1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
13 The acpi_wmi(4) status device /dev/wmistat has been renamed to
17 A new VOP_ADVISE() was added to support posix_fadvise(2). All
18 filesystem modules must be recompiled.
21 The interface of the VOP_VPTOCNP(9) changed, now the returned
22 vnode shall be referenced, previously it was required to be
23 only held. All in-tree filesystems are converted.
29 The broken amd(4) driver has been replaced with esp(4) in the amd64,
30 i386 and pc98 GENERIC kernel configuration files.
33 This commit modifies vfs_register() so that it uses a hash
34 calculation to set vfc_typenum, which is enabled by default.
35 The first time a system is booted after this change, the
36 vfc_typenum values will change for all file systems. The
37 main effect of this is a change to the NFS server file handles
38 for file systems that use vfc_typenum in their fsid, such as ZFS.
39 It will, however, prevent vfc_typenum from changing when file
40 systems are loaded in a different order for subsequent reboots.
41 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
42 until you are ready to remount all NFS clients after a reboot.
45 Bump the shared library version numbers for libraries that
46 do not use symbol versioning, have changed the ABI compared
47 to stable/8 and which shared library version was not bumped.
48 Done as part of 9.0-RELEASE cycle.
51 During the merge of Capsicum features, the fget(9) KPI was modified.
52 This may require the rebuilding of out-of-tree device drivers --
53 issues have been reported specifically with the nVidia device driver.
54 __FreeBSD_version is bumped to 900041.
56 Also, there is a period between 20110811 and 20110814 where the
57 special devices /dev/{stdin,stdout,stderr} did not work correctly.
58 Building world from a kernel during that window may not work.
61 The packet filter (pf) code has been updated to OpenBSD 4.5.
62 You need to update userland tools to be in sync with kernel.
63 This update breaks backward compatibility with earlier pfsync(4)
64 versions. Care must be taken when updating redundant firewall setups.
67 The following sysctls and tunables are retired on x86 platforms:
69 machdep.hlt_logical_cpus
70 The following sysctl is retired:
71 machdep.hyperthreading_allowed
72 The sysctls were supposed to provide a way to dynamically offline and
73 online selected CPUs on x86 platforms, but the implementation has not
74 been reliable especially with SCHED_ULE scheduler.
75 machdep.hyperthreading_allowed tunable is still available to ignore
76 hyperthreading CPUs at OS level.
77 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
78 where X is an APIC ID of a CPU. Be advised, though, that disabling
79 CPUs in non-uniform fashion will result in non-uniform topology and
80 may lead to sub-optimal system performance with SCHED_ULE, which is
84 cpumask_t type is retired and cpuset_t is used in order to describe
88 Changes to ifconfig(8) for dynamic address family detection mandate
89 that you are running a kernel of 20110525 or later. Make sure to
90 follow the update procedure to boot a new kernel before installing
94 Support for sun4v architecture is officially dropped
97 Several KPI breaking changes have been committed to the mii(4) layer,
98 the PHY drivers and consequently some Ethernet drivers using mii(4).
99 This means that miibus.ko and the modules of the affected Ethernet
100 drivers need to be recompiled.
102 Note to kernel developers: Given that the OUI bit reversion problem
103 was fixed as part of these changes all mii(4) commits related to OUIs,
104 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
105 handling, no longer can be merged verbatim to stable/8 and previous
109 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
110 into their kernel configurations along with 'device pci'.
113 The default NFS client is now the new NFS client, so fstype "newnfs"
114 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
115 Although mounts via fstype "nfs" will usually work without userland
116 changes, it is recommended that the mount(8) and mount_nfs(8)
117 commands be rebuilt from sources and that a link to mount_nfs called
118 mount_oldnfs be created. The new client is compiled into the
119 kernel with "options NFSCL" and this is needed for diskless root
120 file systems. The GENERIC kernel configs have been changed to use
121 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
122 To use the regular/old client, you can "mount -t oldnfs ...". For
123 a diskless root file system, you must also include a line like:
125 vfs.root.mountfrom="oldnfs:"
127 in the boot/loader.conf on the root fs on the NFS server to make
128 a diskless root fs use the old client.
131 The GENERIC kernels for all architectures now default to the new
132 CAM-based ATA stack. It means that all legacy ATA drivers were
133 removed and replaced by respective CAM drivers. If you are using
134 ATA device names in /etc/fstab or other places, make sure to update
135 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
136 where 'Y's are the sequential numbers starting from zero for each type
137 in order of detection, unless configured otherwise with tunables,
138 see cam(4)). There will be symbolic links created in /dev/ to map
139 old adX devices to the respective adaY. They should provide basic
140 compatibility for file systems mounting in most cases, but they do
141 not support old user-level APIs and do not have respective providers
142 in GEOM. Consider using updated management tools with new device names.
144 It is possible to load devices ahci, ata, siis and mvs as modules,
145 but option ATA_CAM should remain in kernel configuration to make ata
146 module work as CAM driver supporting legacy ATA controllers. Device ata
147 still can be used in modular fashion (atacore + ...). Modules atadisk
148 and atapi* are not used and won't affect operation in ATA_CAM mode.
149 Note that to use CAM-based ATA kernel should include CAM devices
150 scbus, pass, da (or explicitly ada), cd and optionally others. All of
151 them are parts of the cam module.
153 ataraid(4) functionality is now supported by the RAID GEOM class.
154 To use it you can load geom_raid kernel module and use graid(8) tool
155 for management. Instead of /dev/arX device names, use /dev/raid/rX.
157 No kernel config options or code have been removed, so if a problem
158 arises, please report it and optionally revert to the old ATA stack.
159 In order to do it you can remove from the kernel config:
164 , and instead add back:
165 device atadisk # ATA disk drives
166 device ataraid # ATA RAID drives
167 device atapicd # ATAPI CDROM drives
168 device atapifd # ATAPI floppy drives
169 device atapist # ATAPI tape drives
172 The default NFS server has been changed to the new server, which
173 was referred to as the experimental server. If you need to switch
174 back to the old NFS server, you must now put the "-o" option on
175 both the mountd and nfsd commands. This can be done using the
176 mountd_flags and nfs_server_flags rc.conf variables until an
177 update to the rc scripts is committed, which is coming soon.
180 The GNU Objective-C runtime library (libobjc), and other Objective-C
181 related components have been removed from the base system. If you
182 require an Objective-C library, please use one of the available ports.
185 ath(4) has been split into bus- and device- modules. if_ath contains
186 the HAL, the TX rate control and the network device code. if_ath_pci
187 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
188 contains the AHB glue. Users need to load both if_ath_pci and if_ath
189 in order to use ath on everything else.
191 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
192 need to load if_ath and if_ath_pci for ath(4) operation.
195 As part of the replacement of sysinstall, the process of building
196 release media has changed significantly. For details, please re-read
197 release(7), which has been updated to reflect the new build process.
200 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
201 is the last available version under GPLv2. It brings a number of new
202 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
203 4.1 and SSE 4.2), better support for powerpc64, a number of new
204 directives, and lots of other small improvements. See the ChangeLog
205 file in contrib/binutils for the full details.
208 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
209 compliant, and will now use half of hash for authentication.
210 This will break interoperability with all stacks (including all
211 actual FreeBSD versions) who implement
212 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
214 The only workaround with such peers is to use another HMAC
215 algorithm for IPsec ("phase 2") authentication.
218 Remove the uio_yield prototype and symbol. This function has
219 been misnamed since it was introduced and should not be
220 globally exposed with this name. The equivalent functionality
221 is now available using kern_yield(curthread->td_user_pri).
222 The function remains undocumented.
225 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
226 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
227 for scalar sysctls is defined but disabled. Code that needs
228 UQUAD to pass the type checking that must compile on older
229 systems where the define is not present can check against
230 __FreeBSD_version >= 900030.
232 The system dialog(1) has been replaced with a new version previously
233 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
234 with the previous version, but the libdialog associated with it has
235 a largely incompatible API. As such, the original version of libdialog
236 will be kept temporarily as libodialog, until its base system consumers
237 are replaced or updated. Bump __FreeBSD_version to 900030.
240 If you are trying to run make universe on a -stable system, and you get
241 the following warning:
242 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
243 unknown. config(8) likely too old."
244 or something similar to it, then you must upgrade your -stable system
245 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
246 stable/8) or build the config from the latest stable/8 branch and
247 install it on your system.
249 Prior to this date, building a current universe on 8-stable system from
250 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
251 error in the first kernel build phase. A new config on those old
252 systems will fix that problem for older versions of -current.
255 The TCP stack has been modified to allow Khelp modules to interact with
256 it via helper hook points and store per-connection data in the TCP
257 control block. Bump __FreeBSD_version to 900029. User space tools that
258 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
262 Generic IEEE 802.3 annex 31B full duplex flow control support has been
263 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
264 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
265 to take advantage of it instead of using custom implementations. This
266 means that these drivers now no longer unconditionally advertise
267 support for flow control but only do so if flow control is a selected
268 media option. This was implemented in the generic support that way in
269 order to allow flow control to be switched on and off via ifconfig(8)
270 with the PHY specific default to typically off in order to protect
271 from unwanted effects. Consequently, if you used flow control with
272 one of the above mentioned drivers you now need to explicitly enable
274 ifconfig bge0 media auto mediaopt flowcontrol
276 Along with the above mentioned changes generic support for setting
277 1000baseT master mode also has been added and brgphy(4), ciphy(4),
278 e1000phy(4) as well as ip1000phy(4) have been converted to take
279 advantage of it. This means that these drivers now no longer take the
280 link0 parameter for selecting master mode but the master media option
281 has to be used instead, for example like in the following:
282 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
284 Selection of master mode now is also available with all other PHY
285 drivers supporting 1000baseT.
288 The TCP stack has received a significant update to add support for
289 modularised congestion control and generally improve the clarity of
290 congestion control decisions. Bump __FreeBSD_version to 900025. User
291 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
292 sockstat) need to be recompiled.
295 The man(1) utility has been replaced by a new version that no longer
296 uses /etc/manpath.config. Please consult man.conf(5) for how to
297 migrate local entries to the new format.
300 The copyright strings printed by login(1) and sshd(8) at the time of a
301 new connection have been removed to follow other operating systems and
305 A workaround for a fixed ld bug has been removed in kernel code,
306 so make sure that your system ld is built from sources after
307 revision 210245 from 2010-07-19 (r211583 if building head kernel
308 on stable/8, r211584 for stable/7; both from 2010-08-21).
309 A symptom of incorrect ld version is different addresses for
310 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
313 The $ipv6_prefer variable in rc.conf(5) has been split into
314 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
316 The $ip6addrctl_policy is a variable to choose a pre-defined
317 address selection policy set by ip6addrctl(8). A value
318 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
321 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
322 flag (see an entry of 20090926) is set on an interface with no
323 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
324 security reason. If you want IPv6 link-local address on all
325 interfaces by default, set this to "YES".
327 The old ipv6_prefer="YES" is equivalent to
328 ipv6_activate_all_interfaces="YES" and
329 ip6addrctl_policy="ipv6_prefer".
332 DTrace has grown support for userland tracing. Due to this, DTrace is
333 now i386 and amd64 only.
334 dtruss(1) is now installed by default on those systems and a new
335 kernel module is needed for userland tracing: fasttrap.
336 No changes to your kernel config file are necessary to enable
337 userland tracing, but you might consider adding 'STRIP=' and
338 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
339 to have informative userland stack traces in DTrace (ustack).
342 The acpi_aiboost(4) driver has been removed in favor of the new
343 aibs(4) driver. You should update your kernel configuration file.
346 BSD grep has been imported to the base system and it is built by
347 default. It is completely BSD licensed, highly GNU-compatible, uses
348 less memory than its GNU counterpart and has a small codebase.
349 However, it is slower than its GNU counterpart, which is mostly
350 noticeable for larger searches, for smaller ones it is measurable
351 but not significant. The reason is complex, the most important factor
352 is that we lack a modern and efficient regex library and GNU
353 overcomes this by optimizing the searches internally. Future work
354 on improving the regex performance is planned, for the meantime,
355 users that need better performance, can build GNU grep instead by
356 setting the WITH_GNU_GREP knob.
359 Due to the import of powerpc64 support, all existing powerpc kernel
360 configuration files must be updated with a machine directive like this:
361 machine powerpc powerpc
363 In addition, an updated config(8) is required to build powerpc kernels
367 A new version of ZFS (version 15) has been merged to -HEAD.
368 This version uses a python library for the following subcommands:
369 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
370 For full functionality of these commands the following port must
371 be installed: sysutils/py-zfs
374 'vm_page's are now hashed by physical address to an array of mutexes.
375 Currently this is only used to serialize access to hold_count. Over
376 time the page queue mutex will be peeled away. This changes the size
377 of pmap on every architecture. And requires all callers of vm_page_hold
378 and vm_page_unhold to be updated.
381 WITH_CTF can now be specified in src.conf (not recommended, there
382 are some problems with static executables), make.conf (would also
383 affect ports which do not use GNU make and do not override the
384 compile targets) or in the kernel config (via "makeoptions
386 When WITH_CTF was specified there before this was silently ignored,
387 so make sure that WITH_CTF is not used in places which could lead
388 to unwanted behavior.
391 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
392 to allow 32-bit compatibility on non-x86 platforms. All kernel
393 configurations on amd64 and ia64 platforms using these options must
394 be modified accordingly.
397 The utmp user accounting database has been replaced with utmpx,
398 the user accounting interface standardized by POSIX.
399 Unfortunately the semantics of utmp and utmpx don't match,
400 making it practically impossible to support both interfaces.
401 The user accounting database is used by tools like finger(1),
402 last(1), talk(1), w(1) and ac(8).
404 All applications in the base system use utmpx. This means only
405 local binaries (e.g. from the ports tree) may still use these
406 utmp database files. These applications must be rebuilt to make
409 After the system has been upgraded, it is safe to remove the old
410 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
411 assuming their contents is of no importance anymore. Old wtmp
412 databases can only be used by last(1) and ac(8) after they have
413 been converted to the new format using wtmpcvt(1).
416 Introduce the kernel thread "deadlock resolver" (which can be enabled
417 via the DEADLKRES option, see NOTES for more details) and the
418 sleepq_type() function for sleepqueues.
421 The rc.firewall and rc.firewall6 were unified, and
422 rc.firewall6 and rc.d/ip6fw were removed.
423 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
424 variables are obsoleted. Instead, the following new rc
425 variables are added to rc.d/ipfw:
427 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
428 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
429 firewall_simple_onet_ipv6, firewall_trusted_ipv6
431 The meanings correspond to the relevant IPv4 variables.
437 The default terminal emulation for syscons(4) has been changed
438 from cons25 to xterm on all platforms except pc98. This means
439 that the /etc/ttys file needs to be updated to ensure correct
440 operation of applications on the console.
442 The terminal emulation style can be toggled per window by using
443 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
444 options can be used to change the compile-time default back to
447 To prevent graphical artifacts, make sure the TERM environment
448 variable is set to match the terminal emulation that is being
449 performed by syscons(4).
452 The layout of the structure ieee80211req_scan_result has changed.
453 Applications that require wireless scan results (e.g. ifconfig(8))
454 from net80211 need to be recompiled.
456 Applications such as wpa_supplicant(8) may require a full world
457 build without using NO_CLEAN in order to get synchronized with the
461 The iwn(4) driver has been updated to support the 5000 and 5150 series.
462 There's one kernel module for each firmware. Adding "device iwnfw"
463 to the kernel configuration file means including all three firmware
464 images inside the kernel. If you want to include just the one for
465 your wireless card, use the devices iwn4965fw, iwn5000fw or
469 The rc.d/network_ipv6, IPv6 configuration script has been integrated
470 into rc.d/netif. The changes are the following:
472 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
473 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
474 Note that both variables need the "inet6" keyword at the head.
476 Do not set $ipv6_network_interfaces manually if you do not
477 understand what you are doing. It is not needed in most cases.
479 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
482 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
483 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
485 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
486 all configured interfaces have "inet6 accept_rtadv" in the
487 $ifconfig_IF_ipv6. These are for backward compatibility.
489 3. A new variable $ipv6_prefer has been added. If NO, IPv6
490 functionality of interfaces with no corresponding
491 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
492 and the default address selection policy of ip6addrctl(8)
493 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
494 Note that if you want to configure IPv6 functionality on the
495 disabled interfaces after boot, first you need to clear the flag by
496 using ifconfig(8) like:
498 ifconfig em0 inet6 -ifdisabled
500 If YES, the default address selection policy is set as
503 The default value of $ipv6_prefer is NO.
505 4. If your system need to receive Router Advertisement messages,
506 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
507 scripts automatically invoke rtsol(8) when the interface becomes
508 UP. The Router Advertisement messages are used for SLAAC
509 (State-Less Address AutoConfiguration).
512 802.11s D3.03 support was committed. This is incompatible with the
513 previous code, which was based on D3.0.
516 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
517 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
518 control whether accepting Router Advertisement messages or not.
519 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
520 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
521 The ifconfig(8) utility now supports these flags.
524 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
525 mount(8) and -a option for df(1) to see them.
528 The old tunable hw.bus.devctl_disable has been superseded by
529 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
530 replaced by hw.bus.devctl_queue=0. The default for this new tunable
534 Remove the option STOP_NMI. The default action is now to use NMI only
535 for KDB via the newly introduced function stop_cpus_hard() and
536 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
539 The stable/8 branch created in subversion. This corresponds to the
540 RELENG_8 branch in CVS.
543 Bump the shared library version numbers for all libraries that do not
544 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
545 __FreeBSD_version to 800105.
548 Due to changes in the implementation of virtual network stack support,
549 all network-related kernel modules must be recompiled. As this change
550 breaks the ABI, bump __FreeBSD_version to 800104.
553 The TOE interface to the TCP syncache has been modified to remove
554 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
555 The cxgb driver is the only TOE consumer affected by this change, and
556 needs to be recompiled along with the kernel. As this change breaks
557 the ABI, bump __FreeBSD_version to 800103.
560 Padding has been added to struct tcpcb, sackhint and tcpstat in
561 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
562 maintaining the ABI. However, this change breaks the ABI, so bump
563 __FreeBSD_version to 800102. User space tools that rely on the size of
564 any of these structs (e.g. sockstat) need to be recompiled.
567 The NFS_LEGACYRPC option has been removed along with the old kernel
568 RPC implementation that this option selected. Kernel configurations
569 may need to be adjusted.
572 The network interface device nodes at /dev/net/<interface> have been
573 removed. All ioctl operations can be performed the normal way using
574 routing sockets. The kqueue functionality can generally be replaced
575 with routing sockets.
578 The documentation from the FreeBSD Documentation Project (Handbook,
579 FAQ, etc.) is now installed via packages by sysinstall(8) and under
580 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
583 The ABI of various structures related to the SYSV IPC API have been
584 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
585 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
589 Layout of struct vnet has changed as routing related variables were
590 moved to their own Vimage module. Modules need to be recompiled. Bump
591 __FreeBSD_version to 800099.
594 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
595 respectively. As long as no more than 16 groups per process are used,
596 no changes should be visible. When more than 16 groups are used, old
597 binaries may fail if they call getgroups() or getgrouplist() with
598 statically sized storage. Recompiling will work around this, but
599 applications should be modified to use dynamically allocated storage
600 for group arrays as POSIX.1-2008 does not cap an implementation's
601 number of supported groups at NGROUPS_MAX+1 as previous versions did.
603 NFS and portalfs mounts may also be affected as the list of groups is
604 truncated to 16. Users of NFS who use more than 16 groups, should
605 take care that negative group permissions are not used on the exported
606 file systems as they will not be reliable unless a GSSAPI based
607 authentication method is used.
610 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
611 option compiles in the support for adaptive spinning for lockmgrs
612 which want to enable it. The lockinit() function now accepts the flag
613 LK_ADAPTIVE in order to make the lock object subject to adaptive
614 spinning when both held in write and read mode.
617 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
618 changed. User applications that use this ioctl need to be rebuilt.
621 The layout of struct thread has changed. Kernel and modules need to
625 The layout of structs ifnet, domain, protosw and vnet_net has changed.
626 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
629 window(1) has been removed from the base system. It can now be
630 installed from ports. The port is called misc/window.
633 The way we are storing and accessing `routing table' entries has
634 changed. Programs reading the FIB, like netstat, need to be
638 A new netisr implementation has been added for FreeBSD 8. Network
639 file system modules, such as igmp, ipdivert, and others, should be
641 Bump __FreeBSD_version to 800096.
644 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
648 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
649 Bump __FreeBSD_version to 800094.
652 Add mnt_xflag field to 'struct mount'. File system modules need to be
654 Bump __FreeBSD_version to 800093.
657 The compiling option ADAPTIVE_SX has been retired while it has been
658 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
659 The KPI for sx_init_flags() changes as accepting flags:
660 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
661 been introduced in order to handle the reversed logic.
662 Bump __FreeBSD_version to 800092.
665 Add support for hierarchical jails. Remove global securelevel.
666 Bump __FreeBSD_version to 800091.
669 The layout of struct vnet_net has changed, therefore modules
671 Bump __FreeBSD_version to 800090.
674 The newly imported zic(8) produces a new format in the output. Please
675 run tzsetup(8) to install the newly created data to /etc/localtime.
678 The sysctl tree for the usb stack has renamed from hw.usb2.* to
679 hw.usb.* and is now consistent again with previous releases.
682 802.11 monitor mode support was revised and driver api's were changed.
683 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
684 of DLT_IEEE802_11. No user-visible data structures were changed but
685 applications that use DLT_IEEE802_11 may require changes.
686 Bump __FreeBSD_version to 800088.
689 The layout of the following structs has changed: sysctl_oid,
690 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
691 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
692 panics may be experienced. World rebuild is required for
693 correctly checking networking state from userland.
694 Bump __FreeBSD_version to 800085.
697 MLDv2 and Source-Specific Multicast (SSM) have been merged
698 to the IPv6 stack. VIMAGE hooks are in but not yet used.
699 The implementation of SSM within FreeBSD's IPv6 stack closely
700 follows the IPv4 implementation.
702 For kernel developers:
704 * The most important changes are that the ip6_output() and
705 ip6_input() paths no longer take the IN6_MULTI_LOCK,
706 and this lock has been downgraded to a non-recursive mutex.
708 * As with the changes to the IPv4 stack to support SSM, filtering
709 of inbound multicast traffic must now be performed by transport
710 protocols within the IPv6 stack. This does not apply to TCP and
711 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
713 * The KPIs used by IPv6 multicast are similar to those used by
714 the IPv4 stack, with the following differences:
715 * im6o_mc_filter() is analogous to imo_multicast_filter().
716 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
717 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
718 * IN6_LOOKUP_MULTI() has been deprecated and removed.
719 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
720 for MLDv1 have an additional 'timer' argument which is used to
721 jitter the initial membership report for the solicited-node
722 multicast membership on-link.
723 * This is not strictly needed for MLDv2, which already jitters
724 its report transmissions. However, the 'timer' argument is
725 preserved in case MLDv1 is active on the interface.
727 * The KAME linked-list based IPv6 membership implementation has
728 been refactored to use a vector similar to that used by the IPv4
730 Code which maintains a list of its own multicast memberships
731 internally, e.g. carp, has been updated to reflect the new
734 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
735 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
736 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
737 implementation constraint which needs to be addressed in HEAD.
739 For application developers:
741 * The changes are broadly similar to those made for the IPv4
744 * The use of IPv4 and IPv6 multicast socket options on the same
745 socket, using mapped addresses, HAS NOT been tested or supported.
747 * There are a number of issues with the implementation of various
748 IPv6 multicast APIs which need to be resolved in the API surface
749 before the implementation is fully compatible with KAME userland
750 use, and these are mostly to do with interface index treatment.
752 * The literature available discusses the use of either the delta / ASM
753 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
754 using setsourcefilter(3)/getsourcefilter(3). For more information
755 please refer to RFC 3768, 'Socket Interface Extensions for
756 Multicast Source Filters'.
758 * Applications which use the published RFC 3678 APIs should be fine.
760 For systems administrators:
762 * The mtest(8) utility has been refactored to support IPv6, in
763 addition to IPv4. Interface addresses are no longer accepted
764 as arguments, their names must be used instead. The utility
765 will map the interface name to its first IPv4 address as
766 returned by getifaddrs(3).
768 * The ifmcstat(8) utility has also been updated to print the MLDv2
769 endpoint state and source filter lists via sysctl(3).
771 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
772 loopback of IPv6 multicast datagrams by default; it defaults to 1
773 to preserve the existing behaviour. Disabling multicast loopback is
774 recommended for optimal system performance.
776 * The IPv6 MROUTING code has been changed to examine this sysctl
777 instead of attempting to perform a group lookup before looping
778 back forwarded datagrams.
780 Bump __FreeBSD_version to 800084.
783 Implement low-level Bluetooth HCI API.
784 Bump __FreeBSD_version to 800083.
787 The layout of struct malloc_type, used by modules to register new
788 memory allocation types, has changed. Most modules will need to
789 be rebuilt or panics may be experienced.
790 Bump __FreeBSD_version to 800081.
793 Anticipate overflowing inp_flags - add inp_flags2.
794 This changes most offsets in inpcb, so checking v4 connection
795 state will require a world rebuild.
796 Bump __FreeBSD_version to 800080.
799 Add an llentry to struct route and struct route_in6. Modules
800 embedding a struct route will need to be recompiled.
801 Bump __FreeBSD_version to 800079.
804 The size of rt_metrics_lite and by extension rtentry has changed.
805 Networking administration apps will need to be recompiled.
806 The route command now supports show as an alias for get, weighting
807 of routes, sticky and nostick flags to alter the behavior of stateful
809 Bump __FreeBSD_version to 800078.
812 Do not use Giant for kbdmux(4) locking. This is wrong and
813 apparently causing more problems than it solves. This will
814 re-open the issue where interrupt handlers may race with
815 kbdmux(4) in polling mode. Typical symptoms include (but
816 not limited to) duplicated and/or missing characters when
817 low level console functions (such as gets) are used while
818 interrupts are enabled (for example geli password prompt,
819 mountroot prompt etc.). Disabling kbdmux(4) may help.
822 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
823 kernel modules referencing any of the above need to be recompiled.
824 Bump __FreeBSD_version to 800075.
827 GEOM_PART has become the default partition slicer for storage devices,
828 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
829 introduces some changes:
831 MSDOS/EBR: the devices created from MSDOS extended partition entries
832 (EBR) can be named differently than with GEOM_MBR and are now symlinks
833 to devices with offset-based names. fstabs may need to be modified.
835 BSD: the "geometry does not match label" warning is harmless in most
836 cases but it points to problems in file system misalignment with
837 disk geometry. The "c" partition is now implicit, covers the whole
838 top-level drive and cannot be (mis)used by users.
840 General: Kernel dumps are now not allowed to be written to devices
841 whose partition types indicate they are meant to be used for file
842 systems (or, in case of MSDOS partitions, as something else than
845 Most of these changes date approximately from 200812.
848 The uscanner(4) driver has been removed from the kernel. This follows
849 Linux removing theirs in 2.6 and making libusb the default interface
853 The multicast forwarding code has been cleaned up. netstat(1)
854 only relies on KVM now for printing bandwidth upcall meters.
855 The IPv4 and IPv6 modules are split into ip_mroute_mod and
856 ip6_mroute_mod respectively. The config(5) options for statically
857 compiling this code remain the same, i.e. 'options MROUTING'.
860 Support for the IFF_NEEDSGIANT network interface flag has been
861 removed, which means that non-MPSAFE network device drivers are no
862 longer supported. In particular, if_ar, if_sr, and network device
863 drivers from the old (legacy) USB stack can no longer be built or
867 POSIX.1 Native Language Support (NLS) has been enabled in libc and
868 a bunch of new language catalog files have also been added.
869 This means that some common libc messages are now localized and
870 they depend on the LC_MESSAGES environmental variable.
873 The k8temp(4) driver has been renamed to amdtemp(4) since
874 support for Family 10 and Family 11 CPU families was added.
877 IGMPv3 and Source-Specific Multicast (SSM) have been merged
878 to the IPv4 stack. VIMAGE hooks are in but not yet used.
880 For kernel developers, the most important changes are that the
881 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
882 and this lock has been downgraded to a non-recursive mutex.
884 Transport protocols (UDP, Raw IP) are now responsible for filtering
885 inbound multicast traffic according to group membership and source
886 filters. The imo_multicast_filter() KPI exists for this purpose.
887 Transports which do not use multicast (SCTP, TCP) already reject
888 multicast by default. Forwarding and receive performance may improve
889 as a mutex acquisition is no longer needed in the ip_input()
890 low-level input path. in_addmulti() and in_delmulti() are shimmed
891 to new KPIs which exist to support SSM in-kernel.
893 For application developers, it is recommended that loopback of
894 multicast datagrams be disabled for best performance, as this
895 will still cause the lock to be taken for each looped-back
896 datagram transmission. The net.inet.ip.mcast.loop sysctl may
897 be tuned to 0 to disable loopback by default; it defaults to 1
898 to preserve the existing behaviour.
900 For systems administrators, to obtain best performance with
901 multicast reception and multiple groups, it is always recommended
902 that a card with a suitably precise hash filter is used. Hash
903 collisions will still result in the lock being taken within the
904 transport protocol input path to check group membership.
906 If deploying FreeBSD in an environment with IGMP snooping switches,
907 it is recommended that the net.inet.igmp.sendlocal sysctl remain
908 enabled; this forces 224.0.0.0/24 group membership to be announced
911 The size of 'struct igmpstat' has changed; netstat needs to be
912 recompiled to reflect this.
913 Bump __FreeBSD_version to 800070.
916 libusb20.so.1 is now installed as libusb.so.1 and the ports system
917 updated to use it. This requires a buildworld/installworld in order to
918 update the library and dependencies (usbconfig, etc). Its advisable to
919 rebuild all ports which uses libusb. More specific directions are given
920 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
921 libusb are no longer required and can be removed.
924 A workaround is committed to allow the creation of System V shared
925 memory segment of size > 2 GB on the 64-bit architectures.
926 Due to a limitation of the existing ABI, the shm_segsz member
927 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
928 wrong for large segments. Note that limits must be explicitly
929 raised to allow such segments to be created.
932 The layout of struct ifnet has changed, requiring a rebuild of all
933 network device driver modules.
936 The /dev handling for the new USB stack has changed, a
937 buildworld/installworld is required for libusb20.
940 The new USB2 stack has now been permanently moved in and all kernel and
941 module names reverted to their previous values (eg, usb, ehci, ohci,
942 ums, ...). The old usb stack can be compiled in by prefixing the name
943 with the letter 'o', the old usb modules have been removed.
944 Updating entry 20090216 for xorg and 20090215 for libmap may still
948 The rc.conf(5) option if_up_delay has been renamed to
949 defaultroute_delay to better reflect its purpose. If you have
950 customized this setting in /etc/rc.conf you need to update it to
954 xorg 7.4 wants to configure its input devices via hald which does not
955 yet work with USB2. If the keyboard/mouse does not work in xorg then
957 Option "AllowEmptyInput" "off"
958 to your ServerLayout section. This will cause X to use the configured
959 kbd and mouse sections from your xorg.conf.
962 The GENERIC kernels for all architectures now default to the new USB2
963 stack. No kernel config options or code have been removed so if a
964 problem arises please report it and optionally revert to the old USB
965 stack. If you are loading USB kernel modules or have a custom kernel
966 that includes GENERIC then ensure that usb names are also changed over,
967 eg uftdi -> usb2_serial_ftdi.
969 Older programs linked against the ports libusb 0.1 need to be
970 redirected to the new stack's libusb20. /etc/libmap.conf can
972 # Map old usb library to new one for usb2 stack
973 libusb-0.1.so.8 libusb20.so.1
976 All USB ethernet devices now attach as interfaces under the name ueN
977 (eg. ue0). This is to provide a predictable name as vendors often
978 change usb chipsets in a product without notice.
981 The ichsmb(4) driver has been changed to require SMBus slave
982 addresses be left-justified (xxxxxxx0b) rather than right-justified.
983 All of the other SMBus controller drivers require left-justified
984 slave addresses, so this change makes all the drivers provide the
988 INET6 statistics (struct ip6stat) was updated.
989 netstat(1) needs to be recompiled.
992 NTFS has been removed from GENERIC kernel on amd64 to match
993 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
994 will load ntfs.ko module automatically when NTFS support is
995 actually needed, unless ntfs.ko is not installed or security
996 level prohibits loading kernel modules. If either is the case,
997 "options NTFS" has to be added into kernel config.
1000 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1001 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1002 800061. User space tools that rely on the size of struct tcpcb in
1003 tcp_var.h (e.g. sockstat) need to be recompiled.
1006 ng_tty(4) module updated to match the new TTY subsystem.
1007 Due to API change, user-level applications must be updated.
1008 New API support added to mpd5 CVS and expected to be present
1009 in next mpd5.3 release.
1012 With __FreeBSD_version 800060 the makefs tool is part of
1013 the base system (it was a port).
1016 The afdata and ifnet locks have been changed from mutexes to
1017 rwlocks, network modules will need to be re-compiled.
1020 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1021 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1022 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1023 architecture and 40 bytes on 64-bit architecture. The userland
1024 applications "arp" and "ndp" have been updated accordingly.
1025 The output from "netstat -r" shows only routing entries and
1026 none of the L2 information.
1029 __FreeBSD_version 800057 marks the switchover from the
1030 binary ath hal to source code. Users must add the line:
1032 options AH_SUPPORT_AR5416
1034 to their kernel config files when specifying:
1038 The ath_hal module no longer exists; the code is now compiled
1039 together with the driver in the ath module. It is now
1040 possible to tailor chip support (i.e. reduce the set of chips
1041 and thereby the code size); consult ath_hal(4) for details.
1044 __FreeBSD_version 800054 adds memory barriers to
1045 <machine/atomic.h>, new interfaces to ifnet to facilitate
1046 multiple hardware transmit queues for cards that support
1047 them, and a lock-less ring-buffer implementation to
1048 enable drivers to more efficiently manage queueing of
1052 A new version of ZFS (version 13) has been merged to -HEAD.
1053 This version has zpool attribute "listsnapshots" off by
1054 default, which means "zfs list" does not show snapshots,
1055 and is the same as Solaris behavior.
1058 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1061 The uhci, ohci, ehci and slhci USB Host controller drivers have
1062 been put into separate modules. If you load the usb module
1063 separately through loader.conf you will need to load the
1064 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1065 controller add the following to loader.conf:
1071 The ABI used by the PMC toolset has changed. Please keep
1072 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1076 atapci kernel module now includes only generic PCI ATA
1077 driver. AHCI driver moved to ataahci kernel module.
1078 All vendor-specific code moved into separate kernel modules:
1079 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1080 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1081 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1082 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1085 The TTY subsystem of the kernel has been replaced by a new
1086 implementation, which provides better scalability and an
1087 improved driver model. Most common drivers have been migrated to
1088 the new TTY subsystem, while others have not. The following
1089 drivers have not yet been ported to the new TTY layer:
1092 cy, digi, rc, rp, sio
1098 ng_h4, ng_tty, ppp, sl, snp
1100 Adding these drivers to your kernel configuration file shall
1101 cause compilation to fail.
1104 ntpd has been upgraded to 4.2.4p5.
1107 OpenSSH has been upgraded to 5.1p1.
1109 For many years, FreeBSD's version of OpenSSH preferred DSA
1110 over RSA for host and user authentication keys. With this
1111 upgrade, we've switched to the vendor's default of RSA over
1112 DSA. This may cause upgraded clients to warn about unknown
1113 host keys even for previously known hosts. Users should
1114 follow the usual procedure for verifying host keys before
1115 accepting the RSA key.
1117 This can be circumvented by setting the "HostKeyAlgorithms"
1118 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1121 Please note that the sequence of keys offered for
1122 authentication has been changed as well. You may want to
1123 specify IdentityFile in a different order to revert this
1127 The sio(4) driver has been removed from the i386 and amd64
1128 kernel configuration files. This means uart(4) is now the
1129 default serial port driver on those platforms as well.
1131 To prevent collisions with the sio(4) driver, the uart(4) driver
1132 uses different names for its device nodes. This means the
1133 onboard serial port will now most likely be called "ttyu0"
1134 instead of "ttyd0". You may need to reconfigure applications to
1135 use the new device names.
1137 When using the serial port as a boot console, be sure to update
1138 /boot/device.hints and /etc/ttys before booting the new kernel.
1139 If you forget to do so, you can still manually specify the hints
1140 at the loader prompt:
1142 set hint.uart.0.at="isa"
1143 set hint.uart.0.port="0x3F8"
1144 set hint.uart.0.flags="0x10"
1145 set hint.uart.0.irq="4"
1149 The gpt(8) utility has been removed. Use gpart(8) to partition
1153 The version that Linuxulator emulates was changed from 2.4.2
1154 to 2.6.16. If you experience any problems with Linux binaries
1155 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1156 if it fixes the problem contact emulation mailing list.
1159 ISDN4BSD (I4B) was removed from the src tree. You may need to
1160 update a your kernel configuration and remove relevant entries.
1163 I have checked in code to support multiple routing tables.
1164 See the man pages setfib(1) and setfib(2).
1165 This is a hopefully backwards compatible version,
1166 but to make use of it you need to compile your kernel
1167 with options ROUTETABLES=2 (or more up to 16).
1170 The 802.11 wireless support was redone to enable multi-bss
1171 operation on devices that are capable. The underlying device
1172 is no longer used directly but instead wlanX devices are
1173 cloned with ifconfig. This requires changes to rc.conf files.
1174 For example, change:
1175 ifconfig_ath0="WPA DHCP"
1178 ifconfig_wlan0="WPA DHCP"
1179 see rc.conf(5) for more details. In addition, mergemaster of
1180 /etc/rc.d is highly recommended. Simultaneous update of userland
1181 and kernel wouldn't hurt either.
1183 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1184 modules were merged into the base wlan module. All references
1185 to these modules (e.g. in kernel config files) must be removed.
1188 psm(4) has gained write(2) support in native operation level.
1189 Arbitrary commands can be written to /dev/psm%d and status can
1190 be read back from it. Therefore, an application is responsible
1191 for status validation and error recovery. It is a no-op in
1192 other operation levels.
1195 Support for KSE threading has been removed from the kernel. To
1196 run legacy applications linked against KSE libmap.conf may
1197 be used. The following libmap.conf may be used to ensure
1198 compatibility with any prior release:
1200 libpthread.so.1 libthr.so.1
1201 libpthread.so.2 libthr.so.2
1202 libkse.so.3 libthr.so.3
1205 The layout of struct vmspace has changed. This affects libkvm
1206 and any executables that link against libkvm and use the
1207 kvm_getprocs() function. In particular, but not exclusively,
1208 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1209 The effects are minimal, but it's advisable to upgrade world
1213 The latest em driver no longer has support in it for the
1214 82575 adapter, this is now moved to the igb driver. The
1215 split was done to make new features that are incompatible
1216 with older hardware easier to do.
1219 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1220 likewise the kernel option is now GEOM_LINUX_LVM.
1223 The default NFS mount mode has changed from UDP to TCP for
1224 increased reliability. If you rely on (insecurely) NFS
1225 mounting across a firewall you may need to update your
1229 Belatedly note the addition of m_collapse for compacting
1233 The fts(3) structures have been changed to use adequate
1234 integer types for their members and so to be able to cope
1235 with huge file trees. The old fts(3) ABI is preserved
1236 through symbol versioning in libc, so third-party binaries
1237 using fts(3) should still work, although they will not take
1238 advantage of the extended types. At the same time, some
1239 third-party software might fail to build after this change
1240 due to unportable assumptions made in its source code about
1241 fts(3) structure members. Such software should be fixed
1242 by its vendor or, in the worst case, in the ports tree.
1243 FreeBSD_version 800015 marks this change for the unlikely
1244 case that a portable fix is impossible.
1247 To upgrade to -current after this date, you must be running
1248 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1249 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1252 The ADAPTIVE_GIANT kernel option has been retired because its
1253 functionality is the default now.
1256 The AT keyboard emulation of sunkbd(4) has been turned on
1257 by default. In order to make the special symbols of the Sun
1258 keyboards driven by sunkbd(4) work under X these now have
1259 to be configured the same way as Sun USB keyboards driven
1260 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1262 Option "XkbLayout" "us"
1263 Option "XkbRules" "xorg"
1264 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1267 It has been decided that it is desirable to provide ABI
1268 backwards compatibility to the FreeBSD 4/5/6 versions of the
1269 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1270 broken with the introduction of PCI domain support (see the
1271 20070930 entry). Unfortunately, this required the ABI of
1272 PCIOCGETCONF to be broken again in order to be able to
1273 provide backwards compatibility to the old version of that
1274 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1275 again. As for prominent ports this affects neither pciutils
1276 nor xorg-server this time, the hal port needs to be rebuilt
1280 The misnamed kthread_create() and friends have been renamed
1281 to kproc_create() etc. Many of the callers already
1282 used kproc_start()..
1283 I will return kthread_create() and friends in a while
1284 with implementations that actually create threads, not procs.
1285 Renaming corresponds with version 800002.
1294 Avoid using make -j when upgrading. While generally safe, there are
1295 sometimes problems using -j to upgrade. If your upgrade fails with
1296 -j, please try again without -j. From time to time in the past there
1297 have been problems using -j with buildworld and/or installworld. This
1298 is especially true when upgrading between "distant" versions (eg one
1299 that cross a major release boundary or several minor releases, or when
1300 several months have passed on the -current branch).
1302 Sometimes, obscure build problems are the result of environment
1303 poisoning. This can happen because the make utility reads its
1304 environment when searching for values for global variables. To run
1305 your build attempts in an "environmental clean room", prefix all make
1306 commands with 'env -i '. See the env(1) manual page for more details.
1308 When upgrading from one major version to another it is generally best
1309 to upgrade to the latest code in the currently installed branch first,
1310 then do an upgrade to the new branch. This is the best-tested upgrade
1311 path, and has the highest probability of being successful. Please try
1312 this approach before reporting problems with a major version upgrade.
1316 When upgrading the boot ZFS pool to a new version, always follow
1319 1.) recompile and reinstall the ZFS boot loader and boot block
1320 (this is part of "make buildworld" and "make installworld")
1322 2.) update the ZFS boot block on your boot drive
1324 The following example updates the ZFS boot block on the first
1325 partition (freebsd-boot) of a GPT partitioned drive ad0:
1326 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1328 Non-boot pools do not need these updates.
1332 If you are updating from a prior version of FreeBSD (even one just
1333 a few days old), you should follow this procedure. It is the most
1334 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1336 make kernel-toolchain
1337 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1338 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1340 To test a kernel once
1341 ---------------------
1342 If you just want to boot a kernel once (because you are not sure
1343 if it works, or if you want to boot a known bad kernel to provide
1344 debugging information) run
1345 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1346 nextboot -k testkernel
1348 To just build a kernel when you know that it won't mess you up
1349 --------------------------------------------------------------
1350 This assumes you are already running a CURRENT system. Replace
1351 ${arch} with the architecture of your machine (e.g. "i386",
1352 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1354 cd src/sys/${arch}/conf
1355 config KERNEL_NAME_HERE
1356 cd ../compile/KERNEL_NAME_HERE
1361 If this fails, go to the "To build a kernel" section.
1363 To rebuild everything and install it on the current system.
1364 -----------------------------------------------------------
1365 # Note: sometimes if you are running current you gotta do more than
1366 # is listed here if you are upgrading from a really old current.
1368 <make sure you have good level 0 dumps>
1370 make kernel KERNCONF=YOUR_KERNEL_HERE
1372 <reboot in single user> [3]
1380 To cross-install current onto a separate partition
1381 --------------------------------------------------
1382 # In this approach we use a separate partition to hold
1383 # current's root, 'usr', and 'var' directories. A partition
1384 # holding "/", "/usr" and "/var" should be about 2GB in
1387 <make sure you have good level 0 dumps>
1390 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1391 <maybe newfs current's root partition>
1392 <mount current's root partition on directory ${CURRENT_ROOT}>
1393 make installworld DESTDIR=${CURRENT_ROOT}
1394 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1395 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1396 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1397 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1398 <reboot into current>
1399 <do a "native" rebuild/install as described in the previous section>
1400 <maybe install compatibility libraries from ports/misc/compat*>
1404 To upgrade in-place from 8.x-stable to current
1405 ----------------------------------------------
1406 <make sure you have good level 0 dumps>
1408 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1410 <reboot in single user> [3]
1417 Make sure that you've read the UPDATING file to understand the
1418 tweaks to various things you need. At this point in the life
1419 cycle of current, things change often and you are on your own
1420 to cope. The defaults can also change, so please read ALL of
1421 the UPDATING entries.
1423 Also, if you are tracking -current, you must be subscribed to
1424 freebsd-current@freebsd.org. Make sure that before you update
1425 your sources that you have read and understood all the recent
1426 messages there. If in doubt, please track -stable which has
1427 much fewer pitfalls.
1429 [1] If you have third party modules, such as vmware, you
1430 should disable them at this point so they don't crash your
1433 [3] From the bootblocks, boot -s, and then do
1438 adjkerntz -i # if CMOS is wall time
1439 Also, when doing a major release upgrade, it is required that
1440 you boot into single user mode to do the installworld.
1442 [4] Note: This step is non-optional. Failure to do this step
1443 can result in a significant reduction in the functionality of the
1444 system. Attempting to do it by hand is not recommended and those
1445 that pursue this avenue should read this file carefully, as well
1446 as the archives of freebsd-current and freebsd-hackers mailing lists
1447 for potential gotchas. The -U option is also useful to consider.
1448 See mergemaster(8) for more information.
1450 [5] Usually this step is a noop. However, from time to time
1451 you may need to do this if you get unknown user in the following
1452 step. It never hurts to do it all the time. You may need to
1453 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1454 install) after the buildworld before this step if you last updated
1455 from current before 20020224 or from -stable before 20020408.
1457 [6] This only deletes old files and directories. Old libraries
1458 can be deleted by "make delete-old-libs", but you have to make
1459 sure that no program is using those libraries anymore.
1461 [8] In order to have a kernel that can run the 4.x binaries needed to
1462 do an installworld, you must include the COMPAT_FREEBSD4 option in
1463 your kernel. Failure to do so may leave you with a system that is
1464 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1465 required to run the 5.x binaries on more recent kernels. And so on
1466 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1468 Make sure that you merge any new devices from GENERIC since the
1469 last time you updated your kernel config file.
1471 [9] When checking out sources, you must include the -P flag to have
1472 cvs prune empty directories.
1474 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1475 "?=" instead of the "=" assignment operator, so that buildworld can
1476 override the CPUTYPE if it needs to.
1478 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1479 not on the command line, or in /etc/make.conf. buildworld will
1480 warn if it is improperly defined.
1483 This file contains a list, in reverse chronological order, of major
1484 breakages in tracking -current. It is not guaranteed to be a complete
1485 list of such breakages, and only contains entries since October 10, 2007.
1486 If you need to see UPDATING entries from before that date, you will need
1487 to fetch an UPDATING file from an older FreeBSD release.
1489 Copyright information:
1491 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1493 Redistribution, publication, translation and use, with or without
1494 modification, in full or in part, in any form or format of this
1495 document are permitted without further permission from the author.
1497 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1498 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1499 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1500 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1501 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1502 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1503 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1504 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1505 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1506 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1507 POSSIBILITY OF SUCH DAMAGE.
1509 Contact Warner Losh if you have any questions about your use of