1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
13 The broken amd(4) driver has been replaced with esp(4) in the amd64,
14 i386 and pc98 GENERIC kernel configuration files.
17 This commit modifies vfs_register() so that it uses a hash
18 calculation to set vfc_typenum, which is enabled by default.
19 The first time a system is booted after this change, the
20 vfc_typenum values will change for all file systems. The
21 main effect of this is a change to the NFS server file handles
22 for file systems that use vfc_typenum in their fsid, such as ZFS.
23 It will, however, prevent vfc_typenum from changing when file
24 systems are loaded in a different order for subsequent reboots.
25 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
26 until you are ready to remount all NFS clients after a reboot.
29 Bump the shared library version numbers for libraries that
30 do not use symbol versioning, have changed the ABI compared
31 to stable/8 and which shared library version was not bumped.
32 Done as part of 9.0-RELEASE cycle.
35 During the merge of Capsicum features, the fget(9) KPI was modified.
36 This may require the rebuilding of out-of-tree device drivers --
37 issues have been reported specifically with the nVidia device driver.
38 __FreeBSD_version is bumped to 900041.
40 Also, there is a period between 20110811 and 20110814 where the
41 special devices /dev/{stdin,stdout,stderr} did not work correctly.
42 Building world from a kernel during that window may not work.
45 The packet filter (pf) code has been updated to OpenBSD 4.5.
46 You need to update userland tools to be in sync with kernel.
47 This update breaks backward compatibility with earlier pfsync(4)
48 versions. Care must be taken when updating redundant firewall setups.
51 The following sysctls and tunables are retired on x86 platforms:
53 machdep.hlt_logical_cpus
54 The following sysctl is retired:
55 machdep.hyperthreading_allowed
56 The sysctls were supposed to provide a way to dynamically offline and
57 online selected CPUs on x86 platforms, but the implementation has not
58 been reliable especially with SCHED_ULE scheduler.
59 machdep.hyperthreading_allowed tunable is still available to ignore
60 hyperthreading CPUs at OS level.
61 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
62 where X is an APIC ID of a CPU. Be advised, though, that disabling
63 CPUs in non-uniform fashion will result in non-uniform topology and
64 may lead to sub-optimal system performance with SCHED_ULE, which is
68 cpumask_t type is retired and cpuset_t is used in order to describe
72 Changes to ifconfig(8) for dynamic address family detection mandate
73 that you are running a kernel of 20110525 or later. Make sure to
74 follow the update procedure to boot a new kernel before installing
78 Support for sun4v architecture is officially dropped
81 Several KPI breaking changes have been committed to the mii(4) layer,
82 the PHY drivers and consequently some Ethernet drivers using mii(4).
83 This means that miibus.ko and the modules of the affected Ethernet
84 drivers need to be recompiled.
86 Note to kernel developers: Given that the OUI bit reversion problem
87 was fixed as part of these changes all mii(4) commits related to OUIs,
88 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
89 handling, no longer can be merged verbatim to stable/8 and previous
93 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
94 into their kernel configurations along with 'device pci'.
97 The default NFS client is now the new NFS client, so fstype "newnfs"
98 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
99 Although mounts via fstype "nfs" will usually work without userland
100 changes, it is recommended that the mount(8) and mount_nfs(8)
101 commands be rebuilt from sources and that a link to mount_nfs called
102 mount_oldnfs be created. The new client is compiled into the
103 kernel with "options NFSCL" and this is needed for diskless root
104 file systems. The GENERIC kernel configs have been changed to use
105 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
106 To use the regular/old client, you can "mount -t oldnfs ...". For
107 a diskless root file system, you must also include a line like:
109 vfs.root.mountfrom="oldnfs:"
111 in the boot/loader.conf on the root fs on the NFS server to make
112 a diskless root fs use the old client.
115 The GENERIC kernels for all architectures now default to the new
116 CAM-based ATA stack. It means that all legacy ATA drivers were
117 removed and replaced by respective CAM drivers. If you are using
118 ATA device names in /etc/fstab or other places, make sure to update
119 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
120 where 'Y's are the sequential numbers starting from zero for each type
121 in order of detection, unless configured otherwise with tunables,
122 see cam(4)). There will be symbolic links created in /dev/ to map
123 old adX devices to the respective adaY. They should provide basic
124 compatibility for file systems mounting in most cases, but they do
125 not support old user-level APIs and do not have respective providers
126 in GEOM. Consider using updated management tools with new device names.
128 It is possible to load devices ahci, ata, siis and mvs as modules,
129 but option ATA_CAM should remain in kernel configuration to make ata
130 module work as CAM driver supporting legacy ATA controllers. Device ata
131 still can be used in modular fashion (atacore + ...). Modules atadisk
132 and atapi* are not used and won't affect operation in ATA_CAM mode.
133 Note that to use CAM-based ATA kernel should include CAM devices
134 scbus, pass, da (or explicitly ada), cd and optionally others. All of
135 them are parts of the cam module.
137 ataraid(4) functionality is now supported by the RAID GEOM class.
138 To use it you can load geom_raid kernel module and use graid(8) tool
139 for management. Instead of /dev/arX device names, use /dev/raid/rX.
141 No kernel config options or code have been removed, so if a problem
142 arises, please report it and optionally revert to the old ATA stack.
143 In order to do it you can remove from the kernel config:
148 , and instead add back:
149 device atadisk # ATA disk drives
150 device ataraid # ATA RAID drives
151 device atapicd # ATAPI CDROM drives
152 device atapifd # ATAPI floppy drives
153 device atapist # ATAPI tape drives
156 The default NFS server has been changed to the new server, which
157 was referred to as the experimental server. If you need to switch
158 back to the old NFS server, you must now put the "-o" option on
159 both the mountd and nfsd commands. This can be done using the
160 mountd_flags and nfs_server_flags rc.conf variables until an
161 update to the rc scripts is committed, which is coming soon.
164 The GNU Objective-C runtime library (libobjc), and other Objective-C
165 related components have been removed from the base system. If you
166 require an Objective-C library, please use one of the available ports.
169 ath(4) has been split into bus- and device- modules. if_ath contains
170 the HAL, the TX rate control and the network device code. if_ath_pci
171 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
172 contains the AHB glue. Users need to load both if_ath_pci and if_ath
173 in order to use ath on everything else.
175 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
176 need to load if_ath and if_ath_pci for ath(4) operation.
179 As part of the replacement of sysinstall, the process of building
180 release media has changed significantly. For details, please re-read
181 release(7), which has been updated to reflect the new build process.
184 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
185 is the last available version under GPLv2. It brings a number of new
186 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
187 4.1 and SSE 4.2), better support for powerpc64, a number of new
188 directives, and lots of other small improvements. See the ChangeLog
189 file in contrib/binutils for the full details.
192 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
193 compliant, and will now use half of hash for authentication.
194 This will break interoperability with all stacks (including all
195 actual FreeBSD versions) who implement
196 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
198 The only workaround with such peers is to use another HMAC
199 algorithm for IPsec ("phase 2") authentication.
202 Remove the uio_yield prototype and symbol. This function has
203 been misnamed since it was introduced and should not be
204 globally exposed with this name. The equivalent functionality
205 is now available using kern_yield(curthread->td_user_pri).
206 The function remains undocumented.
209 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
210 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
211 for scalar sysctls is defined but disabled. Code that needs
212 UQUAD to pass the type checking that must compile on older
213 systems where the define is not present can check against
214 __FreeBSD_version >= 900030.
216 The system dialog(1) has been replaced with a new version previously
217 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
218 with the previous version, but the libdialog associated with it has
219 a largely incompatible API. As such, the original version of libdialog
220 will be kept temporarily as libodialog, until its base system consumers
221 are replaced or updated. Bump __FreeBSD_version to 900030.
224 If you are trying to run make universe on a -stable system, and you get
225 the following warning:
226 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
227 unknown. config(8) likely too old."
228 or something similar to it, then you must upgrade your -stable system
229 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
230 stable/8) or build the config from the latest stable/8 branch and
231 install it on your system.
233 Prior to this date, building a current universe on 8-stable system from
234 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
235 error in the first kernel build phase. A new config on those old
236 systems will fix that problem for older versions of -current.
239 The TCP stack has been modified to allow Khelp modules to interact with
240 it via helper hook points and store per-connection data in the TCP
241 control block. Bump __FreeBSD_version to 900029. User space tools that
242 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
246 Generic IEEE 802.3 annex 31B full duplex flow control support has been
247 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
248 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
249 to take advantage of it instead of using custom implementations. This
250 means that these drivers now no longer unconditionally advertise
251 support for flow control but only do so if flow control is a selected
252 media option. This was implemented in the generic support that way in
253 order to allow flow control to be switched on and off via ifconfig(8)
254 with the PHY specific default to typically off in order to protect
255 from unwanted effects. Consequently, if you used flow control with
256 one of the above mentioned drivers you now need to explicitly enable
258 ifconfig bge0 media auto mediaopt flowcontrol
260 Along with the above mentioned changes generic support for setting
261 1000baseT master mode also has been added and brgphy(4), ciphy(4),
262 e1000phy(4) as well as ip1000phy(4) have been converted to take
263 advantage of it. This means that these drivers now no longer take the
264 link0 parameter for selecting master mode but the master media option
265 has to be used instead, for example like in the following:
266 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
268 Selection of master mode now is also available with all other PHY
269 drivers supporting 1000baseT.
272 The TCP stack has received a significant update to add support for
273 modularised congestion control and generally improve the clarity of
274 congestion control decisions. Bump __FreeBSD_version to 900025. User
275 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
276 sockstat) need to be recompiled.
279 The man(1) utility has been replaced by a new version that no longer
280 uses /etc/manpath.config. Please consult man.conf(5) for how to
281 migrate local entries to the new format.
284 The copyright strings printed by login(1) and sshd(8) at the time of a
285 new connection have been removed to follow other operating systems and
289 A workaround for a fixed ld bug has been removed in kernel code,
290 so make sure that your system ld is built from sources after
291 revision 210245 from 2010-07-19 (r211583 if building head kernel
292 on stable/8, r211584 for stable/7; both from 2010-08-21).
293 A symptom of incorrect ld version is different addresses for
294 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
297 The $ipv6_prefer variable in rc.conf(5) has been split into
298 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
300 The $ip6addrctl_policy is a variable to choose a pre-defined
301 address selection policy set by ip6addrctl(8). A value
302 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
305 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
306 flag (see an entry of 20090926) is set on an interface with no
307 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
308 security reason. If you want IPv6 link-local address on all
309 interfaces by default, set this to "YES".
311 The old ipv6_prefer="YES" is equivalent to
312 ipv6_activate_all_interfaces="YES" and
313 ip6addrctl_policy="ipv6_prefer".
316 DTrace has grown support for userland tracing. Due to this, DTrace is
317 now i386 and amd64 only.
318 dtruss(1) is now installed by default on those systems and a new
319 kernel module is needed for userland tracing: fasttrap.
320 No changes to your kernel config file are necessary to enable
321 userland tracing, but you might consider adding 'STRIP=' and
322 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
323 to have informative userland stack traces in DTrace (ustack).
326 The acpi_aiboost(4) driver has been removed in favor of the new
327 aibs(4) driver. You should update your kernel configuration file.
330 BSD grep has been imported to the base system and it is built by
331 default. It is completely BSD licensed, highly GNU-compatible, uses
332 less memory than its GNU counterpart and has a small codebase.
333 However, it is slower than its GNU counterpart, which is mostly
334 noticeable for larger searches, for smaller ones it is measurable
335 but not significant. The reason is complex, the most important factor
336 is that we lack a modern and efficient regex library and GNU
337 overcomes this by optimizing the searches internally. Future work
338 on improving the regex performance is planned, for the meantime,
339 users that need better performance, can build GNU grep instead by
340 setting the WITH_GNU_GREP knob.
343 Due to the import of powerpc64 support, all existing powerpc kernel
344 configuration files must be updated with a machine directive like this:
345 machine powerpc powerpc
347 In addition, an updated config(8) is required to build powerpc kernels
351 A new version of ZFS (version 15) has been merged to -HEAD.
352 This version uses a python library for the following subcommands:
353 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
354 For full functionality of these commands the following port must
355 be installed: sysutils/py-zfs
358 'vm_page's are now hashed by physical address to an array of mutexes.
359 Currently this is only used to serialize access to hold_count. Over
360 time the page queue mutex will be peeled away. This changes the size
361 of pmap on every architecture. And requires all callers of vm_page_hold
362 and vm_page_unhold to be updated.
365 WITH_CTF can now be specified in src.conf (not recommended, there
366 are some problems with static executables), make.conf (would also
367 affect ports which do not use GNU make and do not override the
368 compile targets) or in the kernel config (via "makeoptions
370 When WITH_CTF was specified there before this was silently ignored,
371 so make sure that WITH_CTF is not used in places which could lead
372 to unwanted behavior.
375 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
376 to allow 32-bit compatibility on non-x86 platforms. All kernel
377 configurations on amd64 and ia64 platforms using these options must
378 be modified accordingly.
381 The utmp user accounting database has been replaced with utmpx,
382 the user accounting interface standardized by POSIX.
383 Unfortunately the semantics of utmp and utmpx don't match,
384 making it practically impossible to support both interfaces.
385 The user accounting database is used by tools like finger(1),
386 last(1), talk(1), w(1) and ac(8).
388 All applications in the base system use utmpx. This means only
389 local binaries (e.g. from the ports tree) may still use these
390 utmp database files. These applications must be rebuilt to make
393 After the system has been upgraded, it is safe to remove the old
394 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
395 assuming their contents is of no importance anymore. Old wtmp
396 databases can only be used by last(1) and ac(8) after they have
397 been converted to the new format using wtmpcvt(1).
400 Introduce the kernel thread "deadlock resolver" (which can be enabled
401 via the DEADLKRES option, see NOTES for more details) and the
402 sleepq_type() function for sleepqueues.
405 The rc.firewall and rc.firewall6 were unified, and
406 rc.firewall6 and rc.d/ip6fw were removed.
407 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
408 variables are obsoleted. Instead, the following new rc
409 variables are added to rc.d/ipfw:
411 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
412 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
413 firewall_simple_onet_ipv6, firewall_trusted_ipv6
415 The meanings correspond to the relevant IPv4 variables.
421 The default terminal emulation for syscons(4) has been changed
422 from cons25 to xterm on all platforms except pc98. This means
423 that the /etc/ttys file needs to be updated to ensure correct
424 operation of applications on the console.
426 The terminal emulation style can be toggled per window by using
427 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
428 options can be used to change the compile-time default back to
431 To prevent graphical artifacts, make sure the TERM environment
432 variable is set to match the terminal emulation that is being
433 performed by syscons(4).
436 The layout of the structure ieee80211req_scan_result has changed.
437 Applications that require wireless scan results (e.g. ifconfig(8))
438 from net80211 need to be recompiled.
440 Applications such as wpa_supplicant(8) may require a full world
441 build without using NO_CLEAN in order to get synchronized with the
445 The iwn(4) driver has been updated to support the 5000 and 5150 series.
446 There's one kernel module for each firmware. Adding "device iwnfw"
447 to the kernel configuration file means including all three firmware
448 images inside the kernel. If you want to include just the one for
449 your wireless card, use the the devices iwn4965fw, iwn5000fw or
453 The rc.d/network_ipv6, IPv6 configuration script has been integrated
454 into rc.d/netif. The changes are the following:
456 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
457 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
458 Note that both variables need the "inet6" keyword at the head.
460 Do not set $ipv6_network_interfaces manually if you do not
461 understand what you are doing. It is not needed in most cases.
463 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
466 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
467 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
469 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
470 all configured interfaces have "inet6 accept_rtadv" in the
471 $ifconfig_IF_ipv6. These are for backward compatibility.
473 3. A new variable $ipv6_prefer has been added. If NO, IPv6
474 functionality of interfaces with no corresponding
475 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
476 and the default address selection policy of ip6addrctl(8)
477 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
478 Note that if you want to configure IPv6 functionality on the
479 disabled interfaces after boot, first you need to clear the flag by
480 using ifconfig(8) like:
482 ifconfig em0 inet6 -ifdisabled
484 If YES, the default address selection policy is set as
487 The default value of $ipv6_prefer is NO.
489 4. If your system need to receive Router Advertisement messages,
490 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
491 scripts automatically invoke rtsol(8) when the interface becomes
492 UP. The Router Advertisement messages are used for SLAAC
493 (State-Less Address AutoConfiguration).
496 802.11s D3.03 support was committed. This is incompatible with the
497 previous code, which was based on D3.0.
500 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
501 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
502 control whether accepting Router Advertisement messages or not.
503 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
504 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
505 The ifconfig(8) utility now supports these flags.
508 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
509 mount(8) and -a option for df(1) to see them.
512 The old tunable hw.bus.devctl_disable has been superseded by
513 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
514 replaced by hw.bus.devctl_queue=0. The default for this new tunable
518 Remove the option STOP_NMI. The default action is now to use NMI only
519 for KDB via the newly introduced function stop_cpus_hard() and
520 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
523 The stable/8 branch created in subversion. This corresponds to the
524 RELENG_8 branch in CVS.
527 Bump the shared library version numbers for all libraries that do not
528 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
529 __FreeBSD_version to 800105.
532 Due to changes in the implementation of virtual network stack support,
533 all network-related kernel modules must be recompiled. As this change
534 breaks the ABI, bump __FreeBSD_version to 800104.
537 The TOE interface to the TCP syncache has been modified to remove
538 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
539 The cxgb driver is the only TOE consumer affected by this change, and
540 needs to be recompiled along with the kernel. As this change breaks
541 the ABI, bump __FreeBSD_version to 800103.
544 Padding has been added to struct tcpcb, sackhint and tcpstat in
545 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
546 maintaining the ABI. However, this change breaks the ABI, so bump
547 __FreeBSD_version to 800102. User space tools that rely on the size of
548 any of these structs (e.g. sockstat) need to be recompiled.
551 The NFS_LEGACYRPC option has been removed along with the old kernel
552 RPC implementation that this option selected. Kernel configurations
553 may need to be adjusted.
556 The network interface device nodes at /dev/net/<interface> have been
557 removed. All ioctl operations can be performed the normal way using
558 routing sockets. The kqueue functionality can generally be replaced
559 with routing sockets.
562 The documentation from the FreeBSD Documentation Project (Handbook,
563 FAQ, etc.) is now installed via packages by sysinstall(8) and under
564 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
567 The ABI of various structures related to the SYSV IPC API have been
568 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
569 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
573 Layout of struct vnet has changed as routing related variables were
574 moved to their own Vimage module. Modules need to be recompiled. Bump
575 __FreeBSD_version to 800099.
578 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
579 respectively. As long as no more than 16 groups per process are used,
580 no changes should be visible. When more than 16 groups are used, old
581 binaries may fail if they call getgroups() or getgrouplist() with
582 statically sized storage. Recompiling will work around this, but
583 applications should be modified to use dynamically allocated storage
584 for group arrays as POSIX.1-2008 does not cap an implementation's
585 number of supported groups at NGROUPS_MAX+1 as previous versions did.
587 NFS and portalfs mounts may also be affected as the list of groups is
588 truncated to 16. Users of NFS who use more than 16 groups, should
589 take care that negative group permissions are not used on the exported
590 file systems as they will not be reliable unless a GSSAPI based
591 authentication method is used.
594 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
595 option compiles in the support for adaptive spinning for lockmgrs
596 which want to enable it. The lockinit() function now accepts the flag
597 LK_ADAPTIVE in order to make the lock object subject to adaptive
598 spinning when both held in write and read mode.
601 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
602 changed. User applications that use this ioctl need to be rebuilt.
605 The layout of struct thread has changed. Kernel and modules need to
609 The layout of structs ifnet, domain, protosw and vnet_net has changed.
610 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
613 window(1) has been removed from the base system. It can now be
614 installed from ports. The port is called misc/window.
617 The way we are storing and accessing `routing table' entries has
618 changed. Programs reading the FIB, like netstat, need to be
622 A new netisr implementation has been added for FreeBSD 8. Network
623 file system modules, such as igmp, ipdivert, and others, should be
625 Bump __FreeBSD_version to 800096.
628 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
632 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
633 Bump __FreeBSD_version to 800094.
636 Add mnt_xflag field to 'struct mount'. File system modules need to be
638 Bump __FreeBSD_version to 800093.
641 The compiling option ADAPTIVE_SX has been retired while it has been
642 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
643 The KPI for sx_init_flags() changes as accepting flags:
644 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
645 been introduced in order to handle the reversed logic.
646 Bump __FreeBSD_version to 800092.
649 Add support for hierarchical jails. Remove global securelevel.
650 Bump __FreeBSD_version to 800091.
653 The layout of struct vnet_net has changed, therefore modules
655 Bump __FreeBSD_version to 800090.
658 The newly imported zic(8) produces a new format in the output. Please
659 run tzsetup(8) to install the newly created data to /etc/localtime.
662 The sysctl tree for the usb stack has renamed from hw.usb2.* to
663 hw.usb.* and is now consistent again with previous releases.
666 802.11 monitor mode support was revised and driver api's were changed.
667 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
668 of DLT_IEEE802_11. No user-visible data structures were changed but
669 applications that use DLT_IEEE802_11 may require changes.
670 Bump __FreeBSD_version to 800088.
673 The layout of the following structs has changed: sysctl_oid,
674 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
675 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
676 panics may be experienced. World rebuild is required for
677 correctly checking networking state from userland.
678 Bump __FreeBSD_version to 800085.
681 MLDv2 and Source-Specific Multicast (SSM) have been merged
682 to the IPv6 stack. VIMAGE hooks are in but not yet used.
683 The implementation of SSM within FreeBSD's IPv6 stack closely
684 follows the IPv4 implementation.
686 For kernel developers:
688 * The most important changes are that the ip6_output() and
689 ip6_input() paths no longer take the IN6_MULTI_LOCK,
690 and this lock has been downgraded to a non-recursive mutex.
692 * As with the changes to the IPv4 stack to support SSM, filtering
693 of inbound multicast traffic must now be performed by transport
694 protocols within the IPv6 stack. This does not apply to TCP and
695 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
697 * The KPIs used by IPv6 multicast are similar to those used by
698 the IPv4 stack, with the following differences:
699 * im6o_mc_filter() is analogous to imo_multicast_filter().
700 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
701 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
702 * IN6_LOOKUP_MULTI() has been deprecated and removed.
703 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
704 for MLDv1 have an additional 'timer' argument which is used to
705 jitter the initial membership report for the solicited-node
706 multicast membership on-link.
707 * This is not strictly needed for MLDv2, which already jitters
708 its report transmissions. However, the 'timer' argument is
709 preserved in case MLDv1 is active on the interface.
711 * The KAME linked-list based IPv6 membership implementation has
712 been refactored to use a vector similar to that used by the IPv4
714 Code which maintains a list of its own multicast memberships
715 internally, e.g. carp, has been updated to reflect the new
718 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
719 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
720 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
721 implementation constraint which needs to be addressed in HEAD.
723 For application developers:
725 * The changes are broadly similar to those made for the IPv4
728 * The use of IPv4 and IPv6 multicast socket options on the same
729 socket, using mapped addresses, HAS NOT been tested or supported.
731 * There are a number of issues with the implementation of various
732 IPv6 multicast APIs which need to be resolved in the API surface
733 before the implementation is fully compatible with KAME userland
734 use, and these are mostly to do with interface index treatment.
736 * The literature available discusses the use of either the delta / ASM
737 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
738 using setsourcefilter(3)/getsourcefilter(3). For more information
739 please refer to RFC 3768, 'Socket Interface Extensions for
740 Multicast Source Filters'.
742 * Applications which use the published RFC 3678 APIs should be fine.
744 For systems administrators:
746 * The mtest(8) utility has been refactored to support IPv6, in
747 addition to IPv4. Interface addresses are no longer accepted
748 as arguments, their names must be used instead. The utility
749 will map the interface name to its first IPv4 address as
750 returned by getifaddrs(3).
752 * The ifmcstat(8) utility has also been updated to print the MLDv2
753 endpoint state and source filter lists via sysctl(3).
755 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
756 loopback of IPv6 multicast datagrams by default; it defaults to 1
757 to preserve the existing behaviour. Disabling multicast loopback is
758 recommended for optimal system performance.
760 * The IPv6 MROUTING code has been changed to examine this sysctl
761 instead of attempting to perform a group lookup before looping
762 back forwarded datagrams.
764 Bump __FreeBSD_version to 800084.
767 Implement low-level Bluetooth HCI API.
768 Bump __FreeBSD_version to 800083.
771 The layout of struct malloc_type, used by modules to register new
772 memory allocation types, has changed. Most modules will need to
773 be rebuilt or panics may be experienced.
774 Bump __FreeBSD_version to 800081.
777 Anticipate overflowing inp_flags - add inp_flags2.
778 This changes most offsets in inpcb, so checking v4 connection
779 state will require a world rebuild.
780 Bump __FreeBSD_version to 800080.
783 Add an llentry to struct route and struct route_in6. Modules
784 embedding a struct route will need to be recompiled.
785 Bump __FreeBSD_version to 800079.
788 The size of rt_metrics_lite and by extension rtentry has changed.
789 Networking administration apps will need to be recompiled.
790 The route command now supports show as an alias for get, weighting
791 of routes, sticky and nostick flags to alter the behavior of stateful
793 Bump __FreeBSD_version to 800078.
796 Do not use Giant for kbdmux(4) locking. This is wrong and
797 apparently causing more problems than it solves. This will
798 re-open the issue where interrupt handlers may race with
799 kbdmux(4) in polling mode. Typical symptoms include (but
800 not limited to) duplicated and/or missing characters when
801 low level console functions (such as gets) are used while
802 interrupts are enabled (for example geli password prompt,
803 mountroot prompt etc.). Disabling kbdmux(4) may help.
806 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
807 kernel modules referencing any of the above need to be recompiled.
808 Bump __FreeBSD_version to 800075.
811 GEOM_PART has become the default partition slicer for storage devices,
812 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
813 introduces some changes:
815 MSDOS/EBR: the devices created from MSDOS extended partition entries
816 (EBR) can be named differently than with GEOM_MBR and are now symlinks
817 to devices with offset-based names. fstabs may need to be modified.
819 BSD: the "geometry does not match label" warning is harmless in most
820 cases but it points to problems in file system misalignment with
821 disk geometry. The "c" partition is now implicit, covers the whole
822 top-level drive and cannot be (mis)used by users.
824 General: Kernel dumps are now not allowed to be written to devices
825 whose partition types indicate they are meant to be used for file
826 systems (or, in case of MSDOS partitions, as something else than
829 Most of these changes date approximately from 200812.
832 The uscanner(4) driver has been removed from the kernel. This follows
833 Linux removing theirs in 2.6 and making libusb the default interface
837 The multicast forwarding code has been cleaned up. netstat(1)
838 only relies on KVM now for printing bandwidth upcall meters.
839 The IPv4 and IPv6 modules are split into ip_mroute_mod and
840 ip6_mroute_mod respectively. The config(5) options for statically
841 compiling this code remain the same, i.e. 'options MROUTING'.
844 Support for the IFF_NEEDSGIANT network interface flag has been
845 removed, which means that non-MPSAFE network device drivers are no
846 longer supported. In particular, if_ar, if_sr, and network device
847 drivers from the old (legacy) USB stack can no longer be built or
851 POSIX.1 Native Language Support (NLS) has been enabled in libc and
852 a bunch of new language catalog files have also been added.
853 This means that some common libc messages are now localized and
854 they depend on the LC_MESSAGES environmental variable.
857 The k8temp(4) driver has been renamed to amdtemp(4) since
858 support for Family 10 and Family 11 CPU families was added.
861 IGMPv3 and Source-Specific Multicast (SSM) have been merged
862 to the IPv4 stack. VIMAGE hooks are in but not yet used.
864 For kernel developers, the most important changes are that the
865 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
866 and this lock has been downgraded to a non-recursive mutex.
868 Transport protocols (UDP, Raw IP) are now responsible for filtering
869 inbound multicast traffic according to group membership and source
870 filters. The imo_multicast_filter() KPI exists for this purpose.
871 Transports which do not use multicast (SCTP, TCP) already reject
872 multicast by default. Forwarding and receive performance may improve
873 as a mutex acquisition is no longer needed in the ip_input()
874 low-level input path. in_addmulti() and in_delmulti() are shimmed
875 to new KPIs which exist to support SSM in-kernel.
877 For application developers, it is recommended that loopback of
878 multicast datagrams be disabled for best performance, as this
879 will still cause the lock to be taken for each looped-back
880 datagram transmission. The net.inet.ip.mcast.loop sysctl may
881 be tuned to 0 to disable loopback by default; it defaults to 1
882 to preserve the existing behaviour.
884 For systems administrators, to obtain best performance with
885 multicast reception and multiple groups, it is always recommended
886 that a card with a suitably precise hash filter is used. Hash
887 collisions will still result in the lock being taken within the
888 transport protocol input path to check group membership.
890 If deploying FreeBSD in an environment with IGMP snooping switches,
891 it is recommended that the net.inet.igmp.sendlocal sysctl remain
892 enabled; this forces 224.0.0.0/24 group membership to be announced
895 The size of 'struct igmpstat' has changed; netstat needs to be
896 recompiled to reflect this.
897 Bump __FreeBSD_version to 800070.
900 libusb20.so.1 is now installed as libusb.so.1 and the ports system
901 updated to use it. This requires a buildworld/installworld in order to
902 update the library and dependencies (usbconfig, etc). Its advisable to
903 rebuild all ports which uses libusb. More specific directions are given
904 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
905 libusb are no longer required and can be removed.
908 A workaround is committed to allow the creation of System V shared
909 memory segment of size > 2 GB on the 64-bit architectures.
910 Due to a limitation of the existing ABI, the shm_segsz member
911 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
912 wrong for large segments. Note that limits must be explicitly
913 raised to allow such segments to be created.
916 The layout of struct ifnet has changed, requiring a rebuild of all
917 network device driver modules.
920 The /dev handling for the new USB stack has changed, a
921 buildworld/installworld is required for libusb20.
924 The new USB2 stack has now been permanently moved in and all kernel and
925 module names reverted to their previous values (eg, usb, ehci, ohci,
926 ums, ...). The old usb stack can be compiled in by prefixing the name
927 with the letter 'o', the old usb modules have been removed.
928 Updating entry 20090216 for xorg and 20090215 for libmap may still
932 The rc.conf(5) option if_up_delay has been renamed to
933 defaultroute_delay to better reflect its purpose. If you have
934 customized this setting in /etc/rc.conf you need to update it to
938 xorg 7.4 wants to configure its input devices via hald which does not
939 yet work with USB2. If the keyboard/mouse does not work in xorg then
941 Option "AllowEmptyInput" "off"
942 to your ServerLayout section. This will cause X to use the configured
943 kbd and mouse sections from your xorg.conf.
946 The GENERIC kernels for all architectures now default to the new USB2
947 stack. No kernel config options or code have been removed so if a
948 problem arises please report it and optionally revert to the old USB
949 stack. If you are loading USB kernel modules or have a custom kernel
950 that includes GENERIC then ensure that usb names are also changed over,
951 eg uftdi -> usb2_serial_ftdi.
953 Older programs linked against the ports libusb 0.1 need to be
954 redirected to the new stack's libusb20. /etc/libmap.conf can
956 # Map old usb library to new one for usb2 stack
957 libusb-0.1.so.8 libusb20.so.1
960 All USB ethernet devices now attach as interfaces under the name ueN
961 (eg. ue0). This is to provide a predictable name as vendors often
962 change usb chipsets in a product without notice.
965 The ichsmb(4) driver has been changed to require SMBus slave
966 addresses be left-justified (xxxxxxx0b) rather than right-justified.
967 All of the other SMBus controller drivers require left-justified
968 slave addresses, so this change makes all the drivers provide the
972 INET6 statistics (struct ip6stat) was updated.
973 netstat(1) needs to be recompiled.
976 NTFS has been removed from GENERIC kernel on amd64 to match
977 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
978 will load ntfs.ko module automatically when NTFS support is
979 actually needed, unless ntfs.ko is not installed or security
980 level prohibits loading kernel modules. If either is the case,
981 "options NTFS" has to be added into kernel config.
984 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
985 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
986 800061. User space tools that rely on the size of struct tcpcb in
987 tcp_var.h (e.g. sockstat) need to be recompiled.
990 ng_tty(4) module updated to match the new TTY subsystem.
991 Due to API change, user-level applications must be updated.
992 New API support added to mpd5 CVS and expected to be present
993 in next mpd5.3 release.
996 With __FreeBSD_version 800060 the makefs tool is part of
997 the base system (it was a port).
1000 The afdata and ifnet locks have been changed from mutexes to
1001 rwlocks, network modules will need to be re-compiled.
1004 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1005 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1006 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1007 architecture and 40 bytes on 64-bit architecture. The userland
1008 applications "arp" and "ndp" have been updated accordingly.
1009 The output from "netstat -r" shows only routing entries and
1010 none of the L2 information.
1013 __FreeBSD_version 800057 marks the switchover from the
1014 binary ath hal to source code. Users must add the line:
1016 options AH_SUPPORT_AR5416
1018 to their kernel config files when specifying:
1022 The ath_hal module no longer exists; the code is now compiled
1023 together with the driver in the ath module. It is now
1024 possible to tailor chip support (i.e. reduce the set of chips
1025 and thereby the code size); consult ath_hal(4) for details.
1028 __FreeBSD_version 800054 adds memory barriers to
1029 <machine/atomic.h>, new interfaces to ifnet to facilitate
1030 multiple hardware transmit queues for cards that support
1031 them, and a lock-less ring-buffer implementation to
1032 enable drivers to more efficiently manage queueing of
1036 A new version of ZFS (version 13) has been merged to -HEAD.
1037 This version has zpool attribute "listsnapshots" off by
1038 default, which means "zfs list" does not show snapshots,
1039 and is the same as Solaris behavior.
1042 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1045 The uhci, ohci, ehci and slhci USB Host controller drivers have
1046 been put into separate modules. If you load the usb module
1047 separately through loader.conf you will need to load the
1048 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1049 controller add the following to loader.conf:
1055 The ABI used by the PMC toolset has changed. Please keep
1056 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1060 atapci kernel module now includes only generic PCI ATA
1061 driver. AHCI driver moved to ataahci kernel module.
1062 All vendor-specific code moved into separate kernel modules:
1063 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1064 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1065 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1066 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1069 The TTY subsystem of the kernel has been replaced by a new
1070 implementation, which provides better scalability and an
1071 improved driver model. Most common drivers have been migrated to
1072 the new TTY subsystem, while others have not. The following
1073 drivers have not yet been ported to the new TTY layer:
1076 cy, digi, rc, rp, sio
1082 ng_h4, ng_tty, ppp, sl, snp
1084 Adding these drivers to your kernel configuration file shall
1085 cause compilation to fail.
1088 ntpd has been upgraded to 4.2.4p5.
1091 OpenSSH has been upgraded to 5.1p1.
1093 For many years, FreeBSD's version of OpenSSH preferred DSA
1094 over RSA for host and user authentication keys. With this
1095 upgrade, we've switched to the vendor's default of RSA over
1096 DSA. This may cause upgraded clients to warn about unknown
1097 host keys even for previously known hosts. Users should
1098 follow the usual procedure for verifying host keys before
1099 accepting the RSA key.
1101 This can be circumvented by setting the "HostKeyAlgorithms"
1102 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1105 Please note that the sequence of keys offered for
1106 authentication has been changed as well. You may want to
1107 specify IdentityFile in a different order to revert this
1111 The sio(4) driver has been removed from the i386 and amd64
1112 kernel configuration files. This means uart(4) is now the
1113 default serial port driver on those platforms as well.
1115 To prevent collisions with the sio(4) driver, the uart(4) driver
1116 uses different names for its device nodes. This means the
1117 onboard serial port will now most likely be called "ttyu0"
1118 instead of "ttyd0". You may need to reconfigure applications to
1119 use the new device names.
1121 When using the serial port as a boot console, be sure to update
1122 /boot/device.hints and /etc/ttys before booting the new kernel.
1123 If you forget to do so, you can still manually specify the hints
1124 at the loader prompt:
1126 set hint.uart.0.at="isa"
1127 set hint.uart.0.port="0x3F8"
1128 set hint.uart.0.flags="0x10"
1129 set hint.uart.0.irq="4"
1133 The gpt(8) utility has been removed. Use gpart(8) to partition
1137 The version that Linuxulator emulates was changed from 2.4.2
1138 to 2.6.16. If you experience any problems with Linux binaries
1139 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1140 if it fixes the problem contact emulation mailing list.
1143 ISDN4BSD (I4B) was removed from the src tree. You may need to
1144 update a your kernel configuration and remove relevant entries.
1147 I have checked in code to support multiple routing tables.
1148 See the man pages setfib(1) and setfib(2).
1149 This is a hopefully backwards compatible version,
1150 but to make use of it you need to compile your kernel
1151 with options ROUTETABLES=2 (or more up to 16).
1154 The 802.11 wireless support was redone to enable multi-bss
1155 operation on devices that are capable. The underlying device
1156 is no longer used directly but instead wlanX devices are
1157 cloned with ifconfig. This requires changes to rc.conf files.
1158 For example, change:
1159 ifconfig_ath0="WPA DHCP"
1162 ifconfig_wlan0="WPA DHCP"
1163 see rc.conf(5) for more details. In addition, mergemaster of
1164 /etc/rc.d is highly recommended. Simultaneous update of userland
1165 and kernel wouldn't hurt either.
1167 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1168 modules were merged into the base wlan module. All references
1169 to these modules (e.g. in kernel config files) must be removed.
1172 psm(4) has gained write(2) support in native operation level.
1173 Arbitrary commands can be written to /dev/psm%d and status can
1174 be read back from it. Therefore, an application is responsible
1175 for status validation and error recovery. It is a no-op in
1176 other operation levels.
1179 Support for KSE threading has been removed from the kernel. To
1180 run legacy applications linked against KSE libmap.conf may
1181 be used. The following libmap.conf may be used to ensure
1182 compatibility with any prior release:
1184 libpthread.so.1 libthr.so.1
1185 libpthread.so.2 libthr.so.2
1186 libkse.so.3 libthr.so.3
1189 The layout of struct vmspace has changed. This affects libkvm
1190 and any executables that link against libkvm and use the
1191 kvm_getprocs() function. In particular, but not exclusively,
1192 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1193 The effects are minimal, but it's advisable to upgrade world
1197 The latest em driver no longer has support in it for the
1198 82575 adapter, this is now moved to the igb driver. The
1199 split was done to make new features that are incompatible
1200 with older hardware easier to do.
1203 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1204 likewise the kernel option is now GEOM_LINUX_LVM.
1207 The default NFS mount mode has changed from UDP to TCP for
1208 increased reliability. If you rely on (insecurely) NFS
1209 mounting across a firewall you may need to update your
1213 Belatedly note the addition of m_collapse for compacting
1217 The fts(3) structures have been changed to use adequate
1218 integer types for their members and so to be able to cope
1219 with huge file trees. The old fts(3) ABI is preserved
1220 through symbol versioning in libc, so third-party binaries
1221 using fts(3) should still work, although they will not take
1222 advantage of the extended types. At the same time, some
1223 third-party software might fail to build after this change
1224 due to unportable assumptions made in its source code about
1225 fts(3) structure members. Such software should be fixed
1226 by its vendor or, in the worst case, in the ports tree.
1227 FreeBSD_version 800015 marks this change for the unlikely
1228 case that a portable fix is impossible.
1231 To upgrade to -current after this date, you must be running
1232 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1233 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1236 The ADAPTIVE_GIANT kernel option has been retired because its
1237 functionality is the default now.
1240 The AT keyboard emulation of sunkbd(4) has been turned on
1241 by default. In order to make the special symbols of the Sun
1242 keyboards driven by sunkbd(4) work under X these now have
1243 to be configured the same way as Sun USB keyboards driven
1244 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1246 Option "XkbLayout" "us"
1247 Option "XkbRules" "xorg"
1248 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1251 It has been decided that it is desirable to provide ABI
1252 backwards compatibility to the FreeBSD 4/5/6 versions of the
1253 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1254 broken with the introduction of PCI domain support (see the
1255 20070930 entry). Unfortunately, this required the ABI of
1256 PCIOCGETCONF to be broken again in order to be able to
1257 provide backwards compatibility to the old version of that
1258 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1259 again. As for prominent ports this affects neither pciutils
1260 nor xorg-server this time, the hal port needs to be rebuilt
1264 The misnamed kthread_create() and friends have been renamed
1265 to kproc_create() etc. Many of the callers already
1266 used kproc_start()..
1267 I will return kthread_create() and friends in a while
1268 with implementations that actually create threads, not procs.
1269 Renaming corresponds with version 800002.
1278 Avoid using make -j when upgrading. While generally safe, there are
1279 sometimes problems using -j to upgrade. If your upgrade fails with
1280 -j, please try again without -j. From time to time in the past there
1281 have been problems using -j with buildworld and/or installworld. This
1282 is especially true when upgrading between "distant" versions (eg one
1283 that cross a major release boundary or several minor releases, or when
1284 several months have passed on the -current branch).
1286 Sometimes, obscure build problems are the result of environment
1287 poisoning. This can happen because the make utility reads its
1288 environment when searching for values for global variables. To run
1289 your build attempts in an "environmental clean room", prefix all make
1290 commands with 'env -i '. See the env(1) manual page for more details.
1292 When upgrading from one major version to another it is generally best
1293 to upgrade to the latest code in the currently installed branch first,
1294 then do an upgrade to the new branch. This is the best-tested upgrade
1295 path, and has the highest probability of being successful. Please try
1296 this approach before reporting problems with a major version upgrade.
1300 When upgrading the boot ZFS pool to a new version, always follow
1303 1.) recompile and reinstall the ZFS boot loader and boot block
1304 (this is part of "make buildworld" and "make installworld")
1306 2.) update the ZFS boot block on your boot drive
1308 The following example updates the ZFS boot block on the first
1309 partition (freebsd-boot) of a GPT partitioned drive ad0:
1310 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1312 Non-boot pools do not need these updates.
1316 If you are updating from a prior version of FreeBSD (even one just
1317 a few days old), you should follow this procedure. It is the most
1318 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1320 make kernel-toolchain
1321 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1322 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1324 To test a kernel once
1325 ---------------------
1326 If you just want to boot a kernel once (because you are not sure
1327 if it works, or if you want to boot a known bad kernel to provide
1328 debugging information) run
1329 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1330 nextboot -k testkernel
1332 To just build a kernel when you know that it won't mess you up
1333 --------------------------------------------------------------
1334 This assumes you are already running a CURRENT system. Replace
1335 ${arch} with the architecture of your machine (e.g. "i386",
1336 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1338 cd src/sys/${arch}/conf
1339 config KERNEL_NAME_HERE
1340 cd ../compile/KERNEL_NAME_HERE
1345 If this fails, go to the "To build a kernel" section.
1347 To rebuild everything and install it on the current system.
1348 -----------------------------------------------------------
1349 # Note: sometimes if you are running current you gotta do more than
1350 # is listed here if you are upgrading from a really old current.
1352 <make sure you have good level 0 dumps>
1354 make kernel KERNCONF=YOUR_KERNEL_HERE
1356 <reboot in single user> [3]
1364 To cross-install current onto a separate partition
1365 --------------------------------------------------
1366 # In this approach we use a separate partition to hold
1367 # current's root, 'usr', and 'var' directories. A partition
1368 # holding "/", "/usr" and "/var" should be about 2GB in
1371 <make sure you have good level 0 dumps>
1374 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1375 <maybe newfs current's root partition>
1376 <mount current's root partition on directory ${CURRENT_ROOT}>
1377 make installworld DESTDIR=${CURRENT_ROOT}
1378 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1379 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1380 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1381 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1382 <reboot into current>
1383 <do a "native" rebuild/install as described in the previous section>
1384 <maybe install compatibility libraries from ports/misc/compat*>
1388 To upgrade in-place from 8.x-stable to current
1389 ----------------------------------------------
1390 <make sure you have good level 0 dumps>
1392 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1394 <reboot in single user> [3]
1401 Make sure that you've read the UPDATING file to understand the
1402 tweaks to various things you need. At this point in the life
1403 cycle of current, things change often and you are on your own
1404 to cope. The defaults can also change, so please read ALL of
1405 the UPDATING entries.
1407 Also, if you are tracking -current, you must be subscribed to
1408 freebsd-current@freebsd.org. Make sure that before you update
1409 your sources that you have read and understood all the recent
1410 messages there. If in doubt, please track -stable which has
1411 much fewer pitfalls.
1413 [1] If you have third party modules, such as vmware, you
1414 should disable them at this point so they don't crash your
1417 [3] From the bootblocks, boot -s, and then do
1422 adjkerntz -i # if CMOS is wall time
1423 Also, when doing a major release upgrade, it is required that
1424 you boot into single user mode to do the installworld.
1426 [4] Note: This step is non-optional. Failure to do this step
1427 can result in a significant reduction in the functionality of the
1428 system. Attempting to do it by hand is not recommended and those
1429 that pursue this avenue should read this file carefully, as well
1430 as the archives of freebsd-current and freebsd-hackers mailing lists
1431 for potential gotchas. The -U option is also useful to consider.
1432 See mergemaster(8) for more information.
1434 [5] Usually this step is a noop. However, from time to time
1435 you may need to do this if you get unknown user in the following
1436 step. It never hurts to do it all the time. You may need to
1437 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1438 install) after the buildworld before this step if you last updated
1439 from current before 20020224 or from -stable before 20020408.
1441 [6] This only deletes old files and directories. Old libraries
1442 can be deleted by "make delete-old-libs", but you have to make
1443 sure that no program is using those libraries anymore.
1445 [8] In order to have a kernel that can run the 4.x binaries needed to
1446 do an installworld, you must include the COMPAT_FREEBSD4 option in
1447 your kernel. Failure to do so may leave you with a system that is
1448 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1449 required to run the 5.x binaries on more recent kernels. And so on
1450 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1452 Make sure that you merge any new devices from GENERIC since the
1453 last time you updated your kernel config file.
1455 [9] When checking out sources, you must include the -P flag to have
1456 cvs prune empty directories.
1458 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1459 "?=" instead of the "=" assignment operator, so that buildworld can
1460 override the CPUTYPE if it needs to.
1462 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1463 not on the command line, or in /etc/make.conf. buildworld will
1464 warn if it is improperly defined.
1467 This file contains a list, in reverse chronological order, of major
1468 breakages in tracking -current. It is not guaranteed to be a complete
1469 list of such breakages, and only contains entries since October 10, 2007.
1470 If you need to see UPDATING entries from before that date, you will need
1471 to fetch an UPDATING file from an older FreeBSD release.
1473 Copyright information:
1475 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1477 Redistribution, publication, translation and use, with or without
1478 modification, in full or in part, in any form or format of this
1479 document are permitted without further permission from the author.
1481 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1482 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1483 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1484 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1485 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1486 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1487 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1488 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1489 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1490 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1491 POSSIBILITY OF SUCH DAMAGE.
1493 Contact Warner Losh if you have any questions about your use of