1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
12 NOTE TO PEOPLE WHO THINK THAT FreeBSD 9.x IS SLOW:
13 FreeBSD 9.x has many debugging features turned on, in both the kernel
14 and userland. These features attempt to detect incorrect use of
15 system primitives, and encourage loud failure through extra sanity
16 checking and fail stop semantics. They also substantially impact
17 system performance. If you want to do performance measurement,
18 benchmarking, and optimization, you'll want to turn them off. This
19 includes various WITNESS- related kernel options, INVARIANTS, malloc
20 debugging flags in userland, and various verbose features in the
21 kernel. Many developers choose to disable these features on build
22 machines to maximize performance. (To disable malloc debugging, run
23 ln -s aj /etc/malloc.conf.)
26 The packet filter (pf) code has been updated to OpenBSD 4.5.
27 You need to update userland tools to be in sync with kernel.
28 This update breaks backward compatibility with earlier pfsync(4)
29 versions. Care must be taken when updating redundant firewall setups.
32 The following sysctls and tunables are retired on x86 platforms:
34 machdep.hlt_logical_cpus
35 The following sysctl is retired:
36 machdep.hyperthreading_allowed
37 The sysctls were supposed to provide a way to dynamically offline and
38 online selected CPUs on x86 platforms, but the implementation has not
39 been reliable especially with SCHED_ULE scheduler.
40 machdep.hyperthreading_allowed tunable is still available to ignore
41 hyperthreading CPUs at OS level.
42 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
43 where X is an APIC ID of a CPU. Be advised, though, that disabling
44 CPUs in non-uniform fashion will result in non-uniform topology and
45 may lead to sub-optimal system performance with SCHED_ULE, which is
49 cpumask_t type is retired and cpuset_t is used in order to describe
53 Changes to ifconfig(8) for dynamic address family detection mandate
54 that you are running a kernel of 20110525 or later. Make sure to
55 follow the update procedure to boot a new kernel before installing
59 Support for sun4v architecture is officially dropped
62 Several KPI breaking changes have been committed to the mii(4) layer,
63 the PHY drivers and consequently some Ethernet drivers using mii(4).
64 This means that miibus.ko and the modules of the affected Ethernet
65 drivers need to be recompiled.
67 Note to kernel developers: Given that the OUI bit reversion problem
68 was fixed as part of these changes all mii(4) commits related to OUIs,
69 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
70 handling, no longer can be merged verbatim to stable/8 and previous
74 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
75 into their kernel configurations along with 'device pci'.
78 The default NFS client is now the new NFS client, so fstype "newnfs"
79 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
80 Although mounts via fstype "nfs" will usually work without userland
81 changes, it is recommended that the mount(8) and mount_nfs(8)
82 commands be rebuilt from sources and that a link to mount_nfs called
83 mount_oldnfs be created. The new client is compiled into the
84 kernel with "options NFSCL" and this is needed for diskless root
85 file systems. The GENERIC kernel configs have been changed to use
86 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
87 To use the regular/old client, you can "mount -t oldnfs ...". For
88 a diskless root file system, you must also include a line like:
90 vfs.root.mountfrom="oldnfs:"
92 in the boot/loader.conf on the root fs on the NFS server to make
93 a diskless root fs use the old client.
96 The GENERIC kernels for all architectures now default to the new
97 CAM-based ATA stack. It means that all legacy ATA drivers were
98 removed and replaced by respective CAM drivers. If you are using
99 ATA device names in /etc/fstab or other places, make sure to update
100 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
101 where 'Y's are the sequential numbers starting from zero for each type
102 in order of detection, unless configured otherwise with tunables,
103 see cam(4)). There will be symbolic links created in /dev/ to map
104 old adX devices to the respective adaY. They should provide basic
105 compatibility for file systems mounting in most cases, but they do
106 not support old user-level APIs and do not have respective providers
107 in GEOM. Consider using updated management tools with new device names.
109 It is possible to load devices ahci, ata, siis and mvs as modules,
110 but option ATA_CAM should remain in kernel configuration to make ata
111 module work as CAM driver supporting legacy ATA controllers. Device ata
112 still can be used in modular fashion (atacore + ...). Modules atadisk
113 and atapi* are not used and won't affect operation in ATA_CAM mode.
114 Note that to use CAM-based ATA kernel should include CAM devices
115 scbus, pass, da (or explicitly ada), cd and optionally others. All of
116 them are parts of the cam module.
118 ataraid(4) functionality is now supported by the RAID GEOM class.
119 To use it you can load geom_raid kernel module and use graid(8) tool
120 for management. Instead of /dev/arX device names, use /dev/raid/rX.
122 No kernel config options or code have been removed, so if a problem
123 arises, please report it and optionally revert to the old ATA stack.
124 In order to do it you can remove from the kernel config:
129 , and instead add back:
130 device atadisk # ATA disk drives
131 device ataraid # ATA RAID drives
132 device atapicd # ATAPI CDROM drives
133 device atapifd # ATAPI floppy drives
134 device atapist # ATAPI tape drives
137 The default NFS server has been changed to the new server, which
138 was referred to as the experimental server. If you need to switch
139 back to the old NFS server, you must now put the "-o" option on
140 both the mountd and nfsd commands. This can be done using the
141 mountd_flags and nfs_server_flags rc.conf variables until an
142 update to the rc scripts is committed, which is coming soon.
145 The GNU Objective-C runtime library (libobjc), and other Objective-C
146 related components have been removed from the base system. If you
147 require an Objective-C library, please use one of the available ports.
150 ath(4) has been split into bus- and device- modules. if_ath contains
151 the HAL, the TX rate control and the network device code. if_ath_pci
152 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
153 contains the AHB glue. Users need to load both if_ath_pci and if_ath
154 in order to use ath on everything else.
156 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
157 need to load if_ath and if_ath_pci for ath(4) operation.
160 As part of the replacement of sysinstall, the process of building
161 release media has changed significantly. For details, please re-read
162 release(7), which has been updated to reflect the new build process.
165 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
166 is the last available version under GPLv2. It brings a number of new
167 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
168 4.1 and SSE 4.2), better support for powerpc64, a number of new
169 directives, and lots of other small improvements. See the ChangeLog
170 file in contrib/binutils for the full details.
173 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
174 compliant, and will now use half of hash for authentication.
175 This will break interoperability with all stacks (including all
176 actual FreeBSD versions) who implement
177 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
179 The only workaround with such peers is to use another HMAC
180 algorithm for IPsec ("phase 2") authentication.
183 Remove the uio_yield prototype and symbol. This function has
184 been misnamed since it was introduced and should not be
185 globally exposed with this name. The equivalent functionality
186 is now available using kern_yield(curthread->td_user_pri).
187 The function remains undocumented.
190 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
191 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
192 for scalar sysctls is defined but disabled. Code that needs
193 UQUAD to pass the type checking that must compile on older
194 systems where the define is not present can check against
195 __FreeBSD_version >= 900030.
197 The system dialog(1) has been replaced with a new version previously
198 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
199 with the previous version, but the libdialog associated with it has
200 a largely incompatible API. As such, the original version of libdialog
201 will be kept temporarily as libodialog, until its base system consumers
202 are replaced or updated. Bump __FreeBSD_version to 900030.
205 If you are trying to run make universe on a -stable system, and you get
206 the following warning:
207 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
208 unknown. config(8) likely too old."
209 or something similar to it, then you must upgrade your -stable system
210 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
211 stable/8) or build the config from the latest stable/8 branch and
212 install it on your system.
214 Prior to this date, building a current universe on 8-stable system from
215 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
216 error in the first kernel build phase. A new config on those old
217 systems will fix that problem for older versions of -current.
220 The TCP stack has been modified to allow Khelp modules to interact with
221 it via helper hook points and store per-connection data in the TCP
222 control block. Bump __FreeBSD_version to 900029. User space tools that
223 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
227 Generic IEEE 802.3 annex 31B full duplex flow control support has been
228 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
229 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
230 to take advantage of it instead of using custom implementations. This
231 means that these drivers now no longer unconditionally advertise
232 support for flow control but only do so if flow control is a selected
233 media option. This was implemented in the generic support that way in
234 order to allow flow control to be switched on and off via ifconfig(8)
235 with the PHY specific default to typically off in order to protect
236 from unwanted effects. Consequently, if you used flow control with
237 one of the above mentioned drivers you now need to explicitly enable
239 ifconfig bge0 media auto mediaopt flowcontrol
241 Along with the above mentioned changes generic support for setting
242 1000baseT master mode also has been added and brgphy(4), ciphy(4),
243 e1000phy(4) as well as ip1000phy(4) have been converted to take
244 advantage of it. This means that these drivers now no longer take the
245 link0 parameter for selecting master mode but the master media option
246 has to be used instead, for example like in the following:
247 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
249 Selection of master mode now is also available with all other PHY
250 drivers supporting 1000baseT.
253 The TCP stack has received a significant update to add support for
254 modularised congestion control and generally improve the clarity of
255 congestion control decisions. Bump __FreeBSD_version to 900025. User
256 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
257 sockstat) need to be recompiled.
260 The man(1) utility has been replaced by a new version that no longer
261 uses /etc/manpath.config. Please consult man.conf(5) for how to
262 migrate local entries to the new format.
265 The copyright strings printed by login(1) and sshd(8) at the time of a
266 new connection have been removed to follow other operating systems and
270 A workaround for a fixed ld bug has been removed in kernel code,
271 so make sure that your system ld is built from sources after
272 revision 210245 from 2010-07-19 (r211583 if building head kernel
273 on stable/8, r211584 for stable/7; both from 2010-08-21).
274 A symptom of incorrect ld version is different addresses for
275 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
278 The $ipv6_prefer variable in rc.conf(5) has been split into
279 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
281 The $ip6addrctl_policy is a variable to choose a pre-defined
282 address selection policy set by ip6addrctl(8). A value
283 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
286 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
287 flag (see an entry of 20090926) is set on an interface with no
288 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
289 security reason. If you want IPv6 link-local address on all
290 interfaces by default, set this to "YES".
292 The old ipv6_prefer="YES" is equivalent to
293 ipv6_activate_all_interfaces="YES" and
294 ip6addrctl_policy="ipv6_prefer".
297 DTrace has grown support for userland tracing. Due to this, DTrace is
298 now i386 and amd64 only.
299 dtruss(1) is now installed by default on those systems and a new
300 kernel module is needed for userland tracing: fasttrap.
301 No changes to your kernel config file are necessary to enable
302 userland tracing, but you might consider adding 'STRIP=' and
303 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
304 to have informative userland stack traces in DTrace (ustack).
307 The acpi_aiboost(4) driver has been removed in favor of the new
308 aibs(4) driver. You should update your kernel configuration file.
311 BSD grep has been imported to the base system and it is built by
312 default. It is completely BSD licensed, highly GNU-compatible, uses
313 less memory than its GNU counterpart and has a small codebase.
314 However, it is slower than its GNU counterpart, which is mostly
315 noticeable for larger searches, for smaller ones it is measurable
316 but not significant. The reason is complex, the most important factor
317 is that we lack a modern and efficient regex library and GNU
318 overcomes this by optimizing the searches internally. Future work
319 on improving the regex performance is planned, for the meantime,
320 users that need better performance, can build GNU grep instead by
321 setting the WITH_GNU_GREP knob.
324 Due to the import of powerpc64 support, all existing powerpc kernel
325 configuration files must be updated with a machine directive like this:
326 machine powerpc powerpc
328 In addition, an updated config(8) is required to build powerpc kernels
332 A new version of ZFS (version 15) has been merged to -HEAD.
333 This version uses a python library for the following subcommands:
334 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
335 For full functionality of these commands the following port must
336 be installed: sysutils/py-zfs
339 'vm_page's are now hashed by physical address to an array of mutexes.
340 Currently this is only used to serialize access to hold_count. Over
341 time the page queue mutex will be peeled away. This changes the size
342 of pmap on every architecture. And requires all callers of vm_page_hold
343 and vm_page_unhold to be updated.
346 WITH_CTF can now be specified in src.conf (not recommended, there
347 are some problems with static executables), make.conf (would also
348 affect ports which do not use GNU make and do not override the
349 compile targets) or in the kernel config (via "makeoptions
351 When WITH_CTF was specified there before this was silently ignored,
352 so make sure that WITH_CTF is not used in places which could lead
353 to unwanted behavior.
356 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
357 to allow 32-bit compatibility on non-x86 platforms. All kernel
358 configurations on amd64 and ia64 platforms using these options must
359 be modified accordingly.
362 The utmp user accounting database has been replaced with utmpx,
363 the user accounting interface standardized by POSIX.
364 Unfortunately the semantics of utmp and utmpx don't match,
365 making it practically impossible to support both interfaces.
366 The user accounting database is used by tools like finger(1),
367 last(1), talk(1), w(1) and ac(8).
369 All applications in the base system use utmpx. This means only
370 local binaries (e.g. from the ports tree) may still use these
371 utmp database files. These applications must be rebuilt to make
374 After the system has been upgraded, it is safe to remove the old
375 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
376 assuming their contents is of no importance anymore. Old wtmp
377 databases can only be used by last(1) and ac(8) after they have
378 been converted to the new format using wtmpcvt(1).
381 Introduce the kernel thread "deadlock resolver" (which can be enabled
382 via the DEADLKRES option, see NOTES for more details) and the
383 sleepq_type() function for sleepqueues.
386 The rc.firewall and rc.firewall6 were unified, and
387 rc.firewall6 and rc.d/ip6fw were removed.
388 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
389 variables are obsoleted. Instead, the following new rc
390 variables are added to rc.d/ipfw:
392 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
393 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
394 firewall_simple_onet_ipv6, firewall_trusted_ipv6
396 The meanings correspond to the relevant IPv4 variables.
402 The default terminal emulation for syscons(4) has been changed
403 from cons25 to xterm on all platforms except pc98. This means
404 that the /etc/ttys file needs to be updated to ensure correct
405 operation of applications on the console.
407 The terminal emulation style can be toggled per window by using
408 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
409 options can be used to change the compile-time default back to
412 To prevent graphical artifacts, make sure the TERM environment
413 variable is set to match the terminal emulation that is being
414 performed by syscons(4).
417 The layout of the structure ieee80211req_scan_result has changed.
418 Applications that require wireless scan results (e.g. ifconfig(8))
419 from net80211 need to be recompiled.
421 Applications such as wpa_supplicant(8) may require a full world
422 build without using NO_CLEAN in order to get synchronized with the
426 The iwn(4) driver has been updated to support the 5000 and 5150 series.
427 There's one kernel module for each firmware. Adding "device iwnfw"
428 to the kernel configuration file means including all three firmware
429 images inside the kernel. If you want to include just the one for
430 your wireless card, use the the devices iwn4965fw, iwn5000fw or
434 The rc.d/network_ipv6, IPv6 configuration script has been integrated
435 into rc.d/netif. The changes are the following:
437 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
438 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
439 Note that both variables need the "inet6" keyword at the head.
441 Do not set $ipv6_network_interfaces manually if you do not
442 understand what you are doing. It is not needed in most cases.
444 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
447 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
448 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
450 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
451 all configured interfaces have "inet6 accept_rtadv" in the
452 $ifconfig_IF_ipv6. These are for backward compatibility.
454 3. A new variable $ipv6_prefer has been added. If NO, IPv6
455 functionality of interfaces with no corresponding
456 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
457 and the default address selection policy of ip6addrctl(8)
458 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
459 Note that if you want to configure IPv6 functionality on the
460 disabled interfaces after boot, first you need to clear the flag by
461 using ifconfig(8) like:
463 ifconfig em0 inet6 -ifdisabled
465 If YES, the default address selection policy is set as
468 The default value of $ipv6_prefer is NO.
470 4. If your system need to receive Router Advertisement messages,
471 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
472 scripts automatically invoke rtsol(8) when the interface becomes
473 UP. The Router Advertisement messages are used for SLAAC
474 (State-Less Address AutoConfiguration).
477 802.11s D3.03 support was committed. This is incompatible with the
478 previous code, which was based on D3.0.
481 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
482 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
483 control whether accepting Router Advertisement messages or not.
484 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
485 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
486 The ifconfig(8) utility now supports these flags.
489 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
490 mount(8) and -a option for df(1) to see them.
493 The old tunable hw.bus.devctl_disable has been superseded by
494 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
495 replaced by hw.bus.devctl_queue=0. The default for this new tunable
499 Remove the option STOP_NMI. The default action is now to use NMI only
500 for KDB via the newly introduced function stop_cpus_hard() and
501 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
504 The stable/8 branch created in subversion. This corresponds to the
505 RELENG_8 branch in CVS.
508 Bump the shared library version numbers for all libraries that do not
509 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
510 __FreeBSD_version to 800105.
513 Due to changes in the implementation of virtual network stack support,
514 all network-related kernel modules must be recompiled. As this change
515 breaks the ABI, bump __FreeBSD_version to 800104.
518 The TOE interface to the TCP syncache has been modified to remove
519 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
520 The cxgb driver is the only TOE consumer affected by this change, and
521 needs to be recompiled along with the kernel. As this change breaks
522 the ABI, bump __FreeBSD_version to 800103.
525 Padding has been added to struct tcpcb, sackhint and tcpstat in
526 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
527 maintaining the ABI. However, this change breaks the ABI, so bump
528 __FreeBSD_version to 800102. User space tools that rely on the size of
529 any of these structs (e.g. sockstat) need to be recompiled.
532 The NFS_LEGACYRPC option has been removed along with the old kernel
533 RPC implementation that this option selected. Kernel configurations
534 may need to be adjusted.
537 The network interface device nodes at /dev/net/<interface> have been
538 removed. All ioctl operations can be performed the normal way using
539 routing sockets. The kqueue functionality can generally be replaced
540 with routing sockets.
543 The documentation from the FreeBSD Documentation Project (Handbook,
544 FAQ, etc.) is now installed via packages by sysinstall(8) and under
545 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
548 The ABI of various structures related to the SYSV IPC API have been
549 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
550 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
554 Layout of struct vnet has changed as routing related variables were
555 moved to their own Vimage module. Modules need to be recompiled. Bump
556 __FreeBSD_version to 800099.
559 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
560 respectively. As long as no more than 16 groups per process are used,
561 no changes should be visible. When more than 16 groups are used, old
562 binaries may fail if they call getgroups() or getgrouplist() with
563 statically sized storage. Recompiling will work around this, but
564 applications should be modified to use dynamically allocated storage
565 for group arrays as POSIX.1-2008 does not cap an implementation's
566 number of supported groups at NGROUPS_MAX+1 as previous versions did.
568 NFS and portalfs mounts may also be affected as the list of groups is
569 truncated to 16. Users of NFS who use more than 16 groups, should
570 take care that negative group permissions are not used on the exported
571 file systems as they will not be reliable unless a GSSAPI based
572 authentication method is used.
575 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
576 option compiles in the support for adaptive spinning for lockmgrs
577 which want to enable it. The lockinit() function now accepts the flag
578 LK_ADAPTIVE in order to make the lock object subject to adaptive
579 spinning when both held in write and read mode.
582 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
583 changed. User applications that use this ioctl need to be rebuilt.
586 The layout of struct thread has changed. Kernel and modules need to
590 The layout of structs ifnet, domain, protosw and vnet_net has changed.
591 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
594 window(1) has been removed from the base system. It can now be
595 installed from ports. The port is called misc/window.
598 The way we are storing and accessing `routing table' entries has
599 changed. Programs reading the FIB, like netstat, need to be
603 A new netisr implementation has been added for FreeBSD 8. Network
604 file system modules, such as igmp, ipdivert, and others, should be
606 Bump __FreeBSD_version to 800096.
609 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
613 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
614 Bump __FreeBSD_version to 800094.
617 Add mnt_xflag field to 'struct mount'. File system modules need to be
619 Bump __FreeBSD_version to 800093.
622 The compiling option ADAPTIVE_SX has been retired while it has been
623 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
624 The KPI for sx_init_flags() changes as accepting flags:
625 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
626 been introduced in order to handle the reversed logic.
627 Bump __FreeBSD_version to 800092.
630 Add support for hierarchical jails. Remove global securelevel.
631 Bump __FreeBSD_version to 800091.
634 The layout of struct vnet_net has changed, therefore modules
636 Bump __FreeBSD_version to 800090.
639 The newly imported zic(8) produces a new format in the output. Please
640 run tzsetup(8) to install the newly created data to /etc/localtime.
643 The sysctl tree for the usb stack has renamed from hw.usb2.* to
644 hw.usb.* and is now consistent again with previous releases.
647 802.11 monitor mode support was revised and driver api's were changed.
648 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
649 of DLT_IEEE802_11. No user-visible data structures were changed but
650 applications that use DLT_IEEE802_11 may require changes.
651 Bump __FreeBSD_version to 800088.
654 The layout of the following structs has changed: sysctl_oid,
655 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
656 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
657 panics may be experienced. World rebuild is required for
658 correctly checking networking state from userland.
659 Bump __FreeBSD_version to 800085.
662 MLDv2 and Source-Specific Multicast (SSM) have been merged
663 to the IPv6 stack. VIMAGE hooks are in but not yet used.
664 The implementation of SSM within FreeBSD's IPv6 stack closely
665 follows the IPv4 implementation.
667 For kernel developers:
669 * The most important changes are that the ip6_output() and
670 ip6_input() paths no longer take the IN6_MULTI_LOCK,
671 and this lock has been downgraded to a non-recursive mutex.
673 * As with the changes to the IPv4 stack to support SSM, filtering
674 of inbound multicast traffic must now be performed by transport
675 protocols within the IPv6 stack. This does not apply to TCP and
676 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
678 * The KPIs used by IPv6 multicast are similar to those used by
679 the IPv4 stack, with the following differences:
680 * im6o_mc_filter() is analogous to imo_multicast_filter().
681 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
682 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
683 * IN6_LOOKUP_MULTI() has been deprecated and removed.
684 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
685 for MLDv1 have an additional 'timer' argument which is used to
686 jitter the initial membership report for the solicited-node
687 multicast membership on-link.
688 * This is not strictly needed for MLDv2, which already jitters
689 its report transmissions. However, the 'timer' argument is
690 preserved in case MLDv1 is active on the interface.
692 * The KAME linked-list based IPv6 membership implementation has
693 been refactored to use a vector similar to that used by the IPv4
695 Code which maintains a list of its own multicast memberships
696 internally, e.g. carp, has been updated to reflect the new
699 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
700 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
701 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
702 implementation constraint which needs to be addressed in HEAD.
704 For application developers:
706 * The changes are broadly similar to those made for the IPv4
709 * The use of IPv4 and IPv6 multicast socket options on the same
710 socket, using mapped addresses, HAS NOT been tested or supported.
712 * There are a number of issues with the implementation of various
713 IPv6 multicast APIs which need to be resolved in the API surface
714 before the implementation is fully compatible with KAME userland
715 use, and these are mostly to do with interface index treatment.
717 * The literature available discusses the use of either the delta / ASM
718 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
719 using setsourcefilter(3)/getsourcefilter(3). For more information
720 please refer to RFC 3768, 'Socket Interface Extensions for
721 Multicast Source Filters'.
723 * Applications which use the published RFC 3678 APIs should be fine.
725 For systems administrators:
727 * The mtest(8) utility has been refactored to support IPv6, in
728 addition to IPv4. Interface addresses are no longer accepted
729 as arguments, their names must be used instead. The utility
730 will map the interface name to its first IPv4 address as
731 returned by getifaddrs(3).
733 * The ifmcstat(8) utility has also been updated to print the MLDv2
734 endpoint state and source filter lists via sysctl(3).
736 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
737 loopback of IPv6 multicast datagrams by default; it defaults to 1
738 to preserve the existing behaviour. Disabling multicast loopback is
739 recommended for optimal system performance.
741 * The IPv6 MROUTING code has been changed to examine this sysctl
742 instead of attempting to perform a group lookup before looping
743 back forwarded datagrams.
745 Bump __FreeBSD_version to 800084.
748 Implement low-level Bluetooth HCI API.
749 Bump __FreeBSD_version to 800083.
752 The layout of struct malloc_type, used by modules to register new
753 memory allocation types, has changed. Most modules will need to
754 be rebuilt or panics may be experienced.
755 Bump __FreeBSD_version to 800081.
758 Anticipate overflowing inp_flags - add inp_flags2.
759 This changes most offsets in inpcb, so checking v4 connection
760 state will require a world rebuild.
761 Bump __FreeBSD_version to 800080.
764 Add an llentry to struct route and struct route_in6. Modules
765 embedding a struct route will need to be recompiled.
766 Bump __FreeBSD_version to 800079.
769 The size of rt_metrics_lite and by extension rtentry has changed.
770 Networking administration apps will need to be recompiled.
771 The route command now supports show as an alias for get, weighting
772 of routes, sticky and nostick flags to alter the behavior of stateful
774 Bump __FreeBSD_version to 800078.
777 Do not use Giant for kbdmux(4) locking. This is wrong and
778 apparently causing more problems than it solves. This will
779 re-open the issue where interrupt handlers may race with
780 kbdmux(4) in polling mode. Typical symptoms include (but
781 not limited to) duplicated and/or missing characters when
782 low level console functions (such as gets) are used while
783 interrupts are enabled (for example geli password prompt,
784 mountroot prompt etc.). Disabling kbdmux(4) may help.
787 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
788 kernel modules referencing any of the above need to be recompiled.
789 Bump __FreeBSD_version to 800075.
792 GEOM_PART has become the default partition slicer for storage devices,
793 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
794 introduces some changes:
796 MSDOS/EBR: the devices created from MSDOS extended partition entries
797 (EBR) can be named differently than with GEOM_MBR and are now symlinks
798 to devices with offset-based names. fstabs may need to be modified.
800 BSD: the "geometry does not match label" warning is harmless in most
801 cases but it points to problems in file system misalignment with
802 disk geometry. The "c" partition is now implicit, covers the whole
803 top-level drive and cannot be (mis)used by users.
805 General: Kernel dumps are now not allowed to be written to devices
806 whose partition types indicate they are meant to be used for file
807 systems (or, in case of MSDOS partitions, as something else than
810 Most of these changes date approximately from 200812.
813 The uscanner(4) driver has been removed from the kernel. This follows
814 Linux removing theirs in 2.6 and making libusb the default interface
818 The multicast forwarding code has been cleaned up. netstat(1)
819 only relies on KVM now for printing bandwidth upcall meters.
820 The IPv4 and IPv6 modules are split into ip_mroute_mod and
821 ip6_mroute_mod respectively. The config(5) options for statically
822 compiling this code remain the same, i.e. 'options MROUTING'.
825 Support for the IFF_NEEDSGIANT network interface flag has been
826 removed, which means that non-MPSAFE network device drivers are no
827 longer supported. In particular, if_ar, if_sr, and network device
828 drivers from the old (legacy) USB stack can no longer be built or
832 POSIX.1 Native Language Support (NLS) has been enabled in libc and
833 a bunch of new language catalog files have also been added.
834 This means that some common libc messages are now localized and
835 they depend on the LC_MESSAGES environmental variable.
838 The k8temp(4) driver has been renamed to amdtemp(4) since
839 support for Family 10 and Family 11 CPU families was added.
842 IGMPv3 and Source-Specific Multicast (SSM) have been merged
843 to the IPv4 stack. VIMAGE hooks are in but not yet used.
845 For kernel developers, the most important changes are that the
846 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
847 and this lock has been downgraded to a non-recursive mutex.
849 Transport protocols (UDP, Raw IP) are now responsible for filtering
850 inbound multicast traffic according to group membership and source
851 filters. The imo_multicast_filter() KPI exists for this purpose.
852 Transports which do not use multicast (SCTP, TCP) already reject
853 multicast by default. Forwarding and receive performance may improve
854 as a mutex acquisition is no longer needed in the ip_input()
855 low-level input path. in_addmulti() and in_delmulti() are shimmed
856 to new KPIs which exist to support SSM in-kernel.
858 For application developers, it is recommended that loopback of
859 multicast datagrams be disabled for best performance, as this
860 will still cause the lock to be taken for each looped-back
861 datagram transmission. The net.inet.ip.mcast.loop sysctl may
862 be tuned to 0 to disable loopback by default; it defaults to 1
863 to preserve the existing behaviour.
865 For systems administrators, to obtain best performance with
866 multicast reception and multiple groups, it is always recommended
867 that a card with a suitably precise hash filter is used. Hash
868 collisions will still result in the lock being taken within the
869 transport protocol input path to check group membership.
871 If deploying FreeBSD in an environment with IGMP snooping switches,
872 it is recommended that the net.inet.igmp.sendlocal sysctl remain
873 enabled; this forces 224.0.0.0/24 group membership to be announced
876 The size of 'struct igmpstat' has changed; netstat needs to be
877 recompiled to reflect this.
878 Bump __FreeBSD_version to 800070.
881 libusb20.so.1 is now installed as libusb.so.1 and the ports system
882 updated to use it. This requires a buildworld/installworld in order to
883 update the library and dependencies (usbconfig, etc). Its advisable to
884 rebuild all ports which uses libusb. More specific directions are given
885 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
886 libusb are no longer required and can be removed.
889 A workaround is committed to allow the creation of System V shared
890 memory segment of size > 2 GB on the 64-bit architectures.
891 Due to a limitation of the existing ABI, the shm_segsz member
892 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
893 wrong for large segments. Note that limits must be explicitly
894 raised to allow such segments to be created.
897 The layout of struct ifnet has changed, requiring a rebuild of all
898 network device driver modules.
901 The /dev handling for the new USB stack has changed, a
902 buildworld/installworld is required for libusb20.
905 The new USB2 stack has now been permanently moved in and all kernel and
906 module names reverted to their previous values (eg, usb, ehci, ohci,
907 ums, ...). The old usb stack can be compiled in by prefixing the name
908 with the letter 'o', the old usb modules have been removed.
909 Updating entry 20090216 for xorg and 20090215 for libmap may still
913 The rc.conf(5) option if_up_delay has been renamed to
914 defaultroute_delay to better reflect its purpose. If you have
915 customized this setting in /etc/rc.conf you need to update it to
919 xorg 7.4 wants to configure its input devices via hald which does not
920 yet work with USB2. If the keyboard/mouse does not work in xorg then
922 Option "AllowEmptyInput" "off"
923 to your ServerLayout section. This will cause X to use the configured
924 kbd and mouse sections from your xorg.conf.
927 The GENERIC kernels for all architectures now default to the new USB2
928 stack. No kernel config options or code have been removed so if a
929 problem arises please report it and optionally revert to the old USB
930 stack. If you are loading USB kernel modules or have a custom kernel
931 that includes GENERIC then ensure that usb names are also changed over,
932 eg uftdi -> usb2_serial_ftdi.
934 Older programs linked against the ports libusb 0.1 need to be
935 redirected to the new stack's libusb20. /etc/libmap.conf can
937 # Map old usb library to new one for usb2 stack
938 libusb-0.1.so.8 libusb20.so.1
941 All USB ethernet devices now attach as interfaces under the name ueN
942 (eg. ue0). This is to provide a predictable name as vendors often
943 change usb chipsets in a product without notice.
946 The ichsmb(4) driver has been changed to require SMBus slave
947 addresses be left-justified (xxxxxxx0b) rather than right-justified.
948 All of the other SMBus controller drivers require left-justified
949 slave addresses, so this change makes all the drivers provide the
953 INET6 statistics (struct ip6stat) was updated.
954 netstat(1) needs to be recompiled.
957 NTFS has been removed from GENERIC kernel on amd64 to match
958 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
959 will load ntfs.ko module automatically when NTFS support is
960 actually needed, unless ntfs.ko is not installed or security
961 level prohibits loading kernel modules. If either is the case,
962 "options NTFS" has to be added into kernel config.
965 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
966 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
967 800061. User space tools that rely on the size of struct tcpcb in
968 tcp_var.h (e.g. sockstat) need to be recompiled.
971 ng_tty(4) module updated to match the new TTY subsystem.
972 Due to API change, user-level applications must be updated.
973 New API support added to mpd5 CVS and expected to be present
974 in next mpd5.3 release.
977 With __FreeBSD_version 800060 the makefs tool is part of
978 the base system (it was a port).
981 The afdata and ifnet locks have been changed from mutexes to
982 rwlocks, network modules will need to be re-compiled.
985 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
986 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
987 The new code reduced struct rtentry{} by 16 bytes on 32-bit
988 architecture and 40 bytes on 64-bit architecture. The userland
989 applications "arp" and "ndp" have been updated accordingly.
990 The output from "netstat -r" shows only routing entries and
991 none of the L2 information.
994 __FreeBSD_version 800057 marks the switchover from the
995 binary ath hal to source code. Users must add the line:
997 options AH_SUPPORT_AR5416
999 to their kernel config files when specifying:
1003 The ath_hal module no longer exists; the code is now compiled
1004 together with the driver in the ath module. It is now
1005 possible to tailor chip support (i.e. reduce the set of chips
1006 and thereby the code size); consult ath_hal(4) for details.
1009 __FreeBSD_version 800054 adds memory barriers to
1010 <machine/atomic.h>, new interfaces to ifnet to facilitate
1011 multiple hardware transmit queues for cards that support
1012 them, and a lock-less ring-buffer implementation to
1013 enable drivers to more efficiently manage queueing of
1017 A new version of ZFS (version 13) has been merged to -HEAD.
1018 This version has zpool attribute "listsnapshots" off by
1019 default, which means "zfs list" does not show snapshots,
1020 and is the same as Solaris behavior.
1023 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1026 The uhci, ohci, ehci and slhci USB Host controller drivers have
1027 been put into separate modules. If you load the usb module
1028 separately through loader.conf you will need to load the
1029 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1030 controller add the following to loader.conf:
1036 The ABI used by the PMC toolset has changed. Please keep
1037 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1041 atapci kernel module now includes only generic PCI ATA
1042 driver. AHCI driver moved to ataahci kernel module.
1043 All vendor-specific code moved into separate kernel modules:
1044 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1045 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1046 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1047 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1050 The TTY subsystem of the kernel has been replaced by a new
1051 implementation, which provides better scalability and an
1052 improved driver model. Most common drivers have been migrated to
1053 the new TTY subsystem, while others have not. The following
1054 drivers have not yet been ported to the new TTY layer:
1057 cy, digi, rc, rp, sio
1063 ng_h4, ng_tty, ppp, sl, snp
1065 Adding these drivers to your kernel configuration file shall
1066 cause compilation to fail.
1069 ntpd has been upgraded to 4.2.4p5.
1072 OpenSSH has been upgraded to 5.1p1.
1074 For many years, FreeBSD's version of OpenSSH preferred DSA
1075 over RSA for host and user authentication keys. With this
1076 upgrade, we've switched to the vendor's default of RSA over
1077 DSA. This may cause upgraded clients to warn about unknown
1078 host keys even for previously known hosts. Users should
1079 follow the usual procedure for verifying host keys before
1080 accepting the RSA key.
1082 This can be circumvented by setting the "HostKeyAlgorithms"
1083 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1086 Please note that the sequence of keys offered for
1087 authentication has been changed as well. You may want to
1088 specify IdentityFile in a different order to revert this
1092 The sio(4) driver has been removed from the i386 and amd64
1093 kernel configuration files. This means uart(4) is now the
1094 default serial port driver on those platforms as well.
1096 To prevent collisions with the sio(4) driver, the uart(4) driver
1097 uses different names for its device nodes. This means the
1098 onboard serial port will now most likely be called "ttyu0"
1099 instead of "ttyd0". You may need to reconfigure applications to
1100 use the new device names.
1102 When using the serial port as a boot console, be sure to update
1103 /boot/device.hints and /etc/ttys before booting the new kernel.
1104 If you forget to do so, you can still manually specify the hints
1105 at the loader prompt:
1107 set hint.uart.0.at="isa"
1108 set hint.uart.0.port="0x3F8"
1109 set hint.uart.0.flags="0x10"
1110 set hint.uart.0.irq="4"
1114 The gpt(8) utility has been removed. Use gpart(8) to partition
1118 The version that Linuxulator emulates was changed from 2.4.2
1119 to 2.6.16. If you experience any problems with Linux binaries
1120 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1121 if it fixes the problem contact emulation mailing list.
1124 ISDN4BSD (I4B) was removed from the src tree. You may need to
1125 update a your kernel configuration and remove relevant entries.
1128 I have checked in code to support multiple routing tables.
1129 See the man pages setfib(1) and setfib(2).
1130 This is a hopefully backwards compatible version,
1131 but to make use of it you need to compile your kernel
1132 with options ROUTETABLES=2 (or more up to 16).
1135 The 802.11 wireless support was redone to enable multi-bss
1136 operation on devices that are capable. The underlying device
1137 is no longer used directly but instead wlanX devices are
1138 cloned with ifconfig. This requires changes to rc.conf files.
1139 For example, change:
1140 ifconfig_ath0="WPA DHCP"
1143 ifconfig_wlan0="WPA DHCP"
1144 see rc.conf(5) for more details. In addition, mergemaster of
1145 /etc/rc.d is highly recommended. Simultaneous update of userland
1146 and kernel wouldn't hurt either.
1148 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1149 modules were merged into the base wlan module. All references
1150 to these modules (e.g. in kernel config files) must be removed.
1153 psm(4) has gained write(2) support in native operation level.
1154 Arbitrary commands can be written to /dev/psm%d and status can
1155 be read back from it. Therefore, an application is responsible
1156 for status validation and error recovery. It is a no-op in
1157 other operation levels.
1160 Support for KSE threading has been removed from the kernel. To
1161 run legacy applications linked against KSE libmap.conf may
1162 be used. The following libmap.conf may be used to ensure
1163 compatibility with any prior release:
1165 libpthread.so.1 libthr.so.1
1166 libpthread.so.2 libthr.so.2
1167 libkse.so.3 libthr.so.3
1170 The layout of struct vmspace has changed. This affects libkvm
1171 and any executables that link against libkvm and use the
1172 kvm_getprocs() function. In particular, but not exclusively,
1173 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1174 The effects are minimal, but it's advisable to upgrade world
1178 The latest em driver no longer has support in it for the
1179 82575 adapter, this is now moved to the igb driver. The
1180 split was done to make new features that are incompatible
1181 with older hardware easier to do.
1184 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1185 likewise the kernel option is now GEOM_LINUX_LVM.
1188 The default NFS mount mode has changed from UDP to TCP for
1189 increased reliability. If you rely on (insecurely) NFS
1190 mounting across a firewall you may need to update your
1194 Belatedly note the addition of m_collapse for compacting
1198 The fts(3) structures have been changed to use adequate
1199 integer types for their members and so to be able to cope
1200 with huge file trees. The old fts(3) ABI is preserved
1201 through symbol versioning in libc, so third-party binaries
1202 using fts(3) should still work, although they will not take
1203 advantage of the extended types. At the same time, some
1204 third-party software might fail to build after this change
1205 due to unportable assumptions made in its source code about
1206 fts(3) structure members. Such software should be fixed
1207 by its vendor or, in the worst case, in the ports tree.
1208 FreeBSD_version 800015 marks this change for the unlikely
1209 case that a portable fix is impossible.
1212 To upgrade to -current after this date, you must be running
1213 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1214 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1217 The ADAPTIVE_GIANT kernel option has been retired because its
1218 functionality is the default now.
1221 The AT keyboard emulation of sunkbd(4) has been turned on
1222 by default. In order to make the special symbols of the Sun
1223 keyboards driven by sunkbd(4) work under X these now have
1224 to be configured the same way as Sun USB keyboards driven
1225 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1227 Option "XkbLayout" "us"
1228 Option "XkbRules" "xorg"
1229 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1232 It has been decided that it is desirable to provide ABI
1233 backwards compatibility to the FreeBSD 4/5/6 versions of the
1234 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1235 broken with the introduction of PCI domain support (see the
1236 20070930 entry). Unfortunately, this required the ABI of
1237 PCIOCGETCONF to be broken again in order to be able to
1238 provide backwards compatibility to the old version of that
1239 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1240 again. As for prominent ports this affects neither pciutils
1241 nor xorg-server this time, the hal port needs to be rebuilt
1245 The misnamed kthread_create() and friends have been renamed
1246 to kproc_create() etc. Many of the callers already
1247 used kproc_start()..
1248 I will return kthread_create() and friends in a while
1249 with implementations that actually create threads, not procs.
1250 Renaming corresponds with version 800002.
1259 Avoid using make -j when upgrading. While generally safe, there are
1260 sometimes problems using -j to upgrade. If your upgrade fails with
1261 -j, please try again without -j. From time to time in the past there
1262 have been problems using -j with buildworld and/or installworld. This
1263 is especially true when upgrading between "distant" versions (eg one
1264 that cross a major release boundary or several minor releases, or when
1265 several months have passed on the -current branch).
1267 Sometimes, obscure build problems are the result of environment
1268 poisoning. This can happen because the make utility reads its
1269 environment when searching for values for global variables. To run
1270 your build attempts in an "environmental clean room", prefix all make
1271 commands with 'env -i '. See the env(1) manual page for more details.
1273 When upgrading from one major version to another it is generally best
1274 to upgrade to the latest code in the currently installed branch first,
1275 then do an upgrade to the new branch. This is the best-tested upgrade
1276 path, and has the highest probability of being successful. Please try
1277 this approach before reporting problems with a major version upgrade.
1281 When upgrading the boot ZFS pool to a new version, always follow
1284 1.) recompile and reinstall the ZFS boot loader and boot block
1285 (this is part of "make buildworld" and "make installworld")
1287 2.) update the ZFS boot block on your boot drive
1289 The following example updates the ZFS boot block on the first
1290 partition (freebsd-boot) of a GPT partitioned drive ad0:
1291 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1293 Non-boot pools do not need these updates.
1297 If you are updating from a prior version of FreeBSD (even one just
1298 a few days old), you should follow this procedure. It is the most
1299 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1301 make kernel-toolchain
1302 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1303 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1305 To test a kernel once
1306 ---------------------
1307 If you just want to boot a kernel once (because you are not sure
1308 if it works, or if you want to boot a known bad kernel to provide
1309 debugging information) run
1310 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1311 nextboot -k testkernel
1313 To just build a kernel when you know that it won't mess you up
1314 --------------------------------------------------------------
1315 This assumes you are already running a CURRENT system. Replace
1316 ${arch} with the architecture of your machine (e.g. "i386",
1317 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1319 cd src/sys/${arch}/conf
1320 config KERNEL_NAME_HERE
1321 cd ../compile/KERNEL_NAME_HERE
1326 If this fails, go to the "To build a kernel" section.
1328 To rebuild everything and install it on the current system.
1329 -----------------------------------------------------------
1330 # Note: sometimes if you are running current you gotta do more than
1331 # is listed here if you are upgrading from a really old current.
1333 <make sure you have good level 0 dumps>
1335 make kernel KERNCONF=YOUR_KERNEL_HERE
1337 <reboot in single user> [3]
1345 To cross-install current onto a separate partition
1346 --------------------------------------------------
1347 # In this approach we use a separate partition to hold
1348 # current's root, 'usr', and 'var' directories. A partition
1349 # holding "/", "/usr" and "/var" should be about 2GB in
1352 <make sure you have good level 0 dumps>
1355 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1356 <maybe newfs current's root partition>
1357 <mount current's root partition on directory ${CURRENT_ROOT}>
1358 make installworld DESTDIR=${CURRENT_ROOT}
1359 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1360 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1361 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1362 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1363 <reboot into current>
1364 <do a "native" rebuild/install as described in the previous section>
1365 <maybe install compatibility libraries from ports/misc/compat*>
1369 To upgrade in-place from 8.x-stable to current
1370 ----------------------------------------------
1371 <make sure you have good level 0 dumps>
1373 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1375 <reboot in single user> [3]
1382 Make sure that you've read the UPDATING file to understand the
1383 tweaks to various things you need. At this point in the life
1384 cycle of current, things change often and you are on your own
1385 to cope. The defaults can also change, so please read ALL of
1386 the UPDATING entries.
1388 Also, if you are tracking -current, you must be subscribed to
1389 freebsd-current@freebsd.org. Make sure that before you update
1390 your sources that you have read and understood all the recent
1391 messages there. If in doubt, please track -stable which has
1392 much fewer pitfalls.
1394 [1] If you have third party modules, such as vmware, you
1395 should disable them at this point so they don't crash your
1398 [3] From the bootblocks, boot -s, and then do
1403 adjkerntz -i # if CMOS is wall time
1404 Also, when doing a major release upgrade, it is required that
1405 you boot into single user mode to do the installworld.
1407 [4] Note: This step is non-optional. Failure to do this step
1408 can result in a significant reduction in the functionality of the
1409 system. Attempting to do it by hand is not recommended and those
1410 that pursue this avenue should read this file carefully, as well
1411 as the archives of freebsd-current and freebsd-hackers mailing lists
1412 for potential gotchas. The -U option is also useful to consider.
1413 See mergemaster(8) for more information.
1415 [5] Usually this step is a noop. However, from time to time
1416 you may need to do this if you get unknown user in the following
1417 step. It never hurts to do it all the time. You may need to
1418 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1419 install) after the buildworld before this step if you last updated
1420 from current before 20020224 or from -stable before 20020408.
1422 [6] This only deletes old files and directories. Old libraries
1423 can be deleted by "make delete-old-libs", but you have to make
1424 sure that no program is using those libraries anymore.
1426 [8] In order to have a kernel that can run the 4.x binaries needed to
1427 do an installworld, you must include the COMPAT_FREEBSD4 option in
1428 your kernel. Failure to do so may leave you with a system that is
1429 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1430 required to run the 5.x binaries on more recent kernels. And so on
1431 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1433 Make sure that you merge any new devices from GENERIC since the
1434 last time you updated your kernel config file.
1436 [9] When checking out sources, you must include the -P flag to have
1437 cvs prune empty directories.
1439 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1440 "?=" instead of the "=" assignment operator, so that buildworld can
1441 override the CPUTYPE if it needs to.
1443 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1444 not on the command line, or in /etc/make.conf. buildworld will
1445 warn if it is improperly defined.
1448 This file contains a list, in reverse chronological order, of major
1449 breakages in tracking -current. Not all things will be listed here,
1450 and it only starts on October 16, 2004. Updating files can found in
1451 previous releases if your system is older than this.
1453 Copyright information:
1455 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1457 Redistribution, publication, translation and use, with or without
1458 modification, in full or in part, in any form or format of this
1459 document are permitted without further permission from the author.
1461 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1462 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1463 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1464 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1465 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1466 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1467 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1468 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1469 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1470 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1471 POSSIBILITY OF SUCH DAMAGE.
1473 Contact Warner Losh if you have any questions about your use of