1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
12 NOTE TO PEOPLE WHO THINK THAT FreeBSD 10.x IS SLOW:
13 FreeBSD 10.x has many debugging features turned on, in both the kernel
14 and userland. These features attempt to detect incorrect use of
15 system primitives, and encourage loud failure through extra sanity
16 checking and fail stop semantics. They also substantially impact
17 system performance. If you want to do performance measurement,
18 benchmarking, and optimization, you'll want to turn them off. This
19 includes various WITNESS- related kernel options, INVARIANTS, malloc
20 debugging flags in userland, and various verbose features in the
21 kernel. Many developers choose to disable these features on build
22 machines to maximize performance. (To disable malloc debugging, run
23 ln -s aj /etc/malloc.conf.)
26 The option VFS_ALLOW_NONMPSAFE option has been added in order to
27 explicitely support non-MPSAFE filesystems.
28 It is on by default for all supported platform at this present
32 The broken amd(4) driver has been replaced with esp(4) in the amd64,
33 i386 and pc98 GENERIC kernel configuration files.
36 sysinstall has been removed
39 The stable/9 branch created in subversion. This corresponds to the
40 RELENG_9 branch in CVS.
43 This commit modifies vfs_register() so that it uses a hash
44 calculation to set vfc_typenum, which is enabled by default.
45 The first time a system is booted after this change, the
46 vfc_typenum values will change for all file systems. The
47 main effect of this is a change to the NFS server file handles
48 for file systems that use vfc_typenum in their fsid, such as ZFS.
49 It will, however, prevent vfc_typenum from changing when file
50 systems are loaded in a different order for subsequent reboots.
51 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
52 until you are ready to remount all NFS clients after a reboot.
55 Bump the shared library version numbers for libraries that
56 do not use symbol versioning, have changed the ABI compared
57 to stable/8 and which shared library version was not bumped.
58 Done as part of 9.0-RELEASE cycle.
61 During the merge of Capsicum features, the fget(9) KPI was modified.
62 This may require the rebuilding of out-of-tree device drivers --
63 issues have been reported specifically with the nVidia device driver.
64 __FreeBSD_version is bumped to 900041.
66 Also, there is a period between 20110811 and 20110814 where the
67 special devices /dev/{stdin,stdout,stderr} did not work correctly.
68 Building world from a kernel during that window may not work.
71 The packet filter (pf) code has been updated to OpenBSD 4.5.
72 You need to update userland tools to be in sync with kernel.
73 This update breaks backward compatibility with earlier pfsync(4)
74 versions. Care must be taken when updating redundant firewall setups.
77 The following sysctls and tunables are retired on x86 platforms:
79 machdep.hlt_logical_cpus
80 The following sysctl is retired:
81 machdep.hyperthreading_allowed
82 The sysctls were supposed to provide a way to dynamically offline and
83 online selected CPUs on x86 platforms, but the implementation has not
84 been reliable especially with SCHED_ULE scheduler.
85 machdep.hyperthreading_allowed tunable is still available to ignore
86 hyperthreading CPUs at OS level.
87 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
88 where X is an APIC ID of a CPU. Be advised, though, that disabling
89 CPUs in non-uniform fashion will result in non-uniform topology and
90 may lead to sub-optimal system performance with SCHED_ULE, which is
94 cpumask_t type is retired and cpuset_t is used in order to describe
98 Changes to ifconfig(8) for dynamic address family detection mandate
99 that you are running a kernel of 20110525 or later. Make sure to
100 follow the update procedure to boot a new kernel before installing
104 Support for sun4v architecture is officially dropped
107 Several KPI breaking changes have been committed to the mii(4) layer,
108 the PHY drivers and consequently some Ethernet drivers using mii(4).
109 This means that miibus.ko and the modules of the affected Ethernet
110 drivers need to be recompiled.
112 Note to kernel developers: Given that the OUI bit reversion problem
113 was fixed as part of these changes all mii(4) commits related to OUIs,
114 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
115 handling, no longer can be merged verbatim to stable/8 and previous
119 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
120 into their kernel configurations along with 'device pci'.
123 The default NFS client is now the new NFS client, so fstype "newnfs"
124 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
125 Although mounts via fstype "nfs" will usually work without userland
126 changes, it is recommended that the mount(8) and mount_nfs(8)
127 commands be rebuilt from sources and that a link to mount_nfs called
128 mount_oldnfs be created. The new client is compiled into the
129 kernel with "options NFSCL" and this is needed for diskless root
130 file systems. The GENERIC kernel configs have been changed to use
131 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
132 To use the regular/old client, you can "mount -t oldnfs ...". For
133 a diskless root file system, you must also include a line like:
135 vfs.root.mountfrom="oldnfs:"
137 in the boot/loader.conf on the root fs on the NFS server to make
138 a diskless root fs use the old client.
141 The GENERIC kernels for all architectures now default to the new
142 CAM-based ATA stack. It means that all legacy ATA drivers were
143 removed and replaced by respective CAM drivers. If you are using
144 ATA device names in /etc/fstab or other places, make sure to update
145 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
146 where 'Y's are the sequential numbers starting from zero for each type
147 in order of detection, unless configured otherwise with tunables,
148 see cam(4)). There will be symbolic links created in /dev/ to map
149 old adX devices to the respective adaY. They should provide basic
150 compatibility for file systems mounting in most cases, but they do
151 not support old user-level APIs and do not have respective providers
152 in GEOM. Consider using updated management tools with new device names.
154 It is possible to load devices ahci, ata, siis and mvs as modules,
155 but option ATA_CAM should remain in kernel configuration to make ata
156 module work as CAM driver supporting legacy ATA controllers. Device ata
157 still can be used in modular fashion (atacore + ...). Modules atadisk
158 and atapi* are not used and won't affect operation in ATA_CAM mode.
159 Note that to use CAM-based ATA kernel should include CAM devices
160 scbus, pass, da (or explicitly ada), cd and optionally others. All of
161 them are parts of the cam module.
163 ataraid(4) functionality is now supported by the RAID GEOM class.
164 To use it you can load geom_raid kernel module and use graid(8) tool
165 for management. Instead of /dev/arX device names, use /dev/raid/rX.
167 No kernel config options or code have been removed, so if a problem
168 arises, please report it and optionally revert to the old ATA stack.
169 In order to do it you can remove from the kernel config:
174 , and instead add back:
175 device atadisk # ATA disk drives
176 device ataraid # ATA RAID drives
177 device atapicd # ATAPI CDROM drives
178 device atapifd # ATAPI floppy drives
179 device atapist # ATAPI tape drives
182 The default NFS server has been changed to the new server, which
183 was referred to as the experimental server. If you need to switch
184 back to the old NFS server, you must now put the "-o" option on
185 both the mountd and nfsd commands. This can be done using the
186 mountd_flags and nfs_server_flags rc.conf variables until an
187 update to the rc scripts is committed, which is coming soon.
190 The GNU Objective-C runtime library (libobjc), and other Objective-C
191 related components have been removed from the base system. If you
192 require an Objective-C library, please use one of the available ports.
195 ath(4) has been split into bus- and device- modules. if_ath contains
196 the HAL, the TX rate control and the network device code. if_ath_pci
197 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
198 contains the AHB glue. Users need to load both if_ath_pci and if_ath
199 in order to use ath on everything else.
201 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
202 need to load if_ath and if_ath_pci for ath(4) operation.
205 As part of the replacement of sysinstall, the process of building
206 release media has changed significantly. For details, please re-read
207 release(7), which has been updated to reflect the new build process.
210 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
211 is the last available version under GPLv2. It brings a number of new
212 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
213 4.1 and SSE 4.2), better support for powerpc64, a number of new
214 directives, and lots of other small improvements. See the ChangeLog
215 file in contrib/binutils for the full details.
218 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
219 compliant, and will now use half of hash for authentication.
220 This will break interoperability with all stacks (including all
221 actual FreeBSD versions) who implement
222 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
224 The only workaround with such peers is to use another HMAC
225 algorithm for IPsec ("phase 2") authentication.
228 Remove the uio_yield prototype and symbol. This function has
229 been misnamed since it was introduced and should not be
230 globally exposed with this name. The equivalent functionality
231 is now available using kern_yield(curthread->td_user_pri).
232 The function remains undocumented.
235 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
236 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
237 for scalar sysctls is defined but disabled. Code that needs
238 UQUAD to pass the type checking that must compile on older
239 systems where the define is not present can check against
240 __FreeBSD_version >= 900030.
242 The system dialog(1) has been replaced with a new version previously
243 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
244 with the previous version, but the libdialog associated with it has
245 a largely incompatible API. As such, the original version of libdialog
246 will be kept temporarily as libodialog, until its base system consumers
247 are replaced or updated. Bump __FreeBSD_version to 900030.
250 If you are trying to run make universe on a -stable system, and you get
251 the following warning:
252 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
253 unknown. config(8) likely too old."
254 or something similar to it, then you must upgrade your -stable system
255 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
256 stable/8) or build the config from the latest stable/8 branch and
257 install it on your system.
259 Prior to this date, building a current universe on 8-stable system from
260 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
261 error in the first kernel build phase. A new config on those old
262 systems will fix that problem for older versions of -current.
265 The TCP stack has been modified to allow Khelp modules to interact with
266 it via helper hook points and store per-connection data in the TCP
267 control block. Bump __FreeBSD_version to 900029. User space tools that
268 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
272 Generic IEEE 802.3 annex 31B full duplex flow control support has been
273 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
274 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
275 to take advantage of it instead of using custom implementations. This
276 means that these drivers now no longer unconditionally advertise
277 support for flow control but only do so if flow control is a selected
278 media option. This was implemented in the generic support that way in
279 order to allow flow control to be switched on and off via ifconfig(8)
280 with the PHY specific default to typically off in order to protect
281 from unwanted effects. Consequently, if you used flow control with
282 one of the above mentioned drivers you now need to explicitly enable
284 ifconfig bge0 media auto mediaopt flowcontrol
286 Along with the above mentioned changes generic support for setting
287 1000baseT master mode also has been added and brgphy(4), ciphy(4),
288 e1000phy(4) as well as ip1000phy(4) have been converted to take
289 advantage of it. This means that these drivers now no longer take the
290 link0 parameter for selecting master mode but the master media option
291 has to be used instead, for example like in the following:
292 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
294 Selection of master mode now is also available with all other PHY
295 drivers supporting 1000baseT.
298 The TCP stack has received a significant update to add support for
299 modularised congestion control and generally improve the clarity of
300 congestion control decisions. Bump __FreeBSD_version to 900025. User
301 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
302 sockstat) need to be recompiled.
305 The man(1) utility has been replaced by a new version that no longer
306 uses /etc/manpath.config. Please consult man.conf(5) for how to
307 migrate local entries to the new format.
310 The copyright strings printed by login(1) and sshd(8) at the time of a
311 new connection have been removed to follow other operating systems and
315 A workaround for a fixed ld bug has been removed in kernel code,
316 so make sure that your system ld is built from sources after
317 revision 210245 from 2010-07-19 (r211583 if building head kernel
318 on stable/8, r211584 for stable/7; both from 2010-08-21).
319 A symptom of incorrect ld version is different addresses for
320 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
323 The $ipv6_prefer variable in rc.conf(5) has been split into
324 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
326 The $ip6addrctl_policy is a variable to choose a pre-defined
327 address selection policy set by ip6addrctl(8). A value
328 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
331 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
332 flag (see an entry of 20090926) is set on an interface with no
333 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
334 security reason. If you want IPv6 link-local address on all
335 interfaces by default, set this to "YES".
337 The old ipv6_prefer="YES" is equivalent to
338 ipv6_activate_all_interfaces="YES" and
339 ip6addrctl_policy="ipv6_prefer".
342 DTrace has grown support for userland tracing. Due to this, DTrace is
343 now i386 and amd64 only.
344 dtruss(1) is now installed by default on those systems and a new
345 kernel module is needed for userland tracing: fasttrap.
346 No changes to your kernel config file are necessary to enable
347 userland tracing, but you might consider adding 'STRIP=' and
348 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
349 to have informative userland stack traces in DTrace (ustack).
352 The acpi_aiboost(4) driver has been removed in favor of the new
353 aibs(4) driver. You should update your kernel configuration file.
356 BSD grep has been imported to the base system and it is built by
357 default. It is completely BSD licensed, highly GNU-compatible, uses
358 less memory than its GNU counterpart and has a small codebase.
359 However, it is slower than its GNU counterpart, which is mostly
360 noticeable for larger searches, for smaller ones it is measurable
361 but not significant. The reason is complex, the most important factor
362 is that we lack a modern and efficient regex library and GNU
363 overcomes this by optimizing the searches internally. Future work
364 on improving the regex performance is planned, for the meantime,
365 users that need better performance, can build GNU grep instead by
366 setting the WITH_GNU_GREP knob.
369 Due to the import of powerpc64 support, all existing powerpc kernel
370 configuration files must be updated with a machine directive like this:
371 machine powerpc powerpc
373 In addition, an updated config(8) is required to build powerpc kernels
377 A new version of ZFS (version 15) has been merged to -HEAD.
378 This version uses a python library for the following subcommands:
379 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
380 For full functionality of these commands the following port must
381 be installed: sysutils/py-zfs
384 'vm_page's are now hashed by physical address to an array of mutexes.
385 Currently this is only used to serialize access to hold_count. Over
386 time the page queue mutex will be peeled away. This changes the size
387 of pmap on every architecture. And requires all callers of vm_page_hold
388 and vm_page_unhold to be updated.
391 WITH_CTF can now be specified in src.conf (not recommended, there
392 are some problems with static executables), make.conf (would also
393 affect ports which do not use GNU make and do not override the
394 compile targets) or in the kernel config (via "makeoptions
396 When WITH_CTF was specified there before this was silently ignored,
397 so make sure that WITH_CTF is not used in places which could lead
398 to unwanted behavior.
401 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
402 to allow 32-bit compatibility on non-x86 platforms. All kernel
403 configurations on amd64 and ia64 platforms using these options must
404 be modified accordingly.
407 The utmp user accounting database has been replaced with utmpx,
408 the user accounting interface standardized by POSIX.
409 Unfortunately the semantics of utmp and utmpx don't match,
410 making it practically impossible to support both interfaces.
411 The user accounting database is used by tools like finger(1),
412 last(1), talk(1), w(1) and ac(8).
414 All applications in the base system use utmpx. This means only
415 local binaries (e.g. from the ports tree) may still use these
416 utmp database files. These applications must be rebuilt to make
419 After the system has been upgraded, it is safe to remove the old
420 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
421 assuming their contents is of no importance anymore. Old wtmp
422 databases can only be used by last(1) and ac(8) after they have
423 been converted to the new format using wtmpcvt(1).
426 Introduce the kernel thread "deadlock resolver" (which can be enabled
427 via the DEADLKRES option, see NOTES for more details) and the
428 sleepq_type() function for sleepqueues.
431 The rc.firewall and rc.firewall6 were unified, and
432 rc.firewall6 and rc.d/ip6fw were removed.
433 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
434 variables are obsoleted. Instead, the following new rc
435 variables are added to rc.d/ipfw:
437 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
438 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
439 firewall_simple_onet_ipv6, firewall_trusted_ipv6
441 The meanings correspond to the relevant IPv4 variables.
447 The default terminal emulation for syscons(4) has been changed
448 from cons25 to xterm on all platforms except pc98. This means
449 that the /etc/ttys file needs to be updated to ensure correct
450 operation of applications on the console.
452 The terminal emulation style can be toggled per window by using
453 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
454 options can be used to change the compile-time default back to
457 To prevent graphical artifacts, make sure the TERM environment
458 variable is set to match the terminal emulation that is being
459 performed by syscons(4).
462 The layout of the structure ieee80211req_scan_result has changed.
463 Applications that require wireless scan results (e.g. ifconfig(8))
464 from net80211 need to be recompiled.
466 Applications such as wpa_supplicant(8) may require a full world
467 build without using NO_CLEAN in order to get synchronized with the
471 The iwn(4) driver has been updated to support the 5000 and 5150 series.
472 There's one kernel module for each firmware. Adding "device iwnfw"
473 to the kernel configuration file means including all three firmware
474 images inside the kernel. If you want to include just the one for
475 your wireless card, use the devices iwn4965fw, iwn5000fw or
479 The rc.d/network_ipv6, IPv6 configuration script has been integrated
480 into rc.d/netif. The changes are the following:
482 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
483 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
484 Note that both variables need the "inet6" keyword at the head.
486 Do not set $ipv6_network_interfaces manually if you do not
487 understand what you are doing. It is not needed in most cases.
489 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
492 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
493 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
495 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
496 all configured interfaces have "inet6 accept_rtadv" in the
497 $ifconfig_IF_ipv6. These are for backward compatibility.
499 3. A new variable $ipv6_prefer has been added. If NO, IPv6
500 functionality of interfaces with no corresponding
501 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
502 and the default address selection policy of ip6addrctl(8)
503 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
504 Note that if you want to configure IPv6 functionality on the
505 disabled interfaces after boot, first you need to clear the flag by
506 using ifconfig(8) like:
508 ifconfig em0 inet6 -ifdisabled
510 If YES, the default address selection policy is set as
513 The default value of $ipv6_prefer is NO.
515 4. If your system need to receive Router Advertisement messages,
516 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
517 scripts automatically invoke rtsol(8) when the interface becomes
518 UP. The Router Advertisement messages are used for SLAAC
519 (State-Less Address AutoConfiguration).
522 802.11s D3.03 support was committed. This is incompatible with the
523 previous code, which was based on D3.0.
526 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
527 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
528 control whether accepting Router Advertisement messages or not.
529 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
530 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
531 The ifconfig(8) utility now supports these flags.
534 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
535 mount(8) and -a option for df(1) to see them.
538 The old tunable hw.bus.devctl_disable has been superseded by
539 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
540 replaced by hw.bus.devctl_queue=0. The default for this new tunable
544 Remove the option STOP_NMI. The default action is now to use NMI only
545 for KDB via the newly introduced function stop_cpus_hard() and
546 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
549 The stable/8 branch created in subversion. This corresponds to the
550 RELENG_8 branch in CVS.
553 Bump the shared library version numbers for all libraries that do not
554 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
555 __FreeBSD_version to 800105.
558 Due to changes in the implementation of virtual network stack support,
559 all network-related kernel modules must be recompiled. As this change
560 breaks the ABI, bump __FreeBSD_version to 800104.
563 The TOE interface to the TCP syncache has been modified to remove
564 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
565 The cxgb driver is the only TOE consumer affected by this change, and
566 needs to be recompiled along with the kernel. As this change breaks
567 the ABI, bump __FreeBSD_version to 800103.
570 Padding has been added to struct tcpcb, sackhint and tcpstat in
571 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
572 maintaining the ABI. However, this change breaks the ABI, so bump
573 __FreeBSD_version to 800102. User space tools that rely on the size of
574 any of these structs (e.g. sockstat) need to be recompiled.
577 The NFS_LEGACYRPC option has been removed along with the old kernel
578 RPC implementation that this option selected. Kernel configurations
579 may need to be adjusted.
582 The network interface device nodes at /dev/net/<interface> have been
583 removed. All ioctl operations can be performed the normal way using
584 routing sockets. The kqueue functionality can generally be replaced
585 with routing sockets.
588 The documentation from the FreeBSD Documentation Project (Handbook,
589 FAQ, etc.) is now installed via packages by sysinstall(8) and under
590 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
593 The ABI of various structures related to the SYSV IPC API have been
594 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
595 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
599 Layout of struct vnet has changed as routing related variables were
600 moved to their own Vimage module. Modules need to be recompiled. Bump
601 __FreeBSD_version to 800099.
604 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
605 respectively. As long as no more than 16 groups per process are used,
606 no changes should be visible. When more than 16 groups are used, old
607 binaries may fail if they call getgroups() or getgrouplist() with
608 statically sized storage. Recompiling will work around this, but
609 applications should be modified to use dynamically allocated storage
610 for group arrays as POSIX.1-2008 does not cap an implementation's
611 number of supported groups at NGROUPS_MAX+1 as previous versions did.
613 NFS and portalfs mounts may also be affected as the list of groups is
614 truncated to 16. Users of NFS who use more than 16 groups, should
615 take care that negative group permissions are not used on the exported
616 file systems as they will not be reliable unless a GSSAPI based
617 authentication method is used.
620 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
621 option compiles in the support for adaptive spinning for lockmgrs
622 which want to enable it. The lockinit() function now accepts the flag
623 LK_ADAPTIVE in order to make the lock object subject to adaptive
624 spinning when both held in write and read mode.
627 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
628 changed. User applications that use this ioctl need to be rebuilt.
631 The layout of struct thread has changed. Kernel and modules need to
635 The layout of structs ifnet, domain, protosw and vnet_net has changed.
636 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
639 window(1) has been removed from the base system. It can now be
640 installed from ports. The port is called misc/window.
643 The way we are storing and accessing `routing table' entries has
644 changed. Programs reading the FIB, like netstat, need to be
648 A new netisr implementation has been added for FreeBSD 8. Network
649 file system modules, such as igmp, ipdivert, and others, should be
651 Bump __FreeBSD_version to 800096.
654 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
658 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
659 Bump __FreeBSD_version to 800094.
662 Add mnt_xflag field to 'struct mount'. File system modules need to be
664 Bump __FreeBSD_version to 800093.
667 The compiling option ADAPTIVE_SX has been retired while it has been
668 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
669 The KPI for sx_init_flags() changes as accepting flags:
670 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
671 been introduced in order to handle the reversed logic.
672 Bump __FreeBSD_version to 800092.
675 Add support for hierarchical jails. Remove global securelevel.
676 Bump __FreeBSD_version to 800091.
679 The layout of struct vnet_net has changed, therefore modules
681 Bump __FreeBSD_version to 800090.
684 The newly imported zic(8) produces a new format in the output. Please
685 run tzsetup(8) to install the newly created data to /etc/localtime.
688 The sysctl tree for the usb stack has renamed from hw.usb2.* to
689 hw.usb.* and is now consistent again with previous releases.
692 802.11 monitor mode support was revised and driver api's were changed.
693 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
694 of DLT_IEEE802_11. No user-visible data structures were changed but
695 applications that use DLT_IEEE802_11 may require changes.
696 Bump __FreeBSD_version to 800088.
699 The layout of the following structs has changed: sysctl_oid,
700 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
701 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
702 panics may be experienced. World rebuild is required for
703 correctly checking networking state from userland.
704 Bump __FreeBSD_version to 800085.
707 MLDv2 and Source-Specific Multicast (SSM) have been merged
708 to the IPv6 stack. VIMAGE hooks are in but not yet used.
709 The implementation of SSM within FreeBSD's IPv6 stack closely
710 follows the IPv4 implementation.
712 For kernel developers:
714 * The most important changes are that the ip6_output() and
715 ip6_input() paths no longer take the IN6_MULTI_LOCK,
716 and this lock has been downgraded to a non-recursive mutex.
718 * As with the changes to the IPv4 stack to support SSM, filtering
719 of inbound multicast traffic must now be performed by transport
720 protocols within the IPv6 stack. This does not apply to TCP and
721 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
723 * The KPIs used by IPv6 multicast are similar to those used by
724 the IPv4 stack, with the following differences:
725 * im6o_mc_filter() is analogous to imo_multicast_filter().
726 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
727 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
728 * IN6_LOOKUP_MULTI() has been deprecated and removed.
729 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
730 for MLDv1 have an additional 'timer' argument which is used to
731 jitter the initial membership report for the solicited-node
732 multicast membership on-link.
733 * This is not strictly needed for MLDv2, which already jitters
734 its report transmissions. However, the 'timer' argument is
735 preserved in case MLDv1 is active on the interface.
737 * The KAME linked-list based IPv6 membership implementation has
738 been refactored to use a vector similar to that used by the IPv4
740 Code which maintains a list of its own multicast memberships
741 internally, e.g. carp, has been updated to reflect the new
744 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
745 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
746 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
747 implementation constraint which needs to be addressed in HEAD.
749 For application developers:
751 * The changes are broadly similar to those made for the IPv4
754 * The use of IPv4 and IPv6 multicast socket options on the same
755 socket, using mapped addresses, HAS NOT been tested or supported.
757 * There are a number of issues with the implementation of various
758 IPv6 multicast APIs which need to be resolved in the API surface
759 before the implementation is fully compatible with KAME userland
760 use, and these are mostly to do with interface index treatment.
762 * The literature available discusses the use of either the delta / ASM
763 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
764 using setsourcefilter(3)/getsourcefilter(3). For more information
765 please refer to RFC 3768, 'Socket Interface Extensions for
766 Multicast Source Filters'.
768 * Applications which use the published RFC 3678 APIs should be fine.
770 For systems administrators:
772 * The mtest(8) utility has been refactored to support IPv6, in
773 addition to IPv4. Interface addresses are no longer accepted
774 as arguments, their names must be used instead. The utility
775 will map the interface name to its first IPv4 address as
776 returned by getifaddrs(3).
778 * The ifmcstat(8) utility has also been updated to print the MLDv2
779 endpoint state and source filter lists via sysctl(3).
781 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
782 loopback of IPv6 multicast datagrams by default; it defaults to 1
783 to preserve the existing behaviour. Disabling multicast loopback is
784 recommended for optimal system performance.
786 * The IPv6 MROUTING code has been changed to examine this sysctl
787 instead of attempting to perform a group lookup before looping
788 back forwarded datagrams.
790 Bump __FreeBSD_version to 800084.
793 Implement low-level Bluetooth HCI API.
794 Bump __FreeBSD_version to 800083.
797 The layout of struct malloc_type, used by modules to register new
798 memory allocation types, has changed. Most modules will need to
799 be rebuilt or panics may be experienced.
800 Bump __FreeBSD_version to 800081.
803 Anticipate overflowing inp_flags - add inp_flags2.
804 This changes most offsets in inpcb, so checking v4 connection
805 state will require a world rebuild.
806 Bump __FreeBSD_version to 800080.
809 Add an llentry to struct route and struct route_in6. Modules
810 embedding a struct route will need to be recompiled.
811 Bump __FreeBSD_version to 800079.
814 The size of rt_metrics_lite and by extension rtentry has changed.
815 Networking administration apps will need to be recompiled.
816 The route command now supports show as an alias for get, weighting
817 of routes, sticky and nostick flags to alter the behavior of stateful
819 Bump __FreeBSD_version to 800078.
822 Do not use Giant for kbdmux(4) locking. This is wrong and
823 apparently causing more problems than it solves. This will
824 re-open the issue where interrupt handlers may race with
825 kbdmux(4) in polling mode. Typical symptoms include (but
826 not limited to) duplicated and/or missing characters when
827 low level console functions (such as gets) are used while
828 interrupts are enabled (for example geli password prompt,
829 mountroot prompt etc.). Disabling kbdmux(4) may help.
832 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
833 kernel modules referencing any of the above need to be recompiled.
834 Bump __FreeBSD_version to 800075.
837 GEOM_PART has become the default partition slicer for storage devices,
838 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
839 introduces some changes:
841 MSDOS/EBR: the devices created from MSDOS extended partition entries
842 (EBR) can be named differently than with GEOM_MBR and are now symlinks
843 to devices with offset-based names. fstabs may need to be modified.
845 BSD: the "geometry does not match label" warning is harmless in most
846 cases but it points to problems in file system misalignment with
847 disk geometry. The "c" partition is now implicit, covers the whole
848 top-level drive and cannot be (mis)used by users.
850 General: Kernel dumps are now not allowed to be written to devices
851 whose partition types indicate they are meant to be used for file
852 systems (or, in case of MSDOS partitions, as something else than
855 Most of these changes date approximately from 200812.
858 The uscanner(4) driver has been removed from the kernel. This follows
859 Linux removing theirs in 2.6 and making libusb the default interface
863 The multicast forwarding code has been cleaned up. netstat(1)
864 only relies on KVM now for printing bandwidth upcall meters.
865 The IPv4 and IPv6 modules are split into ip_mroute_mod and
866 ip6_mroute_mod respectively. The config(5) options for statically
867 compiling this code remain the same, i.e. 'options MROUTING'.
870 Support for the IFF_NEEDSGIANT network interface flag has been
871 removed, which means that non-MPSAFE network device drivers are no
872 longer supported. In particular, if_ar, if_sr, and network device
873 drivers from the old (legacy) USB stack can no longer be built or
877 POSIX.1 Native Language Support (NLS) has been enabled in libc and
878 a bunch of new language catalog files have also been added.
879 This means that some common libc messages are now localized and
880 they depend on the LC_MESSAGES environmental variable.
883 The k8temp(4) driver has been renamed to amdtemp(4) since
884 support for Family 10 and Family 11 CPU families was added.
887 IGMPv3 and Source-Specific Multicast (SSM) have been merged
888 to the IPv4 stack. VIMAGE hooks are in but not yet used.
890 For kernel developers, the most important changes are that the
891 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
892 and this lock has been downgraded to a non-recursive mutex.
894 Transport protocols (UDP, Raw IP) are now responsible for filtering
895 inbound multicast traffic according to group membership and source
896 filters. The imo_multicast_filter() KPI exists for this purpose.
897 Transports which do not use multicast (SCTP, TCP) already reject
898 multicast by default. Forwarding and receive performance may improve
899 as a mutex acquisition is no longer needed in the ip_input()
900 low-level input path. in_addmulti() and in_delmulti() are shimmed
901 to new KPIs which exist to support SSM in-kernel.
903 For application developers, it is recommended that loopback of
904 multicast datagrams be disabled for best performance, as this
905 will still cause the lock to be taken for each looped-back
906 datagram transmission. The net.inet.ip.mcast.loop sysctl may
907 be tuned to 0 to disable loopback by default; it defaults to 1
908 to preserve the existing behaviour.
910 For systems administrators, to obtain best performance with
911 multicast reception and multiple groups, it is always recommended
912 that a card with a suitably precise hash filter is used. Hash
913 collisions will still result in the lock being taken within the
914 transport protocol input path to check group membership.
916 If deploying FreeBSD in an environment with IGMP snooping switches,
917 it is recommended that the net.inet.igmp.sendlocal sysctl remain
918 enabled; this forces 224.0.0.0/24 group membership to be announced
921 The size of 'struct igmpstat' has changed; netstat needs to be
922 recompiled to reflect this.
923 Bump __FreeBSD_version to 800070.
926 libusb20.so.1 is now installed as libusb.so.1 and the ports system
927 updated to use it. This requires a buildworld/installworld in order to
928 update the library and dependencies (usbconfig, etc). Its advisable to
929 rebuild all ports which uses libusb. More specific directions are given
930 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
931 libusb are no longer required and can be removed.
934 A workaround is committed to allow the creation of System V shared
935 memory segment of size > 2 GB on the 64-bit architectures.
936 Due to a limitation of the existing ABI, the shm_segsz member
937 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
938 wrong for large segments. Note that limits must be explicitly
939 raised to allow such segments to be created.
942 The layout of struct ifnet has changed, requiring a rebuild of all
943 network device driver modules.
946 The /dev handling for the new USB stack has changed, a
947 buildworld/installworld is required for libusb20.
950 The new USB2 stack has now been permanently moved in and all kernel and
951 module names reverted to their previous values (eg, usb, ehci, ohci,
952 ums, ...). The old usb stack can be compiled in by prefixing the name
953 with the letter 'o', the old usb modules have been removed.
954 Updating entry 20090216 for xorg and 20090215 for libmap may still
958 The rc.conf(5) option if_up_delay has been renamed to
959 defaultroute_delay to better reflect its purpose. If you have
960 customized this setting in /etc/rc.conf you need to update it to
964 xorg 7.4 wants to configure its input devices via hald which does not
965 yet work with USB2. If the keyboard/mouse does not work in xorg then
967 Option "AllowEmptyInput" "off"
968 to your ServerLayout section. This will cause X to use the configured
969 kbd and mouse sections from your xorg.conf.
972 The GENERIC kernels for all architectures now default to the new USB2
973 stack. No kernel config options or code have been removed so if a
974 problem arises please report it and optionally revert to the old USB
975 stack. If you are loading USB kernel modules or have a custom kernel
976 that includes GENERIC then ensure that usb names are also changed over,
977 eg uftdi -> usb2_serial_ftdi.
979 Older programs linked against the ports libusb 0.1 need to be
980 redirected to the new stack's libusb20. /etc/libmap.conf can
982 # Map old usb library to new one for usb2 stack
983 libusb-0.1.so.8 libusb20.so.1
986 All USB ethernet devices now attach as interfaces under the name ueN
987 (eg. ue0). This is to provide a predictable name as vendors often
988 change usb chipsets in a product without notice.
991 The ichsmb(4) driver has been changed to require SMBus slave
992 addresses be left-justified (xxxxxxx0b) rather than right-justified.
993 All of the other SMBus controller drivers require left-justified
994 slave addresses, so this change makes all the drivers provide the
998 INET6 statistics (struct ip6stat) was updated.
999 netstat(1) needs to be recompiled.
1002 NTFS has been removed from GENERIC kernel on amd64 to match
1003 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1004 will load ntfs.ko module automatically when NTFS support is
1005 actually needed, unless ntfs.ko is not installed or security
1006 level prohibits loading kernel modules. If either is the case,
1007 "options NTFS" has to be added into kernel config.
1010 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1011 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1012 800061. User space tools that rely on the size of struct tcpcb in
1013 tcp_var.h (e.g. sockstat) need to be recompiled.
1016 ng_tty(4) module updated to match the new TTY subsystem.
1017 Due to API change, user-level applications must be updated.
1018 New API support added to mpd5 CVS and expected to be present
1019 in next mpd5.3 release.
1022 With __FreeBSD_version 800060 the makefs tool is part of
1023 the base system (it was a port).
1026 The afdata and ifnet locks have been changed from mutexes to
1027 rwlocks, network modules will need to be re-compiled.
1030 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1031 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1032 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1033 architecture and 40 bytes on 64-bit architecture. The userland
1034 applications "arp" and "ndp" have been updated accordingly.
1035 The output from "netstat -r" shows only routing entries and
1036 none of the L2 information.
1039 __FreeBSD_version 800057 marks the switchover from the
1040 binary ath hal to source code. Users must add the line:
1042 options AH_SUPPORT_AR5416
1044 to their kernel config files when specifying:
1048 The ath_hal module no longer exists; the code is now compiled
1049 together with the driver in the ath module. It is now
1050 possible to tailor chip support (i.e. reduce the set of chips
1051 and thereby the code size); consult ath_hal(4) for details.
1054 __FreeBSD_version 800054 adds memory barriers to
1055 <machine/atomic.h>, new interfaces to ifnet to facilitate
1056 multiple hardware transmit queues for cards that support
1057 them, and a lock-less ring-buffer implementation to
1058 enable drivers to more efficiently manage queueing of
1062 A new version of ZFS (version 13) has been merged to -HEAD.
1063 This version has zpool attribute "listsnapshots" off by
1064 default, which means "zfs list" does not show snapshots,
1065 and is the same as Solaris behavior.
1068 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1071 The uhci, ohci, ehci and slhci USB Host controller drivers have
1072 been put into separate modules. If you load the usb module
1073 separately through loader.conf you will need to load the
1074 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1075 controller add the following to loader.conf:
1081 The ABI used by the PMC toolset has changed. Please keep
1082 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1086 atapci kernel module now includes only generic PCI ATA
1087 driver. AHCI driver moved to ataahci kernel module.
1088 All vendor-specific code moved into separate kernel modules:
1089 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1090 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1091 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1092 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1095 The TTY subsystem of the kernel has been replaced by a new
1096 implementation, which provides better scalability and an
1097 improved driver model. Most common drivers have been migrated to
1098 the new TTY subsystem, while others have not. The following
1099 drivers have not yet been ported to the new TTY layer:
1102 cy, digi, rc, rp, sio
1108 ng_h4, ng_tty, ppp, sl, snp
1110 Adding these drivers to your kernel configuration file shall
1111 cause compilation to fail.
1114 ntpd has been upgraded to 4.2.4p5.
1117 OpenSSH has been upgraded to 5.1p1.
1119 For many years, FreeBSD's version of OpenSSH preferred DSA
1120 over RSA for host and user authentication keys. With this
1121 upgrade, we've switched to the vendor's default of RSA over
1122 DSA. This may cause upgraded clients to warn about unknown
1123 host keys even for previously known hosts. Users should
1124 follow the usual procedure for verifying host keys before
1125 accepting the RSA key.
1127 This can be circumvented by setting the "HostKeyAlgorithms"
1128 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1131 Please note that the sequence of keys offered for
1132 authentication has been changed as well. You may want to
1133 specify IdentityFile in a different order to revert this
1137 The sio(4) driver has been removed from the i386 and amd64
1138 kernel configuration files. This means uart(4) is now the
1139 default serial port driver on those platforms as well.
1141 To prevent collisions with the sio(4) driver, the uart(4) driver
1142 uses different names for its device nodes. This means the
1143 onboard serial port will now most likely be called "ttyu0"
1144 instead of "ttyd0". You may need to reconfigure applications to
1145 use the new device names.
1147 When using the serial port as a boot console, be sure to update
1148 /boot/device.hints and /etc/ttys before booting the new kernel.
1149 If you forget to do so, you can still manually specify the hints
1150 at the loader prompt:
1152 set hint.uart.0.at="isa"
1153 set hint.uart.0.port="0x3F8"
1154 set hint.uart.0.flags="0x10"
1155 set hint.uart.0.irq="4"
1159 The gpt(8) utility has been removed. Use gpart(8) to partition
1163 The version that Linuxulator emulates was changed from 2.4.2
1164 to 2.6.16. If you experience any problems with Linux binaries
1165 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1166 if it fixes the problem contact emulation mailing list.
1169 ISDN4BSD (I4B) was removed from the src tree. You may need to
1170 update a your kernel configuration and remove relevant entries.
1173 I have checked in code to support multiple routing tables.
1174 See the man pages setfib(1) and setfib(2).
1175 This is a hopefully backwards compatible version,
1176 but to make use of it you need to compile your kernel
1177 with options ROUTETABLES=2 (or more up to 16).
1180 The 802.11 wireless support was redone to enable multi-bss
1181 operation on devices that are capable. The underlying device
1182 is no longer used directly but instead wlanX devices are
1183 cloned with ifconfig. This requires changes to rc.conf files.
1184 For example, change:
1185 ifconfig_ath0="WPA DHCP"
1188 ifconfig_wlan0="WPA DHCP"
1189 see rc.conf(5) for more details. In addition, mergemaster of
1190 /etc/rc.d is highly recommended. Simultaneous update of userland
1191 and kernel wouldn't hurt either.
1193 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1194 modules were merged into the base wlan module. All references
1195 to these modules (e.g. in kernel config files) must be removed.
1198 psm(4) has gained write(2) support in native operation level.
1199 Arbitrary commands can be written to /dev/psm%d and status can
1200 be read back from it. Therefore, an application is responsible
1201 for status validation and error recovery. It is a no-op in
1202 other operation levels.
1205 Support for KSE threading has been removed from the kernel. To
1206 run legacy applications linked against KSE libmap.conf may
1207 be used. The following libmap.conf may be used to ensure
1208 compatibility with any prior release:
1210 libpthread.so.1 libthr.so.1
1211 libpthread.so.2 libthr.so.2
1212 libkse.so.3 libthr.so.3
1215 The layout of struct vmspace has changed. This affects libkvm
1216 and any executables that link against libkvm and use the
1217 kvm_getprocs() function. In particular, but not exclusively,
1218 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1219 The effects are minimal, but it's advisable to upgrade world
1223 The latest em driver no longer has support in it for the
1224 82575 adapter, this is now moved to the igb driver. The
1225 split was done to make new features that are incompatible
1226 with older hardware easier to do.
1229 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1230 likewise the kernel option is now GEOM_LINUX_LVM.
1233 The default NFS mount mode has changed from UDP to TCP for
1234 increased reliability. If you rely on (insecurely) NFS
1235 mounting across a firewall you may need to update your
1239 Belatedly note the addition of m_collapse for compacting
1243 The fts(3) structures have been changed to use adequate
1244 integer types for their members and so to be able to cope
1245 with huge file trees. The old fts(3) ABI is preserved
1246 through symbol versioning in libc, so third-party binaries
1247 using fts(3) should still work, although they will not take
1248 advantage of the extended types. At the same time, some
1249 third-party software might fail to build after this change
1250 due to unportable assumptions made in its source code about
1251 fts(3) structure members. Such software should be fixed
1252 by its vendor or, in the worst case, in the ports tree.
1253 FreeBSD_version 800015 marks this change for the unlikely
1254 case that a portable fix is impossible.
1257 To upgrade to -current after this date, you must be running
1258 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1259 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1262 The ADAPTIVE_GIANT kernel option has been retired because its
1263 functionality is the default now.
1266 The AT keyboard emulation of sunkbd(4) has been turned on
1267 by default. In order to make the special symbols of the Sun
1268 keyboards driven by sunkbd(4) work under X these now have
1269 to be configured the same way as Sun USB keyboards driven
1270 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1272 Option "XkbLayout" "us"
1273 Option "XkbRules" "xorg"
1274 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1277 It has been decided that it is desirable to provide ABI
1278 backwards compatibility to the FreeBSD 4/5/6 versions of the
1279 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1280 broken with the introduction of PCI domain support (see the
1281 20070930 entry). Unfortunately, this required the ABI of
1282 PCIOCGETCONF to be broken again in order to be able to
1283 provide backwards compatibility to the old version of that
1284 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1285 again. As for prominent ports this affects neither pciutils
1286 nor xorg-server this time, the hal port needs to be rebuilt
1290 The misnamed kthread_create() and friends have been renamed
1291 to kproc_create() etc. Many of the callers already
1292 used kproc_start()..
1293 I will return kthread_create() and friends in a while
1294 with implementations that actually create threads, not procs.
1295 Renaming corresponds with version 800002.
1304 Avoid using make -j when upgrading. While generally safe, there are
1305 sometimes problems using -j to upgrade. If your upgrade fails with
1306 -j, please try again without -j. From time to time in the past there
1307 have been problems using -j with buildworld and/or installworld. This
1308 is especially true when upgrading between "distant" versions (eg one
1309 that cross a major release boundary or several minor releases, or when
1310 several months have passed on the -current branch).
1312 Sometimes, obscure build problems are the result of environment
1313 poisoning. This can happen because the make utility reads its
1314 environment when searching for values for global variables. To run
1315 your build attempts in an "environmental clean room", prefix all make
1316 commands with 'env -i '. See the env(1) manual page for more details.
1318 When upgrading from one major version to another it is generally best
1319 to upgrade to the latest code in the currently installed branch first,
1320 then do an upgrade to the new branch. This is the best-tested upgrade
1321 path, and has the highest probability of being successful. Please try
1322 this approach before reporting problems with a major version upgrade.
1326 When upgrading the boot ZFS pool to a new version, always follow
1329 1.) recompile and reinstall the ZFS boot loader and boot block
1330 (this is part of "make buildworld" and "make installworld")
1332 2.) update the ZFS boot block on your boot drive
1334 The following example updates the ZFS boot block on the first
1335 partition (freebsd-boot) of a GPT partitioned drive ad0:
1336 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1338 Non-boot pools do not need these updates.
1342 If you are updating from a prior version of FreeBSD (even one just
1343 a few days old), you should follow this procedure. It is the most
1344 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1346 make kernel-toolchain
1347 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1348 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1350 To test a kernel once
1351 ---------------------
1352 If you just want to boot a kernel once (because you are not sure
1353 if it works, or if you want to boot a known bad kernel to provide
1354 debugging information) run
1355 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1356 nextboot -k testkernel
1358 To just build a kernel when you know that it won't mess you up
1359 --------------------------------------------------------------
1360 This assumes you are already running a CURRENT system. Replace
1361 ${arch} with the architecture of your machine (e.g. "i386",
1362 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1364 cd src/sys/${arch}/conf
1365 config KERNEL_NAME_HERE
1366 cd ../compile/KERNEL_NAME_HERE
1371 If this fails, go to the "To build a kernel" section.
1373 To rebuild everything and install it on the current system.
1374 -----------------------------------------------------------
1375 # Note: sometimes if you are running current you gotta do more than
1376 # is listed here if you are upgrading from a really old current.
1378 <make sure you have good level 0 dumps>
1380 make kernel KERNCONF=YOUR_KERNEL_HERE
1382 <reboot in single user> [3]
1390 To cross-install current onto a separate partition
1391 --------------------------------------------------
1392 # In this approach we use a separate partition to hold
1393 # current's root, 'usr', and 'var' directories. A partition
1394 # holding "/", "/usr" and "/var" should be about 2GB in
1397 <make sure you have good level 0 dumps>
1400 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1401 <maybe newfs current's root partition>
1402 <mount current's root partition on directory ${CURRENT_ROOT}>
1403 make installworld DESTDIR=${CURRENT_ROOT}
1404 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1405 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1406 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1407 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1408 <reboot into current>
1409 <do a "native" rebuild/install as described in the previous section>
1410 <maybe install compatibility libraries from ports/misc/compat*>
1414 To upgrade in-place from 8.x-stable to current
1415 ----------------------------------------------
1416 <make sure you have good level 0 dumps>
1418 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1420 <reboot in single user> [3]
1427 Make sure that you've read the UPDATING file to understand the
1428 tweaks to various things you need. At this point in the life
1429 cycle of current, things change often and you are on your own
1430 to cope. The defaults can also change, so please read ALL of
1431 the UPDATING entries.
1433 Also, if you are tracking -current, you must be subscribed to
1434 freebsd-current@freebsd.org. Make sure that before you update
1435 your sources that you have read and understood all the recent
1436 messages there. If in doubt, please track -stable which has
1437 much fewer pitfalls.
1439 [1] If you have third party modules, such as vmware, you
1440 should disable them at this point so they don't crash your
1443 [3] From the bootblocks, boot -s, and then do
1448 adjkerntz -i # if CMOS is wall time
1449 Also, when doing a major release upgrade, it is required that
1450 you boot into single user mode to do the installworld.
1452 [4] Note: This step is non-optional. Failure to do this step
1453 can result in a significant reduction in the functionality of the
1454 system. Attempting to do it by hand is not recommended and those
1455 that pursue this avenue should read this file carefully, as well
1456 as the archives of freebsd-current and freebsd-hackers mailing lists
1457 for potential gotchas. The -U option is also useful to consider.
1458 See mergemaster(8) for more information.
1460 [5] Usually this step is a noop. However, from time to time
1461 you may need to do this if you get unknown user in the following
1462 step. It never hurts to do it all the time. You may need to
1463 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1464 install) after the buildworld before this step if you last updated
1465 from current before 20020224 or from -stable before 20020408.
1467 [6] This only deletes old files and directories. Old libraries
1468 can be deleted by "make delete-old-libs", but you have to make
1469 sure that no program is using those libraries anymore.
1471 [8] In order to have a kernel that can run the 4.x binaries needed to
1472 do an installworld, you must include the COMPAT_FREEBSD4 option in
1473 your kernel. Failure to do so may leave you with a system that is
1474 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1475 required to run the 5.x binaries on more recent kernels. And so on
1476 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1478 Make sure that you merge any new devices from GENERIC since the
1479 last time you updated your kernel config file.
1481 [9] When checking out sources, you must include the -P flag to have
1482 cvs prune empty directories.
1484 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1485 "?=" instead of the "=" assignment operator, so that buildworld can
1486 override the CPUTYPE if it needs to.
1488 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1489 not on the command line, or in /etc/make.conf. buildworld will
1490 warn if it is improperly defined.
1493 This file contains a list, in reverse chronological order, of major
1494 breakages in tracking -current. It is not guaranteed to be a complete
1495 list of such breakages, and only contains entries since October 10, 2007.
1496 If you need to see UPDATING entries from before that date, you will need
1497 to fetch an UPDATING file from an older FreeBSD release.
1499 Copyright information:
1501 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1503 Redistribution, publication, translation and use, with or without
1504 modification, in full or in part, in any form or format of this
1505 document are permitted without further permission from the author.
1507 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1508 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1509 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1510 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1511 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1512 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1513 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1514 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1515 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1516 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1517 POSSIBILITY OF SUCH DAMAGE.
1519 Contact Warner Losh if you have any questions about your use of