1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
16 The sparc64 ZFS loader has been changed to no longer try to auto-
17 detect ZFS providers based on diskN aliases but now requires these
18 to be explicitly listed in the OFW boot-device environment variable.
21 Now unix domain sockets behave "as expected" on nullfs(5). Previously
22 nullfs(5) did not pass through all behaviours to the underlying layer,
23 as a result if we bound to a socket on the lower layer we could connect
24 only to the lower path; if we bound to the upper layer we could connect
25 only to the upper path. The new behavior is one can connect to both the
26 lower and the upper paths regardless what layer path one binds to.
29 The acpi_wmi(4) status device /dev/wmistat has been renamed to
33 A new VOP_ADVISE() was added to support posix_fadvise(2). All
34 filesystem modules must be recompiled.
37 The interface of the VOP_VPTOCNP(9) changed, now the returned
38 vnode shall be referenced, previously it was required to be
39 only held. All in-tree filesystems are converted.
45 The broken amd(4) driver has been replaced with esp(4) in the amd64,
46 i386 and pc98 GENERIC kernel configuration files.
49 This commit modifies vfs_register() so that it uses a hash
50 calculation to set vfc_typenum, which is enabled by default.
51 The first time a system is booted after this change, the
52 vfc_typenum values will change for all file systems. The
53 main effect of this is a change to the NFS server file handles
54 for file systems that use vfc_typenum in their fsid, such as ZFS.
55 It will, however, prevent vfc_typenum from changing when file
56 systems are loaded in a different order for subsequent reboots.
57 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
58 until you are ready to remount all NFS clients after a reboot.
61 Bump the shared library version numbers for libraries that
62 do not use symbol versioning, have changed the ABI compared
63 to stable/8 and which shared library version was not bumped.
64 Done as part of 9.0-RELEASE cycle.
67 During the merge of Capsicum features, the fget(9) KPI was modified.
68 This may require the rebuilding of out-of-tree device drivers --
69 issues have been reported specifically with the nVidia device driver.
70 __FreeBSD_version is bumped to 900041.
72 Also, there is a period between 20110811 and 20110814 where the
73 special devices /dev/{stdin,stdout,stderr} did not work correctly.
74 Building world from a kernel during that window may not work.
77 The packet filter (pf) code has been updated to OpenBSD 4.5.
78 You need to update userland tools to be in sync with kernel.
79 This update breaks backward compatibility with earlier pfsync(4)
80 versions. Care must be taken when updating redundant firewall setups.
83 The following sysctls and tunables are retired on x86 platforms:
85 machdep.hlt_logical_cpus
86 The following sysctl is retired:
87 machdep.hyperthreading_allowed
88 The sysctls were supposed to provide a way to dynamically offline and
89 online selected CPUs on x86 platforms, but the implementation has not
90 been reliable especially with SCHED_ULE scheduler.
91 machdep.hyperthreading_allowed tunable is still available to ignore
92 hyperthreading CPUs at OS level.
93 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
94 where X is an APIC ID of a CPU. Be advised, though, that disabling
95 CPUs in non-uniform fashion will result in non-uniform topology and
96 may lead to sub-optimal system performance with SCHED_ULE, which is
100 cpumask_t type is retired and cpuset_t is used in order to describe
104 Changes to ifconfig(8) for dynamic address family detection mandate
105 that you are running a kernel of 20110525 or later. Make sure to
106 follow the update procedure to boot a new kernel before installing
110 Support for sun4v architecture is officially dropped
113 Several KPI breaking changes have been committed to the mii(4) layer,
114 the PHY drivers and consequently some Ethernet drivers using mii(4).
115 This means that miibus.ko and the modules of the affected Ethernet
116 drivers need to be recompiled.
118 Note to kernel developers: Given that the OUI bit reversion problem
119 was fixed as part of these changes all mii(4) commits related to OUIs,
120 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
121 handling, no longer can be merged verbatim to stable/8 and previous
125 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
126 into their kernel configurations along with 'device pci'.
129 The default NFS client is now the new NFS client, so fstype "newnfs"
130 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
131 Although mounts via fstype "nfs" will usually work without userland
132 changes, it is recommended that the mount(8) and mount_nfs(8)
133 commands be rebuilt from sources and that a link to mount_nfs called
134 mount_oldnfs be created. The new client is compiled into the
135 kernel with "options NFSCL" and this is needed for diskless root
136 file systems. The GENERIC kernel configs have been changed to use
137 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
138 To use the regular/old client, you can "mount -t oldnfs ...". For
139 a diskless root file system, you must also include a line like:
141 vfs.root.mountfrom="oldnfs:"
143 in the boot/loader.conf on the root fs on the NFS server to make
144 a diskless root fs use the old client.
147 The GENERIC kernels for all architectures now default to the new
148 CAM-based ATA stack. It means that all legacy ATA drivers were
149 removed and replaced by respective CAM drivers. If you are using
150 ATA device names in /etc/fstab or other places, make sure to update
151 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
152 where 'Y's are the sequential numbers starting from zero for each type
153 in order of detection, unless configured otherwise with tunables,
154 see cam(4)). There will be symbolic links created in /dev/ to map
155 old adX devices to the respective adaY. They should provide basic
156 compatibility for file systems mounting in most cases, but they do
157 not support old user-level APIs and do not have respective providers
158 in GEOM. Consider using updated management tools with new device names.
160 It is possible to load devices ahci, ata, siis and mvs as modules,
161 but option ATA_CAM should remain in kernel configuration to make ata
162 module work as CAM driver supporting legacy ATA controllers. Device ata
163 still can be used in modular fashion (atacore + ...). Modules atadisk
164 and atapi* are not used and won't affect operation in ATA_CAM mode.
165 Note that to use CAM-based ATA kernel should include CAM devices
166 scbus, pass, da (or explicitly ada), cd and optionally others. All of
167 them are parts of the cam module.
169 ataraid(4) functionality is now supported by the RAID GEOM class.
170 To use it you can load geom_raid kernel module and use graid(8) tool
171 for management. Instead of /dev/arX device names, use /dev/raid/rX.
173 No kernel config options or code have been removed, so if a problem
174 arises, please report it and optionally revert to the old ATA stack.
175 In order to do it you can remove from the kernel config:
180 , and instead add back:
181 device atadisk # ATA disk drives
182 device ataraid # ATA RAID drives
183 device atapicd # ATAPI CDROM drives
184 device atapifd # ATAPI floppy drives
185 device atapist # ATAPI tape drives
188 The default NFS server has been changed to the new server, which
189 was referred to as the experimental server. If you need to switch
190 back to the old NFS server, you must now put the "-o" option on
191 both the mountd and nfsd commands. This can be done using the
192 mountd_flags and nfs_server_flags rc.conf variables until an
193 update to the rc scripts is committed, which is coming soon.
196 The GNU Objective-C runtime library (libobjc), and other Objective-C
197 related components have been removed from the base system. If you
198 require an Objective-C library, please use one of the available ports.
201 ath(4) has been split into bus- and device- modules. if_ath contains
202 the HAL, the TX rate control and the network device code. if_ath_pci
203 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
204 contains the AHB glue. Users need to load both if_ath_pci and if_ath
205 in order to use ath on everything else.
207 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
208 need to load if_ath and if_ath_pci for ath(4) operation.
211 As part of the replacement of sysinstall, the process of building
212 release media has changed significantly. For details, please re-read
213 release(7), which has been updated to reflect the new build process.
216 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
217 is the last available version under GPLv2. It brings a number of new
218 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
219 4.1 and SSE 4.2), better support for powerpc64, a number of new
220 directives, and lots of other small improvements. See the ChangeLog
221 file in contrib/binutils for the full details.
224 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
225 compliant, and will now use half of hash for authentication.
226 This will break interoperability with all stacks (including all
227 actual FreeBSD versions) who implement
228 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
230 The only workaround with such peers is to use another HMAC
231 algorithm for IPsec ("phase 2") authentication.
234 Remove the uio_yield prototype and symbol. This function has
235 been misnamed since it was introduced and should not be
236 globally exposed with this name. The equivalent functionality
237 is now available using kern_yield(curthread->td_user_pri).
238 The function remains undocumented.
241 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
242 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
243 for scalar sysctls is defined but disabled. Code that needs
244 UQUAD to pass the type checking that must compile on older
245 systems where the define is not present can check against
246 __FreeBSD_version >= 900030.
248 The system dialog(1) has been replaced with a new version previously
249 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
250 with the previous version, but the libdialog associated with it has
251 a largely incompatible API. As such, the original version of libdialog
252 will be kept temporarily as libodialog, until its base system consumers
253 are replaced or updated. Bump __FreeBSD_version to 900030.
256 If you are trying to run make universe on a -stable system, and you get
257 the following warning:
258 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
259 unknown. config(8) likely too old."
260 or something similar to it, then you must upgrade your -stable system
261 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
262 stable/8) or build the config from the latest stable/8 branch and
263 install it on your system.
265 Prior to this date, building a current universe on 8-stable system from
266 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
267 error in the first kernel build phase. A new config on those old
268 systems will fix that problem for older versions of -current.
271 The TCP stack has been modified to allow Khelp modules to interact with
272 it via helper hook points and store per-connection data in the TCP
273 control block. Bump __FreeBSD_version to 900029. User space tools that
274 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
278 Generic IEEE 802.3 annex 31B full duplex flow control support has been
279 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
280 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
281 to take advantage of it instead of using custom implementations. This
282 means that these drivers now no longer unconditionally advertise
283 support for flow control but only do so if flow control is a selected
284 media option. This was implemented in the generic support that way in
285 order to allow flow control to be switched on and off via ifconfig(8)
286 with the PHY specific default to typically off in order to protect
287 from unwanted effects. Consequently, if you used flow control with
288 one of the above mentioned drivers you now need to explicitly enable
290 ifconfig bge0 media auto mediaopt flowcontrol
292 Along with the above mentioned changes generic support for setting
293 1000baseT master mode also has been added and brgphy(4), ciphy(4),
294 e1000phy(4) as well as ip1000phy(4) have been converted to take
295 advantage of it. This means that these drivers now no longer take the
296 link0 parameter for selecting master mode but the master media option
297 has to be used instead, for example like in the following:
298 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
300 Selection of master mode now is also available with all other PHY
301 drivers supporting 1000baseT.
304 The TCP stack has received a significant update to add support for
305 modularised congestion control and generally improve the clarity of
306 congestion control decisions. Bump __FreeBSD_version to 900025. User
307 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
308 sockstat) need to be recompiled.
311 The man(1) utility has been replaced by a new version that no longer
312 uses /etc/manpath.config. Please consult man.conf(5) for how to
313 migrate local entries to the new format.
316 The copyright strings printed by login(1) and sshd(8) at the time of a
317 new connection have been removed to follow other operating systems and
321 A workaround for a fixed ld bug has been removed in kernel code,
322 so make sure that your system ld is built from sources after
323 revision 210245 from 2010-07-19 (r211583 if building head kernel
324 on stable/8, r211584 for stable/7; both from 2010-08-21).
325 A symptom of incorrect ld version is different addresses for
326 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
329 The $ipv6_prefer variable in rc.conf(5) has been split into
330 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
332 The $ip6addrctl_policy is a variable to choose a pre-defined
333 address selection policy set by ip6addrctl(8). A value
334 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
337 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
338 flag (see an entry of 20090926) is set on an interface with no
339 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
340 security reason. If you want IPv6 link-local address on all
341 interfaces by default, set this to "YES".
343 The old ipv6_prefer="YES" is equivalent to
344 ipv6_activate_all_interfaces="YES" and
345 ip6addrctl_policy="ipv6_prefer".
348 DTrace has grown support for userland tracing. Due to this, DTrace is
349 now i386 and amd64 only.
350 dtruss(1) is now installed by default on those systems and a new
351 kernel module is needed for userland tracing: fasttrap.
352 No changes to your kernel config file are necessary to enable
353 userland tracing, but you might consider adding 'STRIP=' and
354 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
355 to have informative userland stack traces in DTrace (ustack).
358 The acpi_aiboost(4) driver has been removed in favor of the new
359 aibs(4) driver. You should update your kernel configuration file.
362 BSD grep has been imported to the base system and it is built by
363 default. It is completely BSD licensed, highly GNU-compatible, uses
364 less memory than its GNU counterpart and has a small codebase.
365 However, it is slower than its GNU counterpart, which is mostly
366 noticeable for larger searches, for smaller ones it is measurable
367 but not significant. The reason is complex, the most important factor
368 is that we lack a modern and efficient regex library and GNU
369 overcomes this by optimizing the searches internally. Future work
370 on improving the regex performance is planned, for the meantime,
371 users that need better performance, can build GNU grep instead by
372 setting the WITH_GNU_GREP knob.
375 Due to the import of powerpc64 support, all existing powerpc kernel
376 configuration files must be updated with a machine directive like this:
377 machine powerpc powerpc
379 In addition, an updated config(8) is required to build powerpc kernels
383 A new version of ZFS (version 15) has been merged to -HEAD.
384 This version uses a python library for the following subcommands:
385 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
386 For full functionality of these commands the following port must
387 be installed: sysutils/py-zfs
390 'vm_page's are now hashed by physical address to an array of mutexes.
391 Currently this is only used to serialize access to hold_count. Over
392 time the page queue mutex will be peeled away. This changes the size
393 of pmap on every architecture. And requires all callers of vm_page_hold
394 and vm_page_unhold to be updated.
397 WITH_CTF can now be specified in src.conf (not recommended, there
398 are some problems with static executables), make.conf (would also
399 affect ports which do not use GNU make and do not override the
400 compile targets) or in the kernel config (via "makeoptions
402 When WITH_CTF was specified there before this was silently ignored,
403 so make sure that WITH_CTF is not used in places which could lead
404 to unwanted behavior.
407 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
408 to allow 32-bit compatibility on non-x86 platforms. All kernel
409 configurations on amd64 and ia64 platforms using these options must
410 be modified accordingly.
413 The utmp user accounting database has been replaced with utmpx,
414 the user accounting interface standardized by POSIX.
415 Unfortunately the semantics of utmp and utmpx don't match,
416 making it practically impossible to support both interfaces.
417 The user accounting database is used by tools like finger(1),
418 last(1), talk(1), w(1) and ac(8).
420 All applications in the base system use utmpx. This means only
421 local binaries (e.g. from the ports tree) may still use these
422 utmp database files. These applications must be rebuilt to make
425 After the system has been upgraded, it is safe to remove the old
426 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
427 assuming their contents is of no importance anymore. Old wtmp
428 databases can only be used by last(1) and ac(8) after they have
429 been converted to the new format using wtmpcvt(1).
432 Introduce the kernel thread "deadlock resolver" (which can be enabled
433 via the DEADLKRES option, see NOTES for more details) and the
434 sleepq_type() function for sleepqueues.
437 The rc.firewall and rc.firewall6 were unified, and
438 rc.firewall6 and rc.d/ip6fw were removed.
439 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
440 variables are obsoleted. Instead, the following new rc
441 variables are added to rc.d/ipfw:
443 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
444 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
445 firewall_simple_onet_ipv6, firewall_trusted_ipv6
447 The meanings correspond to the relevant IPv4 variables.
453 The default terminal emulation for syscons(4) has been changed
454 from cons25 to xterm on all platforms except pc98. This means
455 that the /etc/ttys file needs to be updated to ensure correct
456 operation of applications on the console.
458 The terminal emulation style can be toggled per window by using
459 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
460 options can be used to change the compile-time default back to
463 To prevent graphical artifacts, make sure the TERM environment
464 variable is set to match the terminal emulation that is being
465 performed by syscons(4).
468 The layout of the structure ieee80211req_scan_result has changed.
469 Applications that require wireless scan results (e.g. ifconfig(8))
470 from net80211 need to be recompiled.
472 Applications such as wpa_supplicant(8) may require a full world
473 build without using NO_CLEAN in order to get synchronized with the
477 The iwn(4) driver has been updated to support the 5000 and 5150 series.
478 There's one kernel module for each firmware. Adding "device iwnfw"
479 to the kernel configuration file means including all three firmware
480 images inside the kernel. If you want to include just the one for
481 your wireless card, use the devices iwn4965fw, iwn5000fw or
485 The rc.d/network_ipv6, IPv6 configuration script has been integrated
486 into rc.d/netif. The changes are the following:
488 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
489 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
490 Note that both variables need the "inet6" keyword at the head.
492 Do not set $ipv6_network_interfaces manually if you do not
493 understand what you are doing. It is not needed in most cases.
495 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
498 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
499 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
501 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
502 all configured interfaces have "inet6 accept_rtadv" in the
503 $ifconfig_IF_ipv6. These are for backward compatibility.
505 3. A new variable $ipv6_prefer has been added. If NO, IPv6
506 functionality of interfaces with no corresponding
507 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
508 and the default address selection policy of ip6addrctl(8)
509 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
510 Note that if you want to configure IPv6 functionality on the
511 disabled interfaces after boot, first you need to clear the flag by
512 using ifconfig(8) like:
514 ifconfig em0 inet6 -ifdisabled
516 If YES, the default address selection policy is set as
519 The default value of $ipv6_prefer is NO.
521 4. If your system need to receive Router Advertisement messages,
522 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
523 scripts automatically invoke rtsol(8) when the interface becomes
524 UP. The Router Advertisement messages are used for SLAAC
525 (State-Less Address AutoConfiguration).
528 802.11s D3.03 support was committed. This is incompatible with the
529 previous code, which was based on D3.0.
532 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
533 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
534 control whether accepting Router Advertisement messages or not.
535 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
536 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
537 The ifconfig(8) utility now supports these flags.
540 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
541 mount(8) and -a option for df(1) to see them.
544 The old tunable hw.bus.devctl_disable has been superseded by
545 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
546 replaced by hw.bus.devctl_queue=0. The default for this new tunable
550 Remove the option STOP_NMI. The default action is now to use NMI only
551 for KDB via the newly introduced function stop_cpus_hard() and
552 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
555 The stable/8 branch created in subversion. This corresponds to the
556 RELENG_8 branch in CVS.
559 Bump the shared library version numbers for all libraries that do not
560 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
561 __FreeBSD_version to 800105.
564 Due to changes in the implementation of virtual network stack support,
565 all network-related kernel modules must be recompiled. As this change
566 breaks the ABI, bump __FreeBSD_version to 800104.
569 The TOE interface to the TCP syncache has been modified to remove
570 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
571 The cxgb driver is the only TOE consumer affected by this change, and
572 needs to be recompiled along with the kernel. As this change breaks
573 the ABI, bump __FreeBSD_version to 800103.
576 Padding has been added to struct tcpcb, sackhint and tcpstat in
577 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
578 maintaining the ABI. However, this change breaks the ABI, so bump
579 __FreeBSD_version to 800102. User space tools that rely on the size of
580 any of these structs (e.g. sockstat) need to be recompiled.
583 The NFS_LEGACYRPC option has been removed along with the old kernel
584 RPC implementation that this option selected. Kernel configurations
585 may need to be adjusted.
588 The network interface device nodes at /dev/net/<interface> have been
589 removed. All ioctl operations can be performed the normal way using
590 routing sockets. The kqueue functionality can generally be replaced
591 with routing sockets.
594 The documentation from the FreeBSD Documentation Project (Handbook,
595 FAQ, etc.) is now installed via packages by sysinstall(8) and under
596 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
599 The ABI of various structures related to the SYSV IPC API have been
600 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
601 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
605 Layout of struct vnet has changed as routing related variables were
606 moved to their own Vimage module. Modules need to be recompiled. Bump
607 __FreeBSD_version to 800099.
610 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
611 respectively. As long as no more than 16 groups per process are used,
612 no changes should be visible. When more than 16 groups are used, old
613 binaries may fail if they call getgroups() or getgrouplist() with
614 statically sized storage. Recompiling will work around this, but
615 applications should be modified to use dynamically allocated storage
616 for group arrays as POSIX.1-2008 does not cap an implementation's
617 number of supported groups at NGROUPS_MAX+1 as previous versions did.
619 NFS and portalfs mounts may also be affected as the list of groups is
620 truncated to 16. Users of NFS who use more than 16 groups, should
621 take care that negative group permissions are not used on the exported
622 file systems as they will not be reliable unless a GSSAPI based
623 authentication method is used.
626 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
627 option compiles in the support for adaptive spinning for lockmgrs
628 which want to enable it. The lockinit() function now accepts the flag
629 LK_ADAPTIVE in order to make the lock object subject to adaptive
630 spinning when both held in write and read mode.
633 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
634 changed. User applications that use this ioctl need to be rebuilt.
637 The layout of struct thread has changed. Kernel and modules need to
641 The layout of structs ifnet, domain, protosw and vnet_net has changed.
642 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
645 window(1) has been removed from the base system. It can now be
646 installed from ports. The port is called misc/window.
649 The way we are storing and accessing `routing table' entries has
650 changed. Programs reading the FIB, like netstat, need to be
654 A new netisr implementation has been added for FreeBSD 8. Network
655 file system modules, such as igmp, ipdivert, and others, should be
657 Bump __FreeBSD_version to 800096.
660 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
664 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
665 Bump __FreeBSD_version to 800094.
668 Add mnt_xflag field to 'struct mount'. File system modules need to be
670 Bump __FreeBSD_version to 800093.
673 The compiling option ADAPTIVE_SX has been retired while it has been
674 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
675 The KPI for sx_init_flags() changes as accepting flags:
676 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
677 been introduced in order to handle the reversed logic.
678 Bump __FreeBSD_version to 800092.
681 Add support for hierarchical jails. Remove global securelevel.
682 Bump __FreeBSD_version to 800091.
685 The layout of struct vnet_net has changed, therefore modules
687 Bump __FreeBSD_version to 800090.
690 The newly imported zic(8) produces a new format in the output. Please
691 run tzsetup(8) to install the newly created data to /etc/localtime.
694 The sysctl tree for the usb stack has renamed from hw.usb2.* to
695 hw.usb.* and is now consistent again with previous releases.
698 802.11 monitor mode support was revised and driver api's were changed.
699 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
700 of DLT_IEEE802_11. No user-visible data structures were changed but
701 applications that use DLT_IEEE802_11 may require changes.
702 Bump __FreeBSD_version to 800088.
705 The layout of the following structs has changed: sysctl_oid,
706 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
707 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
708 panics may be experienced. World rebuild is required for
709 correctly checking networking state from userland.
710 Bump __FreeBSD_version to 800085.
713 MLDv2 and Source-Specific Multicast (SSM) have been merged
714 to the IPv6 stack. VIMAGE hooks are in but not yet used.
715 The implementation of SSM within FreeBSD's IPv6 stack closely
716 follows the IPv4 implementation.
718 For kernel developers:
720 * The most important changes are that the ip6_output() and
721 ip6_input() paths no longer take the IN6_MULTI_LOCK,
722 and this lock has been downgraded to a non-recursive mutex.
724 * As with the changes to the IPv4 stack to support SSM, filtering
725 of inbound multicast traffic must now be performed by transport
726 protocols within the IPv6 stack. This does not apply to TCP and
727 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
729 * The KPIs used by IPv6 multicast are similar to those used by
730 the IPv4 stack, with the following differences:
731 * im6o_mc_filter() is analogous to imo_multicast_filter().
732 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
733 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
734 * IN6_LOOKUP_MULTI() has been deprecated and removed.
735 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
736 for MLDv1 have an additional 'timer' argument which is used to
737 jitter the initial membership report for the solicited-node
738 multicast membership on-link.
739 * This is not strictly needed for MLDv2, which already jitters
740 its report transmissions. However, the 'timer' argument is
741 preserved in case MLDv1 is active on the interface.
743 * The KAME linked-list based IPv6 membership implementation has
744 been refactored to use a vector similar to that used by the IPv4
746 Code which maintains a list of its own multicast memberships
747 internally, e.g. carp, has been updated to reflect the new
750 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
751 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
752 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
753 implementation constraint which needs to be addressed in HEAD.
755 For application developers:
757 * The changes are broadly similar to those made for the IPv4
760 * The use of IPv4 and IPv6 multicast socket options on the same
761 socket, using mapped addresses, HAS NOT been tested or supported.
763 * There are a number of issues with the implementation of various
764 IPv6 multicast APIs which need to be resolved in the API surface
765 before the implementation is fully compatible with KAME userland
766 use, and these are mostly to do with interface index treatment.
768 * The literature available discusses the use of either the delta / ASM
769 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
770 using setsourcefilter(3)/getsourcefilter(3). For more information
771 please refer to RFC 3768, 'Socket Interface Extensions for
772 Multicast Source Filters'.
774 * Applications which use the published RFC 3678 APIs should be fine.
776 For systems administrators:
778 * The mtest(8) utility has been refactored to support IPv6, in
779 addition to IPv4. Interface addresses are no longer accepted
780 as arguments, their names must be used instead. The utility
781 will map the interface name to its first IPv4 address as
782 returned by getifaddrs(3).
784 * The ifmcstat(8) utility has also been updated to print the MLDv2
785 endpoint state and source filter lists via sysctl(3).
787 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
788 loopback of IPv6 multicast datagrams by default; it defaults to 1
789 to preserve the existing behaviour. Disabling multicast loopback is
790 recommended for optimal system performance.
792 * The IPv6 MROUTING code has been changed to examine this sysctl
793 instead of attempting to perform a group lookup before looping
794 back forwarded datagrams.
796 Bump __FreeBSD_version to 800084.
799 Implement low-level Bluetooth HCI API.
800 Bump __FreeBSD_version to 800083.
803 The layout of struct malloc_type, used by modules to register new
804 memory allocation types, has changed. Most modules will need to
805 be rebuilt or panics may be experienced.
806 Bump __FreeBSD_version to 800081.
809 Anticipate overflowing inp_flags - add inp_flags2.
810 This changes most offsets in inpcb, so checking v4 connection
811 state will require a world rebuild.
812 Bump __FreeBSD_version to 800080.
815 Add an llentry to struct route and struct route_in6. Modules
816 embedding a struct route will need to be recompiled.
817 Bump __FreeBSD_version to 800079.
820 The size of rt_metrics_lite and by extension rtentry has changed.
821 Networking administration apps will need to be recompiled.
822 The route command now supports show as an alias for get, weighting
823 of routes, sticky and nostick flags to alter the behavior of stateful
825 Bump __FreeBSD_version to 800078.
828 Do not use Giant for kbdmux(4) locking. This is wrong and
829 apparently causing more problems than it solves. This will
830 re-open the issue where interrupt handlers may race with
831 kbdmux(4) in polling mode. Typical symptoms include (but
832 not limited to) duplicated and/or missing characters when
833 low level console functions (such as gets) are used while
834 interrupts are enabled (for example geli password prompt,
835 mountroot prompt etc.). Disabling kbdmux(4) may help.
838 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
839 kernel modules referencing any of the above need to be recompiled.
840 Bump __FreeBSD_version to 800075.
843 GEOM_PART has become the default partition slicer for storage devices,
844 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
845 introduces some changes:
847 MSDOS/EBR: the devices created from MSDOS extended partition entries
848 (EBR) can be named differently than with GEOM_MBR and are now symlinks
849 to devices with offset-based names. fstabs may need to be modified.
851 BSD: the "geometry does not match label" warning is harmless in most
852 cases but it points to problems in file system misalignment with
853 disk geometry. The "c" partition is now implicit, covers the whole
854 top-level drive and cannot be (mis)used by users.
856 General: Kernel dumps are now not allowed to be written to devices
857 whose partition types indicate they are meant to be used for file
858 systems (or, in case of MSDOS partitions, as something else than
861 Most of these changes date approximately from 200812.
864 The uscanner(4) driver has been removed from the kernel. This follows
865 Linux removing theirs in 2.6 and making libusb the default interface
869 The multicast forwarding code has been cleaned up. netstat(1)
870 only relies on KVM now for printing bandwidth upcall meters.
871 The IPv4 and IPv6 modules are split into ip_mroute_mod and
872 ip6_mroute_mod respectively. The config(5) options for statically
873 compiling this code remain the same, i.e. 'options MROUTING'.
876 Support for the IFF_NEEDSGIANT network interface flag has been
877 removed, which means that non-MPSAFE network device drivers are no
878 longer supported. In particular, if_ar, if_sr, and network device
879 drivers from the old (legacy) USB stack can no longer be built or
883 POSIX.1 Native Language Support (NLS) has been enabled in libc and
884 a bunch of new language catalog files have also been added.
885 This means that some common libc messages are now localized and
886 they depend on the LC_MESSAGES environmental variable.
889 The k8temp(4) driver has been renamed to amdtemp(4) since
890 support for Family 10 and Family 11 CPU families was added.
893 IGMPv3 and Source-Specific Multicast (SSM) have been merged
894 to the IPv4 stack. VIMAGE hooks are in but not yet used.
896 For kernel developers, the most important changes are that the
897 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
898 and this lock has been downgraded to a non-recursive mutex.
900 Transport protocols (UDP, Raw IP) are now responsible for filtering
901 inbound multicast traffic according to group membership and source
902 filters. The imo_multicast_filter() KPI exists for this purpose.
903 Transports which do not use multicast (SCTP, TCP) already reject
904 multicast by default. Forwarding and receive performance may improve
905 as a mutex acquisition is no longer needed in the ip_input()
906 low-level input path. in_addmulti() and in_delmulti() are shimmed
907 to new KPIs which exist to support SSM in-kernel.
909 For application developers, it is recommended that loopback of
910 multicast datagrams be disabled for best performance, as this
911 will still cause the lock to be taken for each looped-back
912 datagram transmission. The net.inet.ip.mcast.loop sysctl may
913 be tuned to 0 to disable loopback by default; it defaults to 1
914 to preserve the existing behaviour.
916 For systems administrators, to obtain best performance with
917 multicast reception and multiple groups, it is always recommended
918 that a card with a suitably precise hash filter is used. Hash
919 collisions will still result in the lock being taken within the
920 transport protocol input path to check group membership.
922 If deploying FreeBSD in an environment with IGMP snooping switches,
923 it is recommended that the net.inet.igmp.sendlocal sysctl remain
924 enabled; this forces 224.0.0.0/24 group membership to be announced
927 The size of 'struct igmpstat' has changed; netstat needs to be
928 recompiled to reflect this.
929 Bump __FreeBSD_version to 800070.
932 libusb20.so.1 is now installed as libusb.so.1 and the ports system
933 updated to use it. This requires a buildworld/installworld in order to
934 update the library and dependencies (usbconfig, etc). Its advisable to
935 rebuild all ports which uses libusb. More specific directions are given
936 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
937 libusb are no longer required and can be removed.
940 A workaround is committed to allow the creation of System V shared
941 memory segment of size > 2 GB on the 64-bit architectures.
942 Due to a limitation of the existing ABI, the shm_segsz member
943 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
944 wrong for large segments. Note that limits must be explicitly
945 raised to allow such segments to be created.
948 The layout of struct ifnet has changed, requiring a rebuild of all
949 network device driver modules.
952 The /dev handling for the new USB stack has changed, a
953 buildworld/installworld is required for libusb20.
956 The new USB2 stack has now been permanently moved in and all kernel and
957 module names reverted to their previous values (eg, usb, ehci, ohci,
958 ums, ...). The old usb stack can be compiled in by prefixing the name
959 with the letter 'o', the old usb modules have been removed.
960 Updating entry 20090216 for xorg and 20090215 for libmap may still
964 The rc.conf(5) option if_up_delay has been renamed to
965 defaultroute_delay to better reflect its purpose. If you have
966 customized this setting in /etc/rc.conf you need to update it to
970 xorg 7.4 wants to configure its input devices via hald which does not
971 yet work with USB2. If the keyboard/mouse does not work in xorg then
973 Option "AllowEmptyInput" "off"
974 to your ServerLayout section. This will cause X to use the configured
975 kbd and mouse sections from your xorg.conf.
978 The GENERIC kernels for all architectures now default to the new USB2
979 stack. No kernel config options or code have been removed so if a
980 problem arises please report it and optionally revert to the old USB
981 stack. If you are loading USB kernel modules or have a custom kernel
982 that includes GENERIC then ensure that usb names are also changed over,
983 eg uftdi -> usb2_serial_ftdi.
985 Older programs linked against the ports libusb 0.1 need to be
986 redirected to the new stack's libusb20. /etc/libmap.conf can
988 # Map old usb library to new one for usb2 stack
989 libusb-0.1.so.8 libusb20.so.1
992 All USB ethernet devices now attach as interfaces under the name ueN
993 (eg. ue0). This is to provide a predictable name as vendors often
994 change usb chipsets in a product without notice.
997 The ichsmb(4) driver has been changed to require SMBus slave
998 addresses be left-justified (xxxxxxx0b) rather than right-justified.
999 All of the other SMBus controller drivers require left-justified
1000 slave addresses, so this change makes all the drivers provide the
1004 INET6 statistics (struct ip6stat) was updated.
1005 netstat(1) needs to be recompiled.
1008 NTFS has been removed from GENERIC kernel on amd64 to match
1009 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1010 will load ntfs.ko module automatically when NTFS support is
1011 actually needed, unless ntfs.ko is not installed or security
1012 level prohibits loading kernel modules. If either is the case,
1013 "options NTFS" has to be added into kernel config.
1016 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1017 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1018 800061. User space tools that rely on the size of struct tcpcb in
1019 tcp_var.h (e.g. sockstat) need to be recompiled.
1022 ng_tty(4) module updated to match the new TTY subsystem.
1023 Due to API change, user-level applications must be updated.
1024 New API support added to mpd5 CVS and expected to be present
1025 in next mpd5.3 release.
1028 With __FreeBSD_version 800060 the makefs tool is part of
1029 the base system (it was a port).
1032 The afdata and ifnet locks have been changed from mutexes to
1033 rwlocks, network modules will need to be re-compiled.
1036 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1037 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1038 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1039 architecture and 40 bytes on 64-bit architecture. The userland
1040 applications "arp" and "ndp" have been updated accordingly.
1041 The output from "netstat -r" shows only routing entries and
1042 none of the L2 information.
1045 __FreeBSD_version 800057 marks the switchover from the
1046 binary ath hal to source code. Users must add the line:
1048 options AH_SUPPORT_AR5416
1050 to their kernel config files when specifying:
1054 The ath_hal module no longer exists; the code is now compiled
1055 together with the driver in the ath module. It is now
1056 possible to tailor chip support (i.e. reduce the set of chips
1057 and thereby the code size); consult ath_hal(4) for details.
1060 __FreeBSD_version 800054 adds memory barriers to
1061 <machine/atomic.h>, new interfaces to ifnet to facilitate
1062 multiple hardware transmit queues for cards that support
1063 them, and a lock-less ring-buffer implementation to
1064 enable drivers to more efficiently manage queueing of
1068 A new version of ZFS (version 13) has been merged to -HEAD.
1069 This version has zpool attribute "listsnapshots" off by
1070 default, which means "zfs list" does not show snapshots,
1071 and is the same as Solaris behavior.
1074 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1077 The uhci, ohci, ehci and slhci USB Host controller drivers have
1078 been put into separate modules. If you load the usb module
1079 separately through loader.conf you will need to load the
1080 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1081 controller add the following to loader.conf:
1087 The ABI used by the PMC toolset has changed. Please keep
1088 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1092 atapci kernel module now includes only generic PCI ATA
1093 driver. AHCI driver moved to ataahci kernel module.
1094 All vendor-specific code moved into separate kernel modules:
1095 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1096 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1097 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1098 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1101 The TTY subsystem of the kernel has been replaced by a new
1102 implementation, which provides better scalability and an
1103 improved driver model. Most common drivers have been migrated to
1104 the new TTY subsystem, while others have not. The following
1105 drivers have not yet been ported to the new TTY layer:
1108 cy, digi, rc, rp, sio
1114 ng_h4, ng_tty, ppp, sl, snp
1116 Adding these drivers to your kernel configuration file shall
1117 cause compilation to fail.
1120 ntpd has been upgraded to 4.2.4p5.
1123 OpenSSH has been upgraded to 5.1p1.
1125 For many years, FreeBSD's version of OpenSSH preferred DSA
1126 over RSA for host and user authentication keys. With this
1127 upgrade, we've switched to the vendor's default of RSA over
1128 DSA. This may cause upgraded clients to warn about unknown
1129 host keys even for previously known hosts. Users should
1130 follow the usual procedure for verifying host keys before
1131 accepting the RSA key.
1133 This can be circumvented by setting the "HostKeyAlgorithms"
1134 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1137 Please note that the sequence of keys offered for
1138 authentication has been changed as well. You may want to
1139 specify IdentityFile in a different order to revert this
1143 The sio(4) driver has been removed from the i386 and amd64
1144 kernel configuration files. This means uart(4) is now the
1145 default serial port driver on those platforms as well.
1147 To prevent collisions with the sio(4) driver, the uart(4) driver
1148 uses different names for its device nodes. This means the
1149 onboard serial port will now most likely be called "ttyu0"
1150 instead of "ttyd0". You may need to reconfigure applications to
1151 use the new device names.
1153 When using the serial port as a boot console, be sure to update
1154 /boot/device.hints and /etc/ttys before booting the new kernel.
1155 If you forget to do so, you can still manually specify the hints
1156 at the loader prompt:
1158 set hint.uart.0.at="isa"
1159 set hint.uart.0.port="0x3F8"
1160 set hint.uart.0.flags="0x10"
1161 set hint.uart.0.irq="4"
1165 The gpt(8) utility has been removed. Use gpart(8) to partition
1169 The version that Linuxulator emulates was changed from 2.4.2
1170 to 2.6.16. If you experience any problems with Linux binaries
1171 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1172 if it fixes the problem contact emulation mailing list.
1175 ISDN4BSD (I4B) was removed from the src tree. You may need to
1176 update a your kernel configuration and remove relevant entries.
1179 I have checked in code to support multiple routing tables.
1180 See the man pages setfib(1) and setfib(2).
1181 This is a hopefully backwards compatible version,
1182 but to make use of it you need to compile your kernel
1183 with options ROUTETABLES=2 (or more up to 16).
1186 The 802.11 wireless support was redone to enable multi-bss
1187 operation on devices that are capable. The underlying device
1188 is no longer used directly but instead wlanX devices are
1189 cloned with ifconfig. This requires changes to rc.conf files.
1190 For example, change:
1191 ifconfig_ath0="WPA DHCP"
1194 ifconfig_wlan0="WPA DHCP"
1195 see rc.conf(5) for more details. In addition, mergemaster of
1196 /etc/rc.d is highly recommended. Simultaneous update of userland
1197 and kernel wouldn't hurt either.
1199 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1200 modules were merged into the base wlan module. All references
1201 to these modules (e.g. in kernel config files) must be removed.
1204 psm(4) has gained write(2) support in native operation level.
1205 Arbitrary commands can be written to /dev/psm%d and status can
1206 be read back from it. Therefore, an application is responsible
1207 for status validation and error recovery. It is a no-op in
1208 other operation levels.
1211 Support for KSE threading has been removed from the kernel. To
1212 run legacy applications linked against KSE libmap.conf may
1213 be used. The following libmap.conf may be used to ensure
1214 compatibility with any prior release:
1216 libpthread.so.1 libthr.so.1
1217 libpthread.so.2 libthr.so.2
1218 libkse.so.3 libthr.so.3
1221 The layout of struct vmspace has changed. This affects libkvm
1222 and any executables that link against libkvm and use the
1223 kvm_getprocs() function. In particular, but not exclusively,
1224 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1225 The effects are minimal, but it's advisable to upgrade world
1229 The latest em driver no longer has support in it for the
1230 82575 adapter, this is now moved to the igb driver. The
1231 split was done to make new features that are incompatible
1232 with older hardware easier to do.
1235 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1236 likewise the kernel option is now GEOM_LINUX_LVM.
1239 The default NFS mount mode has changed from UDP to TCP for
1240 increased reliability. If you rely on (insecurely) NFS
1241 mounting across a firewall you may need to update your
1245 Belatedly note the addition of m_collapse for compacting
1249 The fts(3) structures have been changed to use adequate
1250 integer types for their members and so to be able to cope
1251 with huge file trees. The old fts(3) ABI is preserved
1252 through symbol versioning in libc, so third-party binaries
1253 using fts(3) should still work, although they will not take
1254 advantage of the extended types. At the same time, some
1255 third-party software might fail to build after this change
1256 due to unportable assumptions made in its source code about
1257 fts(3) structure members. Such software should be fixed
1258 by its vendor or, in the worst case, in the ports tree.
1259 FreeBSD_version 800015 marks this change for the unlikely
1260 case that a portable fix is impossible.
1263 To upgrade to -current after this date, you must be running
1264 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1265 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1268 The ADAPTIVE_GIANT kernel option has been retired because its
1269 functionality is the default now.
1272 The AT keyboard emulation of sunkbd(4) has been turned on
1273 by default. In order to make the special symbols of the Sun
1274 keyboards driven by sunkbd(4) work under X these now have
1275 to be configured the same way as Sun USB keyboards driven
1276 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1278 Option "XkbLayout" "us"
1279 Option "XkbRules" "xorg"
1280 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1283 It has been decided that it is desirable to provide ABI
1284 backwards compatibility to the FreeBSD 4/5/6 versions of the
1285 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1286 broken with the introduction of PCI domain support (see the
1287 20070930 entry). Unfortunately, this required the ABI of
1288 PCIOCGETCONF to be broken again in order to be able to
1289 provide backwards compatibility to the old version of that
1290 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1291 again. As for prominent ports this affects neither pciutils
1292 nor xorg-server this time, the hal port needs to be rebuilt
1296 The misnamed kthread_create() and friends have been renamed
1297 to kproc_create() etc. Many of the callers already
1298 used kproc_start()..
1299 I will return kthread_create() and friends in a while
1300 with implementations that actually create threads, not procs.
1301 Renaming corresponds with version 800002.
1310 Avoid using make -j when upgrading. While generally safe, there are
1311 sometimes problems using -j to upgrade. If your upgrade fails with
1312 -j, please try again without -j. From time to time in the past there
1313 have been problems using -j with buildworld and/or installworld. This
1314 is especially true when upgrading between "distant" versions (eg one
1315 that cross a major release boundary or several minor releases, or when
1316 several months have passed on the -current branch).
1318 Sometimes, obscure build problems are the result of environment
1319 poisoning. This can happen because the make utility reads its
1320 environment when searching for values for global variables. To run
1321 your build attempts in an "environmental clean room", prefix all make
1322 commands with 'env -i '. See the env(1) manual page for more details.
1324 When upgrading from one major version to another it is generally best
1325 to upgrade to the latest code in the currently installed branch first,
1326 then do an upgrade to the new branch. This is the best-tested upgrade
1327 path, and has the highest probability of being successful. Please try
1328 this approach before reporting problems with a major version upgrade.
1332 When upgrading the boot ZFS pool to a new version, always follow
1335 1.) recompile and reinstall the ZFS boot loader and boot block
1336 (this is part of "make buildworld" and "make installworld")
1338 2.) update the ZFS boot block on your boot drive
1340 The following example updates the ZFS boot block on the first
1341 partition (freebsd-boot) of a GPT partitioned drive ad0:
1342 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1344 Non-boot pools do not need these updates.
1348 If you are updating from a prior version of FreeBSD (even one just
1349 a few days old), you should follow this procedure. It is the most
1350 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1352 make kernel-toolchain
1353 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1354 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1356 To test a kernel once
1357 ---------------------
1358 If you just want to boot a kernel once (because you are not sure
1359 if it works, or if you want to boot a known bad kernel to provide
1360 debugging information) run
1361 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1362 nextboot -k testkernel
1364 To just build a kernel when you know that it won't mess you up
1365 --------------------------------------------------------------
1366 This assumes you are already running a CURRENT system. Replace
1367 ${arch} with the architecture of your machine (e.g. "i386",
1368 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1370 cd src/sys/${arch}/conf
1371 config KERNEL_NAME_HERE
1372 cd ../compile/KERNEL_NAME_HERE
1377 If this fails, go to the "To build a kernel" section.
1379 To rebuild everything and install it on the current system.
1380 -----------------------------------------------------------
1381 # Note: sometimes if you are running current you gotta do more than
1382 # is listed here if you are upgrading from a really old current.
1384 <make sure you have good level 0 dumps>
1386 make kernel KERNCONF=YOUR_KERNEL_HERE
1388 <reboot in single user> [3]
1396 To cross-install current onto a separate partition
1397 --------------------------------------------------
1398 # In this approach we use a separate partition to hold
1399 # current's root, 'usr', and 'var' directories. A partition
1400 # holding "/", "/usr" and "/var" should be about 2GB in
1403 <make sure you have good level 0 dumps>
1406 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1407 <maybe newfs current's root partition>
1408 <mount current's root partition on directory ${CURRENT_ROOT}>
1409 make installworld DESTDIR=${CURRENT_ROOT}
1410 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1411 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1412 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1413 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1414 <reboot into current>
1415 <do a "native" rebuild/install as described in the previous section>
1416 <maybe install compatibility libraries from ports/misc/compat*>
1420 To upgrade in-place from 8.x-stable to current
1421 ----------------------------------------------
1422 <make sure you have good level 0 dumps>
1424 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1426 <reboot in single user> [3]
1433 Make sure that you've read the UPDATING file to understand the
1434 tweaks to various things you need. At this point in the life
1435 cycle of current, things change often and you are on your own
1436 to cope. The defaults can also change, so please read ALL of
1437 the UPDATING entries.
1439 Also, if you are tracking -current, you must be subscribed to
1440 freebsd-current@freebsd.org. Make sure that before you update
1441 your sources that you have read and understood all the recent
1442 messages there. If in doubt, please track -stable which has
1443 much fewer pitfalls.
1445 [1] If you have third party modules, such as vmware, you
1446 should disable them at this point so they don't crash your
1449 [3] From the bootblocks, boot -s, and then do
1454 adjkerntz -i # if CMOS is wall time
1455 Also, when doing a major release upgrade, it is required that
1456 you boot into single user mode to do the installworld.
1458 [4] Note: This step is non-optional. Failure to do this step
1459 can result in a significant reduction in the functionality of the
1460 system. Attempting to do it by hand is not recommended and those
1461 that pursue this avenue should read this file carefully, as well
1462 as the archives of freebsd-current and freebsd-hackers mailing lists
1463 for potential gotchas. The -U option is also useful to consider.
1464 See mergemaster(8) for more information.
1466 [5] Usually this step is a noop. However, from time to time
1467 you may need to do this if you get unknown user in the following
1468 step. It never hurts to do it all the time. You may need to
1469 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1470 install) after the buildworld before this step if you last updated
1471 from current before 20020224 or from -stable before 20020408.
1473 [6] This only deletes old files and directories. Old libraries
1474 can be deleted by "make delete-old-libs", but you have to make
1475 sure that no program is using those libraries anymore.
1477 [8] In order to have a kernel that can run the 4.x binaries needed to
1478 do an installworld, you must include the COMPAT_FREEBSD4 option in
1479 your kernel. Failure to do so may leave you with a system that is
1480 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1481 required to run the 5.x binaries on more recent kernels. And so on
1482 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1484 Make sure that you merge any new devices from GENERIC since the
1485 last time you updated your kernel config file.
1487 [9] When checking out sources, you must include the -P flag to have
1488 cvs prune empty directories.
1490 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1491 "?=" instead of the "=" assignment operator, so that buildworld can
1492 override the CPUTYPE if it needs to.
1494 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1495 not on the command line, or in /etc/make.conf. buildworld will
1496 warn if it is improperly defined.
1499 This file contains a list, in reverse chronological order, of major
1500 breakages in tracking -current. It is not guaranteed to be a complete
1501 list of such breakages, and only contains entries since October 10, 2007.
1502 If you need to see UPDATING entries from before that date, you will need
1503 to fetch an UPDATING file from an older FreeBSD release.
1505 Copyright information:
1507 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1509 Redistribution, publication, translation and use, with or without
1510 modification, in full or in part, in any form or format of this
1511 document are permitted without further permission from the author.
1513 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1514 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1515 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1516 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1517 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1518 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1519 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1520 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1521 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1522 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1523 POSSIBILITY OF SUCH DAMAGE.
1525 Contact Warner Losh if you have any questions about your use of