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 acpi_wmi(4) status device /dev/wmistat has been renamed to
30 The option VFS_ALLOW_NONMPSAFE option has been added in order to
31 explicitely support non-MPSAFE filesystems.
32 It is on by default for all supported platform at this present
36 The broken amd(4) driver has been replaced with esp(4) in the amd64,
37 i386 and pc98 GENERIC kernel configuration files.
40 sysinstall has been removed
43 The stable/9 branch created in subversion. This corresponds to the
44 RELENG_9 branch in CVS.
47 This commit modifies vfs_register() so that it uses a hash
48 calculation to set vfc_typenum, which is enabled by default.
49 The first time a system is booted after this change, the
50 vfc_typenum values will change for all file systems. The
51 main effect of this is a change to the NFS server file handles
52 for file systems that use vfc_typenum in their fsid, such as ZFS.
53 It will, however, prevent vfc_typenum from changing when file
54 systems are loaded in a different order for subsequent reboots.
55 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
56 until you are ready to remount all NFS clients after a reboot.
59 Bump the shared library version numbers for libraries that
60 do not use symbol versioning, have changed the ABI compared
61 to stable/8 and which shared library version was not bumped.
62 Done as part of 9.0-RELEASE cycle.
65 During the merge of Capsicum features, the fget(9) KPI was modified.
66 This may require the rebuilding of out-of-tree device drivers --
67 issues have been reported specifically with the nVidia device driver.
68 __FreeBSD_version is bumped to 900041.
70 Also, there is a period between 20110811 and 20110814 where the
71 special devices /dev/{stdin,stdout,stderr} did not work correctly.
72 Building world from a kernel during that window may not work.
75 The packet filter (pf) code has been updated to OpenBSD 4.5.
76 You need to update userland tools to be in sync with kernel.
77 This update breaks backward compatibility with earlier pfsync(4)
78 versions. Care must be taken when updating redundant firewall setups.
81 The following sysctls and tunables are retired on x86 platforms:
83 machdep.hlt_logical_cpus
84 The following sysctl is retired:
85 machdep.hyperthreading_allowed
86 The sysctls were supposed to provide a way to dynamically offline and
87 online selected CPUs on x86 platforms, but the implementation has not
88 been reliable especially with SCHED_ULE scheduler.
89 machdep.hyperthreading_allowed tunable is still available to ignore
90 hyperthreading CPUs at OS level.
91 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
92 where X is an APIC ID of a CPU. Be advised, though, that disabling
93 CPUs in non-uniform fashion will result in non-uniform topology and
94 may lead to sub-optimal system performance with SCHED_ULE, which is
98 cpumask_t type is retired and cpuset_t is used in order to describe
102 Changes to ifconfig(8) for dynamic address family detection mandate
103 that you are running a kernel of 20110525 or later. Make sure to
104 follow the update procedure to boot a new kernel before installing
108 Support for sun4v architecture is officially dropped
111 Several KPI breaking changes have been committed to the mii(4) layer,
112 the PHY drivers and consequently some Ethernet drivers using mii(4).
113 This means that miibus.ko and the modules of the affected Ethernet
114 drivers need to be recompiled.
116 Note to kernel developers: Given that the OUI bit reversion problem
117 was fixed as part of these changes all mii(4) commits related to OUIs,
118 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
119 handling, no longer can be merged verbatim to stable/8 and previous
123 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
124 into their kernel configurations along with 'device pci'.
127 The default NFS client is now the new NFS client, so fstype "newnfs"
128 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
129 Although mounts via fstype "nfs" will usually work without userland
130 changes, it is recommended that the mount(8) and mount_nfs(8)
131 commands be rebuilt from sources and that a link to mount_nfs called
132 mount_oldnfs be created. The new client is compiled into the
133 kernel with "options NFSCL" and this is needed for diskless root
134 file systems. The GENERIC kernel configs have been changed to use
135 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
136 To use the regular/old client, you can "mount -t oldnfs ...". For
137 a diskless root file system, you must also include a line like:
139 vfs.root.mountfrom="oldnfs:"
141 in the boot/loader.conf on the root fs on the NFS server to make
142 a diskless root fs use the old client.
145 The GENERIC kernels for all architectures now default to the new
146 CAM-based ATA stack. It means that all legacy ATA drivers were
147 removed and replaced by respective CAM drivers. If you are using
148 ATA device names in /etc/fstab or other places, make sure to update
149 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
150 where 'Y's are the sequential numbers starting from zero for each type
151 in order of detection, unless configured otherwise with tunables,
152 see cam(4)). There will be symbolic links created in /dev/ to map
153 old adX devices to the respective adaY. They should provide basic
154 compatibility for file systems mounting in most cases, but they do
155 not support old user-level APIs and do not have respective providers
156 in GEOM. Consider using updated management tools with new device names.
158 It is possible to load devices ahci, ata, siis and mvs as modules,
159 but option ATA_CAM should remain in kernel configuration to make ata
160 module work as CAM driver supporting legacy ATA controllers. Device ata
161 still can be used in modular fashion (atacore + ...). Modules atadisk
162 and atapi* are not used and won't affect operation in ATA_CAM mode.
163 Note that to use CAM-based ATA kernel should include CAM devices
164 scbus, pass, da (or explicitly ada), cd and optionally others. All of
165 them are parts of the cam module.
167 ataraid(4) functionality is now supported by the RAID GEOM class.
168 To use it you can load geom_raid kernel module and use graid(8) tool
169 for management. Instead of /dev/arX device names, use /dev/raid/rX.
171 No kernel config options or code have been removed, so if a problem
172 arises, please report it and optionally revert to the old ATA stack.
173 In order to do it you can remove from the kernel config:
178 , and instead add back:
179 device atadisk # ATA disk drives
180 device ataraid # ATA RAID drives
181 device atapicd # ATAPI CDROM drives
182 device atapifd # ATAPI floppy drives
183 device atapist # ATAPI tape drives
186 The default NFS server has been changed to the new server, which
187 was referred to as the experimental server. If you need to switch
188 back to the old NFS server, you must now put the "-o" option on
189 both the mountd and nfsd commands. This can be done using the
190 mountd_flags and nfs_server_flags rc.conf variables until an
191 update to the rc scripts is committed, which is coming soon.
194 The GNU Objective-C runtime library (libobjc), and other Objective-C
195 related components have been removed from the base system. If you
196 require an Objective-C library, please use one of the available ports.
199 ath(4) has been split into bus- and device- modules. if_ath contains
200 the HAL, the TX rate control and the network device code. if_ath_pci
201 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
202 contains the AHB glue. Users need to load both if_ath_pci and if_ath
203 in order to use ath on everything else.
205 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
206 need to load if_ath and if_ath_pci for ath(4) operation.
209 As part of the replacement of sysinstall, the process of building
210 release media has changed significantly. For details, please re-read
211 release(7), which has been updated to reflect the new build process.
214 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
215 is the last available version under GPLv2. It brings a number of new
216 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
217 4.1 and SSE 4.2), better support for powerpc64, a number of new
218 directives, and lots of other small improvements. See the ChangeLog
219 file in contrib/binutils for the full details.
222 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
223 compliant, and will now use half of hash for authentication.
224 This will break interoperability with all stacks (including all
225 actual FreeBSD versions) who implement
226 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
228 The only workaround with such peers is to use another HMAC
229 algorithm for IPsec ("phase 2") authentication.
232 Remove the uio_yield prototype and symbol. This function has
233 been misnamed since it was introduced and should not be
234 globally exposed with this name. The equivalent functionality
235 is now available using kern_yield(curthread->td_user_pri).
236 The function remains undocumented.
239 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
240 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
241 for scalar sysctls is defined but disabled. Code that needs
242 UQUAD to pass the type checking that must compile on older
243 systems where the define is not present can check against
244 __FreeBSD_version >= 900030.
246 The system dialog(1) has been replaced with a new version previously
247 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
248 with the previous version, but the libdialog associated with it has
249 a largely incompatible API. As such, the original version of libdialog
250 will be kept temporarily as libodialog, until its base system consumers
251 are replaced or updated. Bump __FreeBSD_version to 900030.
254 If you are trying to run make universe on a -stable system, and you get
255 the following warning:
256 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
257 unknown. config(8) likely too old."
258 or something similar to it, then you must upgrade your -stable system
259 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
260 stable/8) or build the config from the latest stable/8 branch and
261 install it on your system.
263 Prior to this date, building a current universe on 8-stable system from
264 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
265 error in the first kernel build phase. A new config on those old
266 systems will fix that problem for older versions of -current.
269 The TCP stack has been modified to allow Khelp modules to interact with
270 it via helper hook points and store per-connection data in the TCP
271 control block. Bump __FreeBSD_version to 900029. User space tools that
272 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
276 Generic IEEE 802.3 annex 31B full duplex flow control support has been
277 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
278 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
279 to take advantage of it instead of using custom implementations. This
280 means that these drivers now no longer unconditionally advertise
281 support for flow control but only do so if flow control is a selected
282 media option. This was implemented in the generic support that way in
283 order to allow flow control to be switched on and off via ifconfig(8)
284 with the PHY specific default to typically off in order to protect
285 from unwanted effects. Consequently, if you used flow control with
286 one of the above mentioned drivers you now need to explicitly enable
288 ifconfig bge0 media auto mediaopt flowcontrol
290 Along with the above mentioned changes generic support for setting
291 1000baseT master mode also has been added and brgphy(4), ciphy(4),
292 e1000phy(4) as well as ip1000phy(4) have been converted to take
293 advantage of it. This means that these drivers now no longer take the
294 link0 parameter for selecting master mode but the master media option
295 has to be used instead, for example like in the following:
296 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
298 Selection of master mode now is also available with all other PHY
299 drivers supporting 1000baseT.
302 The TCP stack has received a significant update to add support for
303 modularised congestion control and generally improve the clarity of
304 congestion control decisions. Bump __FreeBSD_version to 900025. User
305 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
306 sockstat) need to be recompiled.
309 The man(1) utility has been replaced by a new version that no longer
310 uses /etc/manpath.config. Please consult man.conf(5) for how to
311 migrate local entries to the new format.
314 The copyright strings printed by login(1) and sshd(8) at the time of a
315 new connection have been removed to follow other operating systems and
319 A workaround for a fixed ld bug has been removed in kernel code,
320 so make sure that your system ld is built from sources after
321 revision 210245 from 2010-07-19 (r211583 if building head kernel
322 on stable/8, r211584 for stable/7; both from 2010-08-21).
323 A symptom of incorrect ld version is different addresses for
324 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
327 The $ipv6_prefer variable in rc.conf(5) has been split into
328 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
330 The $ip6addrctl_policy is a variable to choose a pre-defined
331 address selection policy set by ip6addrctl(8). A value
332 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
335 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
336 flag (see an entry of 20090926) is set on an interface with no
337 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
338 security reason. If you want IPv6 link-local address on all
339 interfaces by default, set this to "YES".
341 The old ipv6_prefer="YES" is equivalent to
342 ipv6_activate_all_interfaces="YES" and
343 ip6addrctl_policy="ipv6_prefer".
346 DTrace has grown support for userland tracing. Due to this, DTrace is
347 now i386 and amd64 only.
348 dtruss(1) is now installed by default on those systems and a new
349 kernel module is needed for userland tracing: fasttrap.
350 No changes to your kernel config file are necessary to enable
351 userland tracing, but you might consider adding 'STRIP=' and
352 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
353 to have informative userland stack traces in DTrace (ustack).
356 The acpi_aiboost(4) driver has been removed in favor of the new
357 aibs(4) driver. You should update your kernel configuration file.
360 BSD grep has been imported to the base system and it is built by
361 default. It is completely BSD licensed, highly GNU-compatible, uses
362 less memory than its GNU counterpart and has a small codebase.
363 However, it is slower than its GNU counterpart, which is mostly
364 noticeable for larger searches, for smaller ones it is measurable
365 but not significant. The reason is complex, the most important factor
366 is that we lack a modern and efficient regex library and GNU
367 overcomes this by optimizing the searches internally. Future work
368 on improving the regex performance is planned, for the meantime,
369 users that need better performance, can build GNU grep instead by
370 setting the WITH_GNU_GREP knob.
373 Due to the import of powerpc64 support, all existing powerpc kernel
374 configuration files must be updated with a machine directive like this:
375 machine powerpc powerpc
377 In addition, an updated config(8) is required to build powerpc kernels
381 A new version of ZFS (version 15) has been merged to -HEAD.
382 This version uses a python library for the following subcommands:
383 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
384 For full functionality of these commands the following port must
385 be installed: sysutils/py-zfs
388 'vm_page's are now hashed by physical address to an array of mutexes.
389 Currently this is only used to serialize access to hold_count. Over
390 time the page queue mutex will be peeled away. This changes the size
391 of pmap on every architecture. And requires all callers of vm_page_hold
392 and vm_page_unhold to be updated.
395 WITH_CTF can now be specified in src.conf (not recommended, there
396 are some problems with static executables), make.conf (would also
397 affect ports which do not use GNU make and do not override the
398 compile targets) or in the kernel config (via "makeoptions
400 When WITH_CTF was specified there before this was silently ignored,
401 so make sure that WITH_CTF is not used in places which could lead
402 to unwanted behavior.
405 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
406 to allow 32-bit compatibility on non-x86 platforms. All kernel
407 configurations on amd64 and ia64 platforms using these options must
408 be modified accordingly.
411 The utmp user accounting database has been replaced with utmpx,
412 the user accounting interface standardized by POSIX.
413 Unfortunately the semantics of utmp and utmpx don't match,
414 making it practically impossible to support both interfaces.
415 The user accounting database is used by tools like finger(1),
416 last(1), talk(1), w(1) and ac(8).
418 All applications in the base system use utmpx. This means only
419 local binaries (e.g. from the ports tree) may still use these
420 utmp database files. These applications must be rebuilt to make
423 After the system has been upgraded, it is safe to remove the old
424 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
425 assuming their contents is of no importance anymore. Old wtmp
426 databases can only be used by last(1) and ac(8) after they have
427 been converted to the new format using wtmpcvt(1).
430 Introduce the kernel thread "deadlock resolver" (which can be enabled
431 via the DEADLKRES option, see NOTES for more details) and the
432 sleepq_type() function for sleepqueues.
435 The rc.firewall and rc.firewall6 were unified, and
436 rc.firewall6 and rc.d/ip6fw were removed.
437 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
438 variables are obsoleted. Instead, the following new rc
439 variables are added to rc.d/ipfw:
441 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
442 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
443 firewall_simple_onet_ipv6, firewall_trusted_ipv6
445 The meanings correspond to the relevant IPv4 variables.
451 The default terminal emulation for syscons(4) has been changed
452 from cons25 to xterm on all platforms except pc98. This means
453 that the /etc/ttys file needs to be updated to ensure correct
454 operation of applications on the console.
456 The terminal emulation style can be toggled per window by using
457 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
458 options can be used to change the compile-time default back to
461 To prevent graphical artifacts, make sure the TERM environment
462 variable is set to match the terminal emulation that is being
463 performed by syscons(4).
466 The layout of the structure ieee80211req_scan_result has changed.
467 Applications that require wireless scan results (e.g. ifconfig(8))
468 from net80211 need to be recompiled.
470 Applications such as wpa_supplicant(8) may require a full world
471 build without using NO_CLEAN in order to get synchronized with the
475 The iwn(4) driver has been updated to support the 5000 and 5150 series.
476 There's one kernel module for each firmware. Adding "device iwnfw"
477 to the kernel configuration file means including all three firmware
478 images inside the kernel. If you want to include just the one for
479 your wireless card, use the devices iwn4965fw, iwn5000fw or
483 The rc.d/network_ipv6, IPv6 configuration script has been integrated
484 into rc.d/netif. The changes are the following:
486 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
487 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
488 Note that both variables need the "inet6" keyword at the head.
490 Do not set $ipv6_network_interfaces manually if you do not
491 understand what you are doing. It is not needed in most cases.
493 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
496 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
497 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
499 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
500 all configured interfaces have "inet6 accept_rtadv" in the
501 $ifconfig_IF_ipv6. These are for backward compatibility.
503 3. A new variable $ipv6_prefer has been added. If NO, IPv6
504 functionality of interfaces with no corresponding
505 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
506 and the default address selection policy of ip6addrctl(8)
507 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
508 Note that if you want to configure IPv6 functionality on the
509 disabled interfaces after boot, first you need to clear the flag by
510 using ifconfig(8) like:
512 ifconfig em0 inet6 -ifdisabled
514 If YES, the default address selection policy is set as
517 The default value of $ipv6_prefer is NO.
519 4. If your system need to receive Router Advertisement messages,
520 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
521 scripts automatically invoke rtsol(8) when the interface becomes
522 UP. The Router Advertisement messages are used for SLAAC
523 (State-Less Address AutoConfiguration).
526 802.11s D3.03 support was committed. This is incompatible with the
527 previous code, which was based on D3.0.
530 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
531 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
532 control whether accepting Router Advertisement messages or not.
533 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
534 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
535 The ifconfig(8) utility now supports these flags.
538 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
539 mount(8) and -a option for df(1) to see them.
542 The old tunable hw.bus.devctl_disable has been superseded by
543 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
544 replaced by hw.bus.devctl_queue=0. The default for this new tunable
548 Remove the option STOP_NMI. The default action is now to use NMI only
549 for KDB via the newly introduced function stop_cpus_hard() and
550 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
553 The stable/8 branch created in subversion. This corresponds to the
554 RELENG_8 branch in CVS.
557 Bump the shared library version numbers for all libraries that do not
558 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
559 __FreeBSD_version to 800105.
562 Due to changes in the implementation of virtual network stack support,
563 all network-related kernel modules must be recompiled. As this change
564 breaks the ABI, bump __FreeBSD_version to 800104.
567 The TOE interface to the TCP syncache has been modified to remove
568 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
569 The cxgb driver is the only TOE consumer affected by this change, and
570 needs to be recompiled along with the kernel. As this change breaks
571 the ABI, bump __FreeBSD_version to 800103.
574 Padding has been added to struct tcpcb, sackhint and tcpstat in
575 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
576 maintaining the ABI. However, this change breaks the ABI, so bump
577 __FreeBSD_version to 800102. User space tools that rely on the size of
578 any of these structs (e.g. sockstat) need to be recompiled.
581 The NFS_LEGACYRPC option has been removed along with the old kernel
582 RPC implementation that this option selected. Kernel configurations
583 may need to be adjusted.
586 The network interface device nodes at /dev/net/<interface> have been
587 removed. All ioctl operations can be performed the normal way using
588 routing sockets. The kqueue functionality can generally be replaced
589 with routing sockets.
592 The documentation from the FreeBSD Documentation Project (Handbook,
593 FAQ, etc.) is now installed via packages by sysinstall(8) and under
594 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
597 The ABI of various structures related to the SYSV IPC API have been
598 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
599 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
603 Layout of struct vnet has changed as routing related variables were
604 moved to their own Vimage module. Modules need to be recompiled. Bump
605 __FreeBSD_version to 800099.
608 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
609 respectively. As long as no more than 16 groups per process are used,
610 no changes should be visible. When more than 16 groups are used, old
611 binaries may fail if they call getgroups() or getgrouplist() with
612 statically sized storage. Recompiling will work around this, but
613 applications should be modified to use dynamically allocated storage
614 for group arrays as POSIX.1-2008 does not cap an implementation's
615 number of supported groups at NGROUPS_MAX+1 as previous versions did.
617 NFS and portalfs mounts may also be affected as the list of groups is
618 truncated to 16. Users of NFS who use more than 16 groups, should
619 take care that negative group permissions are not used on the exported
620 file systems as they will not be reliable unless a GSSAPI based
621 authentication method is used.
624 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
625 option compiles in the support for adaptive spinning for lockmgrs
626 which want to enable it. The lockinit() function now accepts the flag
627 LK_ADAPTIVE in order to make the lock object subject to adaptive
628 spinning when both held in write and read mode.
631 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
632 changed. User applications that use this ioctl need to be rebuilt.
635 The layout of struct thread has changed. Kernel and modules need to
639 The layout of structs ifnet, domain, protosw and vnet_net has changed.
640 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
643 window(1) has been removed from the base system. It can now be
644 installed from ports. The port is called misc/window.
647 The way we are storing and accessing `routing table' entries has
648 changed. Programs reading the FIB, like netstat, need to be
652 A new netisr implementation has been added for FreeBSD 8. Network
653 file system modules, such as igmp, ipdivert, and others, should be
655 Bump __FreeBSD_version to 800096.
658 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
662 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
663 Bump __FreeBSD_version to 800094.
666 Add mnt_xflag field to 'struct mount'. File system modules need to be
668 Bump __FreeBSD_version to 800093.
671 The compiling option ADAPTIVE_SX has been retired while it has been
672 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
673 The KPI for sx_init_flags() changes as accepting flags:
674 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
675 been introduced in order to handle the reversed logic.
676 Bump __FreeBSD_version to 800092.
679 Add support for hierarchical jails. Remove global securelevel.
680 Bump __FreeBSD_version to 800091.
683 The layout of struct vnet_net has changed, therefore modules
685 Bump __FreeBSD_version to 800090.
688 The newly imported zic(8) produces a new format in the output. Please
689 run tzsetup(8) to install the newly created data to /etc/localtime.
692 The sysctl tree for the usb stack has renamed from hw.usb2.* to
693 hw.usb.* and is now consistent again with previous releases.
696 802.11 monitor mode support was revised and driver api's were changed.
697 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
698 of DLT_IEEE802_11. No user-visible data structures were changed but
699 applications that use DLT_IEEE802_11 may require changes.
700 Bump __FreeBSD_version to 800088.
703 The layout of the following structs has changed: sysctl_oid,
704 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
705 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
706 panics may be experienced. World rebuild is required for
707 correctly checking networking state from userland.
708 Bump __FreeBSD_version to 800085.
711 MLDv2 and Source-Specific Multicast (SSM) have been merged
712 to the IPv6 stack. VIMAGE hooks are in but not yet used.
713 The implementation of SSM within FreeBSD's IPv6 stack closely
714 follows the IPv4 implementation.
716 For kernel developers:
718 * The most important changes are that the ip6_output() and
719 ip6_input() paths no longer take the IN6_MULTI_LOCK,
720 and this lock has been downgraded to a non-recursive mutex.
722 * As with the changes to the IPv4 stack to support SSM, filtering
723 of inbound multicast traffic must now be performed by transport
724 protocols within the IPv6 stack. This does not apply to TCP and
725 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
727 * The KPIs used by IPv6 multicast are similar to those used by
728 the IPv4 stack, with the following differences:
729 * im6o_mc_filter() is analogous to imo_multicast_filter().
730 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
731 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
732 * IN6_LOOKUP_MULTI() has been deprecated and removed.
733 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
734 for MLDv1 have an additional 'timer' argument which is used to
735 jitter the initial membership report for the solicited-node
736 multicast membership on-link.
737 * This is not strictly needed for MLDv2, which already jitters
738 its report transmissions. However, the 'timer' argument is
739 preserved in case MLDv1 is active on the interface.
741 * The KAME linked-list based IPv6 membership implementation has
742 been refactored to use a vector similar to that used by the IPv4
744 Code which maintains a list of its own multicast memberships
745 internally, e.g. carp, has been updated to reflect the new
748 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
749 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
750 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
751 implementation constraint which needs to be addressed in HEAD.
753 For application developers:
755 * The changes are broadly similar to those made for the IPv4
758 * The use of IPv4 and IPv6 multicast socket options on the same
759 socket, using mapped addresses, HAS NOT been tested or supported.
761 * There are a number of issues with the implementation of various
762 IPv6 multicast APIs which need to be resolved in the API surface
763 before the implementation is fully compatible with KAME userland
764 use, and these are mostly to do with interface index treatment.
766 * The literature available discusses the use of either the delta / ASM
767 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
768 using setsourcefilter(3)/getsourcefilter(3). For more information
769 please refer to RFC 3768, 'Socket Interface Extensions for
770 Multicast Source Filters'.
772 * Applications which use the published RFC 3678 APIs should be fine.
774 For systems administrators:
776 * The mtest(8) utility has been refactored to support IPv6, in
777 addition to IPv4. Interface addresses are no longer accepted
778 as arguments, their names must be used instead. The utility
779 will map the interface name to its first IPv4 address as
780 returned by getifaddrs(3).
782 * The ifmcstat(8) utility has also been updated to print the MLDv2
783 endpoint state and source filter lists via sysctl(3).
785 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
786 loopback of IPv6 multicast datagrams by default; it defaults to 1
787 to preserve the existing behaviour. Disabling multicast loopback is
788 recommended for optimal system performance.
790 * The IPv6 MROUTING code has been changed to examine this sysctl
791 instead of attempting to perform a group lookup before looping
792 back forwarded datagrams.
794 Bump __FreeBSD_version to 800084.
797 Implement low-level Bluetooth HCI API.
798 Bump __FreeBSD_version to 800083.
801 The layout of struct malloc_type, used by modules to register new
802 memory allocation types, has changed. Most modules will need to
803 be rebuilt or panics may be experienced.
804 Bump __FreeBSD_version to 800081.
807 Anticipate overflowing inp_flags - add inp_flags2.
808 This changes most offsets in inpcb, so checking v4 connection
809 state will require a world rebuild.
810 Bump __FreeBSD_version to 800080.
813 Add an llentry to struct route and struct route_in6. Modules
814 embedding a struct route will need to be recompiled.
815 Bump __FreeBSD_version to 800079.
818 The size of rt_metrics_lite and by extension rtentry has changed.
819 Networking administration apps will need to be recompiled.
820 The route command now supports show as an alias for get, weighting
821 of routes, sticky and nostick flags to alter the behavior of stateful
823 Bump __FreeBSD_version to 800078.
826 Do not use Giant for kbdmux(4) locking. This is wrong and
827 apparently causing more problems than it solves. This will
828 re-open the issue where interrupt handlers may race with
829 kbdmux(4) in polling mode. Typical symptoms include (but
830 not limited to) duplicated and/or missing characters when
831 low level console functions (such as gets) are used while
832 interrupts are enabled (for example geli password prompt,
833 mountroot prompt etc.). Disabling kbdmux(4) may help.
836 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
837 kernel modules referencing any of the above need to be recompiled.
838 Bump __FreeBSD_version to 800075.
841 GEOM_PART has become the default partition slicer for storage devices,
842 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
843 introduces some changes:
845 MSDOS/EBR: the devices created from MSDOS extended partition entries
846 (EBR) can be named differently than with GEOM_MBR and are now symlinks
847 to devices with offset-based names. fstabs may need to be modified.
849 BSD: the "geometry does not match label" warning is harmless in most
850 cases but it points to problems in file system misalignment with
851 disk geometry. The "c" partition is now implicit, covers the whole
852 top-level drive and cannot be (mis)used by users.
854 General: Kernel dumps are now not allowed to be written to devices
855 whose partition types indicate they are meant to be used for file
856 systems (or, in case of MSDOS partitions, as something else than
859 Most of these changes date approximately from 200812.
862 The uscanner(4) driver has been removed from the kernel. This follows
863 Linux removing theirs in 2.6 and making libusb the default interface
867 The multicast forwarding code has been cleaned up. netstat(1)
868 only relies on KVM now for printing bandwidth upcall meters.
869 The IPv4 and IPv6 modules are split into ip_mroute_mod and
870 ip6_mroute_mod respectively. The config(5) options for statically
871 compiling this code remain the same, i.e. 'options MROUTING'.
874 Support for the IFF_NEEDSGIANT network interface flag has been
875 removed, which means that non-MPSAFE network device drivers are no
876 longer supported. In particular, if_ar, if_sr, and network device
877 drivers from the old (legacy) USB stack can no longer be built or
881 POSIX.1 Native Language Support (NLS) has been enabled in libc and
882 a bunch of new language catalog files have also been added.
883 This means that some common libc messages are now localized and
884 they depend on the LC_MESSAGES environmental variable.
887 The k8temp(4) driver has been renamed to amdtemp(4) since
888 support for Family 10 and Family 11 CPU families was added.
891 IGMPv3 and Source-Specific Multicast (SSM) have been merged
892 to the IPv4 stack. VIMAGE hooks are in but not yet used.
894 For kernel developers, the most important changes are that the
895 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
896 and this lock has been downgraded to a non-recursive mutex.
898 Transport protocols (UDP, Raw IP) are now responsible for filtering
899 inbound multicast traffic according to group membership and source
900 filters. The imo_multicast_filter() KPI exists for this purpose.
901 Transports which do not use multicast (SCTP, TCP) already reject
902 multicast by default. Forwarding and receive performance may improve
903 as a mutex acquisition is no longer needed in the ip_input()
904 low-level input path. in_addmulti() and in_delmulti() are shimmed
905 to new KPIs which exist to support SSM in-kernel.
907 For application developers, it is recommended that loopback of
908 multicast datagrams be disabled for best performance, as this
909 will still cause the lock to be taken for each looped-back
910 datagram transmission. The net.inet.ip.mcast.loop sysctl may
911 be tuned to 0 to disable loopback by default; it defaults to 1
912 to preserve the existing behaviour.
914 For systems administrators, to obtain best performance with
915 multicast reception and multiple groups, it is always recommended
916 that a card with a suitably precise hash filter is used. Hash
917 collisions will still result in the lock being taken within the
918 transport protocol input path to check group membership.
920 If deploying FreeBSD in an environment with IGMP snooping switches,
921 it is recommended that the net.inet.igmp.sendlocal sysctl remain
922 enabled; this forces 224.0.0.0/24 group membership to be announced
925 The size of 'struct igmpstat' has changed; netstat needs to be
926 recompiled to reflect this.
927 Bump __FreeBSD_version to 800070.
930 libusb20.so.1 is now installed as libusb.so.1 and the ports system
931 updated to use it. This requires a buildworld/installworld in order to
932 update the library and dependencies (usbconfig, etc). Its advisable to
933 rebuild all ports which uses libusb. More specific directions are given
934 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
935 libusb are no longer required and can be removed.
938 A workaround is committed to allow the creation of System V shared
939 memory segment of size > 2 GB on the 64-bit architectures.
940 Due to a limitation of the existing ABI, the shm_segsz member
941 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
942 wrong for large segments. Note that limits must be explicitly
943 raised to allow such segments to be created.
946 The layout of struct ifnet has changed, requiring a rebuild of all
947 network device driver modules.
950 The /dev handling for the new USB stack has changed, a
951 buildworld/installworld is required for libusb20.
954 The new USB2 stack has now been permanently moved in and all kernel and
955 module names reverted to their previous values (eg, usb, ehci, ohci,
956 ums, ...). The old usb stack can be compiled in by prefixing the name
957 with the letter 'o', the old usb modules have been removed.
958 Updating entry 20090216 for xorg and 20090215 for libmap may still
962 The rc.conf(5) option if_up_delay has been renamed to
963 defaultroute_delay to better reflect its purpose. If you have
964 customized this setting in /etc/rc.conf you need to update it to
968 xorg 7.4 wants to configure its input devices via hald which does not
969 yet work with USB2. If the keyboard/mouse does not work in xorg then
971 Option "AllowEmptyInput" "off"
972 to your ServerLayout section. This will cause X to use the configured
973 kbd and mouse sections from your xorg.conf.
976 The GENERIC kernels for all architectures now default to the new USB2
977 stack. No kernel config options or code have been removed so if a
978 problem arises please report it and optionally revert to the old USB
979 stack. If you are loading USB kernel modules or have a custom kernel
980 that includes GENERIC then ensure that usb names are also changed over,
981 eg uftdi -> usb2_serial_ftdi.
983 Older programs linked against the ports libusb 0.1 need to be
984 redirected to the new stack's libusb20. /etc/libmap.conf can
986 # Map old usb library to new one for usb2 stack
987 libusb-0.1.so.8 libusb20.so.1
990 All USB ethernet devices now attach as interfaces under the name ueN
991 (eg. ue0). This is to provide a predictable name as vendors often
992 change usb chipsets in a product without notice.
995 The ichsmb(4) driver has been changed to require SMBus slave
996 addresses be left-justified (xxxxxxx0b) rather than right-justified.
997 All of the other SMBus controller drivers require left-justified
998 slave addresses, so this change makes all the drivers provide the
1002 INET6 statistics (struct ip6stat) was updated.
1003 netstat(1) needs to be recompiled.
1006 NTFS has been removed from GENERIC kernel on amd64 to match
1007 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1008 will load ntfs.ko module automatically when NTFS support is
1009 actually needed, unless ntfs.ko is not installed or security
1010 level prohibits loading kernel modules. If either is the case,
1011 "options NTFS" has to be added into kernel config.
1014 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1015 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1016 800061. User space tools that rely on the size of struct tcpcb in
1017 tcp_var.h (e.g. sockstat) need to be recompiled.
1020 ng_tty(4) module updated to match the new TTY subsystem.
1021 Due to API change, user-level applications must be updated.
1022 New API support added to mpd5 CVS and expected to be present
1023 in next mpd5.3 release.
1026 With __FreeBSD_version 800060 the makefs tool is part of
1027 the base system (it was a port).
1030 The afdata and ifnet locks have been changed from mutexes to
1031 rwlocks, network modules will need to be re-compiled.
1034 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1035 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1036 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1037 architecture and 40 bytes on 64-bit architecture. The userland
1038 applications "arp" and "ndp" have been updated accordingly.
1039 The output from "netstat -r" shows only routing entries and
1040 none of the L2 information.
1043 __FreeBSD_version 800057 marks the switchover from the
1044 binary ath hal to source code. Users must add the line:
1046 options AH_SUPPORT_AR5416
1048 to their kernel config files when specifying:
1052 The ath_hal module no longer exists; the code is now compiled
1053 together with the driver in the ath module. It is now
1054 possible to tailor chip support (i.e. reduce the set of chips
1055 and thereby the code size); consult ath_hal(4) for details.
1058 __FreeBSD_version 800054 adds memory barriers to
1059 <machine/atomic.h>, new interfaces to ifnet to facilitate
1060 multiple hardware transmit queues for cards that support
1061 them, and a lock-less ring-buffer implementation to
1062 enable drivers to more efficiently manage queueing of
1066 A new version of ZFS (version 13) has been merged to -HEAD.
1067 This version has zpool attribute "listsnapshots" off by
1068 default, which means "zfs list" does not show snapshots,
1069 and is the same as Solaris behavior.
1072 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1075 The uhci, ohci, ehci and slhci USB Host controller drivers have
1076 been put into separate modules. If you load the usb module
1077 separately through loader.conf you will need to load the
1078 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1079 controller add the following to loader.conf:
1085 The ABI used by the PMC toolset has changed. Please keep
1086 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1090 atapci kernel module now includes only generic PCI ATA
1091 driver. AHCI driver moved to ataahci kernel module.
1092 All vendor-specific code moved into separate kernel modules:
1093 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1094 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1095 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1096 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1099 The TTY subsystem of the kernel has been replaced by a new
1100 implementation, which provides better scalability and an
1101 improved driver model. Most common drivers have been migrated to
1102 the new TTY subsystem, while others have not. The following
1103 drivers have not yet been ported to the new TTY layer:
1106 cy, digi, rc, rp, sio
1112 ng_h4, ng_tty, ppp, sl, snp
1114 Adding these drivers to your kernel configuration file shall
1115 cause compilation to fail.
1118 ntpd has been upgraded to 4.2.4p5.
1121 OpenSSH has been upgraded to 5.1p1.
1123 For many years, FreeBSD's version of OpenSSH preferred DSA
1124 over RSA for host and user authentication keys. With this
1125 upgrade, we've switched to the vendor's default of RSA over
1126 DSA. This may cause upgraded clients to warn about unknown
1127 host keys even for previously known hosts. Users should
1128 follow the usual procedure for verifying host keys before
1129 accepting the RSA key.
1131 This can be circumvented by setting the "HostKeyAlgorithms"
1132 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1135 Please note that the sequence of keys offered for
1136 authentication has been changed as well. You may want to
1137 specify IdentityFile in a different order to revert this
1141 The sio(4) driver has been removed from the i386 and amd64
1142 kernel configuration files. This means uart(4) is now the
1143 default serial port driver on those platforms as well.
1145 To prevent collisions with the sio(4) driver, the uart(4) driver
1146 uses different names for its device nodes. This means the
1147 onboard serial port will now most likely be called "ttyu0"
1148 instead of "ttyd0". You may need to reconfigure applications to
1149 use the new device names.
1151 When using the serial port as a boot console, be sure to update
1152 /boot/device.hints and /etc/ttys before booting the new kernel.
1153 If you forget to do so, you can still manually specify the hints
1154 at the loader prompt:
1156 set hint.uart.0.at="isa"
1157 set hint.uart.0.port="0x3F8"
1158 set hint.uart.0.flags="0x10"
1159 set hint.uart.0.irq="4"
1163 The gpt(8) utility has been removed. Use gpart(8) to partition
1167 The version that Linuxulator emulates was changed from 2.4.2
1168 to 2.6.16. If you experience any problems with Linux binaries
1169 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1170 if it fixes the problem contact emulation mailing list.
1173 ISDN4BSD (I4B) was removed from the src tree. You may need to
1174 update a your kernel configuration and remove relevant entries.
1177 I have checked in code to support multiple routing tables.
1178 See the man pages setfib(1) and setfib(2).
1179 This is a hopefully backwards compatible version,
1180 but to make use of it you need to compile your kernel
1181 with options ROUTETABLES=2 (or more up to 16).
1184 The 802.11 wireless support was redone to enable multi-bss
1185 operation on devices that are capable. The underlying device
1186 is no longer used directly but instead wlanX devices are
1187 cloned with ifconfig. This requires changes to rc.conf files.
1188 For example, change:
1189 ifconfig_ath0="WPA DHCP"
1192 ifconfig_wlan0="WPA DHCP"
1193 see rc.conf(5) for more details. In addition, mergemaster of
1194 /etc/rc.d is highly recommended. Simultaneous update of userland
1195 and kernel wouldn't hurt either.
1197 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1198 modules were merged into the base wlan module. All references
1199 to these modules (e.g. in kernel config files) must be removed.
1202 psm(4) has gained write(2) support in native operation level.
1203 Arbitrary commands can be written to /dev/psm%d and status can
1204 be read back from it. Therefore, an application is responsible
1205 for status validation and error recovery. It is a no-op in
1206 other operation levels.
1209 Support for KSE threading has been removed from the kernel. To
1210 run legacy applications linked against KSE libmap.conf may
1211 be used. The following libmap.conf may be used to ensure
1212 compatibility with any prior release:
1214 libpthread.so.1 libthr.so.1
1215 libpthread.so.2 libthr.so.2
1216 libkse.so.3 libthr.so.3
1219 The layout of struct vmspace has changed. This affects libkvm
1220 and any executables that link against libkvm and use the
1221 kvm_getprocs() function. In particular, but not exclusively,
1222 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1223 The effects are minimal, but it's advisable to upgrade world
1227 The latest em driver no longer has support in it for the
1228 82575 adapter, this is now moved to the igb driver. The
1229 split was done to make new features that are incompatible
1230 with older hardware easier to do.
1233 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1234 likewise the kernel option is now GEOM_LINUX_LVM.
1237 The default NFS mount mode has changed from UDP to TCP for
1238 increased reliability. If you rely on (insecurely) NFS
1239 mounting across a firewall you may need to update your
1243 Belatedly note the addition of m_collapse for compacting
1247 The fts(3) structures have been changed to use adequate
1248 integer types for their members and so to be able to cope
1249 with huge file trees. The old fts(3) ABI is preserved
1250 through symbol versioning in libc, so third-party binaries
1251 using fts(3) should still work, although they will not take
1252 advantage of the extended types. At the same time, some
1253 third-party software might fail to build after this change
1254 due to unportable assumptions made in its source code about
1255 fts(3) structure members. Such software should be fixed
1256 by its vendor or, in the worst case, in the ports tree.
1257 FreeBSD_version 800015 marks this change for the unlikely
1258 case that a portable fix is impossible.
1261 To upgrade to -current after this date, you must be running
1262 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1263 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1266 The ADAPTIVE_GIANT kernel option has been retired because its
1267 functionality is the default now.
1270 The AT keyboard emulation of sunkbd(4) has been turned on
1271 by default. In order to make the special symbols of the Sun
1272 keyboards driven by sunkbd(4) work under X these now have
1273 to be configured the same way as Sun USB keyboards driven
1274 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1276 Option "XkbLayout" "us"
1277 Option "XkbRules" "xorg"
1278 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1281 It has been decided that it is desirable to provide ABI
1282 backwards compatibility to the FreeBSD 4/5/6 versions of the
1283 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1284 broken with the introduction of PCI domain support (see the
1285 20070930 entry). Unfortunately, this required the ABI of
1286 PCIOCGETCONF to be broken again in order to be able to
1287 provide backwards compatibility to the old version of that
1288 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1289 again. As for prominent ports this affects neither pciutils
1290 nor xorg-server this time, the hal port needs to be rebuilt
1294 The misnamed kthread_create() and friends have been renamed
1295 to kproc_create() etc. Many of the callers already
1296 used kproc_start()..
1297 I will return kthread_create() and friends in a while
1298 with implementations that actually create threads, not procs.
1299 Renaming corresponds with version 800002.
1308 Avoid using make -j when upgrading. While generally safe, there are
1309 sometimes problems using -j to upgrade. If your upgrade fails with
1310 -j, please try again without -j. From time to time in the past there
1311 have been problems using -j with buildworld and/or installworld. This
1312 is especially true when upgrading between "distant" versions (eg one
1313 that cross a major release boundary or several minor releases, or when
1314 several months have passed on the -current branch).
1316 Sometimes, obscure build problems are the result of environment
1317 poisoning. This can happen because the make utility reads its
1318 environment when searching for values for global variables. To run
1319 your build attempts in an "environmental clean room", prefix all make
1320 commands with 'env -i '. See the env(1) manual page for more details.
1322 When upgrading from one major version to another it is generally best
1323 to upgrade to the latest code in the currently installed branch first,
1324 then do an upgrade to the new branch. This is the best-tested upgrade
1325 path, and has the highest probability of being successful. Please try
1326 this approach before reporting problems with a major version upgrade.
1330 When upgrading the boot ZFS pool to a new version, always follow
1333 1.) recompile and reinstall the ZFS boot loader and boot block
1334 (this is part of "make buildworld" and "make installworld")
1336 2.) update the ZFS boot block on your boot drive
1338 The following example updates the ZFS boot block on the first
1339 partition (freebsd-boot) of a GPT partitioned drive ad0:
1340 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1342 Non-boot pools do not need these updates.
1346 If you are updating from a prior version of FreeBSD (even one just
1347 a few days old), you should follow this procedure. It is the most
1348 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1350 make kernel-toolchain
1351 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1352 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1354 To test a kernel once
1355 ---------------------
1356 If you just want to boot a kernel once (because you are not sure
1357 if it works, or if you want to boot a known bad kernel to provide
1358 debugging information) run
1359 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1360 nextboot -k testkernel
1362 To just build a kernel when you know that it won't mess you up
1363 --------------------------------------------------------------
1364 This assumes you are already running a CURRENT system. Replace
1365 ${arch} with the architecture of your machine (e.g. "i386",
1366 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1368 cd src/sys/${arch}/conf
1369 config KERNEL_NAME_HERE
1370 cd ../compile/KERNEL_NAME_HERE
1375 If this fails, go to the "To build a kernel" section.
1377 To rebuild everything and install it on the current system.
1378 -----------------------------------------------------------
1379 # Note: sometimes if you are running current you gotta do more than
1380 # is listed here if you are upgrading from a really old current.
1382 <make sure you have good level 0 dumps>
1384 make kernel KERNCONF=YOUR_KERNEL_HERE
1386 <reboot in single user> [3]
1394 To cross-install current onto a separate partition
1395 --------------------------------------------------
1396 # In this approach we use a separate partition to hold
1397 # current's root, 'usr', and 'var' directories. A partition
1398 # holding "/", "/usr" and "/var" should be about 2GB in
1401 <make sure you have good level 0 dumps>
1404 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1405 <maybe newfs current's root partition>
1406 <mount current's root partition on directory ${CURRENT_ROOT}>
1407 make installworld DESTDIR=${CURRENT_ROOT}
1408 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1409 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1410 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1411 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1412 <reboot into current>
1413 <do a "native" rebuild/install as described in the previous section>
1414 <maybe install compatibility libraries from ports/misc/compat*>
1418 To upgrade in-place from 8.x-stable to current
1419 ----------------------------------------------
1420 <make sure you have good level 0 dumps>
1422 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1424 <reboot in single user> [3]
1431 Make sure that you've read the UPDATING file to understand the
1432 tweaks to various things you need. At this point in the life
1433 cycle of current, things change often and you are on your own
1434 to cope. The defaults can also change, so please read ALL of
1435 the UPDATING entries.
1437 Also, if you are tracking -current, you must be subscribed to
1438 freebsd-current@freebsd.org. Make sure that before you update
1439 your sources that you have read and understood all the recent
1440 messages there. If in doubt, please track -stable which has
1441 much fewer pitfalls.
1443 [1] If you have third party modules, such as vmware, you
1444 should disable them at this point so they don't crash your
1447 [3] From the bootblocks, boot -s, and then do
1452 adjkerntz -i # if CMOS is wall time
1453 Also, when doing a major release upgrade, it is required that
1454 you boot into single user mode to do the installworld.
1456 [4] Note: This step is non-optional. Failure to do this step
1457 can result in a significant reduction in the functionality of the
1458 system. Attempting to do it by hand is not recommended and those
1459 that pursue this avenue should read this file carefully, as well
1460 as the archives of freebsd-current and freebsd-hackers mailing lists
1461 for potential gotchas. The -U option is also useful to consider.
1462 See mergemaster(8) for more information.
1464 [5] Usually this step is a noop. However, from time to time
1465 you may need to do this if you get unknown user in the following
1466 step. It never hurts to do it all the time. You may need to
1467 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1468 install) after the buildworld before this step if you last updated
1469 from current before 20020224 or from -stable before 20020408.
1471 [6] This only deletes old files and directories. Old libraries
1472 can be deleted by "make delete-old-libs", but you have to make
1473 sure that no program is using those libraries anymore.
1475 [8] In order to have a kernel that can run the 4.x binaries needed to
1476 do an installworld, you must include the COMPAT_FREEBSD4 option in
1477 your kernel. Failure to do so may leave you with a system that is
1478 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1479 required to run the 5.x binaries on more recent kernels. And so on
1480 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1482 Make sure that you merge any new devices from GENERIC since the
1483 last time you updated your kernel config file.
1485 [9] When checking out sources, you must include the -P flag to have
1486 cvs prune empty directories.
1488 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1489 "?=" instead of the "=" assignment operator, so that buildworld can
1490 override the CPUTYPE if it needs to.
1492 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1493 not on the command line, or in /etc/make.conf. buildworld will
1494 warn if it is improperly defined.
1497 This file contains a list, in reverse chronological order, of major
1498 breakages in tracking -current. It is not guaranteed to be a complete
1499 list of such breakages, and only contains entries since October 10, 2007.
1500 If you need to see UPDATING entries from before that date, you will need
1501 to fetch an UPDATING file from an older FreeBSD release.
1503 Copyright information:
1505 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1507 Redistribution, publication, translation and use, with or without
1508 modification, in full or in part, in any form or format of this
1509 document are permitted without further permission from the author.
1511 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1512 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1513 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1514 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1515 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1516 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1517 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1518 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1519 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1520 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1521 POSSIBILITY OF SUCH DAMAGE.
1523 Contact Warner Losh if you have any questions about your use of