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 interface of the VOP_VPTOCNP(9) changed, now the returned
14 vnode shall be referenced, previously it was required to be
15 only held. All in-tree filesystems are converted.
21 The broken amd(4) driver has been replaced with esp(4) in the amd64,
22 i386 and pc98 GENERIC kernel configuration files.
25 This commit modifies vfs_register() so that it uses a hash
26 calculation to set vfc_typenum, which is enabled by default.
27 The first time a system is booted after this change, the
28 vfc_typenum values will change for all file systems. The
29 main effect of this is a change to the NFS server file handles
30 for file systems that use vfc_typenum in their fsid, such as ZFS.
31 It will, however, prevent vfc_typenum from changing when file
32 systems are loaded in a different order for subsequent reboots.
33 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
34 until you are ready to remount all NFS clients after a reboot.
37 Bump the shared library version numbers for libraries that
38 do not use symbol versioning, have changed the ABI compared
39 to stable/8 and which shared library version was not bumped.
40 Done as part of 9.0-RELEASE cycle.
43 During the merge of Capsicum features, the fget(9) KPI was modified.
44 This may require the rebuilding of out-of-tree device drivers --
45 issues have been reported specifically with the nVidia device driver.
46 __FreeBSD_version is bumped to 900041.
48 Also, there is a period between 20110811 and 20110814 where the
49 special devices /dev/{stdin,stdout,stderr} did not work correctly.
50 Building world from a kernel during that window may not work.
53 The packet filter (pf) code has been updated to OpenBSD 4.5.
54 You need to update userland tools to be in sync with kernel.
55 This update breaks backward compatibility with earlier pfsync(4)
56 versions. Care must be taken when updating redundant firewall setups.
59 The following sysctls and tunables are retired on x86 platforms:
61 machdep.hlt_logical_cpus
62 The following sysctl is retired:
63 machdep.hyperthreading_allowed
64 The sysctls were supposed to provide a way to dynamically offline and
65 online selected CPUs on x86 platforms, but the implementation has not
66 been reliable especially with SCHED_ULE scheduler.
67 machdep.hyperthreading_allowed tunable is still available to ignore
68 hyperthreading CPUs at OS level.
69 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
70 where X is an APIC ID of a CPU. Be advised, though, that disabling
71 CPUs in non-uniform fashion will result in non-uniform topology and
72 may lead to sub-optimal system performance with SCHED_ULE, which is
76 cpumask_t type is retired and cpuset_t is used in order to describe
80 Changes to ifconfig(8) for dynamic address family detection mandate
81 that you are running a kernel of 20110525 or later. Make sure to
82 follow the update procedure to boot a new kernel before installing
86 Support for sun4v architecture is officially dropped
89 Several KPI breaking changes have been committed to the mii(4) layer,
90 the PHY drivers and consequently some Ethernet drivers using mii(4).
91 This means that miibus.ko and the modules of the affected Ethernet
92 drivers need to be recompiled.
94 Note to kernel developers: Given that the OUI bit reversion problem
95 was fixed as part of these changes all mii(4) commits related to OUIs,
96 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
97 handling, no longer can be merged verbatim to stable/8 and previous
101 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
102 into their kernel configurations along with 'device pci'.
105 The default NFS client is now the new NFS client, so fstype "newnfs"
106 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
107 Although mounts via fstype "nfs" will usually work without userland
108 changes, it is recommended that the mount(8) and mount_nfs(8)
109 commands be rebuilt from sources and that a link to mount_nfs called
110 mount_oldnfs be created. The new client is compiled into the
111 kernel with "options NFSCL" and this is needed for diskless root
112 file systems. The GENERIC kernel configs have been changed to use
113 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
114 To use the regular/old client, you can "mount -t oldnfs ...". For
115 a diskless root file system, you must also include a line like:
117 vfs.root.mountfrom="oldnfs:"
119 in the boot/loader.conf on the root fs on the NFS server to make
120 a diskless root fs use the old client.
123 The GENERIC kernels for all architectures now default to the new
124 CAM-based ATA stack. It means that all legacy ATA drivers were
125 removed and replaced by respective CAM drivers. If you are using
126 ATA device names in /etc/fstab or other places, make sure to update
127 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
128 where 'Y's are the sequential numbers starting from zero for each type
129 in order of detection, unless configured otherwise with tunables,
130 see cam(4)). There will be symbolic links created in /dev/ to map
131 old adX devices to the respective adaY. They should provide basic
132 compatibility for file systems mounting in most cases, but they do
133 not support old user-level APIs and do not have respective providers
134 in GEOM. Consider using updated management tools with new device names.
136 It is possible to load devices ahci, ata, siis and mvs as modules,
137 but option ATA_CAM should remain in kernel configuration to make ata
138 module work as CAM driver supporting legacy ATA controllers. Device ata
139 still can be used in modular fashion (atacore + ...). Modules atadisk
140 and atapi* are not used and won't affect operation in ATA_CAM mode.
141 Note that to use CAM-based ATA kernel should include CAM devices
142 scbus, pass, da (or explicitly ada), cd and optionally others. All of
143 them are parts of the cam module.
145 ataraid(4) functionality is now supported by the RAID GEOM class.
146 To use it you can load geom_raid kernel module and use graid(8) tool
147 for management. Instead of /dev/arX device names, use /dev/raid/rX.
149 No kernel config options or code have been removed, so if a problem
150 arises, please report it and optionally revert to the old ATA stack.
151 In order to do it you can remove from the kernel config:
156 , and instead add back:
157 device atadisk # ATA disk drives
158 device ataraid # ATA RAID drives
159 device atapicd # ATAPI CDROM drives
160 device atapifd # ATAPI floppy drives
161 device atapist # ATAPI tape drives
164 The default NFS server has been changed to the new server, which
165 was referred to as the experimental server. If you need to switch
166 back to the old NFS server, you must now put the "-o" option on
167 both the mountd and nfsd commands. This can be done using the
168 mountd_flags and nfs_server_flags rc.conf variables until an
169 update to the rc scripts is committed, which is coming soon.
172 The GNU Objective-C runtime library (libobjc), and other Objective-C
173 related components have been removed from the base system. If you
174 require an Objective-C library, please use one of the available ports.
177 ath(4) has been split into bus- and device- modules. if_ath contains
178 the HAL, the TX rate control and the network device code. if_ath_pci
179 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
180 contains the AHB glue. Users need to load both if_ath_pci and if_ath
181 in order to use ath on everything else.
183 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
184 need to load if_ath and if_ath_pci for ath(4) operation.
187 As part of the replacement of sysinstall, the process of building
188 release media has changed significantly. For details, please re-read
189 release(7), which has been updated to reflect the new build process.
192 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
193 is the last available version under GPLv2. It brings a number of new
194 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
195 4.1 and SSE 4.2), better support for powerpc64, a number of new
196 directives, and lots of other small improvements. See the ChangeLog
197 file in contrib/binutils for the full details.
200 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
201 compliant, and will now use half of hash for authentication.
202 This will break interoperability with all stacks (including all
203 actual FreeBSD versions) who implement
204 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
206 The only workaround with such peers is to use another HMAC
207 algorithm for IPsec ("phase 2") authentication.
210 Remove the uio_yield prototype and symbol. This function has
211 been misnamed since it was introduced and should not be
212 globally exposed with this name. The equivalent functionality
213 is now available using kern_yield(curthread->td_user_pri).
214 The function remains undocumented.
217 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
218 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
219 for scalar sysctls is defined but disabled. Code that needs
220 UQUAD to pass the type checking that must compile on older
221 systems where the define is not present can check against
222 __FreeBSD_version >= 900030.
224 The system dialog(1) has been replaced with a new version previously
225 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
226 with the previous version, but the libdialog associated with it has
227 a largely incompatible API. As such, the original version of libdialog
228 will be kept temporarily as libodialog, until its base system consumers
229 are replaced or updated. Bump __FreeBSD_version to 900030.
232 If you are trying to run make universe on a -stable system, and you get
233 the following warning:
234 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
235 unknown. config(8) likely too old."
236 or something similar to it, then you must upgrade your -stable system
237 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
238 stable/8) or build the config from the latest stable/8 branch and
239 install it on your system.
241 Prior to this date, building a current universe on 8-stable system from
242 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
243 error in the first kernel build phase. A new config on those old
244 systems will fix that problem for older versions of -current.
247 The TCP stack has been modified to allow Khelp modules to interact with
248 it via helper hook points and store per-connection data in the TCP
249 control block. Bump __FreeBSD_version to 900029. User space tools that
250 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
254 Generic IEEE 802.3 annex 31B full duplex flow control support has been
255 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
256 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
257 to take advantage of it instead of using custom implementations. This
258 means that these drivers now no longer unconditionally advertise
259 support for flow control but only do so if flow control is a selected
260 media option. This was implemented in the generic support that way in
261 order to allow flow control to be switched on and off via ifconfig(8)
262 with the PHY specific default to typically off in order to protect
263 from unwanted effects. Consequently, if you used flow control with
264 one of the above mentioned drivers you now need to explicitly enable
266 ifconfig bge0 media auto mediaopt flowcontrol
268 Along with the above mentioned changes generic support for setting
269 1000baseT master mode also has been added and brgphy(4), ciphy(4),
270 e1000phy(4) as well as ip1000phy(4) have been converted to take
271 advantage of it. This means that these drivers now no longer take the
272 link0 parameter for selecting master mode but the master media option
273 has to be used instead, for example like in the following:
274 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
276 Selection of master mode now is also available with all other PHY
277 drivers supporting 1000baseT.
280 The TCP stack has received a significant update to add support for
281 modularised congestion control and generally improve the clarity of
282 congestion control decisions. Bump __FreeBSD_version to 900025. User
283 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
284 sockstat) need to be recompiled.
287 The man(1) utility has been replaced by a new version that no longer
288 uses /etc/manpath.config. Please consult man.conf(5) for how to
289 migrate local entries to the new format.
292 The copyright strings printed by login(1) and sshd(8) at the time of a
293 new connection have been removed to follow other operating systems and
297 A workaround for a fixed ld bug has been removed in kernel code,
298 so make sure that your system ld is built from sources after
299 revision 210245 from 2010-07-19 (r211583 if building head kernel
300 on stable/8, r211584 for stable/7; both from 2010-08-21).
301 A symptom of incorrect ld version is different addresses for
302 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
305 The $ipv6_prefer variable in rc.conf(5) has been split into
306 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
308 The $ip6addrctl_policy is a variable to choose a pre-defined
309 address selection policy set by ip6addrctl(8). A value
310 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
313 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
314 flag (see an entry of 20090926) is set on an interface with no
315 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
316 security reason. If you want IPv6 link-local address on all
317 interfaces by default, set this to "YES".
319 The old ipv6_prefer="YES" is equivalent to
320 ipv6_activate_all_interfaces="YES" and
321 ip6addrctl_policy="ipv6_prefer".
324 DTrace has grown support for userland tracing. Due to this, DTrace is
325 now i386 and amd64 only.
326 dtruss(1) is now installed by default on those systems and a new
327 kernel module is needed for userland tracing: fasttrap.
328 No changes to your kernel config file are necessary to enable
329 userland tracing, but you might consider adding 'STRIP=' and
330 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
331 to have informative userland stack traces in DTrace (ustack).
334 The acpi_aiboost(4) driver has been removed in favor of the new
335 aibs(4) driver. You should update your kernel configuration file.
338 BSD grep has been imported to the base system and it is built by
339 default. It is completely BSD licensed, highly GNU-compatible, uses
340 less memory than its GNU counterpart and has a small codebase.
341 However, it is slower than its GNU counterpart, which is mostly
342 noticeable for larger searches, for smaller ones it is measurable
343 but not significant. The reason is complex, the most important factor
344 is that we lack a modern and efficient regex library and GNU
345 overcomes this by optimizing the searches internally. Future work
346 on improving the regex performance is planned, for the meantime,
347 users that need better performance, can build GNU grep instead by
348 setting the WITH_GNU_GREP knob.
351 Due to the import of powerpc64 support, all existing powerpc kernel
352 configuration files must be updated with a machine directive like this:
353 machine powerpc powerpc
355 In addition, an updated config(8) is required to build powerpc kernels
359 A new version of ZFS (version 15) has been merged to -HEAD.
360 This version uses a python library for the following subcommands:
361 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
362 For full functionality of these commands the following port must
363 be installed: sysutils/py-zfs
366 'vm_page's are now hashed by physical address to an array of mutexes.
367 Currently this is only used to serialize access to hold_count. Over
368 time the page queue mutex will be peeled away. This changes the size
369 of pmap on every architecture. And requires all callers of vm_page_hold
370 and vm_page_unhold to be updated.
373 WITH_CTF can now be specified in src.conf (not recommended, there
374 are some problems with static executables), make.conf (would also
375 affect ports which do not use GNU make and do not override the
376 compile targets) or in the kernel config (via "makeoptions
378 When WITH_CTF was specified there before this was silently ignored,
379 so make sure that WITH_CTF is not used in places which could lead
380 to unwanted behavior.
383 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
384 to allow 32-bit compatibility on non-x86 platforms. All kernel
385 configurations on amd64 and ia64 platforms using these options must
386 be modified accordingly.
389 The utmp user accounting database has been replaced with utmpx,
390 the user accounting interface standardized by POSIX.
391 Unfortunately the semantics of utmp and utmpx don't match,
392 making it practically impossible to support both interfaces.
393 The user accounting database is used by tools like finger(1),
394 last(1), talk(1), w(1) and ac(8).
396 All applications in the base system use utmpx. This means only
397 local binaries (e.g. from the ports tree) may still use these
398 utmp database files. These applications must be rebuilt to make
401 After the system has been upgraded, it is safe to remove the old
402 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
403 assuming their contents is of no importance anymore. Old wtmp
404 databases can only be used by last(1) and ac(8) after they have
405 been converted to the new format using wtmpcvt(1).
408 Introduce the kernel thread "deadlock resolver" (which can be enabled
409 via the DEADLKRES option, see NOTES for more details) and the
410 sleepq_type() function for sleepqueues.
413 The rc.firewall and rc.firewall6 were unified, and
414 rc.firewall6 and rc.d/ip6fw were removed.
415 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
416 variables are obsoleted. Instead, the following new rc
417 variables are added to rc.d/ipfw:
419 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
420 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
421 firewall_simple_onet_ipv6, firewall_trusted_ipv6
423 The meanings correspond to the relevant IPv4 variables.
429 The default terminal emulation for syscons(4) has been changed
430 from cons25 to xterm on all platforms except pc98. This means
431 that the /etc/ttys file needs to be updated to ensure correct
432 operation of applications on the console.
434 The terminal emulation style can be toggled per window by using
435 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
436 options can be used to change the compile-time default back to
439 To prevent graphical artifacts, make sure the TERM environment
440 variable is set to match the terminal emulation that is being
441 performed by syscons(4).
444 The layout of the structure ieee80211req_scan_result has changed.
445 Applications that require wireless scan results (e.g. ifconfig(8))
446 from net80211 need to be recompiled.
448 Applications such as wpa_supplicant(8) may require a full world
449 build without using NO_CLEAN in order to get synchronized with the
453 The iwn(4) driver has been updated to support the 5000 and 5150 series.
454 There's one kernel module for each firmware. Adding "device iwnfw"
455 to the kernel configuration file means including all three firmware
456 images inside the kernel. If you want to include just the one for
457 your wireless card, use the devices iwn4965fw, iwn5000fw or
461 The rc.d/network_ipv6, IPv6 configuration script has been integrated
462 into rc.d/netif. The changes are the following:
464 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
465 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
466 Note that both variables need the "inet6" keyword at the head.
468 Do not set $ipv6_network_interfaces manually if you do not
469 understand what you are doing. It is not needed in most cases.
471 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
474 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
475 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
477 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
478 all configured interfaces have "inet6 accept_rtadv" in the
479 $ifconfig_IF_ipv6. These are for backward compatibility.
481 3. A new variable $ipv6_prefer has been added. If NO, IPv6
482 functionality of interfaces with no corresponding
483 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
484 and the default address selection policy of ip6addrctl(8)
485 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
486 Note that if you want to configure IPv6 functionality on the
487 disabled interfaces after boot, first you need to clear the flag by
488 using ifconfig(8) like:
490 ifconfig em0 inet6 -ifdisabled
492 If YES, the default address selection policy is set as
495 The default value of $ipv6_prefer is NO.
497 4. If your system need to receive Router Advertisement messages,
498 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
499 scripts automatically invoke rtsol(8) when the interface becomes
500 UP. The Router Advertisement messages are used for SLAAC
501 (State-Less Address AutoConfiguration).
504 802.11s D3.03 support was committed. This is incompatible with the
505 previous code, which was based on D3.0.
508 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
509 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
510 control whether accepting Router Advertisement messages or not.
511 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
512 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
513 The ifconfig(8) utility now supports these flags.
516 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
517 mount(8) and -a option for df(1) to see them.
520 The old tunable hw.bus.devctl_disable has been superseded by
521 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
522 replaced by hw.bus.devctl_queue=0. The default for this new tunable
526 Remove the option STOP_NMI. The default action is now to use NMI only
527 for KDB via the newly introduced function stop_cpus_hard() and
528 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
531 The stable/8 branch created in subversion. This corresponds to the
532 RELENG_8 branch in CVS.
535 Bump the shared library version numbers for all libraries that do not
536 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
537 __FreeBSD_version to 800105.
540 Due to changes in the implementation of virtual network stack support,
541 all network-related kernel modules must be recompiled. As this change
542 breaks the ABI, bump __FreeBSD_version to 800104.
545 The TOE interface to the TCP syncache has been modified to remove
546 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
547 The cxgb driver is the only TOE consumer affected by this change, and
548 needs to be recompiled along with the kernel. As this change breaks
549 the ABI, bump __FreeBSD_version to 800103.
552 Padding has been added to struct tcpcb, sackhint and tcpstat in
553 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
554 maintaining the ABI. However, this change breaks the ABI, so bump
555 __FreeBSD_version to 800102. User space tools that rely on the size of
556 any of these structs (e.g. sockstat) need to be recompiled.
559 The NFS_LEGACYRPC option has been removed along with the old kernel
560 RPC implementation that this option selected. Kernel configurations
561 may need to be adjusted.
564 The network interface device nodes at /dev/net/<interface> have been
565 removed. All ioctl operations can be performed the normal way using
566 routing sockets. The kqueue functionality can generally be replaced
567 with routing sockets.
570 The documentation from the FreeBSD Documentation Project (Handbook,
571 FAQ, etc.) is now installed via packages by sysinstall(8) and under
572 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
575 The ABI of various structures related to the SYSV IPC API have been
576 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
577 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
581 Layout of struct vnet has changed as routing related variables were
582 moved to their own Vimage module. Modules need to be recompiled. Bump
583 __FreeBSD_version to 800099.
586 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
587 respectively. As long as no more than 16 groups per process are used,
588 no changes should be visible. When more than 16 groups are used, old
589 binaries may fail if they call getgroups() or getgrouplist() with
590 statically sized storage. Recompiling will work around this, but
591 applications should be modified to use dynamically allocated storage
592 for group arrays as POSIX.1-2008 does not cap an implementation's
593 number of supported groups at NGROUPS_MAX+1 as previous versions did.
595 NFS and portalfs mounts may also be affected as the list of groups is
596 truncated to 16. Users of NFS who use more than 16 groups, should
597 take care that negative group permissions are not used on the exported
598 file systems as they will not be reliable unless a GSSAPI based
599 authentication method is used.
602 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
603 option compiles in the support for adaptive spinning for lockmgrs
604 which want to enable it. The lockinit() function now accepts the flag
605 LK_ADAPTIVE in order to make the lock object subject to adaptive
606 spinning when both held in write and read mode.
609 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
610 changed. User applications that use this ioctl need to be rebuilt.
613 The layout of struct thread has changed. Kernel and modules need to
617 The layout of structs ifnet, domain, protosw and vnet_net has changed.
618 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
621 window(1) has been removed from the base system. It can now be
622 installed from ports. The port is called misc/window.
625 The way we are storing and accessing `routing table' entries has
626 changed. Programs reading the FIB, like netstat, need to be
630 A new netisr implementation has been added for FreeBSD 8. Network
631 file system modules, such as igmp, ipdivert, and others, should be
633 Bump __FreeBSD_version to 800096.
636 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
640 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
641 Bump __FreeBSD_version to 800094.
644 Add mnt_xflag field to 'struct mount'. File system modules need to be
646 Bump __FreeBSD_version to 800093.
649 The compiling option ADAPTIVE_SX has been retired while it has been
650 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
651 The KPI for sx_init_flags() changes as accepting flags:
652 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
653 been introduced in order to handle the reversed logic.
654 Bump __FreeBSD_version to 800092.
657 Add support for hierarchical jails. Remove global securelevel.
658 Bump __FreeBSD_version to 800091.
661 The layout of struct vnet_net has changed, therefore modules
663 Bump __FreeBSD_version to 800090.
666 The newly imported zic(8) produces a new format in the output. Please
667 run tzsetup(8) to install the newly created data to /etc/localtime.
670 The sysctl tree for the usb stack has renamed from hw.usb2.* to
671 hw.usb.* and is now consistent again with previous releases.
674 802.11 monitor mode support was revised and driver api's were changed.
675 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
676 of DLT_IEEE802_11. No user-visible data structures were changed but
677 applications that use DLT_IEEE802_11 may require changes.
678 Bump __FreeBSD_version to 800088.
681 The layout of the following structs has changed: sysctl_oid,
682 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
683 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
684 panics may be experienced. World rebuild is required for
685 correctly checking networking state from userland.
686 Bump __FreeBSD_version to 800085.
689 MLDv2 and Source-Specific Multicast (SSM) have been merged
690 to the IPv6 stack. VIMAGE hooks are in but not yet used.
691 The implementation of SSM within FreeBSD's IPv6 stack closely
692 follows the IPv4 implementation.
694 For kernel developers:
696 * The most important changes are that the ip6_output() and
697 ip6_input() paths no longer take the IN6_MULTI_LOCK,
698 and this lock has been downgraded to a non-recursive mutex.
700 * As with the changes to the IPv4 stack to support SSM, filtering
701 of inbound multicast traffic must now be performed by transport
702 protocols within the IPv6 stack. This does not apply to TCP and
703 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
705 * The KPIs used by IPv6 multicast are similar to those used by
706 the IPv4 stack, with the following differences:
707 * im6o_mc_filter() is analogous to imo_multicast_filter().
708 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
709 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
710 * IN6_LOOKUP_MULTI() has been deprecated and removed.
711 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
712 for MLDv1 have an additional 'timer' argument which is used to
713 jitter the initial membership report for the solicited-node
714 multicast membership on-link.
715 * This is not strictly needed for MLDv2, which already jitters
716 its report transmissions. However, the 'timer' argument is
717 preserved in case MLDv1 is active on the interface.
719 * The KAME linked-list based IPv6 membership implementation has
720 been refactored to use a vector similar to that used by the IPv4
722 Code which maintains a list of its own multicast memberships
723 internally, e.g. carp, has been updated to reflect the new
726 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
727 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
728 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
729 implementation constraint which needs to be addressed in HEAD.
731 For application developers:
733 * The changes are broadly similar to those made for the IPv4
736 * The use of IPv4 and IPv6 multicast socket options on the same
737 socket, using mapped addresses, HAS NOT been tested or supported.
739 * There are a number of issues with the implementation of various
740 IPv6 multicast APIs which need to be resolved in the API surface
741 before the implementation is fully compatible with KAME userland
742 use, and these are mostly to do with interface index treatment.
744 * The literature available discusses the use of either the delta / ASM
745 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
746 using setsourcefilter(3)/getsourcefilter(3). For more information
747 please refer to RFC 3768, 'Socket Interface Extensions for
748 Multicast Source Filters'.
750 * Applications which use the published RFC 3678 APIs should be fine.
752 For systems administrators:
754 * The mtest(8) utility has been refactored to support IPv6, in
755 addition to IPv4. Interface addresses are no longer accepted
756 as arguments, their names must be used instead. The utility
757 will map the interface name to its first IPv4 address as
758 returned by getifaddrs(3).
760 * The ifmcstat(8) utility has also been updated to print the MLDv2
761 endpoint state and source filter lists via sysctl(3).
763 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
764 loopback of IPv6 multicast datagrams by default; it defaults to 1
765 to preserve the existing behaviour. Disabling multicast loopback is
766 recommended for optimal system performance.
768 * The IPv6 MROUTING code has been changed to examine this sysctl
769 instead of attempting to perform a group lookup before looping
770 back forwarded datagrams.
772 Bump __FreeBSD_version to 800084.
775 Implement low-level Bluetooth HCI API.
776 Bump __FreeBSD_version to 800083.
779 The layout of struct malloc_type, used by modules to register new
780 memory allocation types, has changed. Most modules will need to
781 be rebuilt or panics may be experienced.
782 Bump __FreeBSD_version to 800081.
785 Anticipate overflowing inp_flags - add inp_flags2.
786 This changes most offsets in inpcb, so checking v4 connection
787 state will require a world rebuild.
788 Bump __FreeBSD_version to 800080.
791 Add an llentry to struct route and struct route_in6. Modules
792 embedding a struct route will need to be recompiled.
793 Bump __FreeBSD_version to 800079.
796 The size of rt_metrics_lite and by extension rtentry has changed.
797 Networking administration apps will need to be recompiled.
798 The route command now supports show as an alias for get, weighting
799 of routes, sticky and nostick flags to alter the behavior of stateful
801 Bump __FreeBSD_version to 800078.
804 Do not use Giant for kbdmux(4) locking. This is wrong and
805 apparently causing more problems than it solves. This will
806 re-open the issue where interrupt handlers may race with
807 kbdmux(4) in polling mode. Typical symptoms include (but
808 not limited to) duplicated and/or missing characters when
809 low level console functions (such as gets) are used while
810 interrupts are enabled (for example geli password prompt,
811 mountroot prompt etc.). Disabling kbdmux(4) may help.
814 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
815 kernel modules referencing any of the above need to be recompiled.
816 Bump __FreeBSD_version to 800075.
819 GEOM_PART has become the default partition slicer for storage devices,
820 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
821 introduces some changes:
823 MSDOS/EBR: the devices created from MSDOS extended partition entries
824 (EBR) can be named differently than with GEOM_MBR and are now symlinks
825 to devices with offset-based names. fstabs may need to be modified.
827 BSD: the "geometry does not match label" warning is harmless in most
828 cases but it points to problems in file system misalignment with
829 disk geometry. The "c" partition is now implicit, covers the whole
830 top-level drive and cannot be (mis)used by users.
832 General: Kernel dumps are now not allowed to be written to devices
833 whose partition types indicate they are meant to be used for file
834 systems (or, in case of MSDOS partitions, as something else than
837 Most of these changes date approximately from 200812.
840 The uscanner(4) driver has been removed from the kernel. This follows
841 Linux removing theirs in 2.6 and making libusb the default interface
845 The multicast forwarding code has been cleaned up. netstat(1)
846 only relies on KVM now for printing bandwidth upcall meters.
847 The IPv4 and IPv6 modules are split into ip_mroute_mod and
848 ip6_mroute_mod respectively. The config(5) options for statically
849 compiling this code remain the same, i.e. 'options MROUTING'.
852 Support for the IFF_NEEDSGIANT network interface flag has been
853 removed, which means that non-MPSAFE network device drivers are no
854 longer supported. In particular, if_ar, if_sr, and network device
855 drivers from the old (legacy) USB stack can no longer be built or
859 POSIX.1 Native Language Support (NLS) has been enabled in libc and
860 a bunch of new language catalog files have also been added.
861 This means that some common libc messages are now localized and
862 they depend on the LC_MESSAGES environmental variable.
865 The k8temp(4) driver has been renamed to amdtemp(4) since
866 support for Family 10 and Family 11 CPU families was added.
869 IGMPv3 and Source-Specific Multicast (SSM) have been merged
870 to the IPv4 stack. VIMAGE hooks are in but not yet used.
872 For kernel developers, the most important changes are that the
873 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
874 and this lock has been downgraded to a non-recursive mutex.
876 Transport protocols (UDP, Raw IP) are now responsible for filtering
877 inbound multicast traffic according to group membership and source
878 filters. The imo_multicast_filter() KPI exists for this purpose.
879 Transports which do not use multicast (SCTP, TCP) already reject
880 multicast by default. Forwarding and receive performance may improve
881 as a mutex acquisition is no longer needed in the ip_input()
882 low-level input path. in_addmulti() and in_delmulti() are shimmed
883 to new KPIs which exist to support SSM in-kernel.
885 For application developers, it is recommended that loopback of
886 multicast datagrams be disabled for best performance, as this
887 will still cause the lock to be taken for each looped-back
888 datagram transmission. The net.inet.ip.mcast.loop sysctl may
889 be tuned to 0 to disable loopback by default; it defaults to 1
890 to preserve the existing behaviour.
892 For systems administrators, to obtain best performance with
893 multicast reception and multiple groups, it is always recommended
894 that a card with a suitably precise hash filter is used. Hash
895 collisions will still result in the lock being taken within the
896 transport protocol input path to check group membership.
898 If deploying FreeBSD in an environment with IGMP snooping switches,
899 it is recommended that the net.inet.igmp.sendlocal sysctl remain
900 enabled; this forces 224.0.0.0/24 group membership to be announced
903 The size of 'struct igmpstat' has changed; netstat needs to be
904 recompiled to reflect this.
905 Bump __FreeBSD_version to 800070.
908 libusb20.so.1 is now installed as libusb.so.1 and the ports system
909 updated to use it. This requires a buildworld/installworld in order to
910 update the library and dependencies (usbconfig, etc). Its advisable to
911 rebuild all ports which uses libusb. More specific directions are given
912 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
913 libusb are no longer required and can be removed.
916 A workaround is committed to allow the creation of System V shared
917 memory segment of size > 2 GB on the 64-bit architectures.
918 Due to a limitation of the existing ABI, the shm_segsz member
919 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
920 wrong for large segments. Note that limits must be explicitly
921 raised to allow such segments to be created.
924 The layout of struct ifnet has changed, requiring a rebuild of all
925 network device driver modules.
928 The /dev handling for the new USB stack has changed, a
929 buildworld/installworld is required for libusb20.
932 The new USB2 stack has now been permanently moved in and all kernel and
933 module names reverted to their previous values (eg, usb, ehci, ohci,
934 ums, ...). The old usb stack can be compiled in by prefixing the name
935 with the letter 'o', the old usb modules have been removed.
936 Updating entry 20090216 for xorg and 20090215 for libmap may still
940 The rc.conf(5) option if_up_delay has been renamed to
941 defaultroute_delay to better reflect its purpose. If you have
942 customized this setting in /etc/rc.conf you need to update it to
946 xorg 7.4 wants to configure its input devices via hald which does not
947 yet work with USB2. If the keyboard/mouse does not work in xorg then
949 Option "AllowEmptyInput" "off"
950 to your ServerLayout section. This will cause X to use the configured
951 kbd and mouse sections from your xorg.conf.
954 The GENERIC kernels for all architectures now default to the new USB2
955 stack. No kernel config options or code have been removed so if a
956 problem arises please report it and optionally revert to the old USB
957 stack. If you are loading USB kernel modules or have a custom kernel
958 that includes GENERIC then ensure that usb names are also changed over,
959 eg uftdi -> usb2_serial_ftdi.
961 Older programs linked against the ports libusb 0.1 need to be
962 redirected to the new stack's libusb20. /etc/libmap.conf can
964 # Map old usb library to new one for usb2 stack
965 libusb-0.1.so.8 libusb20.so.1
968 All USB ethernet devices now attach as interfaces under the name ueN
969 (eg. ue0). This is to provide a predictable name as vendors often
970 change usb chipsets in a product without notice.
973 The ichsmb(4) driver has been changed to require SMBus slave
974 addresses be left-justified (xxxxxxx0b) rather than right-justified.
975 All of the other SMBus controller drivers require left-justified
976 slave addresses, so this change makes all the drivers provide the
980 INET6 statistics (struct ip6stat) was updated.
981 netstat(1) needs to be recompiled.
984 NTFS has been removed from GENERIC kernel on amd64 to match
985 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
986 will load ntfs.ko module automatically when NTFS support is
987 actually needed, unless ntfs.ko is not installed or security
988 level prohibits loading kernel modules. If either is the case,
989 "options NTFS" has to be added into kernel config.
992 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
993 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
994 800061. User space tools that rely on the size of struct tcpcb in
995 tcp_var.h (e.g. sockstat) need to be recompiled.
998 ng_tty(4) module updated to match the new TTY subsystem.
999 Due to API change, user-level applications must be updated.
1000 New API support added to mpd5 CVS and expected to be present
1001 in next mpd5.3 release.
1004 With __FreeBSD_version 800060 the makefs tool is part of
1005 the base system (it was a port).
1008 The afdata and ifnet locks have been changed from mutexes to
1009 rwlocks, network modules will need to be re-compiled.
1012 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1013 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1014 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1015 architecture and 40 bytes on 64-bit architecture. The userland
1016 applications "arp" and "ndp" have been updated accordingly.
1017 The output from "netstat -r" shows only routing entries and
1018 none of the L2 information.
1021 __FreeBSD_version 800057 marks the switchover from the
1022 binary ath hal to source code. Users must add the line:
1024 options AH_SUPPORT_AR5416
1026 to their kernel config files when specifying:
1030 The ath_hal module no longer exists; the code is now compiled
1031 together with the driver in the ath module. It is now
1032 possible to tailor chip support (i.e. reduce the set of chips
1033 and thereby the code size); consult ath_hal(4) for details.
1036 __FreeBSD_version 800054 adds memory barriers to
1037 <machine/atomic.h>, new interfaces to ifnet to facilitate
1038 multiple hardware transmit queues for cards that support
1039 them, and a lock-less ring-buffer implementation to
1040 enable drivers to more efficiently manage queueing of
1044 A new version of ZFS (version 13) has been merged to -HEAD.
1045 This version has zpool attribute "listsnapshots" off by
1046 default, which means "zfs list" does not show snapshots,
1047 and is the same as Solaris behavior.
1050 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1053 The uhci, ohci, ehci and slhci USB Host controller drivers have
1054 been put into separate modules. If you load the usb module
1055 separately through loader.conf you will need to load the
1056 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1057 controller add the following to loader.conf:
1063 The ABI used by the PMC toolset has changed. Please keep
1064 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1068 atapci kernel module now includes only generic PCI ATA
1069 driver. AHCI driver moved to ataahci kernel module.
1070 All vendor-specific code moved into separate kernel modules:
1071 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1072 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1073 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1074 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1077 The TTY subsystem of the kernel has been replaced by a new
1078 implementation, which provides better scalability and an
1079 improved driver model. Most common drivers have been migrated to
1080 the new TTY subsystem, while others have not. The following
1081 drivers have not yet been ported to the new TTY layer:
1084 cy, digi, rc, rp, sio
1090 ng_h4, ng_tty, ppp, sl, snp
1092 Adding these drivers to your kernel configuration file shall
1093 cause compilation to fail.
1096 ntpd has been upgraded to 4.2.4p5.
1099 OpenSSH has been upgraded to 5.1p1.
1101 For many years, FreeBSD's version of OpenSSH preferred DSA
1102 over RSA for host and user authentication keys. With this
1103 upgrade, we've switched to the vendor's default of RSA over
1104 DSA. This may cause upgraded clients to warn about unknown
1105 host keys even for previously known hosts. Users should
1106 follow the usual procedure for verifying host keys before
1107 accepting the RSA key.
1109 This can be circumvented by setting the "HostKeyAlgorithms"
1110 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1113 Please note that the sequence of keys offered for
1114 authentication has been changed as well. You may want to
1115 specify IdentityFile in a different order to revert this
1119 The sio(4) driver has been removed from the i386 and amd64
1120 kernel configuration files. This means uart(4) is now the
1121 default serial port driver on those platforms as well.
1123 To prevent collisions with the sio(4) driver, the uart(4) driver
1124 uses different names for its device nodes. This means the
1125 onboard serial port will now most likely be called "ttyu0"
1126 instead of "ttyd0". You may need to reconfigure applications to
1127 use the new device names.
1129 When using the serial port as a boot console, be sure to update
1130 /boot/device.hints and /etc/ttys before booting the new kernel.
1131 If you forget to do so, you can still manually specify the hints
1132 at the loader prompt:
1134 set hint.uart.0.at="isa"
1135 set hint.uart.0.port="0x3F8"
1136 set hint.uart.0.flags="0x10"
1137 set hint.uart.0.irq="4"
1141 The gpt(8) utility has been removed. Use gpart(8) to partition
1145 The version that Linuxulator emulates was changed from 2.4.2
1146 to 2.6.16. If you experience any problems with Linux binaries
1147 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1148 if it fixes the problem contact emulation mailing list.
1151 ISDN4BSD (I4B) was removed from the src tree. You may need to
1152 update a your kernel configuration and remove relevant entries.
1155 I have checked in code to support multiple routing tables.
1156 See the man pages setfib(1) and setfib(2).
1157 This is a hopefully backwards compatible version,
1158 but to make use of it you need to compile your kernel
1159 with options ROUTETABLES=2 (or more up to 16).
1162 The 802.11 wireless support was redone to enable multi-bss
1163 operation on devices that are capable. The underlying device
1164 is no longer used directly but instead wlanX devices are
1165 cloned with ifconfig. This requires changes to rc.conf files.
1166 For example, change:
1167 ifconfig_ath0="WPA DHCP"
1170 ifconfig_wlan0="WPA DHCP"
1171 see rc.conf(5) for more details. In addition, mergemaster of
1172 /etc/rc.d is highly recommended. Simultaneous update of userland
1173 and kernel wouldn't hurt either.
1175 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1176 modules were merged into the base wlan module. All references
1177 to these modules (e.g. in kernel config files) must be removed.
1180 psm(4) has gained write(2) support in native operation level.
1181 Arbitrary commands can be written to /dev/psm%d and status can
1182 be read back from it. Therefore, an application is responsible
1183 for status validation and error recovery. It is a no-op in
1184 other operation levels.
1187 Support for KSE threading has been removed from the kernel. To
1188 run legacy applications linked against KSE libmap.conf may
1189 be used. The following libmap.conf may be used to ensure
1190 compatibility with any prior release:
1192 libpthread.so.1 libthr.so.1
1193 libpthread.so.2 libthr.so.2
1194 libkse.so.3 libthr.so.3
1197 The layout of struct vmspace has changed. This affects libkvm
1198 and any executables that link against libkvm and use the
1199 kvm_getprocs() function. In particular, but not exclusively,
1200 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1201 The effects are minimal, but it's advisable to upgrade world
1205 The latest em driver no longer has support in it for the
1206 82575 adapter, this is now moved to the igb driver. The
1207 split was done to make new features that are incompatible
1208 with older hardware easier to do.
1211 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1212 likewise the kernel option is now GEOM_LINUX_LVM.
1215 The default NFS mount mode has changed from UDP to TCP for
1216 increased reliability. If you rely on (insecurely) NFS
1217 mounting across a firewall you may need to update your
1221 Belatedly note the addition of m_collapse for compacting
1225 The fts(3) structures have been changed to use adequate
1226 integer types for their members and so to be able to cope
1227 with huge file trees. The old fts(3) ABI is preserved
1228 through symbol versioning in libc, so third-party binaries
1229 using fts(3) should still work, although they will not take
1230 advantage of the extended types. At the same time, some
1231 third-party software might fail to build after this change
1232 due to unportable assumptions made in its source code about
1233 fts(3) structure members. Such software should be fixed
1234 by its vendor or, in the worst case, in the ports tree.
1235 FreeBSD_version 800015 marks this change for the unlikely
1236 case that a portable fix is impossible.
1239 To upgrade to -current after this date, you must be running
1240 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1241 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1244 The ADAPTIVE_GIANT kernel option has been retired because its
1245 functionality is the default now.
1248 The AT keyboard emulation of sunkbd(4) has been turned on
1249 by default. In order to make the special symbols of the Sun
1250 keyboards driven by sunkbd(4) work under X these now have
1251 to be configured the same way as Sun USB keyboards driven
1252 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1254 Option "XkbLayout" "us"
1255 Option "XkbRules" "xorg"
1256 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1259 It has been decided that it is desirable to provide ABI
1260 backwards compatibility to the FreeBSD 4/5/6 versions of the
1261 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1262 broken with the introduction of PCI domain support (see the
1263 20070930 entry). Unfortunately, this required the ABI of
1264 PCIOCGETCONF to be broken again in order to be able to
1265 provide backwards compatibility to the old version of that
1266 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1267 again. As for prominent ports this affects neither pciutils
1268 nor xorg-server this time, the hal port needs to be rebuilt
1272 The misnamed kthread_create() and friends have been renamed
1273 to kproc_create() etc. Many of the callers already
1274 used kproc_start()..
1275 I will return kthread_create() and friends in a while
1276 with implementations that actually create threads, not procs.
1277 Renaming corresponds with version 800002.
1286 Avoid using make -j when upgrading. While generally safe, there are
1287 sometimes problems using -j to upgrade. If your upgrade fails with
1288 -j, please try again without -j. From time to time in the past there
1289 have been problems using -j with buildworld and/or installworld. This
1290 is especially true when upgrading between "distant" versions (eg one
1291 that cross a major release boundary or several minor releases, or when
1292 several months have passed on the -current branch).
1294 Sometimes, obscure build problems are the result of environment
1295 poisoning. This can happen because the make utility reads its
1296 environment when searching for values for global variables. To run
1297 your build attempts in an "environmental clean room", prefix all make
1298 commands with 'env -i '. See the env(1) manual page for more details.
1300 When upgrading from one major version to another it is generally best
1301 to upgrade to the latest code in the currently installed branch first,
1302 then do an upgrade to the new branch. This is the best-tested upgrade
1303 path, and has the highest probability of being successful. Please try
1304 this approach before reporting problems with a major version upgrade.
1308 When upgrading the boot ZFS pool to a new version, always follow
1311 1.) recompile and reinstall the ZFS boot loader and boot block
1312 (this is part of "make buildworld" and "make installworld")
1314 2.) update the ZFS boot block on your boot drive
1316 The following example updates the ZFS boot block on the first
1317 partition (freebsd-boot) of a GPT partitioned drive ad0:
1318 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1320 Non-boot pools do not need these updates.
1324 If you are updating from a prior version of FreeBSD (even one just
1325 a few days old), you should follow this procedure. It is the most
1326 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1328 make kernel-toolchain
1329 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1330 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1332 To test a kernel once
1333 ---------------------
1334 If you just want to boot a kernel once (because you are not sure
1335 if it works, or if you want to boot a known bad kernel to provide
1336 debugging information) run
1337 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1338 nextboot -k testkernel
1340 To just build a kernel when you know that it won't mess you up
1341 --------------------------------------------------------------
1342 This assumes you are already running a CURRENT system. Replace
1343 ${arch} with the architecture of your machine (e.g. "i386",
1344 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1346 cd src/sys/${arch}/conf
1347 config KERNEL_NAME_HERE
1348 cd ../compile/KERNEL_NAME_HERE
1353 If this fails, go to the "To build a kernel" section.
1355 To rebuild everything and install it on the current system.
1356 -----------------------------------------------------------
1357 # Note: sometimes if you are running current you gotta do more than
1358 # is listed here if you are upgrading from a really old current.
1360 <make sure you have good level 0 dumps>
1362 make kernel KERNCONF=YOUR_KERNEL_HERE
1364 <reboot in single user> [3]
1372 To cross-install current onto a separate partition
1373 --------------------------------------------------
1374 # In this approach we use a separate partition to hold
1375 # current's root, 'usr', and 'var' directories. A partition
1376 # holding "/", "/usr" and "/var" should be about 2GB in
1379 <make sure you have good level 0 dumps>
1382 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1383 <maybe newfs current's root partition>
1384 <mount current's root partition on directory ${CURRENT_ROOT}>
1385 make installworld DESTDIR=${CURRENT_ROOT}
1386 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1387 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1388 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1389 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1390 <reboot into current>
1391 <do a "native" rebuild/install as described in the previous section>
1392 <maybe install compatibility libraries from ports/misc/compat*>
1396 To upgrade in-place from 8.x-stable to current
1397 ----------------------------------------------
1398 <make sure you have good level 0 dumps>
1400 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1402 <reboot in single user> [3]
1409 Make sure that you've read the UPDATING file to understand the
1410 tweaks to various things you need. At this point in the life
1411 cycle of current, things change often and you are on your own
1412 to cope. The defaults can also change, so please read ALL of
1413 the UPDATING entries.
1415 Also, if you are tracking -current, you must be subscribed to
1416 freebsd-current@freebsd.org. Make sure that before you update
1417 your sources that you have read and understood all the recent
1418 messages there. If in doubt, please track -stable which has
1419 much fewer pitfalls.
1421 [1] If you have third party modules, such as vmware, you
1422 should disable them at this point so they don't crash your
1425 [3] From the bootblocks, boot -s, and then do
1430 adjkerntz -i # if CMOS is wall time
1431 Also, when doing a major release upgrade, it is required that
1432 you boot into single user mode to do the installworld.
1434 [4] Note: This step is non-optional. Failure to do this step
1435 can result in a significant reduction in the functionality of the
1436 system. Attempting to do it by hand is not recommended and those
1437 that pursue this avenue should read this file carefully, as well
1438 as the archives of freebsd-current and freebsd-hackers mailing lists
1439 for potential gotchas. The -U option is also useful to consider.
1440 See mergemaster(8) for more information.
1442 [5] Usually this step is a noop. However, from time to time
1443 you may need to do this if you get unknown user in the following
1444 step. It never hurts to do it all the time. You may need to
1445 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1446 install) after the buildworld before this step if you last updated
1447 from current before 20020224 or from -stable before 20020408.
1449 [6] This only deletes old files and directories. Old libraries
1450 can be deleted by "make delete-old-libs", but you have to make
1451 sure that no program is using those libraries anymore.
1453 [8] In order to have a kernel that can run the 4.x binaries needed to
1454 do an installworld, you must include the COMPAT_FREEBSD4 option in
1455 your kernel. Failure to do so may leave you with a system that is
1456 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1457 required to run the 5.x binaries on more recent kernels. And so on
1458 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1460 Make sure that you merge any new devices from GENERIC since the
1461 last time you updated your kernel config file.
1463 [9] When checking out sources, you must include the -P flag to have
1464 cvs prune empty directories.
1466 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1467 "?=" instead of the "=" assignment operator, so that buildworld can
1468 override the CPUTYPE if it needs to.
1470 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1471 not on the command line, or in /etc/make.conf. buildworld will
1472 warn if it is improperly defined.
1475 This file contains a list, in reverse chronological order, of major
1476 breakages in tracking -current. It is not guaranteed to be a complete
1477 list of such breakages, and only contains entries since October 10, 2007.
1478 If you need to see UPDATING entries from before that date, you will need
1479 to fetch an UPDATING file from an older FreeBSD release.
1481 Copyright information:
1483 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1485 Redistribution, publication, translation and use, with or without
1486 modification, in full or in part, in any form or format of this
1487 document are permitted without further permission from the author.
1489 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1490 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1491 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1492 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1493 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1494 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1495 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1496 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1497 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1498 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1499 POSSIBILITY OF SUCH DAMAGE.
1501 Contact Warner Losh if you have any questions about your use of