1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
13 A new version of ZFS (pool version 5000) has been merged to 9-STABLE.
14 Starting with this version the old system of ZFS pool versioning
15 is superseded by "feature flags". This concept enables forward
16 compatibility against certain future changes in functionality of ZFS
17 pools. The first two read-only compatible "feature flags" for ZFS
18 pools are "com.delphix:async_destroy" and "com.delphix:empty_bpobj".
19 For more information read the new zpool-features(7) manual page.
20 Please refer to the "ZFS notes" section of this file for information
21 on upgrading boot ZFS pools.
24 The IPFIREWALL_FORWARD kernel option has been removed. Its
25 functionality now turned on by default.
28 The random(4) support for the VIA hardware random number
29 generator (`PADLOCK') is no longer enabled unconditionally.
30 Add the PADLOCK_RNG option in the custom kernel config if
31 needed. The GENERIC kernels on i386 and amd64 do include the
32 option, so the change only affects the custom kernel
36 The sparc64 ZFS loader has been changed to no longer try to auto-
37 detect ZFS providers based on diskN aliases but now requires these
38 to be explicitly listed in the OFW boot-device environment variable.
41 Now unix domain sockets behave "as expected" on nullfs(5). Previously
42 nullfs(5) did not pass through all behaviours to the underlying layer,
43 as a result if we bound to a socket on the lower layer we could connect
44 only to the lower path; if we bound to the upper layer we could connect
45 only to the upper path. The new behavior is one can connect to both the
46 lower and the upper paths regardless what layer path one binds to.
49 The acpi_wmi(4) status device /dev/wmistat has been renamed to
53 A new VOP_ADVISE() was added to support posix_fadvise(2). All
54 filesystem modules must be recompiled.
57 The interface of the VOP_VPTOCNP(9) changed, now the returned
58 vnode shall be referenced, previously it was required to be
59 only held. All in-tree filesystems are converted.
65 The broken amd(4) driver has been replaced with esp(4) in the amd64,
66 i386 and pc98 GENERIC kernel configuration files.
69 This commit modifies vfs_register() so that it uses a hash
70 calculation to set vfc_typenum, which is enabled by default.
71 The first time a system is booted after this change, the
72 vfc_typenum values will change for all file systems. The
73 main effect of this is a change to the NFS server file handles
74 for file systems that use vfc_typenum in their fsid, such as ZFS.
75 It will, however, prevent vfc_typenum from changing when file
76 systems are loaded in a different order for subsequent reboots.
77 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
78 until you are ready to remount all NFS clients after a reboot.
81 Bump the shared library version numbers for libraries that
82 do not use symbol versioning, have changed the ABI compared
83 to stable/8 and which shared library version was not bumped.
84 Done as part of 9.0-RELEASE cycle.
87 During the merge of Capsicum features, the fget(9) KPI was modified.
88 This may require the rebuilding of out-of-tree device drivers --
89 issues have been reported specifically with the nVidia device driver.
90 __FreeBSD_version is bumped to 900041.
92 Also, there is a period between 20110811 and 20110814 where the
93 special devices /dev/{stdin,stdout,stderr} did not work correctly.
94 Building world from a kernel during that window may not work.
97 The packet filter (pf) code has been updated to OpenBSD 4.5.
98 You need to update userland tools to be in sync with kernel.
99 This update breaks backward compatibility with earlier pfsync(4)
100 versions. Care must be taken when updating redundant firewall setups.
103 The following sysctls and tunables are retired on x86 platforms:
105 machdep.hlt_logical_cpus
106 The following sysctl is retired:
107 machdep.hyperthreading_allowed
108 The sysctls were supposed to provide a way to dynamically offline and
109 online selected CPUs on x86 platforms, but the implementation has not
110 been reliable especially with SCHED_ULE scheduler.
111 machdep.hyperthreading_allowed tunable is still available to ignore
112 hyperthreading CPUs at OS level.
113 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
114 where X is an APIC ID of a CPU. Be advised, though, that disabling
115 CPUs in non-uniform fashion will result in non-uniform topology and
116 may lead to sub-optimal system performance with SCHED_ULE, which is
120 cpumask_t type is retired and cpuset_t is used in order to describe
124 Changes to ifconfig(8) for dynamic address family detection mandate
125 that you are running a kernel of 20110525 or later. Make sure to
126 follow the update procedure to boot a new kernel before installing
130 Support for sun4v architecture is officially dropped
133 Several KPI breaking changes have been committed to the mii(4) layer,
134 the PHY drivers and consequently some Ethernet drivers using mii(4).
135 This means that miibus.ko and the modules of the affected Ethernet
136 drivers need to be recompiled.
138 Note to kernel developers: Given that the OUI bit reversion problem
139 was fixed as part of these changes all mii(4) commits related to OUIs,
140 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
141 handling, no longer can be merged verbatim to stable/8 and previous
145 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
146 into their kernel configurations along with 'device pci'.
149 The default NFS client is now the new NFS client, so fstype "newnfs"
150 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
151 Although mounts via fstype "nfs" will usually work without userland
152 changes, it is recommended that the mount(8) and mount_nfs(8)
153 commands be rebuilt from sources and that a link to mount_nfs called
154 mount_oldnfs be created. The new client is compiled into the
155 kernel with "options NFSCL" and this is needed for diskless root
156 file systems. The GENERIC kernel configs have been changed to use
157 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
158 To use the regular/old client, you can "mount -t oldnfs ...". For
159 a diskless root file system, you must also include a line like:
161 vfs.root.mountfrom="oldnfs:"
163 in the boot/loader.conf on the root fs on the NFS server to make
164 a diskless root fs use the old client.
167 The GENERIC kernels for all architectures now default to the new
168 CAM-based ATA stack. It means that all legacy ATA drivers were
169 removed and replaced by respective CAM drivers. If you are using
170 ATA device names in /etc/fstab or other places, make sure to update
171 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
172 where 'Y's are the sequential numbers starting from zero for each type
173 in order of detection, unless configured otherwise with tunables,
174 see cam(4)). There will be symbolic links created in /dev/ to map
175 old adX devices to the respective adaY. They should provide basic
176 compatibility for file systems mounting in most cases, but they do
177 not support old user-level APIs and do not have respective providers
178 in GEOM. Consider using updated management tools with new device names.
180 It is possible to load devices ahci, ata, siis and mvs as modules,
181 but option ATA_CAM should remain in kernel configuration to make ata
182 module work as CAM driver supporting legacy ATA controllers. Device ata
183 still can be used in modular fashion (atacore + ...). Modules atadisk
184 and atapi* are not used and won't affect operation in ATA_CAM mode.
185 Note that to use CAM-based ATA kernel should include CAM devices
186 scbus, pass, da (or explicitly ada), cd and optionally others. All of
187 them are parts of the cam module.
189 ataraid(4) functionality is now supported by the RAID GEOM class.
190 To use it you can load geom_raid kernel module and use graid(8) tool
191 for management. Instead of /dev/arX device names, use /dev/raid/rX.
193 No kernel config options or code have been removed, so if a problem
194 arises, please report it and optionally revert to the old ATA stack.
195 In order to do it you can remove from the kernel config:
200 , and instead add back:
201 device atadisk # ATA disk drives
202 device ataraid # ATA RAID drives
203 device atapicd # ATAPI CDROM drives
204 device atapifd # ATAPI floppy drives
205 device atapist # ATAPI tape drives
208 The default NFS server has been changed to the new server, which
209 was referred to as the experimental server. If you need to switch
210 back to the old NFS server, you must now put the "-o" option on
211 both the mountd and nfsd commands. This can be done using the
212 mountd_flags and nfs_server_flags rc.conf variables until an
213 update to the rc scripts is committed, which is coming soon.
216 The GNU Objective-C runtime library (libobjc), and other Objective-C
217 related components have been removed from the base system. If you
218 require an Objective-C library, please use one of the available ports.
221 ath(4) has been split into bus- and device- modules. if_ath contains
222 the HAL, the TX rate control and the network device code. if_ath_pci
223 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
224 contains the AHB glue. Users need to load both if_ath_pci and if_ath
225 in order to use ath on everything else.
227 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
228 need to load if_ath and if_ath_pci for ath(4) operation.
231 As part of the replacement of sysinstall, the process of building
232 release media has changed significantly. For details, please re-read
233 release(7), which has been updated to reflect the new build process.
236 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
237 is the last available version under GPLv2. It brings a number of new
238 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
239 4.1 and SSE 4.2), better support for powerpc64, a number of new
240 directives, and lots of other small improvements. See the ChangeLog
241 file in contrib/binutils for the full details.
244 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
245 compliant, and will now use half of hash for authentication.
246 This will break interoperability with all stacks (including all
247 actual FreeBSD versions) who implement
248 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
250 The only workaround with such peers is to use another HMAC
251 algorithm for IPsec ("phase 2") authentication.
254 Remove the uio_yield prototype and symbol. This function has
255 been misnamed since it was introduced and should not be
256 globally exposed with this name. The equivalent functionality
257 is now available using kern_yield(curthread->td_user_pri).
258 The function remains undocumented.
261 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
262 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
263 for scalar sysctls is defined but disabled. Code that needs
264 UQUAD to pass the type checking that must compile on older
265 systems where the define is not present can check against
266 __FreeBSD_version >= 900030.
268 The system dialog(1) has been replaced with a new version previously
269 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
270 with the previous version, but the libdialog associated with it has
271 a largely incompatible API. As such, the original version of libdialog
272 will be kept temporarily as libodialog, until its base system consumers
273 are replaced or updated. Bump __FreeBSD_version to 900030.
276 If you are trying to run make universe on a -stable system, and you get
277 the following warning:
278 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
279 unknown. config(8) likely too old."
280 or something similar to it, then you must upgrade your -stable system
281 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
282 stable/8) or build the config from the latest stable/8 branch and
283 install it on your system.
285 Prior to this date, building a current universe on 8-stable system from
286 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
287 error in the first kernel build phase. A new config on those old
288 systems will fix that problem for older versions of -current.
291 The TCP stack has been modified to allow Khelp modules to interact with
292 it via helper hook points and store per-connection data in the TCP
293 control block. Bump __FreeBSD_version to 900029. User space tools that
294 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
298 Generic IEEE 802.3 annex 31B full duplex flow control support has been
299 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
300 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
301 to take advantage of it instead of using custom implementations. This
302 means that these drivers now no longer unconditionally advertise
303 support for flow control but only do so if flow control is a selected
304 media option. This was implemented in the generic support that way in
305 order to allow flow control to be switched on and off via ifconfig(8)
306 with the PHY specific default to typically off in order to protect
307 from unwanted effects. Consequently, if you used flow control with
308 one of the above mentioned drivers you now need to explicitly enable
310 ifconfig bge0 media auto mediaopt flowcontrol
312 Along with the above mentioned changes generic support for setting
313 1000baseT master mode also has been added and brgphy(4), ciphy(4),
314 e1000phy(4) as well as ip1000phy(4) have been converted to take
315 advantage of it. This means that these drivers now no longer take the
316 link0 parameter for selecting master mode but the master media option
317 has to be used instead, for example like in the following:
318 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
320 Selection of master mode now is also available with all other PHY
321 drivers supporting 1000baseT.
324 The TCP stack has received a significant update to add support for
325 modularised congestion control and generally improve the clarity of
326 congestion control decisions. Bump __FreeBSD_version to 900025. User
327 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
328 sockstat) need to be recompiled.
331 The man(1) utility has been replaced by a new version that no longer
332 uses /etc/manpath.config. Please consult man.conf(5) for how to
333 migrate local entries to the new format.
336 The copyright strings printed by login(1) and sshd(8) at the time of a
337 new connection have been removed to follow other operating systems and
341 A workaround for a fixed ld bug has been removed in kernel code,
342 so make sure that your system ld is built from sources after
343 revision 210245 from 2010-07-19 (r211583 if building head kernel
344 on stable/8, r211584 for stable/7; both from 2010-08-21).
345 A symptom of incorrect ld version is different addresses for
346 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
349 The $ipv6_prefer variable in rc.conf(5) has been split into
350 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
352 The $ip6addrctl_policy is a variable to choose a pre-defined
353 address selection policy set by ip6addrctl(8). A value
354 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
357 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
358 flag (see an entry of 20090926) is set on an interface with no
359 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
360 security reason. If you want IPv6 link-local address on all
361 interfaces by default, set this to "YES".
363 The old ipv6_prefer="YES" is equivalent to
364 ipv6_activate_all_interfaces="YES" and
365 ip6addrctl_policy="ipv6_prefer".
368 DTrace has grown support for userland tracing. Due to this, DTrace is
369 now i386 and amd64 only.
370 dtruss(1) is now installed by default on those systems and a new
371 kernel module is needed for userland tracing: fasttrap.
372 No changes to your kernel config file are necessary to enable
373 userland tracing, but you might consider adding 'STRIP=' and
374 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
375 to have informative userland stack traces in DTrace (ustack).
378 The acpi_aiboost(4) driver has been removed in favor of the new
379 aibs(4) driver. You should update your kernel configuration file.
382 BSD grep has been imported to the base system and it is built by
383 default. It is completely BSD licensed, highly GNU-compatible, uses
384 less memory than its GNU counterpart and has a small codebase.
385 However, it is slower than its GNU counterpart, which is mostly
386 noticeable for larger searches, for smaller ones it is measurable
387 but not significant. The reason is complex, the most important factor
388 is that we lack a modern and efficient regex library and GNU
389 overcomes this by optimizing the searches internally. Future work
390 on improving the regex performance is planned, for the meantime,
391 users that need better performance, can build GNU grep instead by
392 setting the WITH_GNU_GREP knob.
395 Due to the import of powerpc64 support, all existing powerpc kernel
396 configuration files must be updated with a machine directive like this:
397 machine powerpc powerpc
399 In addition, an updated config(8) is required to build powerpc kernels
403 A new version of ZFS (version 15) has been merged to -HEAD.
404 This version uses a python library for the following subcommands:
405 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
406 For full functionality of these commands the following port must
407 be installed: sysutils/py-zfs
410 'vm_page's are now hashed by physical address to an array of mutexes.
411 Currently this is only used to serialize access to hold_count. Over
412 time the page queue mutex will be peeled away. This changes the size
413 of pmap on every architecture. And requires all callers of vm_page_hold
414 and vm_page_unhold to be updated.
417 WITH_CTF can now be specified in src.conf (not recommended, there
418 are some problems with static executables), make.conf (would also
419 affect ports which do not use GNU make and do not override the
420 compile targets) or in the kernel config (via "makeoptions
422 When WITH_CTF was specified there before this was silently ignored,
423 so make sure that WITH_CTF is not used in places which could lead
424 to unwanted behavior.
427 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
428 to allow 32-bit compatibility on non-x86 platforms. All kernel
429 configurations on amd64 and ia64 platforms using these options must
430 be modified accordingly.
433 The utmp user accounting database has been replaced with utmpx,
434 the user accounting interface standardized by POSIX.
435 Unfortunately the semantics of utmp and utmpx don't match,
436 making it practically impossible to support both interfaces.
437 The user accounting database is used by tools like finger(1),
438 last(1), talk(1), w(1) and ac(8).
440 All applications in the base system use utmpx. This means only
441 local binaries (e.g. from the ports tree) may still use these
442 utmp database files. These applications must be rebuilt to make
445 After the system has been upgraded, it is safe to remove the old
446 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
447 assuming their contents is of no importance anymore. Old wtmp
448 databases can only be used by last(1) and ac(8) after they have
449 been converted to the new format using wtmpcvt(1).
452 Introduce the kernel thread "deadlock resolver" (which can be enabled
453 via the DEADLKRES option, see NOTES for more details) and the
454 sleepq_type() function for sleepqueues.
457 The rc.firewall and rc.firewall6 were unified, and
458 rc.firewall6 and rc.d/ip6fw were removed.
459 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
460 variables are obsoleted. Instead, the following new rc
461 variables are added to rc.d/ipfw:
463 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
464 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
465 firewall_simple_onet_ipv6, firewall_trusted_ipv6
467 The meanings correspond to the relevant IPv4 variables.
473 The default terminal emulation for syscons(4) has been changed
474 from cons25 to xterm on all platforms except pc98. This means
475 that the /etc/ttys file needs to be updated to ensure correct
476 operation of applications on the console.
478 The terminal emulation style can be toggled per window by using
479 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
480 options can be used to change the compile-time default back to
483 To prevent graphical artifacts, make sure the TERM environment
484 variable is set to match the terminal emulation that is being
485 performed by syscons(4).
488 The layout of the structure ieee80211req_scan_result has changed.
489 Applications that require wireless scan results (e.g. ifconfig(8))
490 from net80211 need to be recompiled.
492 Applications such as wpa_supplicant(8) may require a full world
493 build without using NO_CLEAN in order to get synchronized with the
497 The iwn(4) driver has been updated to support the 5000 and 5150 series.
498 There's one kernel module for each firmware. Adding "device iwnfw"
499 to the kernel configuration file means including all three firmware
500 images inside the kernel. If you want to include just the one for
501 your wireless card, use the devices iwn4965fw, iwn5000fw or
505 The rc.d/network_ipv6, IPv6 configuration script has been integrated
506 into rc.d/netif. The changes are the following:
508 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
509 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
510 Note that both variables need the "inet6" keyword at the head.
512 Do not set $ipv6_network_interfaces manually if you do not
513 understand what you are doing. It is not needed in most cases.
515 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
518 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
519 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
521 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
522 all configured interfaces have "inet6 accept_rtadv" in the
523 $ifconfig_IF_ipv6. These are for backward compatibility.
525 3. A new variable $ipv6_prefer has been added. If NO, IPv6
526 functionality of interfaces with no corresponding
527 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
528 and the default address selection policy of ip6addrctl(8)
529 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
530 Note that if you want to configure IPv6 functionality on the
531 disabled interfaces after boot, first you need to clear the flag by
532 using ifconfig(8) like:
534 ifconfig em0 inet6 -ifdisabled
536 If YES, the default address selection policy is set as
539 The default value of $ipv6_prefer is NO.
541 4. If your system need to receive Router Advertisement messages,
542 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
543 scripts automatically invoke rtsol(8) when the interface becomes
544 UP. The Router Advertisement messages are used for SLAAC
545 (State-Less Address AutoConfiguration).
548 802.11s D3.03 support was committed. This is incompatible with the
549 previous code, which was based on D3.0.
552 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
553 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
554 control whether accepting Router Advertisement messages or not.
555 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
556 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
557 The ifconfig(8) utility now supports these flags.
560 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
561 mount(8) and -a option for df(1) to see them.
564 The old tunable hw.bus.devctl_disable has been superseded by
565 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
566 replaced by hw.bus.devctl_queue=0. The default for this new tunable
570 Remove the option STOP_NMI. The default action is now to use NMI only
571 for KDB via the newly introduced function stop_cpus_hard() and
572 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
575 The stable/8 branch created in subversion. This corresponds to the
576 RELENG_8 branch in CVS.
579 Bump the shared library version numbers for all libraries that do not
580 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
581 __FreeBSD_version to 800105.
584 Due to changes in the implementation of virtual network stack support,
585 all network-related kernel modules must be recompiled. As this change
586 breaks the ABI, bump __FreeBSD_version to 800104.
589 The TOE interface to the TCP syncache has been modified to remove
590 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
591 The cxgb driver is the only TOE consumer affected by this change, and
592 needs to be recompiled along with the kernel. As this change breaks
593 the ABI, bump __FreeBSD_version to 800103.
596 Padding has been added to struct tcpcb, sackhint and tcpstat in
597 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
598 maintaining the ABI. However, this change breaks the ABI, so bump
599 __FreeBSD_version to 800102. User space tools that rely on the size of
600 any of these structs (e.g. sockstat) need to be recompiled.
603 The NFS_LEGACYRPC option has been removed along with the old kernel
604 RPC implementation that this option selected. Kernel configurations
605 may need to be adjusted.
608 The network interface device nodes at /dev/net/<interface> have been
609 removed. All ioctl operations can be performed the normal way using
610 routing sockets. The kqueue functionality can generally be replaced
611 with routing sockets.
614 The documentation from the FreeBSD Documentation Project (Handbook,
615 FAQ, etc.) is now installed via packages by sysinstall(8) and under
616 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
619 The ABI of various structures related to the SYSV IPC API have been
620 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
621 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
625 Layout of struct vnet has changed as routing related variables were
626 moved to their own Vimage module. Modules need to be recompiled. Bump
627 __FreeBSD_version to 800099.
630 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
631 respectively. As long as no more than 16 groups per process are used,
632 no changes should be visible. When more than 16 groups are used, old
633 binaries may fail if they call getgroups() or getgrouplist() with
634 statically sized storage. Recompiling will work around this, but
635 applications should be modified to use dynamically allocated storage
636 for group arrays as POSIX.1-2008 does not cap an implementation's
637 number of supported groups at NGROUPS_MAX+1 as previous versions did.
639 NFS and portalfs mounts may also be affected as the list of groups is
640 truncated to 16. Users of NFS who use more than 16 groups, should
641 take care that negative group permissions are not used on the exported
642 file systems as they will not be reliable unless a GSSAPI based
643 authentication method is used.
646 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
647 option compiles in the support for adaptive spinning for lockmgrs
648 which want to enable it. The lockinit() function now accepts the flag
649 LK_ADAPTIVE in order to make the lock object subject to adaptive
650 spinning when both held in write and read mode.
653 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
654 changed. User applications that use this ioctl need to be rebuilt.
657 The layout of struct thread has changed. Kernel and modules need to
661 The layout of structs ifnet, domain, protosw and vnet_net has changed.
662 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
665 window(1) has been removed from the base system. It can now be
666 installed from ports. The port is called misc/window.
669 The way we are storing and accessing `routing table' entries has
670 changed. Programs reading the FIB, like netstat, need to be
674 A new netisr implementation has been added for FreeBSD 8. Network
675 file system modules, such as igmp, ipdivert, and others, should be
677 Bump __FreeBSD_version to 800096.
680 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
684 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
685 Bump __FreeBSD_version to 800094.
688 Add mnt_xflag field to 'struct mount'. File system modules need to be
690 Bump __FreeBSD_version to 800093.
693 The compiling option ADAPTIVE_SX has been retired while it has been
694 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
695 The KPI for sx_init_flags() changes as accepting flags:
696 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
697 been introduced in order to handle the reversed logic.
698 Bump __FreeBSD_version to 800092.
701 Add support for hierarchical jails. Remove global securelevel.
702 Bump __FreeBSD_version to 800091.
705 The layout of struct vnet_net has changed, therefore modules
707 Bump __FreeBSD_version to 800090.
710 The newly imported zic(8) produces a new format in the output. Please
711 run tzsetup(8) to install the newly created data to /etc/localtime.
714 The sysctl tree for the usb stack has renamed from hw.usb2.* to
715 hw.usb.* and is now consistent again with previous releases.
718 802.11 monitor mode support was revised and driver api's were changed.
719 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
720 of DLT_IEEE802_11. No user-visible data structures were changed but
721 applications that use DLT_IEEE802_11 may require changes.
722 Bump __FreeBSD_version to 800088.
725 The layout of the following structs has changed: sysctl_oid,
726 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
727 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
728 panics may be experienced. World rebuild is required for
729 correctly checking networking state from userland.
730 Bump __FreeBSD_version to 800085.
733 MLDv2 and Source-Specific Multicast (SSM) have been merged
734 to the IPv6 stack. VIMAGE hooks are in but not yet used.
735 The implementation of SSM within FreeBSD's IPv6 stack closely
736 follows the IPv4 implementation.
738 For kernel developers:
740 * The most important changes are that the ip6_output() and
741 ip6_input() paths no longer take the IN6_MULTI_LOCK,
742 and this lock has been downgraded to a non-recursive mutex.
744 * As with the changes to the IPv4 stack to support SSM, filtering
745 of inbound multicast traffic must now be performed by transport
746 protocols within the IPv6 stack. This does not apply to TCP and
747 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
749 * The KPIs used by IPv6 multicast are similar to those used by
750 the IPv4 stack, with the following differences:
751 * im6o_mc_filter() is analogous to imo_multicast_filter().
752 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
753 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
754 * IN6_LOOKUP_MULTI() has been deprecated and removed.
755 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
756 for MLDv1 have an additional 'timer' argument which is used to
757 jitter the initial membership report for the solicited-node
758 multicast membership on-link.
759 * This is not strictly needed for MLDv2, which already jitters
760 its report transmissions. However, the 'timer' argument is
761 preserved in case MLDv1 is active on the interface.
763 * The KAME linked-list based IPv6 membership implementation has
764 been refactored to use a vector similar to that used by the IPv4
766 Code which maintains a list of its own multicast memberships
767 internally, e.g. carp, has been updated to reflect the new
770 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
771 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
772 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
773 implementation constraint which needs to be addressed in HEAD.
775 For application developers:
777 * The changes are broadly similar to those made for the IPv4
780 * The use of IPv4 and IPv6 multicast socket options on the same
781 socket, using mapped addresses, HAS NOT been tested or supported.
783 * There are a number of issues with the implementation of various
784 IPv6 multicast APIs which need to be resolved in the API surface
785 before the implementation is fully compatible with KAME userland
786 use, and these are mostly to do with interface index treatment.
788 * The literature available discusses the use of either the delta / ASM
789 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
790 using setsourcefilter(3)/getsourcefilter(3). For more information
791 please refer to RFC 3768, 'Socket Interface Extensions for
792 Multicast Source Filters'.
794 * Applications which use the published RFC 3678 APIs should be fine.
796 For systems administrators:
798 * The mtest(8) utility has been refactored to support IPv6, in
799 addition to IPv4. Interface addresses are no longer accepted
800 as arguments, their names must be used instead. The utility
801 will map the interface name to its first IPv4 address as
802 returned by getifaddrs(3).
804 * The ifmcstat(8) utility has also been updated to print the MLDv2
805 endpoint state and source filter lists via sysctl(3).
807 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
808 loopback of IPv6 multicast datagrams by default; it defaults to 1
809 to preserve the existing behaviour. Disabling multicast loopback is
810 recommended for optimal system performance.
812 * The IPv6 MROUTING code has been changed to examine this sysctl
813 instead of attempting to perform a group lookup before looping
814 back forwarded datagrams.
816 Bump __FreeBSD_version to 800084.
819 Implement low-level Bluetooth HCI API.
820 Bump __FreeBSD_version to 800083.
823 The layout of struct malloc_type, used by modules to register new
824 memory allocation types, has changed. Most modules will need to
825 be rebuilt or panics may be experienced.
826 Bump __FreeBSD_version to 800081.
829 Anticipate overflowing inp_flags - add inp_flags2.
830 This changes most offsets in inpcb, so checking v4 connection
831 state will require a world rebuild.
832 Bump __FreeBSD_version to 800080.
835 Add an llentry to struct route and struct route_in6. Modules
836 embedding a struct route will need to be recompiled.
837 Bump __FreeBSD_version to 800079.
840 The size of rt_metrics_lite and by extension rtentry has changed.
841 Networking administration apps will need to be recompiled.
842 The route command now supports show as an alias for get, weighting
843 of routes, sticky and nostick flags to alter the behavior of stateful
845 Bump __FreeBSD_version to 800078.
848 Do not use Giant for kbdmux(4) locking. This is wrong and
849 apparently causing more problems than it solves. This will
850 re-open the issue where interrupt handlers may race with
851 kbdmux(4) in polling mode. Typical symptoms include (but
852 not limited to) duplicated and/or missing characters when
853 low level console functions (such as gets) are used while
854 interrupts are enabled (for example geli password prompt,
855 mountroot prompt etc.). Disabling kbdmux(4) may help.
858 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
859 kernel modules referencing any of the above need to be recompiled.
860 Bump __FreeBSD_version to 800075.
863 GEOM_PART has become the default partition slicer for storage devices,
864 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
865 introduces some changes:
867 MSDOS/EBR: the devices created from MSDOS extended partition entries
868 (EBR) can be named differently than with GEOM_MBR and are now symlinks
869 to devices with offset-based names. fstabs may need to be modified.
871 BSD: the "geometry does not match label" warning is harmless in most
872 cases but it points to problems in file system misalignment with
873 disk geometry. The "c" partition is now implicit, covers the whole
874 top-level drive and cannot be (mis)used by users.
876 General: Kernel dumps are now not allowed to be written to devices
877 whose partition types indicate they are meant to be used for file
878 systems (or, in case of MSDOS partitions, as something else than
881 Most of these changes date approximately from 200812.
884 The uscanner(4) driver has been removed from the kernel. This follows
885 Linux removing theirs in 2.6 and making libusb the default interface
889 The multicast forwarding code has been cleaned up. netstat(1)
890 only relies on KVM now for printing bandwidth upcall meters.
891 The IPv4 and IPv6 modules are split into ip_mroute_mod and
892 ip6_mroute_mod respectively. The config(5) options for statically
893 compiling this code remain the same, i.e. 'options MROUTING'.
896 Support for the IFF_NEEDSGIANT network interface flag has been
897 removed, which means that non-MPSAFE network device drivers are no
898 longer supported. In particular, if_ar, if_sr, and network device
899 drivers from the old (legacy) USB stack can no longer be built or
903 POSIX.1 Native Language Support (NLS) has been enabled in libc and
904 a bunch of new language catalog files have also been added.
905 This means that some common libc messages are now localized and
906 they depend on the LC_MESSAGES environmental variable.
909 The k8temp(4) driver has been renamed to amdtemp(4) since
910 support for Family 10 and Family 11 CPU families was added.
913 IGMPv3 and Source-Specific Multicast (SSM) have been merged
914 to the IPv4 stack. VIMAGE hooks are in but not yet used.
916 For kernel developers, the most important changes are that the
917 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
918 and this lock has been downgraded to a non-recursive mutex.
920 Transport protocols (UDP, Raw IP) are now responsible for filtering
921 inbound multicast traffic according to group membership and source
922 filters. The imo_multicast_filter() KPI exists for this purpose.
923 Transports which do not use multicast (SCTP, TCP) already reject
924 multicast by default. Forwarding and receive performance may improve
925 as a mutex acquisition is no longer needed in the ip_input()
926 low-level input path. in_addmulti() and in_delmulti() are shimmed
927 to new KPIs which exist to support SSM in-kernel.
929 For application developers, it is recommended that loopback of
930 multicast datagrams be disabled for best performance, as this
931 will still cause the lock to be taken for each looped-back
932 datagram transmission. The net.inet.ip.mcast.loop sysctl may
933 be tuned to 0 to disable loopback by default; it defaults to 1
934 to preserve the existing behaviour.
936 For systems administrators, to obtain best performance with
937 multicast reception and multiple groups, it is always recommended
938 that a card with a suitably precise hash filter is used. Hash
939 collisions will still result in the lock being taken within the
940 transport protocol input path to check group membership.
942 If deploying FreeBSD in an environment with IGMP snooping switches,
943 it is recommended that the net.inet.igmp.sendlocal sysctl remain
944 enabled; this forces 224.0.0.0/24 group membership to be announced
947 The size of 'struct igmpstat' has changed; netstat needs to be
948 recompiled to reflect this.
949 Bump __FreeBSD_version to 800070.
952 libusb20.so.1 is now installed as libusb.so.1 and the ports system
953 updated to use it. This requires a buildworld/installworld in order to
954 update the library and dependencies (usbconfig, etc). Its advisable to
955 rebuild all ports which uses libusb. More specific directions are given
956 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
957 libusb are no longer required and can be removed.
960 A workaround is committed to allow the creation of System V shared
961 memory segment of size > 2 GB on the 64-bit architectures.
962 Due to a limitation of the existing ABI, the shm_segsz member
963 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
964 wrong for large segments. Note that limits must be explicitly
965 raised to allow such segments to be created.
968 The layout of struct ifnet has changed, requiring a rebuild of all
969 network device driver modules.
972 The /dev handling for the new USB stack has changed, a
973 buildworld/installworld is required for libusb20.
976 The new USB2 stack has now been permanently moved in and all kernel and
977 module names reverted to their previous values (eg, usb, ehci, ohci,
978 ums, ...). The old usb stack can be compiled in by prefixing the name
979 with the letter 'o', the old usb modules have been removed.
980 Updating entry 20090216 for xorg and 20090215 for libmap may still
984 The rc.conf(5) option if_up_delay has been renamed to
985 defaultroute_delay to better reflect its purpose. If you have
986 customized this setting in /etc/rc.conf you need to update it to
990 xorg 7.4 wants to configure its input devices via hald which does not
991 yet work with USB2. If the keyboard/mouse does not work in xorg then
993 Option "AllowEmptyInput" "off"
994 to your ServerLayout section. This will cause X to use the configured
995 kbd and mouse sections from your xorg.conf.
998 The GENERIC kernels for all architectures now default to the new USB2
999 stack. No kernel config options or code have been removed so if a
1000 problem arises please report it and optionally revert to the old USB
1001 stack. If you are loading USB kernel modules or have a custom kernel
1002 that includes GENERIC then ensure that usb names are also changed over,
1003 eg uftdi -> usb2_serial_ftdi.
1005 Older programs linked against the ports libusb 0.1 need to be
1006 redirected to the new stack's libusb20. /etc/libmap.conf can
1008 # Map old usb library to new one for usb2 stack
1009 libusb-0.1.so.8 libusb20.so.1
1012 All USB ethernet devices now attach as interfaces under the name ueN
1013 (eg. ue0). This is to provide a predictable name as vendors often
1014 change usb chipsets in a product without notice.
1017 The ichsmb(4) driver has been changed to require SMBus slave
1018 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1019 All of the other SMBus controller drivers require left-justified
1020 slave addresses, so this change makes all the drivers provide the
1024 INET6 statistics (struct ip6stat) was updated.
1025 netstat(1) needs to be recompiled.
1028 NTFS has been removed from GENERIC kernel on amd64 to match
1029 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1030 will load ntfs.ko module automatically when NTFS support is
1031 actually needed, unless ntfs.ko is not installed or security
1032 level prohibits loading kernel modules. If either is the case,
1033 "options NTFS" has to be added into kernel config.
1036 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1037 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1038 800061. User space tools that rely on the size of struct tcpcb in
1039 tcp_var.h (e.g. sockstat) need to be recompiled.
1042 ng_tty(4) module updated to match the new TTY subsystem.
1043 Due to API change, user-level applications must be updated.
1044 New API support added to mpd5 CVS and expected to be present
1045 in next mpd5.3 release.
1048 With __FreeBSD_version 800060 the makefs tool is part of
1049 the base system (it was a port).
1052 The afdata and ifnet locks have been changed from mutexes to
1053 rwlocks, network modules will need to be re-compiled.
1056 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1057 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1058 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1059 architecture and 40 bytes on 64-bit architecture. The userland
1060 applications "arp" and "ndp" have been updated accordingly.
1061 The output from "netstat -r" shows only routing entries and
1062 none of the L2 information.
1065 __FreeBSD_version 800057 marks the switchover from the
1066 binary ath hal to source code. Users must add the line:
1068 options AH_SUPPORT_AR5416
1070 to their kernel config files when specifying:
1074 The ath_hal module no longer exists; the code is now compiled
1075 together with the driver in the ath module. It is now
1076 possible to tailor chip support (i.e. reduce the set of chips
1077 and thereby the code size); consult ath_hal(4) for details.
1080 __FreeBSD_version 800054 adds memory barriers to
1081 <machine/atomic.h>, new interfaces to ifnet to facilitate
1082 multiple hardware transmit queues for cards that support
1083 them, and a lock-less ring-buffer implementation to
1084 enable drivers to more efficiently manage queueing of
1088 A new version of ZFS (version 13) has been merged to -HEAD.
1089 This version has zpool attribute "listsnapshots" off by
1090 default, which means "zfs list" does not show snapshots,
1091 and is the same as Solaris behavior.
1094 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1097 The uhci, ohci, ehci and slhci USB Host controller drivers have
1098 been put into separate modules. If you load the usb module
1099 separately through loader.conf you will need to load the
1100 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1101 controller add the following to loader.conf:
1107 The ABI used by the PMC toolset has changed. Please keep
1108 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1112 atapci kernel module now includes only generic PCI ATA
1113 driver. AHCI driver moved to ataahci kernel module.
1114 All vendor-specific code moved into separate kernel modules:
1115 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1116 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1117 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1118 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1121 The TTY subsystem of the kernel has been replaced by a new
1122 implementation, which provides better scalability and an
1123 improved driver model. Most common drivers have been migrated to
1124 the new TTY subsystem, while others have not. The following
1125 drivers have not yet been ported to the new TTY layer:
1128 cy, digi, rc, rp, sio
1134 ng_h4, ng_tty, ppp, sl, snp
1136 Adding these drivers to your kernel configuration file shall
1137 cause compilation to fail.
1140 ntpd has been upgraded to 4.2.4p5.
1143 OpenSSH has been upgraded to 5.1p1.
1145 For many years, FreeBSD's version of OpenSSH preferred DSA
1146 over RSA for host and user authentication keys. With this
1147 upgrade, we've switched to the vendor's default of RSA over
1148 DSA. This may cause upgraded clients to warn about unknown
1149 host keys even for previously known hosts. Users should
1150 follow the usual procedure for verifying host keys before
1151 accepting the RSA key.
1153 This can be circumvented by setting the "HostKeyAlgorithms"
1154 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1157 Please note that the sequence of keys offered for
1158 authentication has been changed as well. You may want to
1159 specify IdentityFile in a different order to revert this
1163 The sio(4) driver has been removed from the i386 and amd64
1164 kernel configuration files. This means uart(4) is now the
1165 default serial port driver on those platforms as well.
1167 To prevent collisions with the sio(4) driver, the uart(4) driver
1168 uses different names for its device nodes. This means the
1169 onboard serial port will now most likely be called "ttyu0"
1170 instead of "ttyd0". You may need to reconfigure applications to
1171 use the new device names.
1173 When using the serial port as a boot console, be sure to update
1174 /boot/device.hints and /etc/ttys before booting the new kernel.
1175 If you forget to do so, you can still manually specify the hints
1176 at the loader prompt:
1178 set hint.uart.0.at="isa"
1179 set hint.uart.0.port="0x3F8"
1180 set hint.uart.0.flags="0x10"
1181 set hint.uart.0.irq="4"
1185 The gpt(8) utility has been removed. Use gpart(8) to partition
1189 The version that Linuxulator emulates was changed from 2.4.2
1190 to 2.6.16. If you experience any problems with Linux binaries
1191 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1192 if it fixes the problem contact emulation mailing list.
1195 ISDN4BSD (I4B) was removed from the src tree. You may need to
1196 update a your kernel configuration and remove relevant entries.
1199 I have checked in code to support multiple routing tables.
1200 See the man pages setfib(1) and setfib(2).
1201 This is a hopefully backwards compatible version,
1202 but to make use of it you need to compile your kernel
1203 with options ROUTETABLES=2 (or more up to 16).
1206 The 802.11 wireless support was redone to enable multi-bss
1207 operation on devices that are capable. The underlying device
1208 is no longer used directly but instead wlanX devices are
1209 cloned with ifconfig. This requires changes to rc.conf files.
1210 For example, change:
1211 ifconfig_ath0="WPA DHCP"
1214 ifconfig_wlan0="WPA DHCP"
1215 see rc.conf(5) for more details. In addition, mergemaster of
1216 /etc/rc.d is highly recommended. Simultaneous update of userland
1217 and kernel wouldn't hurt either.
1219 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1220 modules were merged into the base wlan module. All references
1221 to these modules (e.g. in kernel config files) must be removed.
1224 psm(4) has gained write(2) support in native operation level.
1225 Arbitrary commands can be written to /dev/psm%d and status can
1226 be read back from it. Therefore, an application is responsible
1227 for status validation and error recovery. It is a no-op in
1228 other operation levels.
1231 Support for KSE threading has been removed from the kernel. To
1232 run legacy applications linked against KSE libmap.conf may
1233 be used. The following libmap.conf may be used to ensure
1234 compatibility with any prior release:
1236 libpthread.so.1 libthr.so.1
1237 libpthread.so.2 libthr.so.2
1238 libkse.so.3 libthr.so.3
1241 The layout of struct vmspace has changed. This affects libkvm
1242 and any executables that link against libkvm and use the
1243 kvm_getprocs() function. In particular, but not exclusively,
1244 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1245 The effects are minimal, but it's advisable to upgrade world
1249 The latest em driver no longer has support in it for the
1250 82575 adapter, this is now moved to the igb driver. The
1251 split was done to make new features that are incompatible
1252 with older hardware easier to do.
1255 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1256 likewise the kernel option is now GEOM_LINUX_LVM.
1259 The default NFS mount mode has changed from UDP to TCP for
1260 increased reliability. If you rely on (insecurely) NFS
1261 mounting across a firewall you may need to update your
1265 Belatedly note the addition of m_collapse for compacting
1269 The fts(3) structures have been changed to use adequate
1270 integer types for their members and so to be able to cope
1271 with huge file trees. The old fts(3) ABI is preserved
1272 through symbol versioning in libc, so third-party binaries
1273 using fts(3) should still work, although they will not take
1274 advantage of the extended types. At the same time, some
1275 third-party software might fail to build after this change
1276 due to unportable assumptions made in its source code about
1277 fts(3) structure members. Such software should be fixed
1278 by its vendor or, in the worst case, in the ports tree.
1279 FreeBSD_version 800015 marks this change for the unlikely
1280 case that a portable fix is impossible.
1283 To upgrade to -current after this date, you must be running
1284 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1285 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1288 The ADAPTIVE_GIANT kernel option has been retired because its
1289 functionality is the default now.
1292 The AT keyboard emulation of sunkbd(4) has been turned on
1293 by default. In order to make the special symbols of the Sun
1294 keyboards driven by sunkbd(4) work under X these now have
1295 to be configured the same way as Sun USB keyboards driven
1296 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1298 Option "XkbLayout" "us"
1299 Option "XkbRules" "xorg"
1300 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1303 It has been decided that it is desirable to provide ABI
1304 backwards compatibility to the FreeBSD 4/5/6 versions of the
1305 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1306 broken with the introduction of PCI domain support (see the
1307 20070930 entry). Unfortunately, this required the ABI of
1308 PCIOCGETCONF to be broken again in order to be able to
1309 provide backwards compatibility to the old version of that
1310 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1311 again. As for prominent ports this affects neither pciutils
1312 nor xorg-server this time, the hal port needs to be rebuilt
1316 The misnamed kthread_create() and friends have been renamed
1317 to kproc_create() etc. Many of the callers already
1318 used kproc_start()..
1319 I will return kthread_create() and friends in a while
1320 with implementations that actually create threads, not procs.
1321 Renaming corresponds with version 800002.
1330 Avoid using make -j when upgrading. While generally safe, there are
1331 sometimes problems using -j to upgrade. If your upgrade fails with
1332 -j, please try again without -j. From time to time in the past there
1333 have been problems using -j with buildworld and/or installworld. This
1334 is especially true when upgrading between "distant" versions (eg one
1335 that cross a major release boundary or several minor releases, or when
1336 several months have passed on the -current branch).
1338 Sometimes, obscure build problems are the result of environment
1339 poisoning. This can happen because the make utility reads its
1340 environment when searching for values for global variables. To run
1341 your build attempts in an "environmental clean room", prefix all make
1342 commands with 'env -i '. See the env(1) manual page for more details.
1344 When upgrading from one major version to another it is generally best
1345 to upgrade to the latest code in the currently installed branch first,
1346 then do an upgrade to the new branch. This is the best-tested upgrade
1347 path, and has the highest probability of being successful. Please try
1348 this approach before reporting problems with a major version upgrade.
1352 When upgrading the boot ZFS pool to a new version, always follow
1355 1.) recompile and reinstall the ZFS boot loader and boot block
1356 (this is part of "make buildworld" and "make installworld")
1358 2.) update the ZFS boot block on your boot drive
1360 The following example updates the ZFS boot block on the first
1361 partition (freebsd-boot) of a GPT partitioned drive ad0:
1362 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1364 Non-boot pools do not need these updates.
1368 If you are updating from a prior version of FreeBSD (even one just
1369 a few days old), you should follow this procedure. It is the most
1370 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1372 make kernel-toolchain
1373 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1374 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1376 To test a kernel once
1377 ---------------------
1378 If you just want to boot a kernel once (because you are not sure
1379 if it works, or if you want to boot a known bad kernel to provide
1380 debugging information) run
1381 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1382 nextboot -k testkernel
1384 To just build a kernel when you know that it won't mess you up
1385 --------------------------------------------------------------
1386 This assumes you are already running a CURRENT system. Replace
1387 ${arch} with the architecture of your machine (e.g. "i386",
1388 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1390 cd src/sys/${arch}/conf
1391 config KERNEL_NAME_HERE
1392 cd ../compile/KERNEL_NAME_HERE
1397 If this fails, go to the "To build a kernel" section.
1399 To rebuild everything and install it on the current system.
1400 -----------------------------------------------------------
1401 # Note: sometimes if you are running current you gotta do more than
1402 # is listed here if you are upgrading from a really old current.
1404 <make sure you have good level 0 dumps>
1406 make kernel KERNCONF=YOUR_KERNEL_HERE
1408 <reboot in single user> [3]
1416 To cross-install current onto a separate partition
1417 --------------------------------------------------
1418 # In this approach we use a separate partition to hold
1419 # current's root, 'usr', and 'var' directories. A partition
1420 # holding "/", "/usr" and "/var" should be about 2GB in
1423 <make sure you have good level 0 dumps>
1426 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1427 <maybe newfs current's root partition>
1428 <mount current's root partition on directory ${CURRENT_ROOT}>
1429 make installworld DESTDIR=${CURRENT_ROOT}
1430 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1431 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1432 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1433 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1434 <reboot into current>
1435 <do a "native" rebuild/install as described in the previous section>
1436 <maybe install compatibility libraries from ports/misc/compat*>
1440 To upgrade in-place from stable to current
1441 ----------------------------------------------
1442 <make sure you have good level 0 dumps>
1444 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1446 <reboot in single user> [3]
1453 Make sure that you've read the UPDATING file to understand the
1454 tweaks to various things you need. At this point in the life
1455 cycle of current, things change often and you are on your own
1456 to cope. The defaults can also change, so please read ALL of
1457 the UPDATING entries.
1459 Also, if you are tracking -current, you must be subscribed to
1460 freebsd-current@freebsd.org. Make sure that before you update
1461 your sources that you have read and understood all the recent
1462 messages there. If in doubt, please track -stable which has
1463 much fewer pitfalls.
1465 [1] If you have third party modules, such as vmware, you
1466 should disable them at this point so they don't crash your
1469 [3] From the bootblocks, boot -s, and then do
1474 adjkerntz -i # if CMOS is wall time
1475 Also, when doing a major release upgrade, it is required that
1476 you boot into single user mode to do the installworld.
1478 [4] Note: This step is non-optional. Failure to do this step
1479 can result in a significant reduction in the functionality of the
1480 system. Attempting to do it by hand is not recommended and those
1481 that pursue this avenue should read this file carefully, as well
1482 as the archives of freebsd-current and freebsd-hackers mailing lists
1483 for potential gotchas. The -U option is also useful to consider.
1484 See mergemaster(8) for more information.
1486 [5] Usually this step is a noop. However, from time to time
1487 you may need to do this if you get unknown user in the following
1488 step. It never hurts to do it all the time. You may need to
1489 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1490 install) after the buildworld before this step if you last updated
1491 from current before 20020224 or from -stable before 20020408.
1493 [6] This only deletes old files and directories. Old libraries
1494 can be deleted by "make delete-old-libs", but you have to make
1495 sure that no program is using those libraries anymore.
1497 [8] In order to have a kernel that can run the 4.x binaries needed to
1498 do an installworld, you must include the COMPAT_FREEBSD4 option in
1499 your kernel. Failure to do so may leave you with a system that is
1500 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1501 required to run the 5.x binaries on more recent kernels. And so on
1502 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1504 Make sure that you merge any new devices from GENERIC since the
1505 last time you updated your kernel config file.
1507 [9] When checking out sources, you must include the -P flag to have
1508 cvs prune empty directories.
1510 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1511 "?=" instead of the "=" assignment operator, so that buildworld can
1512 override the CPUTYPE if it needs to.
1514 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1515 not on the command line, or in /etc/make.conf. buildworld will
1516 warn if it is improperly defined.
1519 This file contains a list, in reverse chronological order, of major
1520 breakages in tracking -current. It is not guaranteed to be a complete
1521 list of such breakages, and only contains entries since October 10, 2007.
1522 If you need to see UPDATING entries from before that date, you will need
1523 to fetch an UPDATING file from an older FreeBSD release.
1525 Copyright information:
1527 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1529 Redistribution, publication, translation and use, with or without
1530 modification, in full or in part, in any form or format of this
1531 document are permitted without further permission from the author.
1533 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1534 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1535 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1536 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1537 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1538 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1539 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1540 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1541 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1542 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1543 POSSIBILITY OF SUCH DAMAGE.
1545 Contact Warner Losh if you have any questions about your use of