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.
16 A new version of ZFS (pool version 5000) has been merged to 9-STABLE.
17 Starting with this version the old system of ZFS pool versioning
18 is superseded by "feature flags". This concept enables forward
19 compatibility against certain future changes in functionality of ZFS
20 pools. The first two read-only compatible "feature flags" for ZFS
21 pools are "com.delphix:async_destroy" and "com.delphix:empty_bpobj".
22 For more information read the new zpool-features(7) manual page.
23 Please refer to the "ZFS notes" section of this file for information
24 on upgrading boot ZFS pools.
27 The IPFIREWALL_FORWARD kernel option has been removed. Its
28 functionality now turned on by default.
31 The random(4) support for the VIA hardware random number
32 generator (`PADLOCK') is no longer enabled unconditionally.
33 Add the PADLOCK_RNG option in the custom kernel config if
34 needed. The GENERIC kernels on i386 and amd64 do include the
35 option, so the change only affects the custom kernel
39 The sparc64 ZFS loader has been changed to no longer try to auto-
40 detect ZFS providers based on diskN aliases but now requires these
41 to be explicitly listed in the OFW boot-device environment variable.
44 Now unix domain sockets behave "as expected" on nullfs(5). Previously
45 nullfs(5) did not pass through all behaviours to the underlying layer,
46 as a result if we bound to a socket on the lower layer we could connect
47 only to the lower path; if we bound to the upper layer we could connect
48 only to the upper path. The new behavior is one can connect to both the
49 lower and the upper paths regardless what layer path one binds to.
52 The acpi_wmi(4) status device /dev/wmistat has been renamed to
56 A new VOP_ADVISE() was added to support posix_fadvise(2). All
57 filesystem modules must be recompiled.
60 The interface of the VOP_VPTOCNP(9) changed, now the returned
61 vnode shall be referenced, previously it was required to be
62 only held. All in-tree filesystems are converted.
68 The broken amd(4) driver has been replaced with esp(4) in the amd64,
69 i386 and pc98 GENERIC kernel configuration files.
72 This commit modifies vfs_register() so that it uses a hash
73 calculation to set vfc_typenum, which is enabled by default.
74 The first time a system is booted after this change, the
75 vfc_typenum values will change for all file systems. The
76 main effect of this is a change to the NFS server file handles
77 for file systems that use vfc_typenum in their fsid, such as ZFS.
78 It will, however, prevent vfc_typenum from changing when file
79 systems are loaded in a different order for subsequent reboots.
80 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
81 until you are ready to remount all NFS clients after a reboot.
84 Bump the shared library version numbers for libraries that
85 do not use symbol versioning, have changed the ABI compared
86 to stable/8 and which shared library version was not bumped.
87 Done as part of 9.0-RELEASE cycle.
90 During the merge of Capsicum features, the fget(9) KPI was modified.
91 This may require the rebuilding of out-of-tree device drivers --
92 issues have been reported specifically with the nVidia device driver.
93 __FreeBSD_version is bumped to 900041.
95 Also, there is a period between 20110811 and 20110814 where the
96 special devices /dev/{stdin,stdout,stderr} did not work correctly.
97 Building world from a kernel during that window may not work.
100 The packet filter (pf) code has been updated to OpenBSD 4.5.
101 You need to update userland tools to be in sync with kernel.
102 This update breaks backward compatibility with earlier pfsync(4)
103 versions. Care must be taken when updating redundant firewall setups.
106 The following sysctls and tunables are retired on x86 platforms:
108 machdep.hlt_logical_cpus
109 The following sysctl is retired:
110 machdep.hyperthreading_allowed
111 The sysctls were supposed to provide a way to dynamically offline and
112 online selected CPUs on x86 platforms, but the implementation has not
113 been reliable especially with SCHED_ULE scheduler.
114 machdep.hyperthreading_allowed tunable is still available to ignore
115 hyperthreading CPUs at OS level.
116 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
117 where X is an APIC ID of a CPU. Be advised, though, that disabling
118 CPUs in non-uniform fashion will result in non-uniform topology and
119 may lead to sub-optimal system performance with SCHED_ULE, which is
123 cpumask_t type is retired and cpuset_t is used in order to describe
127 Changes to ifconfig(8) for dynamic address family detection mandate
128 that you are running a kernel of 20110525 or later. Make sure to
129 follow the update procedure to boot a new kernel before installing
133 Support for sun4v architecture is officially dropped
136 Several KPI breaking changes have been committed to the mii(4) layer,
137 the PHY drivers and consequently some Ethernet drivers using mii(4).
138 This means that miibus.ko and the modules of the affected Ethernet
139 drivers need to be recompiled.
141 Note to kernel developers: Given that the OUI bit reversion problem
142 was fixed as part of these changes all mii(4) commits related to OUIs,
143 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
144 handling, no longer can be merged verbatim to stable/8 and previous
148 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
149 into their kernel configurations along with 'device pci'.
152 The default NFS client is now the new NFS client, so fstype "newnfs"
153 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
154 Although mounts via fstype "nfs" will usually work without userland
155 changes, it is recommended that the mount(8) and mount_nfs(8)
156 commands be rebuilt from sources and that a link to mount_nfs called
157 mount_oldnfs be created. The new client is compiled into the
158 kernel with "options NFSCL" and this is needed for diskless root
159 file systems. The GENERIC kernel configs have been changed to use
160 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
161 To use the regular/old client, you can "mount -t oldnfs ...". For
162 a diskless root file system, you must also include a line like:
164 vfs.root.mountfrom="oldnfs:"
166 in the boot/loader.conf on the root fs on the NFS server to make
167 a diskless root fs use the old client.
170 The GENERIC kernels for all architectures now default to the new
171 CAM-based ATA stack. It means that all legacy ATA drivers were
172 removed and replaced by respective CAM drivers. If you are using
173 ATA device names in /etc/fstab or other places, make sure to update
174 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
175 where 'Y's are the sequential numbers starting from zero for each type
176 in order of detection, unless configured otherwise with tunables,
177 see cam(4)). There will be symbolic links created in /dev/ to map
178 old adX devices to the respective adaY. They should provide basic
179 compatibility for file systems mounting in most cases, but they do
180 not support old user-level APIs and do not have respective providers
181 in GEOM. Consider using updated management tools with new device names.
183 It is possible to load devices ahci, ata, siis and mvs as modules,
184 but option ATA_CAM should remain in kernel configuration to make ata
185 module work as CAM driver supporting legacy ATA controllers. Device ata
186 still can be used in modular fashion (atacore + ...). Modules atadisk
187 and atapi* are not used and won't affect operation in ATA_CAM mode.
188 Note that to use CAM-based ATA kernel should include CAM devices
189 scbus, pass, da (or explicitly ada), cd and optionally others. All of
190 them are parts of the cam module.
192 ataraid(4) functionality is now supported by the RAID GEOM class.
193 To use it you can load geom_raid kernel module and use graid(8) tool
194 for management. Instead of /dev/arX device names, use /dev/raid/rX.
196 No kernel config options or code have been removed, so if a problem
197 arises, please report it and optionally revert to the old ATA stack.
198 In order to do it you can remove from the kernel config:
203 , and instead add back:
204 device atadisk # ATA disk drives
205 device ataraid # ATA RAID drives
206 device atapicd # ATAPI CDROM drives
207 device atapifd # ATAPI floppy drives
208 device atapist # ATAPI tape drives
211 The default NFS server has been changed to the new server, which
212 was referred to as the experimental server. If you need to switch
213 back to the old NFS server, you must now put the "-o" option on
214 both the mountd and nfsd commands. This can be done using the
215 mountd_flags and nfs_server_flags rc.conf variables until an
216 update to the rc scripts is committed, which is coming soon.
219 The GNU Objective-C runtime library (libobjc), and other Objective-C
220 related components have been removed from the base system. If you
221 require an Objective-C library, please use one of the available ports.
224 ath(4) has been split into bus- and device- modules. if_ath contains
225 the HAL, the TX rate control and the network device code. if_ath_pci
226 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
227 contains the AHB glue. Users need to load both if_ath_pci and if_ath
228 in order to use ath on everything else.
230 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
231 need to load if_ath and if_ath_pci for ath(4) operation.
234 As part of the replacement of sysinstall, the process of building
235 release media has changed significantly. For details, please re-read
236 release(7), which has been updated to reflect the new build process.
239 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
240 is the last available version under GPLv2. It brings a number of new
241 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
242 4.1 and SSE 4.2), better support for powerpc64, a number of new
243 directives, and lots of other small improvements. See the ChangeLog
244 file in contrib/binutils for the full details.
247 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
248 compliant, and will now use half of hash for authentication.
249 This will break interoperability with all stacks (including all
250 actual FreeBSD versions) who implement
251 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
253 The only workaround with such peers is to use another HMAC
254 algorithm for IPsec ("phase 2") authentication.
257 Remove the uio_yield prototype and symbol. This function has
258 been misnamed since it was introduced and should not be
259 globally exposed with this name. The equivalent functionality
260 is now available using kern_yield(curthread->td_user_pri).
261 The function remains undocumented.
264 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
265 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
266 for scalar sysctls is defined but disabled. Code that needs
267 UQUAD to pass the type checking that must compile on older
268 systems where the define is not present can check against
269 __FreeBSD_version >= 900030.
271 The system dialog(1) has been replaced with a new version previously
272 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
273 with the previous version, but the libdialog associated with it has
274 a largely incompatible API. As such, the original version of libdialog
275 will be kept temporarily as libodialog, until its base system consumers
276 are replaced or updated. Bump __FreeBSD_version to 900030.
279 If you are trying to run make universe on a -stable system, and you get
280 the following warning:
281 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
282 unknown. config(8) likely too old."
283 or something similar to it, then you must upgrade your -stable system
284 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
285 stable/8) or build the config from the latest stable/8 branch and
286 install it on your system.
288 Prior to this date, building a current universe on 8-stable system from
289 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
290 error in the first kernel build phase. A new config on those old
291 systems will fix that problem for older versions of -current.
294 The TCP stack has been modified to allow Khelp modules to interact with
295 it via helper hook points and store per-connection data in the TCP
296 control block. Bump __FreeBSD_version to 900029. User space tools that
297 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
301 Generic IEEE 802.3 annex 31B full duplex flow control support has been
302 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
303 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
304 to take advantage of it instead of using custom implementations. This
305 means that these drivers now no longer unconditionally advertise
306 support for flow control but only do so if flow control is a selected
307 media option. This was implemented in the generic support that way in
308 order to allow flow control to be switched on and off via ifconfig(8)
309 with the PHY specific default to typically off in order to protect
310 from unwanted effects. Consequently, if you used flow control with
311 one of the above mentioned drivers you now need to explicitly enable
313 ifconfig bge0 media auto mediaopt flowcontrol
315 Along with the above mentioned changes generic support for setting
316 1000baseT master mode also has been added and brgphy(4), ciphy(4),
317 e1000phy(4) as well as ip1000phy(4) have been converted to take
318 advantage of it. This means that these drivers now no longer take the
319 link0 parameter for selecting master mode but the master media option
320 has to be used instead, for example like in the following:
321 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
323 Selection of master mode now is also available with all other PHY
324 drivers supporting 1000baseT.
327 The TCP stack has received a significant update to add support for
328 modularised congestion control and generally improve the clarity of
329 congestion control decisions. Bump __FreeBSD_version to 900025. User
330 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
331 sockstat) need to be recompiled.
334 The man(1) utility has been replaced by a new version that no longer
335 uses /etc/manpath.config. Please consult man.conf(5) for how to
336 migrate local entries to the new format.
339 The copyright strings printed by login(1) and sshd(8) at the time of a
340 new connection have been removed to follow other operating systems and
344 A workaround for a fixed ld bug has been removed in kernel code,
345 so make sure that your system ld is built from sources after
346 revision 210245 from 2010-07-19 (r211583 if building head kernel
347 on stable/8, r211584 for stable/7; both from 2010-08-21).
348 A symptom of incorrect ld version is different addresses for
349 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
352 The $ipv6_prefer variable in rc.conf(5) has been split into
353 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
355 The $ip6addrctl_policy is a variable to choose a pre-defined
356 address selection policy set by ip6addrctl(8). A value
357 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
360 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
361 flag (see an entry of 20090926) is set on an interface with no
362 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
363 security reason. If you want IPv6 link-local address on all
364 interfaces by default, set this to "YES".
366 The old ipv6_prefer="YES" is equivalent to
367 ipv6_activate_all_interfaces="YES" and
368 ip6addrctl_policy="ipv6_prefer".
371 DTrace has grown support for userland tracing. Due to this, DTrace is
372 now i386 and amd64 only.
373 dtruss(1) is now installed by default on those systems and a new
374 kernel module is needed for userland tracing: fasttrap.
375 No changes to your kernel config file are necessary to enable
376 userland tracing, but you might consider adding 'STRIP=' and
377 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
378 to have informative userland stack traces in DTrace (ustack).
381 The acpi_aiboost(4) driver has been removed in favor of the new
382 aibs(4) driver. You should update your kernel configuration file.
385 BSD grep has been imported to the base system and it is built by
386 default. It is completely BSD licensed, highly GNU-compatible, uses
387 less memory than its GNU counterpart and has a small codebase.
388 However, it is slower than its GNU counterpart, which is mostly
389 noticeable for larger searches, for smaller ones it is measurable
390 but not significant. The reason is complex, the most important factor
391 is that we lack a modern and efficient regex library and GNU
392 overcomes this by optimizing the searches internally. Future work
393 on improving the regex performance is planned, for the meantime,
394 users that need better performance, can build GNU grep instead by
395 setting the WITH_GNU_GREP knob.
398 Due to the import of powerpc64 support, all existing powerpc kernel
399 configuration files must be updated with a machine directive like this:
400 machine powerpc powerpc
402 In addition, an updated config(8) is required to build powerpc kernels
406 A new version of ZFS (version 15) has been merged to -HEAD.
407 This version uses a python library for the following subcommands:
408 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
409 For full functionality of these commands the following port must
410 be installed: sysutils/py-zfs
413 'vm_page's are now hashed by physical address to an array of mutexes.
414 Currently this is only used to serialize access to hold_count. Over
415 time the page queue mutex will be peeled away. This changes the size
416 of pmap on every architecture. And requires all callers of vm_page_hold
417 and vm_page_unhold to be updated.
420 WITH_CTF can now be specified in src.conf (not recommended, there
421 are some problems with static executables), make.conf (would also
422 affect ports which do not use GNU make and do not override the
423 compile targets) or in the kernel config (via "makeoptions
425 When WITH_CTF was specified there before this was silently ignored,
426 so make sure that WITH_CTF is not used in places which could lead
427 to unwanted behavior.
430 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
431 to allow 32-bit compatibility on non-x86 platforms. All kernel
432 configurations on amd64 and ia64 platforms using these options must
433 be modified accordingly.
436 The utmp user accounting database has been replaced with utmpx,
437 the user accounting interface standardized by POSIX.
438 Unfortunately the semantics of utmp and utmpx don't match,
439 making it practically impossible to support both interfaces.
440 The user accounting database is used by tools like finger(1),
441 last(1), talk(1), w(1) and ac(8).
443 All applications in the base system use utmpx. This means only
444 local binaries (e.g. from the ports tree) may still use these
445 utmp database files. These applications must be rebuilt to make
448 After the system has been upgraded, it is safe to remove the old
449 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
450 assuming their contents is of no importance anymore. Old wtmp
451 databases can only be used by last(1) and ac(8) after they have
452 been converted to the new format using wtmpcvt(1).
455 Introduce the kernel thread "deadlock resolver" (which can be enabled
456 via the DEADLKRES option, see NOTES for more details) and the
457 sleepq_type() function for sleepqueues.
460 The rc.firewall and rc.firewall6 were unified, and
461 rc.firewall6 and rc.d/ip6fw were removed.
462 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
463 variables are obsoleted. Instead, the following new rc
464 variables are added to rc.d/ipfw:
466 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
467 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
468 firewall_simple_onet_ipv6, firewall_trusted_ipv6
470 The meanings correspond to the relevant IPv4 variables.
476 The default terminal emulation for syscons(4) has been changed
477 from cons25 to xterm on all platforms except pc98. This means
478 that the /etc/ttys file needs to be updated to ensure correct
479 operation of applications on the console.
481 The terminal emulation style can be toggled per window by using
482 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
483 options can be used to change the compile-time default back to
486 To prevent graphical artifacts, make sure the TERM environment
487 variable is set to match the terminal emulation that is being
488 performed by syscons(4).
491 The layout of the structure ieee80211req_scan_result has changed.
492 Applications that require wireless scan results (e.g. ifconfig(8))
493 from net80211 need to be recompiled.
495 Applications such as wpa_supplicant(8) may require a full world
496 build without using NO_CLEAN in order to get synchronized with the
500 The iwn(4) driver has been updated to support the 5000 and 5150 series.
501 There's one kernel module for each firmware. Adding "device iwnfw"
502 to the kernel configuration file means including all three firmware
503 images inside the kernel. If you want to include just the one for
504 your wireless card, use the devices iwn4965fw, iwn5000fw or
508 The rc.d/network_ipv6, IPv6 configuration script has been integrated
509 into rc.d/netif. The changes are the following:
511 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
512 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
513 Note that both variables need the "inet6" keyword at the head.
515 Do not set $ipv6_network_interfaces manually if you do not
516 understand what you are doing. It is not needed in most cases.
518 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
521 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
522 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
524 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
525 all configured interfaces have "inet6 accept_rtadv" in the
526 $ifconfig_IF_ipv6. These are for backward compatibility.
528 3. A new variable $ipv6_prefer has been added. If NO, IPv6
529 functionality of interfaces with no corresponding
530 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
531 and the default address selection policy of ip6addrctl(8)
532 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
533 Note that if you want to configure IPv6 functionality on the
534 disabled interfaces after boot, first you need to clear the flag by
535 using ifconfig(8) like:
537 ifconfig em0 inet6 -ifdisabled
539 If YES, the default address selection policy is set as
542 The default value of $ipv6_prefer is NO.
544 4. If your system need to receive Router Advertisement messages,
545 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
546 scripts automatically invoke rtsol(8) when the interface becomes
547 UP. The Router Advertisement messages are used for SLAAC
548 (State-Less Address AutoConfiguration).
551 802.11s D3.03 support was committed. This is incompatible with the
552 previous code, which was based on D3.0.
555 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
556 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
557 control whether accepting Router Advertisement messages or not.
558 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
559 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
560 The ifconfig(8) utility now supports these flags.
563 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
564 mount(8) and -a option for df(1) to see them.
567 The old tunable hw.bus.devctl_disable has been superseded by
568 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
569 replaced by hw.bus.devctl_queue=0. The default for this new tunable
573 Remove the option STOP_NMI. The default action is now to use NMI only
574 for KDB via the newly introduced function stop_cpus_hard() and
575 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
578 The stable/8 branch created in subversion. This corresponds to the
579 RELENG_8 branch in CVS.
582 Bump the shared library version numbers for all libraries that do not
583 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
584 __FreeBSD_version to 800105.
587 Due to changes in the implementation of virtual network stack support,
588 all network-related kernel modules must be recompiled. As this change
589 breaks the ABI, bump __FreeBSD_version to 800104.
592 The TOE interface to the TCP syncache has been modified to remove
593 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
594 The cxgb driver is the only TOE consumer affected by this change, and
595 needs to be recompiled along with the kernel. As this change breaks
596 the ABI, bump __FreeBSD_version to 800103.
599 Padding has been added to struct tcpcb, sackhint and tcpstat in
600 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
601 maintaining the ABI. However, this change breaks the ABI, so bump
602 __FreeBSD_version to 800102. User space tools that rely on the size of
603 any of these structs (e.g. sockstat) need to be recompiled.
606 The NFS_LEGACYRPC option has been removed along with the old kernel
607 RPC implementation that this option selected. Kernel configurations
608 may need to be adjusted.
611 The network interface device nodes at /dev/net/<interface> have been
612 removed. All ioctl operations can be performed the normal way using
613 routing sockets. The kqueue functionality can generally be replaced
614 with routing sockets.
617 The documentation from the FreeBSD Documentation Project (Handbook,
618 FAQ, etc.) is now installed via packages by sysinstall(8) and under
619 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
622 The ABI of various structures related to the SYSV IPC API have been
623 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
624 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
628 Layout of struct vnet has changed as routing related variables were
629 moved to their own Vimage module. Modules need to be recompiled. Bump
630 __FreeBSD_version to 800099.
633 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
634 respectively. As long as no more than 16 groups per process are used,
635 no changes should be visible. When more than 16 groups are used, old
636 binaries may fail if they call getgroups() or getgrouplist() with
637 statically sized storage. Recompiling will work around this, but
638 applications should be modified to use dynamically allocated storage
639 for group arrays as POSIX.1-2008 does not cap an implementation's
640 number of supported groups at NGROUPS_MAX+1 as previous versions did.
642 NFS and portalfs mounts may also be affected as the list of groups is
643 truncated to 16. Users of NFS who use more than 16 groups, should
644 take care that negative group permissions are not used on the exported
645 file systems as they will not be reliable unless a GSSAPI based
646 authentication method is used.
649 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
650 option compiles in the support for adaptive spinning for lockmgrs
651 which want to enable it. The lockinit() function now accepts the flag
652 LK_ADAPTIVE in order to make the lock object subject to adaptive
653 spinning when both held in write and read mode.
656 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
657 changed. User applications that use this ioctl need to be rebuilt.
660 The layout of struct thread has changed. Kernel and modules need to
664 The layout of structs ifnet, domain, protosw and vnet_net has changed.
665 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
668 window(1) has been removed from the base system. It can now be
669 installed from ports. The port is called misc/window.
672 The way we are storing and accessing `routing table' entries has
673 changed. Programs reading the FIB, like netstat, need to be
677 A new netisr implementation has been added for FreeBSD 8. Network
678 file system modules, such as igmp, ipdivert, and others, should be
680 Bump __FreeBSD_version to 800096.
683 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
687 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
688 Bump __FreeBSD_version to 800094.
691 Add mnt_xflag field to 'struct mount'. File system modules need to be
693 Bump __FreeBSD_version to 800093.
696 The compiling option ADAPTIVE_SX has been retired while it has been
697 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
698 The KPI for sx_init_flags() changes as accepting flags:
699 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
700 been introduced in order to handle the reversed logic.
701 Bump __FreeBSD_version to 800092.
704 Add support for hierarchical jails. Remove global securelevel.
705 Bump __FreeBSD_version to 800091.
708 The layout of struct vnet_net has changed, therefore modules
710 Bump __FreeBSD_version to 800090.
713 The newly imported zic(8) produces a new format in the output. Please
714 run tzsetup(8) to install the newly created data to /etc/localtime.
717 The sysctl tree for the usb stack has renamed from hw.usb2.* to
718 hw.usb.* and is now consistent again with previous releases.
721 802.11 monitor mode support was revised and driver api's were changed.
722 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
723 of DLT_IEEE802_11. No user-visible data structures were changed but
724 applications that use DLT_IEEE802_11 may require changes.
725 Bump __FreeBSD_version to 800088.
728 The layout of the following structs has changed: sysctl_oid,
729 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
730 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
731 panics may be experienced. World rebuild is required for
732 correctly checking networking state from userland.
733 Bump __FreeBSD_version to 800085.
736 MLDv2 and Source-Specific Multicast (SSM) have been merged
737 to the IPv6 stack. VIMAGE hooks are in but not yet used.
738 The implementation of SSM within FreeBSD's IPv6 stack closely
739 follows the IPv4 implementation.
741 For kernel developers:
743 * The most important changes are that the ip6_output() and
744 ip6_input() paths no longer take the IN6_MULTI_LOCK,
745 and this lock has been downgraded to a non-recursive mutex.
747 * As with the changes to the IPv4 stack to support SSM, filtering
748 of inbound multicast traffic must now be performed by transport
749 protocols within the IPv6 stack. This does not apply to TCP and
750 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
752 * The KPIs used by IPv6 multicast are similar to those used by
753 the IPv4 stack, with the following differences:
754 * im6o_mc_filter() is analogous to imo_multicast_filter().
755 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
756 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
757 * IN6_LOOKUP_MULTI() has been deprecated and removed.
758 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
759 for MLDv1 have an additional 'timer' argument which is used to
760 jitter the initial membership report for the solicited-node
761 multicast membership on-link.
762 * This is not strictly needed for MLDv2, which already jitters
763 its report transmissions. However, the 'timer' argument is
764 preserved in case MLDv1 is active on the interface.
766 * The KAME linked-list based IPv6 membership implementation has
767 been refactored to use a vector similar to that used by the IPv4
769 Code which maintains a list of its own multicast memberships
770 internally, e.g. carp, has been updated to reflect the new
773 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
774 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
775 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
776 implementation constraint which needs to be addressed in HEAD.
778 For application developers:
780 * The changes are broadly similar to those made for the IPv4
783 * The use of IPv4 and IPv6 multicast socket options on the same
784 socket, using mapped addresses, HAS NOT been tested or supported.
786 * There are a number of issues with the implementation of various
787 IPv6 multicast APIs which need to be resolved in the API surface
788 before the implementation is fully compatible with KAME userland
789 use, and these are mostly to do with interface index treatment.
791 * The literature available discusses the use of either the delta / ASM
792 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
793 using setsourcefilter(3)/getsourcefilter(3). For more information
794 please refer to RFC 3768, 'Socket Interface Extensions for
795 Multicast Source Filters'.
797 * Applications which use the published RFC 3678 APIs should be fine.
799 For systems administrators:
801 * The mtest(8) utility has been refactored to support IPv6, in
802 addition to IPv4. Interface addresses are no longer accepted
803 as arguments, their names must be used instead. The utility
804 will map the interface name to its first IPv4 address as
805 returned by getifaddrs(3).
807 * The ifmcstat(8) utility has also been updated to print the MLDv2
808 endpoint state and source filter lists via sysctl(3).
810 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
811 loopback of IPv6 multicast datagrams by default; it defaults to 1
812 to preserve the existing behaviour. Disabling multicast loopback is
813 recommended for optimal system performance.
815 * The IPv6 MROUTING code has been changed to examine this sysctl
816 instead of attempting to perform a group lookup before looping
817 back forwarded datagrams.
819 Bump __FreeBSD_version to 800084.
822 Implement low-level Bluetooth HCI API.
823 Bump __FreeBSD_version to 800083.
826 The layout of struct malloc_type, used by modules to register new
827 memory allocation types, has changed. Most modules will need to
828 be rebuilt or panics may be experienced.
829 Bump __FreeBSD_version to 800081.
832 Anticipate overflowing inp_flags - add inp_flags2.
833 This changes most offsets in inpcb, so checking v4 connection
834 state will require a world rebuild.
835 Bump __FreeBSD_version to 800080.
838 Add an llentry to struct route and struct route_in6. Modules
839 embedding a struct route will need to be recompiled.
840 Bump __FreeBSD_version to 800079.
843 The size of rt_metrics_lite and by extension rtentry has changed.
844 Networking administration apps will need to be recompiled.
845 The route command now supports show as an alias for get, weighting
846 of routes, sticky and nostick flags to alter the behavior of stateful
848 Bump __FreeBSD_version to 800078.
851 Do not use Giant for kbdmux(4) locking. This is wrong and
852 apparently causing more problems than it solves. This will
853 re-open the issue where interrupt handlers may race with
854 kbdmux(4) in polling mode. Typical symptoms include (but
855 not limited to) duplicated and/or missing characters when
856 low level console functions (such as gets) are used while
857 interrupts are enabled (for example geli password prompt,
858 mountroot prompt etc.). Disabling kbdmux(4) may help.
861 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
862 kernel modules referencing any of the above need to be recompiled.
863 Bump __FreeBSD_version to 800075.
866 GEOM_PART has become the default partition slicer for storage devices,
867 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
868 introduces some changes:
870 MSDOS/EBR: the devices created from MSDOS extended partition entries
871 (EBR) can be named differently than with GEOM_MBR and are now symlinks
872 to devices with offset-based names. fstabs may need to be modified.
874 BSD: the "geometry does not match label" warning is harmless in most
875 cases but it points to problems in file system misalignment with
876 disk geometry. The "c" partition is now implicit, covers the whole
877 top-level drive and cannot be (mis)used by users.
879 General: Kernel dumps are now not allowed to be written to devices
880 whose partition types indicate they are meant to be used for file
881 systems (or, in case of MSDOS partitions, as something else than
884 Most of these changes date approximately from 200812.
887 The uscanner(4) driver has been removed from the kernel. This follows
888 Linux removing theirs in 2.6 and making libusb the default interface
892 The multicast forwarding code has been cleaned up. netstat(1)
893 only relies on KVM now for printing bandwidth upcall meters.
894 The IPv4 and IPv6 modules are split into ip_mroute_mod and
895 ip6_mroute_mod respectively. The config(5) options for statically
896 compiling this code remain the same, i.e. 'options MROUTING'.
899 Support for the IFF_NEEDSGIANT network interface flag has been
900 removed, which means that non-MPSAFE network device drivers are no
901 longer supported. In particular, if_ar, if_sr, and network device
902 drivers from the old (legacy) USB stack can no longer be built or
906 POSIX.1 Native Language Support (NLS) has been enabled in libc and
907 a bunch of new language catalog files have also been added.
908 This means that some common libc messages are now localized and
909 they depend on the LC_MESSAGES environmental variable.
912 The k8temp(4) driver has been renamed to amdtemp(4) since
913 support for Family 10 and Family 11 CPU families was added.
916 IGMPv3 and Source-Specific Multicast (SSM) have been merged
917 to the IPv4 stack. VIMAGE hooks are in but not yet used.
919 For kernel developers, the most important changes are that the
920 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
921 and this lock has been downgraded to a non-recursive mutex.
923 Transport protocols (UDP, Raw IP) are now responsible for filtering
924 inbound multicast traffic according to group membership and source
925 filters. The imo_multicast_filter() KPI exists for this purpose.
926 Transports which do not use multicast (SCTP, TCP) already reject
927 multicast by default. Forwarding and receive performance may improve
928 as a mutex acquisition is no longer needed in the ip_input()
929 low-level input path. in_addmulti() and in_delmulti() are shimmed
930 to new KPIs which exist to support SSM in-kernel.
932 For application developers, it is recommended that loopback of
933 multicast datagrams be disabled for best performance, as this
934 will still cause the lock to be taken for each looped-back
935 datagram transmission. The net.inet.ip.mcast.loop sysctl may
936 be tuned to 0 to disable loopback by default; it defaults to 1
937 to preserve the existing behaviour.
939 For systems administrators, to obtain best performance with
940 multicast reception and multiple groups, it is always recommended
941 that a card with a suitably precise hash filter is used. Hash
942 collisions will still result in the lock being taken within the
943 transport protocol input path to check group membership.
945 If deploying FreeBSD in an environment with IGMP snooping switches,
946 it is recommended that the net.inet.igmp.sendlocal sysctl remain
947 enabled; this forces 224.0.0.0/24 group membership to be announced
950 The size of 'struct igmpstat' has changed; netstat needs to be
951 recompiled to reflect this.
952 Bump __FreeBSD_version to 800070.
955 libusb20.so.1 is now installed as libusb.so.1 and the ports system
956 updated to use it. This requires a buildworld/installworld in order to
957 update the library and dependencies (usbconfig, etc). Its advisable to
958 rebuild all ports which uses libusb. More specific directions are given
959 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
960 libusb are no longer required and can be removed.
963 A workaround is committed to allow the creation of System V shared
964 memory segment of size > 2 GB on the 64-bit architectures.
965 Due to a limitation of the existing ABI, the shm_segsz member
966 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
967 wrong for large segments. Note that limits must be explicitly
968 raised to allow such segments to be created.
971 The layout of struct ifnet has changed, requiring a rebuild of all
972 network device driver modules.
975 The /dev handling for the new USB stack has changed, a
976 buildworld/installworld is required for libusb20.
979 The new USB2 stack has now been permanently moved in and all kernel and
980 module names reverted to their previous values (eg, usb, ehci, ohci,
981 ums, ...). The old usb stack can be compiled in by prefixing the name
982 with the letter 'o', the old usb modules have been removed.
983 Updating entry 20090216 for xorg and 20090215 for libmap may still
987 The rc.conf(5) option if_up_delay has been renamed to
988 defaultroute_delay to better reflect its purpose. If you have
989 customized this setting in /etc/rc.conf you need to update it to
993 xorg 7.4 wants to configure its input devices via hald which does not
994 yet work with USB2. If the keyboard/mouse does not work in xorg then
996 Option "AllowEmptyInput" "off"
997 to your ServerLayout section. This will cause X to use the configured
998 kbd and mouse sections from your xorg.conf.
1001 The GENERIC kernels for all architectures now default to the new USB2
1002 stack. No kernel config options or code have been removed so if a
1003 problem arises please report it and optionally revert to the old USB
1004 stack. If you are loading USB kernel modules or have a custom kernel
1005 that includes GENERIC then ensure that usb names are also changed over,
1006 eg uftdi -> usb2_serial_ftdi.
1008 Older programs linked against the ports libusb 0.1 need to be
1009 redirected to the new stack's libusb20. /etc/libmap.conf can
1011 # Map old usb library to new one for usb2 stack
1012 libusb-0.1.so.8 libusb20.so.1
1015 All USB ethernet devices now attach as interfaces under the name ueN
1016 (eg. ue0). This is to provide a predictable name as vendors often
1017 change usb chipsets in a product without notice.
1020 The ichsmb(4) driver has been changed to require SMBus slave
1021 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1022 All of the other SMBus controller drivers require left-justified
1023 slave addresses, so this change makes all the drivers provide the
1027 INET6 statistics (struct ip6stat) was updated.
1028 netstat(1) needs to be recompiled.
1031 NTFS has been removed from GENERIC kernel on amd64 to match
1032 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1033 will load ntfs.ko module automatically when NTFS support is
1034 actually needed, unless ntfs.ko is not installed or security
1035 level prohibits loading kernel modules. If either is the case,
1036 "options NTFS" has to be added into kernel config.
1039 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1040 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1041 800061. User space tools that rely on the size of struct tcpcb in
1042 tcp_var.h (e.g. sockstat) need to be recompiled.
1045 ng_tty(4) module updated to match the new TTY subsystem.
1046 Due to API change, user-level applications must be updated.
1047 New API support added to mpd5 CVS and expected to be present
1048 in next mpd5.3 release.
1051 With __FreeBSD_version 800060 the makefs tool is part of
1052 the base system (it was a port).
1055 The afdata and ifnet locks have been changed from mutexes to
1056 rwlocks, network modules will need to be re-compiled.
1059 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1060 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1061 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1062 architecture and 40 bytes on 64-bit architecture. The userland
1063 applications "arp" and "ndp" have been updated accordingly.
1064 The output from "netstat -r" shows only routing entries and
1065 none of the L2 information.
1068 __FreeBSD_version 800057 marks the switchover from the
1069 binary ath hal to source code. Users must add the line:
1071 options AH_SUPPORT_AR5416
1073 to their kernel config files when specifying:
1077 The ath_hal module no longer exists; the code is now compiled
1078 together with the driver in the ath module. It is now
1079 possible to tailor chip support (i.e. reduce the set of chips
1080 and thereby the code size); consult ath_hal(4) for details.
1083 __FreeBSD_version 800054 adds memory barriers to
1084 <machine/atomic.h>, new interfaces to ifnet to facilitate
1085 multiple hardware transmit queues for cards that support
1086 them, and a lock-less ring-buffer implementation to
1087 enable drivers to more efficiently manage queueing of
1091 A new version of ZFS (version 13) has been merged to -HEAD.
1092 This version has zpool attribute "listsnapshots" off by
1093 default, which means "zfs list" does not show snapshots,
1094 and is the same as Solaris behavior.
1097 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1100 The uhci, ohci, ehci and slhci USB Host controller drivers have
1101 been put into separate modules. If you load the usb module
1102 separately through loader.conf you will need to load the
1103 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1104 controller add the following to loader.conf:
1110 The ABI used by the PMC toolset has changed. Please keep
1111 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1115 atapci kernel module now includes only generic PCI ATA
1116 driver. AHCI driver moved to ataahci kernel module.
1117 All vendor-specific code moved into separate kernel modules:
1118 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1119 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1120 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1121 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1124 The TTY subsystem of the kernel has been replaced by a new
1125 implementation, which provides better scalability and an
1126 improved driver model. Most common drivers have been migrated to
1127 the new TTY subsystem, while others have not. The following
1128 drivers have not yet been ported to the new TTY layer:
1131 cy, digi, rc, rp, sio
1137 ng_h4, ng_tty, ppp, sl, snp
1139 Adding these drivers to your kernel configuration file shall
1140 cause compilation to fail.
1143 ntpd has been upgraded to 4.2.4p5.
1146 OpenSSH has been upgraded to 5.1p1.
1148 For many years, FreeBSD's version of OpenSSH preferred DSA
1149 over RSA for host and user authentication keys. With this
1150 upgrade, we've switched to the vendor's default of RSA over
1151 DSA. This may cause upgraded clients to warn about unknown
1152 host keys even for previously known hosts. Users should
1153 follow the usual procedure for verifying host keys before
1154 accepting the RSA key.
1156 This can be circumvented by setting the "HostKeyAlgorithms"
1157 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1160 Please note that the sequence of keys offered for
1161 authentication has been changed as well. You may want to
1162 specify IdentityFile in a different order to revert this
1166 The sio(4) driver has been removed from the i386 and amd64
1167 kernel configuration files. This means uart(4) is now the
1168 default serial port driver on those platforms as well.
1170 To prevent collisions with the sio(4) driver, the uart(4) driver
1171 uses different names for its device nodes. This means the
1172 onboard serial port will now most likely be called "ttyu0"
1173 instead of "ttyd0". You may need to reconfigure applications to
1174 use the new device names.
1176 When using the serial port as a boot console, be sure to update
1177 /boot/device.hints and /etc/ttys before booting the new kernel.
1178 If you forget to do so, you can still manually specify the hints
1179 at the loader prompt:
1181 set hint.uart.0.at="isa"
1182 set hint.uart.0.port="0x3F8"
1183 set hint.uart.0.flags="0x10"
1184 set hint.uart.0.irq="4"
1188 The gpt(8) utility has been removed. Use gpart(8) to partition
1192 The version that Linuxulator emulates was changed from 2.4.2
1193 to 2.6.16. If you experience any problems with Linux binaries
1194 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1195 if it fixes the problem contact emulation mailing list.
1198 ISDN4BSD (I4B) was removed from the src tree. You may need to
1199 update a your kernel configuration and remove relevant entries.
1202 I have checked in code to support multiple routing tables.
1203 See the man pages setfib(1) and setfib(2).
1204 This is a hopefully backwards compatible version,
1205 but to make use of it you need to compile your kernel
1206 with options ROUTETABLES=2 (or more up to 16).
1209 The 802.11 wireless support was redone to enable multi-bss
1210 operation on devices that are capable. The underlying device
1211 is no longer used directly but instead wlanX devices are
1212 cloned with ifconfig. This requires changes to rc.conf files.
1213 For example, change:
1214 ifconfig_ath0="WPA DHCP"
1217 ifconfig_wlan0="WPA DHCP"
1218 see rc.conf(5) for more details. In addition, mergemaster of
1219 /etc/rc.d is highly recommended. Simultaneous update of userland
1220 and kernel wouldn't hurt either.
1222 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1223 modules were merged into the base wlan module. All references
1224 to these modules (e.g. in kernel config files) must be removed.
1227 psm(4) has gained write(2) support in native operation level.
1228 Arbitrary commands can be written to /dev/psm%d and status can
1229 be read back from it. Therefore, an application is responsible
1230 for status validation and error recovery. It is a no-op in
1231 other operation levels.
1234 Support for KSE threading has been removed from the kernel. To
1235 run legacy applications linked against KSE libmap.conf may
1236 be used. The following libmap.conf may be used to ensure
1237 compatibility with any prior release:
1239 libpthread.so.1 libthr.so.1
1240 libpthread.so.2 libthr.so.2
1241 libkse.so.3 libthr.so.3
1244 The layout of struct vmspace has changed. This affects libkvm
1245 and any executables that link against libkvm and use the
1246 kvm_getprocs() function. In particular, but not exclusively,
1247 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1248 The effects are minimal, but it's advisable to upgrade world
1252 The latest em driver no longer has support in it for the
1253 82575 adapter, this is now moved to the igb driver. The
1254 split was done to make new features that are incompatible
1255 with older hardware easier to do.
1258 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1259 likewise the kernel option is now GEOM_LINUX_LVM.
1262 The default NFS mount mode has changed from UDP to TCP for
1263 increased reliability. If you rely on (insecurely) NFS
1264 mounting across a firewall you may need to update your
1268 Belatedly note the addition of m_collapse for compacting
1272 The fts(3) structures have been changed to use adequate
1273 integer types for their members and so to be able to cope
1274 with huge file trees. The old fts(3) ABI is preserved
1275 through symbol versioning in libc, so third-party binaries
1276 using fts(3) should still work, although they will not take
1277 advantage of the extended types. At the same time, some
1278 third-party software might fail to build after this change
1279 due to unportable assumptions made in its source code about
1280 fts(3) structure members. Such software should be fixed
1281 by its vendor or, in the worst case, in the ports tree.
1282 FreeBSD_version 800015 marks this change for the unlikely
1283 case that a portable fix is impossible.
1286 To upgrade to -current after this date, you must be running
1287 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1288 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1291 The ADAPTIVE_GIANT kernel option has been retired because its
1292 functionality is the default now.
1295 The AT keyboard emulation of sunkbd(4) has been turned on
1296 by default. In order to make the special symbols of the Sun
1297 keyboards driven by sunkbd(4) work under X these now have
1298 to be configured the same way as Sun USB keyboards driven
1299 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1301 Option "XkbLayout" "us"
1302 Option "XkbRules" "xorg"
1303 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1306 It has been decided that it is desirable to provide ABI
1307 backwards compatibility to the FreeBSD 4/5/6 versions of the
1308 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1309 broken with the introduction of PCI domain support (see the
1310 20070930 entry). Unfortunately, this required the ABI of
1311 PCIOCGETCONF to be broken again in order to be able to
1312 provide backwards compatibility to the old version of that
1313 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1314 again. As for prominent ports this affects neither pciutils
1315 nor xorg-server this time, the hal port needs to be rebuilt
1319 The misnamed kthread_create() and friends have been renamed
1320 to kproc_create() etc. Many of the callers already
1321 used kproc_start()..
1322 I will return kthread_create() and friends in a while
1323 with implementations that actually create threads, not procs.
1324 Renaming corresponds with version 800002.
1333 Avoid using make -j when upgrading. While generally safe, there are
1334 sometimes problems using -j to upgrade. If your upgrade fails with
1335 -j, please try again without -j. From time to time in the past there
1336 have been problems using -j with buildworld and/or installworld. This
1337 is especially true when upgrading between "distant" versions (eg one
1338 that cross a major release boundary or several minor releases, or when
1339 several months have passed on the -current branch).
1341 Sometimes, obscure build problems are the result of environment
1342 poisoning. This can happen because the make utility reads its
1343 environment when searching for values for global variables. To run
1344 your build attempts in an "environmental clean room", prefix all make
1345 commands with 'env -i '. See the env(1) manual page for more details.
1347 When upgrading from one major version to another it is generally best
1348 to upgrade to the latest code in the currently installed branch first,
1349 then do an upgrade to the new branch. This is the best-tested upgrade
1350 path, and has the highest probability of being successful. Please try
1351 this approach before reporting problems with a major version upgrade.
1355 When upgrading the boot ZFS pool to a new version, always follow
1358 1.) recompile and reinstall the ZFS boot loader and boot block
1359 (this is part of "make buildworld" and "make installworld")
1361 2.) update the ZFS boot block on your boot drive
1363 The following example updates the ZFS boot block on the first
1364 partition (freebsd-boot) of a GPT partitioned drive ad0:
1365 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1367 Non-boot pools do not need these updates.
1371 If you are updating from a prior version of FreeBSD (even one just
1372 a few days old), you should follow this procedure. It is the most
1373 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1375 make kernel-toolchain
1376 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1377 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1379 To test a kernel once
1380 ---------------------
1381 If you just want to boot a kernel once (because you are not sure
1382 if it works, or if you want to boot a known bad kernel to provide
1383 debugging information) run
1384 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1385 nextboot -k testkernel
1387 To just build a kernel when you know that it won't mess you up
1388 --------------------------------------------------------------
1389 This assumes you are already running a CURRENT system. Replace
1390 ${arch} with the architecture of your machine (e.g. "i386",
1391 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1393 cd src/sys/${arch}/conf
1394 config KERNEL_NAME_HERE
1395 cd ../compile/KERNEL_NAME_HERE
1400 If this fails, go to the "To build a kernel" section.
1402 To rebuild everything and install it on the current system.
1403 -----------------------------------------------------------
1404 # Note: sometimes if you are running current you gotta do more than
1405 # is listed here if you are upgrading from a really old current.
1407 <make sure you have good level 0 dumps>
1409 make kernel KERNCONF=YOUR_KERNEL_HERE
1411 <reboot in single user> [3]
1419 To cross-install current onto a separate partition
1420 --------------------------------------------------
1421 # In this approach we use a separate partition to hold
1422 # current's root, 'usr', and 'var' directories. A partition
1423 # holding "/", "/usr" and "/var" should be about 2GB in
1426 <make sure you have good level 0 dumps>
1429 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1430 <maybe newfs current's root partition>
1431 <mount current's root partition on directory ${CURRENT_ROOT}>
1432 make installworld DESTDIR=${CURRENT_ROOT}
1433 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1434 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1435 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1436 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1437 <reboot into current>
1438 <do a "native" rebuild/install as described in the previous section>
1439 <maybe install compatibility libraries from ports/misc/compat*>
1443 To upgrade in-place from stable to current
1444 ----------------------------------------------
1445 <make sure you have good level 0 dumps>
1447 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1449 <reboot in single user> [3]
1456 Make sure that you've read the UPDATING file to understand the
1457 tweaks to various things you need. At this point in the life
1458 cycle of current, things change often and you are on your own
1459 to cope. The defaults can also change, so please read ALL of
1460 the UPDATING entries.
1462 Also, if you are tracking -current, you must be subscribed to
1463 freebsd-current@freebsd.org. Make sure that before you update
1464 your sources that you have read and understood all the recent
1465 messages there. If in doubt, please track -stable which has
1466 much fewer pitfalls.
1468 [1] If you have third party modules, such as vmware, you
1469 should disable them at this point so they don't crash your
1472 [3] From the bootblocks, boot -s, and then do
1477 adjkerntz -i # if CMOS is wall time
1478 Also, when doing a major release upgrade, it is required that
1479 you boot into single user mode to do the installworld.
1481 [4] Note: This step is non-optional. Failure to do this step
1482 can result in a significant reduction in the functionality of the
1483 system. Attempting to do it by hand is not recommended and those
1484 that pursue this avenue should read this file carefully, as well
1485 as the archives of freebsd-current and freebsd-hackers mailing lists
1486 for potential gotchas. The -U option is also useful to consider.
1487 See mergemaster(8) for more information.
1489 [5] Usually this step is a noop. However, from time to time
1490 you may need to do this if you get unknown user in the following
1491 step. It never hurts to do it all the time. You may need to
1492 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1493 install) after the buildworld before this step if you last updated
1494 from current before 20020224 or from -stable before 20020408.
1496 [6] This only deletes old files and directories. Old libraries
1497 can be deleted by "make delete-old-libs", but you have to make
1498 sure that no program is using those libraries anymore.
1500 [8] In order to have a kernel that can run the 4.x binaries needed to
1501 do an installworld, you must include the COMPAT_FREEBSD4 option in
1502 your kernel. Failure to do so may leave you with a system that is
1503 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1504 required to run the 5.x binaries on more recent kernels. And so on
1505 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1507 Make sure that you merge any new devices from GENERIC since the
1508 last time you updated your kernel config file.
1510 [9] When checking out sources, you must include the -P flag to have
1511 cvs prune empty directories.
1513 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1514 "?=" instead of the "=" assignment operator, so that buildworld can
1515 override the CPUTYPE if it needs to.
1517 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1518 not on the command line, or in /etc/make.conf. buildworld will
1519 warn if it is improperly defined.
1522 This file contains a list, in reverse chronological order, of major
1523 breakages in tracking -current. It is not guaranteed to be a complete
1524 list of such breakages, and only contains entries since October 10, 2007.
1525 If you need to see UPDATING entries from before that date, you will need
1526 to fetch an UPDATING file from an older FreeBSD release.
1528 Copyright information:
1530 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1532 Redistribution, publication, translation and use, with or without
1533 modification, in full or in part, in any form or format of this
1534 document are permitted without further permission from the author.
1536 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1537 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1538 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1539 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1540 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1541 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1542 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1543 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1544 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1545 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1546 POSSIBILITY OF SUCH DAMAGE.
1548 Contact Warner Losh if you have any questions about your use of