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.
30 The following sysctls and tunables are retired on x86 platforms:
32 machdep.hlt_logical_cpus
33 The following sysctl is retired:
34 machdep.hyperthreading_allowed
35 The sysctls were supposed to provide a way to dynamically offline and
36 online selected CPUs on x86 platforms, but the implementation has not
37 been reliable especially with SCHED_ULE scheduler.
38 machdep.hyperthreading_allowed tunable is still available to ignore
39 hyperthreading CPUs at OS level.
40 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
41 where X is an APIC ID of a CPU. Be advised, though, that disabling
42 CPUs in non-uniform fashion will result in non-uniform topology and
43 may lead to sub-optimal system performance with SCHED_ULE, which is
47 cpumask_t type is retired and cpuset_t is used in order to describe
51 Changes to ifconfig(8) for dynamic address family detection mandate
52 that you are running a kernel of 20110525 or later. Make sure to
53 follow the update procedure to boot a new kernel before installing
57 Support for sun4v architecture is officially dropped
60 Several KPI breaking changes have been committed to the mii(4) layer,
61 the PHY drivers and consequently some Ethernet drivers using mii(4).
62 This means that miibus.ko and the modules of the affected Ethernet
63 drivers need to be recompiled.
65 Note to kernel developers: Given that the OUI bit reversion problem
66 was fixed as part of these changes all mii(4) commits related to OUIs,
67 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
68 handling, no longer can be merged verbatim to stable/8 and previous
72 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
73 into their kernel configurations along with 'device pci'.
76 The default NFS client is now the new NFS client, so fstype "newnfs"
77 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
78 Although mounts via fstype "nfs" will usually work without userland
79 changes, it is recommended that the mount(8) and mount_nfs(8)
80 commands be rebuilt from sources and that a link to mount_nfs called
81 mount_oldnfs be created. The new client is compiled into the
82 kernel with "options NFSCL" and this is needed for diskless root
83 file systems. The GENERIC kernel configs have been changed to use
84 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
85 To use the regular/old client, you can "mount -t oldnfs ...". For
86 a diskless root file system, you must also include a line like:
88 vfs.root.mountfrom="oldnfs:"
90 in the boot/loader.conf on the root fs on the NFS server to make
91 a diskless root fs use the old client.
94 The GENERIC kernels for all architectures now default to the new
95 CAM-based ATA stack. It means that all legacy ATA drivers were
96 removed and replaced by respective CAM drivers. If you are using
97 ATA device names in /etc/fstab or other places, make sure to update
98 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
99 where 'Y's are the sequential numbers starting from zero for each type
100 in order of detection, unless configured otherwise with tunables,
101 see cam(4)). There will be symbolic links created in /dev/ to map
102 old adX devices to the respective adaY. They should provide basic
103 compatibility for file systems mounting in most cases, but they do
104 not support old user-level APIs and do not have respective providers
105 in GEOM. Consider using updated management tools with new device names.
107 It is possible to load devices ahci, ata, siis and mvs as modules,
108 but option ATA_CAM should remain in kernel configuration to make ata
109 module work as CAM driver supporting legacy ATA controllers. Device ata
110 still can be used in modular fashion (atacore + ...). Modules atadisk
111 and atapi* are not used and won't affect operation in ATA_CAM mode.
112 Note that to use CAM-based ATA kernel should include CAM devices
113 scbus, pass, da (or explicitly ada), cd and optionally others. All of
114 them are parts of the cam module.
116 ataraid(4) functionality is now supported by the RAID GEOM class.
117 To use it you can load geom_raid kernel module and use graid(8) tool
118 for management. Instead of /dev/arX device names, use /dev/raid/rX.
120 No kernel config options or code have been removed, so if a problem
121 arises, please report it and optionally revert to the old ATA stack.
122 In order to do it you can remove from the kernel config:
127 , and instead add back:
128 device atadisk # ATA disk drives
129 device ataraid # ATA RAID drives
130 device atapicd # ATAPI CDROM drives
131 device atapifd # ATAPI floppy drives
132 device atapist # ATAPI tape drives
135 The default NFS server has been changed to the new server, which
136 was referred to as the experimental server. If you need to switch
137 back to the old NFS server, you must now put the "-o" option on
138 both the mountd and nfsd commands. This can be done using the
139 mountd_flags and nfs_server_flags rc.conf variables until an
140 update to the rc scripts is committed, which is coming soon.
143 The GNU Objective-C runtime library (libobjc), and other Objective-C
144 related components have been removed from the base system. If you
145 require an Objective-C library, please use one of the available ports.
148 ath(4) has been split into bus- and device- modules. if_ath contains
149 the HAL, the TX rate control and the network device code. if_ath_pci
150 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
151 contains the AHB glue. Users need to load both if_ath_pci and if_ath
152 in order to use ath on everything else.
154 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
155 need to load if_ath and if_ath_pci for ath(4) operation.
158 As part of the replacement of sysinstall, the process of building
159 release media has changed significantly. For details, please re-read
160 release(7), which has been updated to reflect the new build process.
163 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
164 is the last available version under GPLv2. It brings a number of new
165 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
166 4.1 and SSE 4.2), better support for powerpc64, a number of new
167 directives, and lots of other small improvements. See the ChangeLog
168 file in contrib/binutils for the full details.
171 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
172 compliant, and will now use half of hash for authentication.
173 This will break interoperability with all stacks (including all
174 actual FreeBSD versions) who implement
175 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
177 The only workaround with such peers is to use another HMAC
178 algorithm for IPsec ("phase 2") authentication.
181 Remove the uio_yield prototype and symbol. This function has
182 been misnamed since it was introduced and should not be
183 globally exposed with this name. The equivalent functionality
184 is now available using kern_yield(curthread->td_user_pri).
185 The function remains undocumented.
188 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
189 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
190 for scalar sysctls is defined but disabled. Code that needs
191 UQUAD to pass the type checking that must compile on older
192 systems where the define is not present can check against
193 __FreeBSD_version >= 900030.
195 The system dialog(1) has been replaced with a new version previously
196 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
197 with the previous version, but the libdialog associated with it has
198 a largely incompatible API. As such, the original version of libdialog
199 will be kept temporarily as libodialog, until its base system consumers
200 are replaced or updated. Bump __FreeBSD_version to 900030.
203 If you are trying to run make universe on a -stable system, and you get
204 the following warning:
205 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
206 unknown. config(8) likely too old."
207 or something similar to it, then you must upgrade your -stable system
208 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
209 stable/8) or build the config from the latest stable/8 branch and
210 install it on your system.
212 Prior to this date, building a current universe on 8-stable system from
213 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
214 error in the first kernel build phase. A new config on those old
215 systems will fix that problem for older versions of -current.
218 The TCP stack has been modified to allow Khelp modules to interact with
219 it via helper hook points and store per-connection data in the TCP
220 control block. Bump __FreeBSD_version to 900029. User space tools that
221 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
225 Generic IEEE 802.3 annex 31B full duplex flow control support has been
226 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
227 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
228 to take advantage of it instead of using custom implementations. This
229 means that these drivers now no longer unconditionally advertise
230 support for flow control but only do so if flow control is a selected
231 media option. This was implemented in the generic support that way in
232 order to allow flow control to be switched on and off via ifconfig(8)
233 with the PHY specific default to typically off in order to protect
234 from unwanted effects. Consequently, if you used flow control with
235 one of the above mentioned drivers you now need to explicitly enable
237 ifconfig bge0 media auto mediaopt flowcontrol
239 Along with the above mentioned changes generic support for setting
240 1000baseT master mode also has been added and brgphy(4), ciphy(4),
241 e1000phy(4) as well as ip1000phy(4) have been converted to take
242 advantage of it. This means that these drivers now no longer take the
243 link0 parameter for selecting master mode but the master media option
244 has to be used instead, for example like in the following:
245 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
247 Selection of master mode now is also available with all other PHY
248 drivers supporting 1000baseT.
251 The TCP stack has received a significant update to add support for
252 modularised congestion control and generally improve the clarity of
253 congestion control decisions. Bump __FreeBSD_version to 900025. User
254 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
255 sockstat) need to be recompiled.
258 The man(1) utility has been replaced by a new version that no longer
259 uses /etc/manpath.config. Please consult man.conf(5) for how to
260 migrate local entries to the new format.
263 The copyright strings printed by login(1) and sshd(8) at the time of a
264 new connection have been removed to follow other operating systems and
268 A workaround for a fixed ld bug has been removed in kernel code,
269 so make sure that your system ld is built from sources after
270 revision 210245 from 2010-07-19 (r211583 if building head kernel
271 on stable/8, r211584 for stable/7; both from 2010-08-21).
272 A symptom of incorrect ld version is different addresses for
273 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
276 The $ipv6_prefer variable in rc.conf(5) has been split into
277 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
279 The $ip6addrctl_policy is a variable to choose a pre-defined
280 address selection policy set by ip6addrctl(8). A value
281 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
284 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
285 flag (see an entry of 20090926) is set on an interface with no
286 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
287 security reason. If you want IPv6 link-local address on all
288 interfaces by default, set this to "YES".
290 The old ipv6_prefer="YES" is equivalent to
291 ipv6_activate_all_interfaces="YES" and
292 ip6addrctl_policy="ipv6_prefer".
295 DTrace has grown support for userland tracing. Due to this, DTrace is
296 now i386 and amd64 only.
297 dtruss(1) is now installed by default on those systems and a new
298 kernel module is needed for userland tracing: fasttrap.
299 No changes to your kernel config file are necessary to enable
300 userland tracing, but you might consider adding 'STRIP=' and
301 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
302 to have informative userland stack traces in DTrace (ustack).
305 The acpi_aiboost(4) driver has been removed in favor of the new
306 aibs(4) driver. You should update your kernel configuration file.
309 BSD grep has been imported to the base system and it is built by
310 default. It is completely BSD licensed, highly GNU-compatible, uses
311 less memory than its GNU counterpart and has a small codebase.
312 However, it is slower than its GNU counterpart, which is mostly
313 noticeable for larger searches, for smaller ones it is measurable
314 but not significant. The reason is complex, the most important factor
315 is that we lack a modern and efficient regex library and GNU
316 overcomes this by optimizing the searches internally. Future work
317 on improving the regex performance is planned, for the meantime,
318 users that need better performance, can build GNU grep instead by
319 setting the WITH_GNU_GREP knob.
322 Due to the import of powerpc64 support, all existing powerpc kernel
323 configuration files must be updated with a machine directive like this:
324 machine powerpc powerpc
326 In addition, an updated config(8) is required to build powerpc kernels
330 A new version of ZFS (version 15) has been merged to -HEAD.
331 This version uses a python library for the following subcommands:
332 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
333 For full functionality of these commands the following port must
334 be installed: sysutils/py-zfs
337 'vm_page's are now hashed by physical address to an array of mutexes.
338 Currently this is only used to serialize access to hold_count. Over
339 time the page queue mutex will be peeled away. This changes the size
340 of pmap on every architecture. And requires all callers of vm_page_hold
341 and vm_page_unhold to be updated.
344 WITH_CTF can now be specified in src.conf (not recommended, there
345 are some problems with static executables), make.conf (would also
346 affect ports which do not use GNU make and do not override the
347 compile targets) or in the kernel config (via "makeoptions
349 When WITH_CTF was specified there before this was silently ignored,
350 so make sure that WITH_CTF is not used in places which could lead
351 to unwanted behavior.
354 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
355 to allow 32-bit compatibility on non-x86 platforms. All kernel
356 configurations on amd64 and ia64 platforms using these options must
357 be modified accordingly.
360 The utmp user accounting database has been replaced with utmpx,
361 the user accounting interface standardized by POSIX.
362 Unfortunately the semantics of utmp and utmpx don't match,
363 making it practically impossible to support both interfaces.
364 The user accounting database is used by tools like finger(1),
365 last(1), talk(1), w(1) and ac(8).
367 All applications in the base system use utmpx. This means only
368 local binaries (e.g. from the ports tree) may still use these
369 utmp database files. These applications must be rebuilt to make
372 After the system has been upgraded, it is safe to remove the old
373 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
374 assuming their contents is of no importance anymore. Old wtmp
375 databases can only be used by last(1) and ac(8) after they have
376 been converted to the new format using wtmpcvt(1).
379 Introduce the kernel thread "deadlock resolver" (which can be enabled
380 via the DEADLKRES option, see NOTES for more details) and the
381 sleepq_type() function for sleepqueues.
384 The rc.firewall and rc.firewall6 were unified, and
385 rc.firewall6 and rc.d/ip6fw were removed.
386 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
387 variables are obsoleted. Instead, the following new rc
388 variables are added to rc.d/ipfw:
390 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
391 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
392 firewall_simple_onet_ipv6, firewall_trusted_ipv6
394 The meanings correspond to the relevant IPv4 variables.
400 The default terminal emulation for syscons(4) has been changed
401 from cons25 to xterm on all platforms except pc98. This means
402 that the /etc/ttys file needs to be updated to ensure correct
403 operation of applications on the console.
405 The terminal emulation style can be toggled per window by using
406 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
407 options can be used to change the compile-time default back to
410 To prevent graphical artifacts, make sure the TERM environment
411 variable is set to match the terminal emulation that is being
412 performed by syscons(4).
415 The layout of the structure ieee80211req_scan_result has changed.
416 Applications that require wireless scan results (e.g. ifconfig(8))
417 from net80211 need to be recompiled.
419 Applications such as wpa_supplicant(8) may require a full world
420 build without using NO_CLEAN in order to get synchronized with the
424 The iwn(4) driver has been updated to support the 5000 and 5150 series.
425 There's one kernel module for each firmware. Adding "device iwnfw"
426 to the kernel configuration file means including all three firmware
427 images inside the kernel. If you want to include just the one for
428 your wireless card, use the the devices iwn4965fw, iwn5000fw or
432 The rc.d/network_ipv6, IPv6 configuration script has been integrated
433 into rc.d/netif. The changes are the following:
435 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
436 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
437 Note that both variables need the "inet6" keyword at the head.
439 Do not set $ipv6_network_interfaces manually if you do not
440 understand what you are doing. It is not needed in most cases.
442 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
445 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
446 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
448 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
449 all configured interfaces have "inet6 accept_rtadv" in the
450 $ifconfig_IF_ipv6. These are for backward compatibility.
452 3. A new variable $ipv6_prefer has been added. If NO, IPv6
453 functionality of interfaces with no corresponding
454 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
455 and the default address selection policy of ip6addrctl(8)
456 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
457 Note that if you want to configure IPv6 functionality on the
458 disabled interfaces after boot, first you need to clear the flag by
459 using ifconfig(8) like:
461 ifconfig em0 inet6 -ifdisabled
463 If YES, the default address selection policy is set as
466 The default value of $ipv6_prefer is NO.
468 4. If your system need to receive Router Advertisement messages,
469 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
470 scripts automatically invoke rtsol(8) when the interface becomes
471 UP. The Router Advertisement messages are used for SLAAC
472 (State-Less Address AutoConfiguration).
475 802.11s D3.03 support was committed. This is incompatible with the
476 previous code, which was based on D3.0.
479 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
480 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
481 control whether accepting Router Advertisement messages or not.
482 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
483 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
484 The ifconfig(8) utility now supports these flags.
487 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
488 mount(8) and -a option for df(1) to see them.
491 The old tunable hw.bus.devctl_disable has been superseded by
492 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
493 replaced by hw.bus.devctl_queue=0. The default for this new tunable
497 Remove the option STOP_NMI. The default action is now to use NMI only
498 for KDB via the newly introduced function stop_cpus_hard() and
499 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
502 The stable/8 branch created in subversion. This corresponds to the
503 RELENG_8 branch in CVS.
506 Bump the shared library version numbers for all libraries that do not
507 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
508 __FreeBSD_version to 800105.
511 Due to changes in the implementation of virtual network stack support,
512 all network-related kernel modules must be recompiled. As this change
513 breaks the ABI, bump __FreeBSD_version to 800104.
516 The TOE interface to the TCP syncache has been modified to remove
517 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
518 The cxgb driver is the only TOE consumer affected by this change, and
519 needs to be recompiled along with the kernel. As this change breaks
520 the ABI, bump __FreeBSD_version to 800103.
523 Padding has been added to struct tcpcb, sackhint and tcpstat in
524 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
525 maintaining the ABI. However, this change breaks the ABI, so bump
526 __FreeBSD_version to 800102. User space tools that rely on the size of
527 any of these structs (e.g. sockstat) need to be recompiled.
530 The NFS_LEGACYRPC option has been removed along with the old kernel
531 RPC implementation that this option selected. Kernel configurations
532 may need to be adjusted.
535 The network interface device nodes at /dev/net/<interface> have been
536 removed. All ioctl operations can be performed the normal way using
537 routing sockets. The kqueue functionality can generally be replaced
538 with routing sockets.
541 The documentation from the FreeBSD Documentation Project (Handbook,
542 FAQ, etc.) is now installed via packages by sysinstall(8) and under
543 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
546 The ABI of various structures related to the SYSV IPC API have been
547 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
548 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
552 Layout of struct vnet has changed as routing related variables were
553 moved to their own Vimage module. Modules need to be recompiled. Bump
554 __FreeBSD_version to 800099.
557 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
558 respectively. As long as no more than 16 groups per process are used,
559 no changes should be visible. When more than 16 groups are used, old
560 binaries may fail if they call getgroups() or getgrouplist() with
561 statically sized storage. Recompiling will work around this, but
562 applications should be modified to use dynamically allocated storage
563 for group arrays as POSIX.1-2008 does not cap an implementation's
564 number of supported groups at NGROUPS_MAX+1 as previous versions did.
566 NFS and portalfs mounts may also be affected as the list of groups is
567 truncated to 16. Users of NFS who use more than 16 groups, should
568 take care that negative group permissions are not used on the exported
569 file systems as they will not be reliable unless a GSSAPI based
570 authentication method is used.
573 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
574 option compiles in the support for adaptive spinning for lockmgrs
575 which want to enable it. The lockinit() function now accepts the flag
576 LK_ADAPTIVE in order to make the lock object subject to adaptive
577 spinning when both held in write and read mode.
580 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
581 changed. User applications that use this ioctl need to be rebuilt.
584 The layout of struct thread has changed. Kernel and modules need to
588 The layout of structs ifnet, domain, protosw and vnet_net has changed.
589 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
592 window(1) has been removed from the base system. It can now be
593 installed from ports. The port is called misc/window.
596 The way we are storing and accessing `routing table' entries has
597 changed. Programs reading the FIB, like netstat, need to be
601 A new netisr implementation has been added for FreeBSD 8. Network
602 file system modules, such as igmp, ipdivert, and others, should be
604 Bump __FreeBSD_version to 800096.
607 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
611 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
612 Bump __FreeBSD_version to 800094.
615 Add mnt_xflag field to 'struct mount'. File system modules need to be
617 Bump __FreeBSD_version to 800093.
620 The compiling option ADAPTIVE_SX has been retired while it has been
621 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
622 The KPI for sx_init_flags() changes as accepting flags:
623 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
624 been introduced in order to handle the reversed logic.
625 Bump __FreeBSD_version to 800092.
628 Add support for hierarchical jails. Remove global securelevel.
629 Bump __FreeBSD_version to 800091.
632 The layout of struct vnet_net has changed, therefore modules
634 Bump __FreeBSD_version to 800090.
637 The newly imported zic(8) produces a new format in the output. Please
638 run tzsetup(8) to install the newly created data to /etc/localtime.
641 The sysctl tree for the usb stack has renamed from hw.usb2.* to
642 hw.usb.* and is now consistent again with previous releases.
645 802.11 monitor mode support was revised and driver api's were changed.
646 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
647 of DLT_IEEE802_11. No user-visible data structures were changed but
648 applications that use DLT_IEEE802_11 may require changes.
649 Bump __FreeBSD_version to 800088.
652 The layout of the following structs has changed: sysctl_oid,
653 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
654 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
655 panics may be experienced. World rebuild is required for
656 correctly checking networking state from userland.
657 Bump __FreeBSD_version to 800085.
660 MLDv2 and Source-Specific Multicast (SSM) have been merged
661 to the IPv6 stack. VIMAGE hooks are in but not yet used.
662 The implementation of SSM within FreeBSD's IPv6 stack closely
663 follows the IPv4 implementation.
665 For kernel developers:
667 * The most important changes are that the ip6_output() and
668 ip6_input() paths no longer take the IN6_MULTI_LOCK,
669 and this lock has been downgraded to a non-recursive mutex.
671 * As with the changes to the IPv4 stack to support SSM, filtering
672 of inbound multicast traffic must now be performed by transport
673 protocols within the IPv6 stack. This does not apply to TCP and
674 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
676 * The KPIs used by IPv6 multicast are similar to those used by
677 the IPv4 stack, with the following differences:
678 * im6o_mc_filter() is analogous to imo_multicast_filter().
679 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
680 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
681 * IN6_LOOKUP_MULTI() has been deprecated and removed.
682 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
683 for MLDv1 have an additional 'timer' argument which is used to
684 jitter the initial membership report for the solicited-node
685 multicast membership on-link.
686 * This is not strictly needed for MLDv2, which already jitters
687 its report transmissions. However, the 'timer' argument is
688 preserved in case MLDv1 is active on the interface.
690 * The KAME linked-list based IPv6 membership implementation has
691 been refactored to use a vector similar to that used by the IPv4
693 Code which maintains a list of its own multicast memberships
694 internally, e.g. carp, has been updated to reflect the new
697 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
698 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
699 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
700 implementation constraint which needs to be addressed in HEAD.
702 For application developers:
704 * The changes are broadly similar to those made for the IPv4
707 * The use of IPv4 and IPv6 multicast socket options on the same
708 socket, using mapped addresses, HAS NOT been tested or supported.
710 * There are a number of issues with the implementation of various
711 IPv6 multicast APIs which need to be resolved in the API surface
712 before the implementation is fully compatible with KAME userland
713 use, and these are mostly to do with interface index treatment.
715 * The literature available discusses the use of either the delta / ASM
716 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
717 using setsourcefilter(3)/getsourcefilter(3). For more information
718 please refer to RFC 3768, 'Socket Interface Extensions for
719 Multicast Source Filters'.
721 * Applications which use the published RFC 3678 APIs should be fine.
723 For systems administrators:
725 * The mtest(8) utility has been refactored to support IPv6, in
726 addition to IPv4. Interface addresses are no longer accepted
727 as arguments, their names must be used instead. The utility
728 will map the interface name to its first IPv4 address as
729 returned by getifaddrs(3).
731 * The ifmcstat(8) utility has also been updated to print the MLDv2
732 endpoint state and source filter lists via sysctl(3).
734 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
735 loopback of IPv6 multicast datagrams by default; it defaults to 1
736 to preserve the existing behaviour. Disabling multicast loopback is
737 recommended for optimal system performance.
739 * The IPv6 MROUTING code has been changed to examine this sysctl
740 instead of attempting to perform a group lookup before looping
741 back forwarded datagrams.
743 Bump __FreeBSD_version to 800084.
746 Implement low-level Bluetooth HCI API.
747 Bump __FreeBSD_version to 800083.
750 The layout of struct malloc_type, used by modules to register new
751 memory allocation types, has changed. Most modules will need to
752 be rebuilt or panics may be experienced.
753 Bump __FreeBSD_version to 800081.
756 Anticipate overflowing inp_flags - add inp_flags2.
757 This changes most offsets in inpcb, so checking v4 connection
758 state will require a world rebuild.
759 Bump __FreeBSD_version to 800080.
762 Add an llentry to struct route and struct route_in6. Modules
763 embedding a struct route will need to be recompiled.
764 Bump __FreeBSD_version to 800079.
767 The size of rt_metrics_lite and by extension rtentry has changed.
768 Networking administration apps will need to be recompiled.
769 The route command now supports show as an alias for get, weighting
770 of routes, sticky and nostick flags to alter the behavior of stateful
772 Bump __FreeBSD_version to 800078.
775 Do not use Giant for kbdmux(4) locking. This is wrong and
776 apparently causing more problems than it solves. This will
777 re-open the issue where interrupt handlers may race with
778 kbdmux(4) in polling mode. Typical symptoms include (but
779 not limited to) duplicated and/or missing characters when
780 low level console functions (such as gets) are used while
781 interrupts are enabled (for example geli password prompt,
782 mountroot prompt etc.). Disabling kbdmux(4) may help.
785 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
786 kernel modules referencing any of the above need to be recompiled.
787 Bump __FreeBSD_version to 800075.
790 GEOM_PART has become the default partition slicer for storage devices,
791 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
792 introduces some changes:
794 MSDOS/EBR: the devices created from MSDOS extended partition entries
795 (EBR) can be named differently than with GEOM_MBR and are now symlinks
796 to devices with offset-based names. fstabs may need to be modified.
798 BSD: the "geometry does not match label" warning is harmless in most
799 cases but it points to problems in file system misalignment with
800 disk geometry. The "c" partition is now implicit, covers the whole
801 top-level drive and cannot be (mis)used by users.
803 General: Kernel dumps are now not allowed to be written to devices
804 whose partition types indicate they are meant to be used for file
805 systems (or, in case of MSDOS partitions, as something else than
808 Most of these changes date approximately from 200812.
811 The uscanner(4) driver has been removed from the kernel. This follows
812 Linux removing theirs in 2.6 and making libusb the default interface
816 The multicast forwarding code has been cleaned up. netstat(1)
817 only relies on KVM now for printing bandwidth upcall meters.
818 The IPv4 and IPv6 modules are split into ip_mroute_mod and
819 ip6_mroute_mod respectively. The config(5) options for statically
820 compiling this code remain the same, i.e. 'options MROUTING'.
823 Support for the IFF_NEEDSGIANT network interface flag has been
824 removed, which means that non-MPSAFE network device drivers are no
825 longer supported. In particular, if_ar, if_sr, and network device
826 drivers from the old (legacy) USB stack can no longer be built or
830 POSIX.1 Native Language Support (NLS) has been enabled in libc and
831 a bunch of new language catalog files have also been added.
832 This means that some common libc messages are now localized and
833 they depend on the LC_MESSAGES environmental variable.
836 The k8temp(4) driver has been renamed to amdtemp(4) since
837 support for Family 10 and Family 11 CPU families was added.
840 IGMPv3 and Source-Specific Multicast (SSM) have been merged
841 to the IPv4 stack. VIMAGE hooks are in but not yet used.
843 For kernel developers, the most important changes are that the
844 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
845 and this lock has been downgraded to a non-recursive mutex.
847 Transport protocols (UDP, Raw IP) are now responsible for filtering
848 inbound multicast traffic according to group membership and source
849 filters. The imo_multicast_filter() KPI exists for this purpose.
850 Transports which do not use multicast (SCTP, TCP) already reject
851 multicast by default. Forwarding and receive performance may improve
852 as a mutex acquisition is no longer needed in the ip_input()
853 low-level input path. in_addmulti() and in_delmulti() are shimmed
854 to new KPIs which exist to support SSM in-kernel.
856 For application developers, it is recommended that loopback of
857 multicast datagrams be disabled for best performance, as this
858 will still cause the lock to be taken for each looped-back
859 datagram transmission. The net.inet.ip.mcast.loop sysctl may
860 be tuned to 0 to disable loopback by default; it defaults to 1
861 to preserve the existing behaviour.
863 For systems administrators, to obtain best performance with
864 multicast reception and multiple groups, it is always recommended
865 that a card with a suitably precise hash filter is used. Hash
866 collisions will still result in the lock being taken within the
867 transport protocol input path to check group membership.
869 If deploying FreeBSD in an environment with IGMP snooping switches,
870 it is recommended that the net.inet.igmp.sendlocal sysctl remain
871 enabled; this forces 224.0.0.0/24 group membership to be announced
874 The size of 'struct igmpstat' has changed; netstat needs to be
875 recompiled to reflect this.
876 Bump __FreeBSD_version to 800070.
879 libusb20.so.1 is now installed as libusb.so.1 and the ports system
880 updated to use it. This requires a buildworld/installworld in order to
881 update the library and dependencies (usbconfig, etc). Its advisable to
882 rebuild all ports which uses libusb. More specific directions are given
883 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
884 libusb are no longer required and can be removed.
887 A workaround is committed to allow the creation of System V shared
888 memory segment of size > 2 GB on the 64-bit architectures.
889 Due to a limitation of the existing ABI, the shm_segsz member
890 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
891 wrong for large segments. Note that limits must be explicitly
892 raised to allow such segments to be created.
895 The layout of struct ifnet has changed, requiring a rebuild of all
896 network device driver modules.
899 The /dev handling for the new USB stack has changed, a
900 buildworld/installworld is required for libusb20.
903 The new USB2 stack has now been permanently moved in and all kernel and
904 module names reverted to their previous values (eg, usb, ehci, ohci,
905 ums, ...). The old usb stack can be compiled in by prefixing the name
906 with the letter 'o', the old usb modules have been removed.
907 Updating entry 20090216 for xorg and 20090215 for libmap may still
911 The rc.conf(5) option if_up_delay has been renamed to
912 defaultroute_delay to better reflect its purpose. If you have
913 customized this setting in /etc/rc.conf you need to update it to
917 xorg 7.4 wants to configure its input devices via hald which does not
918 yet work with USB2. If the keyboard/mouse does not work in xorg then
920 Option "AllowEmptyInput" "off"
921 to your ServerLayout section. This will cause X to use the configured
922 kbd and mouse sections from your xorg.conf.
925 The GENERIC kernels for all architectures now default to the new USB2
926 stack. No kernel config options or code have been removed so if a
927 problem arises please report it and optionally revert to the old USB
928 stack. If you are loading USB kernel modules or have a custom kernel
929 that includes GENERIC then ensure that usb names are also changed over,
930 eg uftdi -> usb2_serial_ftdi.
932 Older programs linked against the ports libusb 0.1 need to be
933 redirected to the new stack's libusb20. /etc/libmap.conf can
935 # Map old usb library to new one for usb2 stack
936 libusb-0.1.so.8 libusb20.so.1
939 All USB ethernet devices now attach as interfaces under the name ueN
940 (eg. ue0). This is to provide a predictable name as vendors often
941 change usb chipsets in a product without notice.
944 The ichsmb(4) driver has been changed to require SMBus slave
945 addresses be left-justified (xxxxxxx0b) rather than right-justified.
946 All of the other SMBus controller drivers require left-justified
947 slave addresses, so this change makes all the drivers provide the
951 INET6 statistics (struct ip6stat) was updated.
952 netstat(1) needs to be recompiled.
955 NTFS has been removed from GENERIC kernel on amd64 to match
956 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
957 will load ntfs.ko module automatically when NTFS support is
958 actually needed, unless ntfs.ko is not installed or security
959 level prohibits loading kernel modules. If either is the case,
960 "options NTFS" has to be added into kernel config.
963 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
964 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
965 800061. User space tools that rely on the size of struct tcpcb in
966 tcp_var.h (e.g. sockstat) need to be recompiled.
969 ng_tty(4) module updated to match the new TTY subsystem.
970 Due to API change, user-level applications must be updated.
971 New API support added to mpd5 CVS and expected to be present
972 in next mpd5.3 release.
975 With __FreeBSD_version 800060 the makefs tool is part of
976 the base system (it was a port).
979 The afdata and ifnet locks have been changed from mutexes to
980 rwlocks, network modules will need to be re-compiled.
983 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
984 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
985 The new code reduced struct rtentry{} by 16 bytes on 32-bit
986 architecture and 40 bytes on 64-bit architecture. The userland
987 applications "arp" and "ndp" have been updated accordingly.
988 The output from "netstat -r" shows only routing entries and
989 none of the L2 information.
992 __FreeBSD_version 800057 marks the switchover from the
993 binary ath hal to source code. Users must add the line:
995 options AH_SUPPORT_AR5416
997 to their kernel config files when specifying:
1001 The ath_hal module no longer exists; the code is now compiled
1002 together with the driver in the ath module. It is now
1003 possible to tailor chip support (i.e. reduce the set of chips
1004 and thereby the code size); consult ath_hal(4) for details.
1007 __FreeBSD_version 800054 adds memory barriers to
1008 <machine/atomic.h>, new interfaces to ifnet to facilitate
1009 multiple hardware transmit queues for cards that support
1010 them, and a lock-less ring-buffer implementation to
1011 enable drivers to more efficiently manage queueing of
1015 A new version of ZFS (version 13) has been merged to -HEAD.
1016 This version has zpool attribute "listsnapshots" off by
1017 default, which means "zfs list" does not show snapshots,
1018 and is the same as Solaris behavior.
1021 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1024 The uhci, ohci, ehci and slhci USB Host controller drivers have
1025 been put into separate modules. If you load the usb module
1026 separately through loader.conf you will need to load the
1027 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1028 controller add the following to loader.conf:
1034 The ABI used by the PMC toolset has changed. Please keep
1035 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1039 atapci kernel module now includes only generic PCI ATA
1040 driver. AHCI driver moved to ataahci kernel module.
1041 All vendor-specific code moved into separate kernel modules:
1042 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1043 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1044 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1045 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1048 The TTY subsystem of the kernel has been replaced by a new
1049 implementation, which provides better scalability and an
1050 improved driver model. Most common drivers have been migrated to
1051 the new TTY subsystem, while others have not. The following
1052 drivers have not yet been ported to the new TTY layer:
1055 cy, digi, rc, rp, sio
1061 ng_h4, ng_tty, ppp, sl, snp
1063 Adding these drivers to your kernel configuration file shall
1064 cause compilation to fail.
1067 ntpd has been upgraded to 4.2.4p5.
1070 OpenSSH has been upgraded to 5.1p1.
1072 For many years, FreeBSD's version of OpenSSH preferred DSA
1073 over RSA for host and user authentication keys. With this
1074 upgrade, we've switched to the vendor's default of RSA over
1075 DSA. This may cause upgraded clients to warn about unknown
1076 host keys even for previously known hosts. Users should
1077 follow the usual procedure for verifying host keys before
1078 accepting the RSA key.
1080 This can be circumvented by setting the "HostKeyAlgorithms"
1081 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1084 Please note that the sequence of keys offered for
1085 authentication has been changed as well. You may want to
1086 specify IdentityFile in a different order to revert this
1090 The sio(4) driver has been removed from the i386 and amd64
1091 kernel configuration files. This means uart(4) is now the
1092 default serial port driver on those platforms as well.
1094 To prevent collisions with the sio(4) driver, the uart(4) driver
1095 uses different names for its device nodes. This means the
1096 onboard serial port will now most likely be called "ttyu0"
1097 instead of "ttyd0". You may need to reconfigure applications to
1098 use the new device names.
1100 When using the serial port as a boot console, be sure to update
1101 /boot/device.hints and /etc/ttys before booting the new kernel.
1102 If you forget to do so, you can still manually specify the hints
1103 at the loader prompt:
1105 set hint.uart.0.at="isa"
1106 set hint.uart.0.port="0x3F8"
1107 set hint.uart.0.flags="0x10"
1108 set hint.uart.0.irq="4"
1112 The gpt(8) utility has been removed. Use gpart(8) to partition
1116 The version that Linuxulator emulates was changed from 2.4.2
1117 to 2.6.16. If you experience any problems with Linux binaries
1118 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1119 if it fixes the problem contact emulation mailing list.
1122 ISDN4BSD (I4B) was removed from the src tree. You may need to
1123 update a your kernel configuration and remove relevant entries.
1126 I have checked in code to support multiple routing tables.
1127 See the man pages setfib(1) and setfib(2).
1128 This is a hopefully backwards compatible version,
1129 but to make use of it you need to compile your kernel
1130 with options ROUTETABLES=2 (or more up to 16).
1133 The 802.11 wireless support was redone to enable multi-bss
1134 operation on devices that are capable. The underlying device
1135 is no longer used directly but instead wlanX devices are
1136 cloned with ifconfig. This requires changes to rc.conf files.
1137 For example, change:
1138 ifconfig_ath0="WPA DHCP"
1141 ifconfig_wlan0="WPA DHCP"
1142 see rc.conf(5) for more details. In addition, mergemaster of
1143 /etc/rc.d is highly recommended. Simultaneous update of userland
1144 and kernel wouldn't hurt either.
1146 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1147 modules were merged into the base wlan module. All references
1148 to these modules (e.g. in kernel config files) must be removed.
1151 psm(4) has gained write(2) support in native operation level.
1152 Arbitrary commands can be written to /dev/psm%d and status can
1153 be read back from it. Therefore, an application is responsible
1154 for status validation and error recovery. It is a no-op in
1155 other operation levels.
1158 Support for KSE threading has been removed from the kernel. To
1159 run legacy applications linked against KSE libmap.conf may
1160 be used. The following libmap.conf may be used to ensure
1161 compatibility with any prior release:
1163 libpthread.so.1 libthr.so.1
1164 libpthread.so.2 libthr.so.2
1165 libkse.so.3 libthr.so.3
1168 The layout of struct vmspace has changed. This affects libkvm
1169 and any executables that link against libkvm and use the
1170 kvm_getprocs() function. In particular, but not exclusively,
1171 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1172 The effects are minimal, but it's advisable to upgrade world
1176 The latest em driver no longer has support in it for the
1177 82575 adapter, this is now moved to the igb driver. The
1178 split was done to make new features that are incompatible
1179 with older hardware easier to do.
1182 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1183 likewise the kernel option is now GEOM_LINUX_LVM.
1186 The default NFS mount mode has changed from UDP to TCP for
1187 increased reliability. If you rely on (insecurely) NFS
1188 mounting across a firewall you may need to update your
1192 Belatedly note the addition of m_collapse for compacting
1196 The fts(3) structures have been changed to use adequate
1197 integer types for their members and so to be able to cope
1198 with huge file trees. The old fts(3) ABI is preserved
1199 through symbol versioning in libc, so third-party binaries
1200 using fts(3) should still work, although they will not take
1201 advantage of the extended types. At the same time, some
1202 third-party software might fail to build after this change
1203 due to unportable assumptions made in its source code about
1204 fts(3) structure members. Such software should be fixed
1205 by its vendor or, in the worst case, in the ports tree.
1206 FreeBSD_version 800015 marks this change for the unlikely
1207 case that a portable fix is impossible.
1210 To upgrade to -current after this date, you must be running
1211 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1212 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1215 The ADAPTIVE_GIANT kernel option has been retired because its
1216 functionality is the default now.
1219 The AT keyboard emulation of sunkbd(4) has been turned on
1220 by default. In order to make the special symbols of the Sun
1221 keyboards driven by sunkbd(4) work under X these now have
1222 to be configured the same way as Sun USB keyboards driven
1223 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1225 Option "XkbLayout" "us"
1226 Option "XkbRules" "xorg"
1227 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1230 It has been decided that it is desirable to provide ABI
1231 backwards compatibility to the FreeBSD 4/5/6 versions of the
1232 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1233 broken with the introduction of PCI domain support (see the
1234 20070930 entry). Unfortunately, this required the ABI of
1235 PCIOCGETCONF to be broken again in order to be able to
1236 provide backwards compatibility to the old version of that
1237 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1238 again. As for prominent ports this affects neither pciutils
1239 nor xorg-server this time, the hal port needs to be rebuilt
1243 The misnamed kthread_create() and friends have been renamed
1244 to kproc_create() etc. Many of the callers already
1245 used kproc_start()..
1246 I will return kthread_create() and friends in a while
1247 with implementations that actually create threads, not procs.
1248 Renaming corresponds with version 800002.
1257 Avoid using make -j when upgrading. While generally safe, there are
1258 sometimes problems using -j to upgrade. If your upgrade fails with
1259 -j, please try again without -j. From time to time in the past there
1260 have been problems using -j with buildworld and/or installworld. This
1261 is especially true when upgrading between "distant" versions (eg one
1262 that cross a major release boundary or several minor releases, or when
1263 several months have passed on the -current branch).
1265 Sometimes, obscure build problems are the result of environment
1266 poisoning. This can happen because the make utility reads its
1267 environment when searching for values for global variables. To run
1268 your build attempts in an "environmental clean room", prefix all make
1269 commands with 'env -i '. See the env(1) manual page for more details.
1271 When upgrading from one major version to another it is generally best
1272 to upgrade to the latest code in the currently installed branch first,
1273 then do an upgrade to the new branch. This is the best-tested upgrade
1274 path, and has the highest probability of being successful. Please try
1275 this approach before reporting problems with a major version upgrade.
1279 When upgrading the boot ZFS pool to a new version, always follow
1282 1.) recompile and reinstall the ZFS boot loader and boot block
1283 (this is part of "make buildworld" and "make installworld")
1285 2.) update the ZFS boot block on your boot drive
1287 The following example updates the ZFS boot block on the first
1288 partition (freebsd-boot) of a GPT partitioned drive ad0:
1289 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1291 Non-boot pools do not need these updates.
1295 If you are updating from a prior version of FreeBSD (even one just
1296 a few days old), you should follow this procedure. It is the most
1297 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1299 make kernel-toolchain
1300 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1301 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1303 To test a kernel once
1304 ---------------------
1305 If you just want to boot a kernel once (because you are not sure
1306 if it works, or if you want to boot a known bad kernel to provide
1307 debugging information) run
1308 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1309 nextboot -k testkernel
1311 To just build a kernel when you know that it won't mess you up
1312 --------------------------------------------------------------
1313 This assumes you are already running a CURRENT system. Replace
1314 ${arch} with the architecture of your machine (e.g. "i386",
1315 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1317 cd src/sys/${arch}/conf
1318 config KERNEL_NAME_HERE
1319 cd ../compile/KERNEL_NAME_HERE
1324 If this fails, go to the "To build a kernel" section.
1326 To rebuild everything and install it on the current system.
1327 -----------------------------------------------------------
1328 # Note: sometimes if you are running current you gotta do more than
1329 # is listed here if you are upgrading from a really old current.
1331 <make sure you have good level 0 dumps>
1333 make kernel KERNCONF=YOUR_KERNEL_HERE
1335 <reboot in single user> [3]
1343 To cross-install current onto a separate partition
1344 --------------------------------------------------
1345 # In this approach we use a separate partition to hold
1346 # current's root, 'usr', and 'var' directories. A partition
1347 # holding "/", "/usr" and "/var" should be about 2GB in
1350 <make sure you have good level 0 dumps>
1353 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1354 <maybe newfs current's root partition>
1355 <mount current's root partition on directory ${CURRENT_ROOT}>
1356 make installworld DESTDIR=${CURRENT_ROOT}
1357 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1358 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1359 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1360 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1361 <reboot into current>
1362 <do a "native" rebuild/install as described in the previous section>
1363 <maybe install compatibility libraries from ports/misc/compat*>
1367 To upgrade in-place from 8.x-stable to current
1368 ----------------------------------------------
1369 <make sure you have good level 0 dumps>
1371 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1373 <reboot in single user> [3]
1380 Make sure that you've read the UPDATING file to understand the
1381 tweaks to various things you need. At this point in the life
1382 cycle of current, things change often and you are on your own
1383 to cope. The defaults can also change, so please read ALL of
1384 the UPDATING entries.
1386 Also, if you are tracking -current, you must be subscribed to
1387 freebsd-current@freebsd.org. Make sure that before you update
1388 your sources that you have read and understood all the recent
1389 messages there. If in doubt, please track -stable which has
1390 much fewer pitfalls.
1392 [1] If you have third party modules, such as vmware, you
1393 should disable them at this point so they don't crash your
1396 [3] From the bootblocks, boot -s, and then do
1401 adjkerntz -i # if CMOS is wall time
1402 Also, when doing a major release upgrade, it is required that
1403 you boot into single user mode to do the installworld.
1405 [4] Note: This step is non-optional. Failure to do this step
1406 can result in a significant reduction in the functionality of the
1407 system. Attempting to do it by hand is not recommended and those
1408 that pursue this avenue should read this file carefully, as well
1409 as the archives of freebsd-current and freebsd-hackers mailing lists
1410 for potential gotchas. The -U option is also useful to consider.
1411 See mergemaster(8) for more information.
1413 [5] Usually this step is a noop. However, from time to time
1414 you may need to do this if you get unknown user in the following
1415 step. It never hurts to do it all the time. You may need to
1416 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1417 install) after the buildworld before this step if you last updated
1418 from current before 20020224 or from -stable before 20020408.
1420 [6] This only deletes old files and directories. Old libraries
1421 can be deleted by "make delete-old-libs", but you have to make
1422 sure that no program is using those libraries anymore.
1424 [8] In order to have a kernel that can run the 4.x binaries needed to
1425 do an installworld, you must include the COMPAT_FREEBSD4 option in
1426 your kernel. Failure to do so may leave you with a system that is
1427 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1428 required to run the 5.x binaries on more recent kernels. And so on
1429 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1431 Make sure that you merge any new devices from GENERIC since the
1432 last time you updated your kernel config file.
1434 [9] When checking out sources, you must include the -P flag to have
1435 cvs prune empty directories.
1437 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1438 "?=" instead of the "=" assignment operator, so that buildworld can
1439 override the CPUTYPE if it needs to.
1441 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1442 not on the command line, or in /etc/make.conf. buildworld will
1443 warn if it is improperly defined.
1446 This file contains a list, in reverse chronological order, of major
1447 breakages in tracking -current. Not all things will be listed here,
1448 and it only starts on October 16, 2004. Updating files can found in
1449 previous releases if your system is older than this.
1451 Copyright information:
1453 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1455 Redistribution, publication, translation and use, with or without
1456 modification, in full or in part, in any form or format of this
1457 document are permitted without further permission from the author.
1459 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1460 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1461 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1462 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1463 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1464 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1465 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1466 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1467 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1468 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1469 POSSIBILITY OF SUCH DAMAGE.
1471 Contact Warner Losh if you have any questions about your use of