel wouldn't hurt either.
1168 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1169 modules were merged into the base wlan module. All references
1170 to these modules (e.g. in kernel config files) must be removed.
1173 psm(4) has gained write(2) support in native operation level.
1174 Arbitrary commands can be written to /dev/psm%d and status can
1175 be read back from it. Therefore, an application is responsible
1176 for status validation and error recovery. It is a no-op in
1177 other operation levels.
1180 Support for KSE threading has been removed from the kernel. To
1181 run legacy applications linked against KSE libmap.conf may
1182 be used. The following libmap.conf may be used to ensure
1183 compatibility with any prior release:
1185 libpthread.so.1 libthr.so.1
1186 libpthread.so.2 libthr.so.2
1187 libkse.so.3 libthr.so.3
1190 The layout of struct vmspace has changed. This affects libkvm
1191 and any executables that link against libkvm and use the
1192 kvm_getprocs() function. In particular, but not exclusively,
1193 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1194 The effects are minimal, but it's advisable to upgrade world
1198 The latest em driver no longer has support in it for the
1199 82575 adapter, this is now moved to the igb driver. The
1200 split was done to make new features that are incompatible
1201 with older hardware easier to do.
1204 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1205 likewise the kernel option is now GEOM_LINUX_LVM.
1208 The default NFS mount mode has changed from UDP to TCP for
1209 increased reliability. If you rely on (insecurely) NFS
1210 mounting across a firewall you may need to update your
1214 Belatedly note the addition of m_collapse for compacting
1218 The fts(3) structures have been changed to use adequate
1219 integer types for their members and so to be able to cope
1220 with huge file trees. The old fts(3) ABI is preserved
1221 through symbol versioning in libc, so third-party binaries
1222 using fts(3) should still work, although they will not take
1223 advantage of the extended types. At the same time, some
1224 third-party software might fail to build after this change
1225 due to unportable assumptions made in its source code about
1226 fts(3) structure members. Such software should be fixed
1227 by its vendor or, in the worst case, in the ports tree.
1228 FreeBSD_version 800015 marks this change for the unlikely
1229 case that a portable fix is impossible.
1232 To upgrade to -current after this date, you must be running
1233 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1234 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1237 The ADAPTIVE_GIANT kernel option has been retired because its
1238 functionality is the default now.
1241 The AT keyboard emulation of sunkbd(4) has been turned on
1242 by default. In order to make the special symbols of the Sun
1243 keyboards driven by sunkbd(4) work under X these now have
1244 to be configured the same way as Sun USB keyboards driven
1245 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1247 Option "XkbLayout" "us"
1248 Option "XkbRules" "xorg"
1249 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1252 It has been decided that it is desirable to provide ABI
1253 backwards compatibility to the FreeBSD 4/5/6 versions of the
1254 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1255 broken with the introduction of PCI domain support (see the
1256 20070930 entry). Unfortunately, this required the ABI of
1257 PCIOCGETCONF to be broken again in order to be able to
1258 provide backwards compatibility to the old version of that
1259 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1260 again. As for prominent ports this affects neither pciutils
1261 nor xorg-server this time, the hal port needs to be rebuilt
1265 The misnamed kthread_create() and friends have been renamed
1266 to kproc_create() etc. Many of the callers already
1267 used kproc_start()..
1268 I will return kthread_create() and friends in a while
1269 with implementations that actually create threads, not procs.
1270 Renaming corresponds with version 800002.
1279 Avoid using make -j when upgrading. While generally safe, there are
1280 sometimes problems using -j to upgrade. If your upgrade fails with
1281 -j, please try again without -j. From time to time in the past there
1282 have been problems using -j with buildworld and/or installworld. This
1283 is especially true when upgrading between "distant" versions (eg one
1284 that cross a major release boundary or several minor releases, or when
1285 several months have passed on the -current branch).
1287 Sometimes, obscure build problems are the result of environment
1288 poisoning. This can happen because the make utility reads its
1289 environment when searching for values for global variables. To run
1290 your build attempts in an "environmental clean room", prefix all make
1291 commands with 'env -i '. See the env(1) manual page for more details.
1293 When upgrading from one major version to another it is generally best
1294 to upgrade to the latest code in the currently installed branch first,
1295 then do an upgrade to the new branch. This is the best-tested upgrade
1296 path, and has the highest probability of being successful. Please try
1297 this approach before reporting problems with a major version upgrade.
1301 When upgrading the boot ZFS pool to a new version, always follow
1304 1.) recompile and reinstall the ZFS boot loader and boot block
1305 (this is part of "make buildworld" and "make installworld")
1307 2.) update the ZFS boot block on your boot drive
1309 The following example updates the ZFS boot block on the first
1310 partition (freebsd-boot) of a GPT partitioned drive ad0:
1311 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1313 Non-boot pools do not need these updates.
1317 If you are updating from a prior version of FreeBSD (even one just
1318 a few days old), you should follow this procedure. It is the most
1319 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1321 make kernel-toolchain
1322 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1323 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1325 To test a kernel once
1326 ---------------------
1327 If you just want to boot a kernel once (because you are not sure
1328 if it works, or if you want to boot a known bad kernel to provide
1329 debugging information) run
1330 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1331 nextboot -k testkernel
1333 To just build a kernel when you know that it won't mess you up
1334 --------------------------------------------------------------
1335 This assumes you are already running a CURRENT system. Replace
1336 ${arch} with the architecture of your machine (e.g. "i386",
1337 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1339 cd src/sys/${arch}/conf
1340 config KERNEL_NAME_HERE
1341 cd ../compile/KERNEL_NAME_HERE
1346 If this fails, go to the "To build a kernel" section.
1348 To rebuild everything and install it on the current system.
1349 -----------------------------------------------------------
1350 # Note: sometimes if you are running current you gotta do more than
1351 # is listed here if you are upgrading from a really old current.
1353 <make sure you have good level 0 dumps>
1355 make kernel KERNCONF=YOUR_KERNEL_HERE
1357 <reboot in single user> [3]
1365 To cross-install current onto a separate partition
1366 --------------------------------------------------
1367 # In this approach we use a separate partition to hold
1368 # current's root, 'usr', and 'var' directories. A partition
1369 # holding "/", "/usr" and "/var" should be about 2GB in
1372 <make sure you have good level 0 dumps>
1375 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1376 <maybe newfs current's root partition>
1377 <mount current's root partition on directory ${CURRENT_ROOT}>
1378 make installworld DESTDIR=${CURRENT_ROOT}
1379 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1380 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1381 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1382 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1383 <reboot into current>
1384 <do a "native" rebuild/install as described in the previous section>
1385 <maybe install compatibility libraries from ports/misc/compat*>
1389 To upgrade in-place from 8.x-stable to current
1390 ----------------------------------------------
1391 <make sure you have good level 0 dumps>
1393 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1395 <reboot in single user> [3]
1402 Make sure that you've read the UPDATING file to understand the
1403 tweaks to various things you need. At this point in the life
1404 cycle of current, things change often and you are on your own
1405 to cope. The defaults can also change, so please read ALL of
1406 the UPDATING entries.
1408 Also, if you are tracking -current, you must be subscribed to
1409 freebsd-current@freebsd.org. Make sure that before you update
1410 your sources that you have read and understood all the recent
1411 messages there. If in doubt, please track -stable which has
1412 much fewer pitfalls.
1414 [1] If you have third party modules, such as vmware, you
1415 should disable them at this point so they don't crash your
1418 [3] From the bootblocks, boot -s, and then do
1423 adjkerntz -i # if CMOS is wall time
1424 Also, when doing a major release upgrade, it is required that
1425 you boot into single user mode to do the installworld.
1427 [4] Note: This step is non-optional. Failure to do this step
1428 can result in a significant reduction in the functionality of the
1429 system. Attempting to do it by hand is not recommended and those
1430 that pursue this avenue should read this file carefully, as well
1431 as the archives of freebsd-current and freebsd-hackers mailing lists
1432 for potential gotchas. The -U option is also useful to consider.
1433 See mergemaster(8) for more information.
1435 [5] Usually this step is a noop. However, from time to time
1436 you may need to do this if you get unknown user in the following
1437 step. It never hurts to do it all the time. You may need to
1438 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1439 install) after the buildworld before this step if you last updated
1440 from current before 20020224 or from -stable before 20020408.
1442 [6] This only deletes old files and directories. Old libraries
1443 can be deleted by "make delete-old-libs", but you have to make
1444 sure that no program is using those libraries anymore.
1446 [8] In order to have a kernel that can run the 4.x binaries needed to
1447 do an installworld, you must include the COMPAT_FREEBSD4 option in
1448 your kernel. Failure to do so may leave you with a system that is
1449 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1450 required to run the 5.x binaries on more recent kernels. And so on
1451 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1453 Make sure that you merge any new devices from GENERIC since the
1454 last time you updated your kernel config file.
1456 [9] When checking out sources, you must include the -P flag to have
1457 cvs prune empty directories.
1459 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1460 "?=" instead of the "=" assignment operator, so that buildworld can
1461 override the CPUTYPE if it needs to.
1463 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1464 not on the command line, or in /etc/make.conf. buildworld will
1465 warn if it is improperly defined.
1468 This file contains a list, in reverse chronological order, of major
1469 breakages in tracking -current. It is not guaranteed to be a complete
1470 list of such breakages, and only contains entries since October 10, 2007.
1471 If you need to see UPDATING entries from before that date, you will need
1472 to fetch an UPDATING file from an older FreeBSD release.
1474 Copyright information:
1476 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1478 Redistribution, publication, translation and use, with or without
1479 modification, in full or in part, in any form or format of this
1480 document are permitted without further permission from the author.
1482 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1483 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1484 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1485 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1486 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1487 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1488 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1489 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1490 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1491 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1492 POSSIBILITY OF SUCH DAMAGE.
1494 Contact Warner Losh if you have any questions about your use of