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.
16 The broken amd(4) driver has been replaced with esp(4) in the amd64,
17 i386 and pc98 GENERIC kernel configuration files.
20 This commit modifies vfs_register() so that it uses a hash
21 calculation to set vfc_typenum, which is enabled by default.
22 The first time a system is booted after this change, the
23 vfc_typenum values will change for all file systems. The
24 main effect of this is a change to the NFS server file handles
25 for file systems that use vfc_typenum in their fsid, such as ZFS.
26 It will, however, prevent vfc_typenum from changing when file
27 systems are loaded in a different order for subsequent reboots.
28 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
29 until you are ready to remount all NFS clients after a reboot.
32 Bump the shared library version numbers for libraries that
33 do not use symbol versioning, have changed the ABI compared
34 to stable/8 and which shared library version was not bumped.
35 Done as part of 9.0-RELEASE cycle.
38 During the merge of Capsicum features, the fget(9) KPI was modified.
39 This may require the rebuilding of out-of-tree device drivers --
40 issues have been reported specifically with the nVidia device driver.
41 __FreeBSD_version is bumped to 900041.
43 Also, there is a period between 20110811 and 20110814 where the
44 special devices /dev/{stdin,stdout,stderr} did not work correctly.
45 Building world from a kernel during that window may not work.
48 The packet filter (pf) code has been updated to OpenBSD 4.5.
49 You need to update userland tools to be in sync with kernel.
50 This update breaks backward compatibility with earlier pfsync(4)
51 versions. Care must be taken when updating redundant firewall setups.
54 The following sysctls and tunables are retired on x86 platforms:
56 machdep.hlt_logical_cpus
57 The following sysctl is retired:
58 machdep.hyperthreading_allowed
59 The sysctls were supposed to provide a way to dynamically offline and
60 online selected CPUs on x86 platforms, but the implementation has not
61 been reliable especially with SCHED_ULE scheduler.
62 machdep.hyperthreading_allowed tunable is still available to ignore
63 hyperthreading CPUs at OS level.
64 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
65 where X is an APIC ID of a CPU. Be advised, though, that disabling
66 CPUs in non-uniform fashion will result in non-uniform topology and
67 may lead to sub-optimal system performance with SCHED_ULE, which is
71 cpumask_t type is retired and cpuset_t is used in order to describe
75 Changes to ifconfig(8) for dynamic address family detection mandate
76 that you are running a kernel of 20110525 or later. Make sure to
77 follow the update procedure to boot a new kernel before installing
81 Support for sun4v architecture is officially dropped
84 Several KPI breaking changes have been committed to the mii(4) layer,
85 the PHY drivers and consequently some Ethernet drivers using mii(4).
86 This means that miibus.ko and the modules of the affected Ethernet
87 drivers need to be recompiled.
89 Note to kernel developers: Given that the OUI bit reversion problem
90 was fixed as part of these changes all mii(4) commits related to OUIs,
91 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
92 handling, no longer can be merged verbatim to stable/8 and previous
96 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
97 into their kernel configurations along with 'device pci'.
100 The default NFS client is now the new NFS client, so fstype "newnfs"
101 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
102 Although mounts via fstype "nfs" will usually work without userland
103 changes, it is recommended that the mount(8) and mount_nfs(8)
104 commands be rebuilt from sources and that a link to mount_nfs called
105 mount_oldnfs be created. The new client is compiled into the
106 kernel with "options NFSCL" and this is needed for diskless root
107 file systems. The GENERIC kernel configs have been changed to use
108 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
109 To use the regular/old client, you can "mount -t oldnfs ...". For
110 a diskless root file system, you must also include a line like:
112 vfs.root.mountfrom="oldnfs:"
114 in the boot/loader.conf on the root fs on the NFS server to make
115 a diskless root fs use the old client.
118 The GENERIC kernels for all architectures now default to the new
119 CAM-based ATA stack. It means that all legacy ATA drivers were
120 removed and replaced by respective CAM drivers. If you are using
121 ATA device names in /etc/fstab or other places, make sure to update
122 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
123 where 'Y's are the sequential numbers starting from zero for each type
124 in order of detection, unless configured otherwise with tunables,
125 see cam(4)). There will be symbolic links created in /dev/ to map
126 old adX devices to the respective adaY. They should provide basic
127 compatibility for file systems mounting in most cases, but they do
128 not support old user-level APIs and do not have respective providers
129 in GEOM. Consider using updated management tools with new device names.
131 It is possible to load devices ahci, ata, siis and mvs as modules,
132 but option ATA_CAM should remain in kernel configuration to make ata
133 module work as CAM driver supporting legacy ATA controllers. Device ata
134 still can be used in modular fashion (atacore + ...). Modules atadisk
135 and atapi* are not used and won't affect operation in ATA_CAM mode.
136 Note that to use CAM-based ATA kernel should include CAM devices
137 scbus, pass, da (or explicitly ada), cd and optionally others. All of
138 them are parts of the cam module.
140 ataraid(4) functionality is now supported by the RAID GEOM class.
141 To use it you can load geom_raid kernel module and use graid(8) tool
142 for management. Instead of /dev/arX device names, use /dev/raid/rX.
144 No kernel config options or code have been removed, so if a problem
145 arises, please report it and optionally revert to the old ATA stack.
146 In order to do it you can remove from the kernel config:
151 , and instead add back:
152 device atadisk # ATA disk drives
153 device ataraid # ATA RAID drives
154 device atapicd # ATAPI CDROM drives
155 device atapifd # ATAPI floppy drives
156 device atapist # ATAPI tape drives
159 The default NFS server has been changed to the new server, which
160 was referred to as the experimental server. If you need to switch
161 back to the old NFS server, you must now put the "-o" option on
162 both the mountd and nfsd commands. This can be done using the
163 mountd_flags and nfs_server_flags rc.conf variables until an
164 update to the rc scripts is committed, which is coming soon.
167 The GNU Objective-C runtime library (libobjc), and other Objective-C
168 related components have been removed from the base system. If you
169 require an Objective-C library, please use one of the available ports.
172 ath(4) has been split into bus- and device- modules. if_ath contains
173 the HAL, the TX rate control and the network device code. if_ath_pci
174 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
175 contains the AHB glue. Users need to load both if_ath_pci and if_ath
176 in order to use ath on everything else.
178 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
179 need to load if_ath and if_ath_pci for ath(4) operation.
182 As part of the replacement of sysinstall, the process of building
183 release media has changed significantly. For details, please re-read
184 release(7), which has been updated to reflect the new build process.
187 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
188 is the last available version under GPLv2. It brings a number of new
189 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
190 4.1 and SSE 4.2), better support for powerpc64, a number of new
191 directives, and lots of other small improvements. See the ChangeLog
192 file in contrib/binutils for the full details.
195 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
196 compliant, and will now use half of hash for authentication.
197 This will break interoperability with all stacks (including all
198 actual FreeBSD versions) who implement
199 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
201 The only workaround with such peers is to use another HMAC
202 algorithm for IPsec ("phase 2") authentication.
205 Remove the uio_yield prototype and symbol. This function has
206 been misnamed since it was introduced and should not be
207 globally exposed with this name. The equivalent functionality
208 is now available using kern_yield(curthread->td_user_pri).
209 The function remains undocumented.
212 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
213 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
214 for scalar sysctls is defined but disabled. Code that needs
215 UQUAD to pass the type checking that must compile on older
216 systems where the define is not present can check against
217 __FreeBSD_version >= 900030.
219 The system dialog(1) has been replaced with a new version previously
220 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
221 with the previous version, but the libdialog associated with it has
222 a largely incompatible API. As such, the original version of libdialog
223 will be kept temporarily as libodialog, until its base system consumers
224 are replaced or updated. Bump __FreeBSD_version to 900030.
227 If you are trying to run make universe on a -stable system, and you get
228 the following warning:
229 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
230 unknown. config(8) likely too old."
231 or something similar to it, then you must upgrade your -stable system
232 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
233 stable/8) or build the config from the latest stable/8 branch and
234 install it on your system.
236 Prior to this date, building a current universe on 8-stable system from
237 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
238 error in the first kernel build phase. A new config on those old
239 systems will fix that problem for older versions of -current.
242 The TCP stack has been modified to allow Khelp modules to interact with
243 it via helper hook points and store per-connection data in the TCP
244 control block. Bump __FreeBSD_version to 900029. User space tools that
245 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
249 Generic IEEE 802.3 annex 31B full duplex flow control support has been
250 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
251 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
252 to take advantage of it instead of using custom implementations. This
253 means that these drivers now no longer unconditionally advertise
254 support for flow control but only do so if flow control is a selected
255 media option. This was implemented in the generic support that way in
256 order to allow flow control to be switched on and off via ifconfig(8)
257 with the PHY specific default to typically off in order to protect
258 from unwanted effects. Consequently, if you used flow control with
259 one of the above mentioned drivers you now need to explicitly enable
261 ifconfig bge0 media auto mediaopt flowcontrol
263 Along with the above mentioned changes generic support for setting
264 1000baseT master mode also has been added and brgphy(4), ciphy(4),
265 e1000phy(4) as well as ip1000phy(4) have been converted to take
266 advantage of it. This means that these drivers now no longer take the
267 link0 parameter for selecting master mode but the master media option
268 has to be used instead, for example like in the following:
269 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
271 Selection of master mode now is also available with all other PHY
272 drivers supporting 1000baseT.
275 The TCP stack has received a significant update to add support for
276 modularised congestion control and generally improve the clarity of
277 congestion control decisions. Bump __FreeBSD_version to 900025. User
278 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
279 sockstat) need to be recompiled.
282 The man(1) utility has been replaced by a new version that no longer
283 uses /etc/manpath.config. Please consult man.conf(5) for how to
284 migrate local entries to the new format.
287 The copyright strings printed by login(1) and sshd(8) at the time of a
288 new connection have been removed to follow other operating systems and
292 A workaround for a fixed ld bug has been removed in kernel code,
293 so make sure that your system ld is built from sources after
294 revision 210245 from 2010-07-19 (r211583 if building head kernel
295 on stable/8, r211584 for stable/7; both from 2010-08-21).
296 A symptom of incorrect ld version is different addresses for
297 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
300 The $ipv6_prefer variable in rc.conf(5) has been split into
301 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
303 The $ip6addrctl_policy is a variable to choose a pre-defined
304 address selection policy set by ip6addrctl(8). A value
305 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
308 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
309 flag (see an entry of 20090926) is set on an interface with no
310 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
311 security reason. If you want IPv6 link-local address on all
312 interfaces by default, set this to "YES".
314 The old ipv6_prefer="YES" is equivalent to
315 ipv6_activate_all_interfaces="YES" and
316 ip6addrctl_policy="ipv6_prefer".
319 DTrace has grown support for userland tracing. Due to this, DTrace is
320 now i386 and amd64 only.
321 dtruss(1) is now installed by default on those systems and a new
322 kernel module is needed for userland tracing: fasttrap.
323 No changes to your kernel config file are necessary to enable
324 userland tracing, but you might consider adding 'STRIP=' and
325 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
326 to have informative userland stack traces in DTrace (ustack).
329 The acpi_aiboost(4) driver has been removed in favor of the new
330 aibs(4) driver. You should update your kernel configuration file.
333 BSD grep has been imported to the base system and it is built by
334 default. It is completely BSD licensed, highly GNU-compatible, uses
335 less memory than its GNU counterpart and has a small codebase.
336 However, it is slower than its GNU counterpart, which is mostly
337 noticeable for larger searches, for smaller ones it is measurable
338 but not significant. The reason is complex, the most important factor
339 is that we lack a modern and efficient regex library and GNU
340 overcomes this by optimizing the searches internally. Future work
341 on improving the regex performance is planned, for the meantime,
342 users that need better performance, can build GNU grep instead by
343 setting the WITH_GNU_GREP knob.
346 Due to the import of powerpc64 support, all existing powerpc kernel
347 configuration files must be updated with a machine directive like this:
348 machine powerpc powerpc
350 In addition, an updated config(8) is required to build powerpc kernels
354 A new version of ZFS (version 15) has been merged to -HEAD.
355 This version uses a python library for the following subcommands:
356 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
357 For full functionality of these commands the following port must
358 be installed: sysutils/py-zfs
361 'vm_page's are now hashed by physical address to an array of mutexes.
362 Currently this is only used to serialize access to hold_count. Over
363 time the page queue mutex will be peeled away. This changes the size
364 of pmap on every architecture. And requires all callers of vm_page_hold
365 and vm_page_unhold to be updated.
368 WITH_CTF can now be specified in src.conf (not recommended, there
369 are some problems with static executables), make.conf (would also
370 affect ports which do not use GNU make and do not override the
371 compile targets) or in the kernel config (via "makeoptions
373 When WITH_CTF was specified there before this was silently ignored,
374 so make sure that WITH_CTF is not used in places which could lead
375 to unwanted behavior.
378 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
379 to allow 32-bit compatibility on non-x86 platforms. All kernel
380 configurations on amd64 and ia64 platforms using these options must
381 be modified accordingly.
384 The utmp user accounting database has been replaced with utmpx,
385 the user accounting interface standardized by POSIX.
386 Unfortunately the semantics of utmp and utmpx don't match,
387 making it practically impossible to support both interfaces.
388 The user accounting database is used by tools like finger(1),
389 last(1), talk(1), w(1) and ac(8).
391 All applications in the base system use utmpx. This means only
392 local binaries (e.g. from the ports tree) may still use these
393 utmp database files. These applications must be rebuilt to make
396 After the system has been upgraded, it is safe to remove the old
397 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
398 assuming their contents is of no importance anymore. Old wtmp
399 databases can only be used by last(1) and ac(8) after they have
400 been converted to the new format using wtmpcvt(1).
403 Introduce the kernel thread "deadlock resolver" (which can be enabled
404 via the DEADLKRES option, see NOTES for more details) and the
405 sleepq_type() function for sleepqueues.
408 The rc.firewall and rc.firewall6 were unified, and
409 rc.firewall6 and rc.d/ip6fw were removed.
410 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
411 variables are obsoleted. Instead, the following new rc
412 variables are added to rc.d/ipfw:
414 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
415 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
416 firewall_simple_onet_ipv6, firewall_trusted_ipv6
418 The meanings correspond to the relevant IPv4 variables.
424 The default terminal emulation for syscons(4) has been changed
425 from cons25 to xterm on all platforms except pc98. This means
426 that the /etc/ttys file needs to be updated to ensure correct
427 operation of applications on the console.
429 The terminal emulation style can be toggled per window by using
430 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
431 options can be used to change the compile-time default back to
434 To prevent graphical artifacts, make sure the TERM environment
435 variable is set to match the terminal emulation that is being
436 performed by syscons(4).
439 The layout of the structure ieee80211req_scan_result has changed.
440 Applications that require wireless scan results (e.g. ifconfig(8))
441 from net80211 need to be recompiled.
443 Applications such as wpa_supplicant(8) may require a full world
444 build without using NO_CLEAN in order to get synchronized with the
448 The iwn(4) driver has been updated to support the 5000 and 5150 series.
449 There's one kernel module for each firmware. Adding "device iwnfw"
450 to the kernel configuration file means including all three firmware
451 images inside the kernel. If you want to include just the one for
452 your wireless card, use the the devices iwn4965fw, iwn5000fw or
456 The rc.d/network_ipv6, IPv6 configuration script has been integrated
457 into rc.d/netif. The changes are the following:
459 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
460 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
461 Note that both variables need the "inet6" keyword at the head.
463 Do not set $ipv6_network_interfaces manually if you do not
464 understand what you are doing. It is not needed in most cases.
466 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
469 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
470 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
472 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
473 all configured interfaces have "inet6 accept_rtadv" in the
474 $ifconfig_IF_ipv6. These are for backward compatibility.
476 3. A new variable $ipv6_prefer has been added. If NO, IPv6
477 functionality of interfaces with no corresponding
478 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
479 and the default address selection policy of ip6addrctl(8)
480 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
481 Note that if you want to configure IPv6 functionality on the
482 disabled interfaces after boot, first you need to clear the flag by
483 using ifconfig(8) like:
485 ifconfig em0 inet6 -ifdisabled
487 If YES, the default address selection policy is set as
490 The default value of $ipv6_prefer is NO.
492 4. If your system need to receive Router Advertisement messages,
493 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
494 scripts automatically invoke rtsol(8) when the interface becomes
495 UP. The Router Advertisement messages are used for SLAAC
496 (State-Less Address AutoConfiguration).
499 802.11s D3.03 support was committed. This is incompatible with the
500 previous code, which was based on D3.0.
503 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
504 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
505 control whether accepting Router Advertisement messages or not.
506 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
507 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
508 The ifconfig(8) utility now supports these flags.
511 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
512 mount(8) and -a option for df(1) to see them.
515 The old tunable hw.bus.devctl_disable has been superseded by
516 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
517 replaced by hw.bus.devctl_queue=0. The default for this new tunable
521 Remove the option STOP_NMI. The default action is now to use NMI only
522 for KDB via the newly introduced function stop_cpus_hard() and
523 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
526 The stable/8 branch created in subversion. This corresponds to the
527 RELENG_8 branch in CVS.
530 Bump the shared library version numbers for all libraries that do not
531 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
532 __FreeBSD_version to 800105.
535 Due to changes in the implementation of virtual network stack support,
536 all network-related kernel modules must be recompiled. As this change
537 breaks the ABI, bump __FreeBSD_version to 800104.
540 The TOE interface to the TCP syncache has been modified to remove
541 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
542 The cxgb driver is the only TOE consumer affected by this change, and
543 needs to be recompiled along with the kernel. As this change breaks
544 the ABI, bump __FreeBSD_version to 800103.
547 Padding has been added to struct tcpcb, sackhint and tcpstat in
548 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
549 maintaining the ABI. However, this change breaks the ABI, so bump
550 __FreeBSD_version to 800102. User space tools that rely on the size of
551 any of these structs (e.g. sockstat) need to be recompiled.
554 The NFS_LEGACYRPC option has been removed along with the old kernel
555 RPC implementation that this option selected. Kernel configurations
556 may need to be adjusted.
559 The network interface device nodes at /dev/net/<interface> have been
560 removed. All ioctl operations can be performed the normal way using
561 routing sockets. The kqueue functionality can generally be replaced
562 with routing sockets.
565 The documentation from the FreeBSD Documentation Project (Handbook,
566 FAQ, etc.) is now installed via packages by sysinstall(8) and under
567 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
570 The ABI of various structures related to the SYSV IPC API have been
571 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
572 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
576 Layout of struct vnet has changed as routing related variables were
577 moved to their own Vimage module. Modules need to be recompiled. Bump
578 __FreeBSD_version to 800099.
581 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
582 respectively. As long as no more than 16 groups per process are used,
583 no changes should be visible. When more than 16 groups are used, old
584 binaries may fail if they call getgroups() or getgrouplist() with
585 statically sized storage. Recompiling will work around this, but
586 applications should be modified to use dynamically allocated storage
587 for group arrays as POSIX.1-2008 does not cap an implementation's
588 number of supported groups at NGROUPS_MAX+1 as previous versions did.
590 NFS and portalfs mounts may also be affected as the list of groups is
591 truncated to 16. Users of NFS who use more than 16 groups, should
592 take care that negative group permissions are not used on the exported
593 file systems as they will not be reliable unless a GSSAPI based
594 authentication method is used.
597 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
598 option compiles in the support for adaptive spinning for lockmgrs
599 which want to enable it. The lockinit() function now accepts the flag
600 LK_ADAPTIVE in order to make the lock object subject to adaptive
601 spinning when both held in write and read mode.
604 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
605 changed. User applications that use this ioctl need to be rebuilt.
608 The layout of struct thread has changed. Kernel and modules need to
612 The layout of structs ifnet, domain, protosw and vnet_net has changed.
613 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
616 window(1) has been removed from the base system. It can now be
617 installed from ports. The port is called misc/window.
620 The way we are storing and accessing `routing table' entries has
621 changed. Programs reading the FIB, like netstat, need to be
625 A new netisr implementation has been added for FreeBSD 8. Network
626 file system modules, such as igmp, ipdivert, and others, should be
628 Bump __FreeBSD_version to 800096.
631 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
635 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
636 Bump __FreeBSD_version to 800094.
639 Add mnt_xflag field to 'struct mount'. File system modules need to be
641 Bump __FreeBSD_version to 800093.
644 The compiling option ADAPTIVE_SX has been retired while it has been
645 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
646 The KPI for sx_init_flags() changes as accepting flags:
647 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
648 been introduced in order to handle the reversed logic.
649 Bump __FreeBSD_version to 800092.
652 Add support for hierarchical jails. Remove global securelevel.
653 Bump __FreeBSD_version to 800091.
656 The layout of struct vnet_net has changed, therefore modules
658 Bump __FreeBSD_version to 800090.
661 The newly imported zic(8) produces a new format in the output. Please
662 run tzsetup(8) to install the newly created data to /etc/localtime.
665 The sysctl tree for the usb stack has renamed from hw.usb2.* to
666 hw.usb.* and is now consistent again with previous releases.
669 802.11 monitor mode support was revised and driver api's were changed.
670 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
671 of DLT_IEEE802_11. No user-visible data structures were changed but
672 applications that use DLT_IEEE802_11 may require changes.
673 Bump __FreeBSD_version to 800088.
676 The layout of the following structs has changed: sysctl_oid,
677 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
678 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
679 panics may be experienced. World rebuild is required for
680 correctly checking networking state from userland.
681 Bump __FreeBSD_version to 800085.
684 MLDv2 and Source-Specific Multicast (SSM) have been merged
685 to the IPv6 stack. VIMAGE hooks are in but not yet used.
686 The implementation of SSM within FreeBSD's IPv6 stack closely
687 follows the IPv4 implementation.
689 For kernel developers:
691 * The most important changes are that the ip6_output() and
692 ip6_input() paths no longer take the IN6_MULTI_LOCK,
693 and this lock has been downgraded to a non-recursive mutex.
695 * As with the changes to the IPv4 stack to support SSM, filtering
696 of inbound multicast traffic must now be performed by transport
697 protocols within the IPv6 stack. This does not apply to TCP and
698 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
700 * The KPIs used by IPv6 multicast are similar to those used by
701 the IPv4 stack, with the following differences:
702 * im6o_mc_filter() is analogous to imo_multicast_filter().
703 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
704 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
705 * IN6_LOOKUP_MULTI() has been deprecated and removed.
706 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
707 for MLDv1 have an additional 'timer' argument which is used to
708 jitter the initial membership report for the solicited-node
709 multicast membership on-link.
710 * This is not strictly needed for MLDv2, which already jitters
711 its report transmissions. However, the 'timer' argument is
712 preserved in case MLDv1 is active on the interface.
714 * The KAME linked-list based IPv6 membership implementation has
715 been refactored to use a vector similar to that used by the IPv4
717 Code which maintains a list of its own multicast memberships
718 internally, e.g. carp, has been updated to reflect the new
721 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
722 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
723 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
724 implementation constraint which needs to be addressed in HEAD.
726 For application developers:
728 * The changes are broadly similar to those made for the IPv4
731 * The use of IPv4 and IPv6 multicast socket options on the same
732 socket, using mapped addresses, HAS NOT been tested or supported.
734 * There are a number of issues with the implementation of various
735 IPv6 multicast APIs which need to be resolved in the API surface
736 before the implementation is fully compatible with KAME userland
737 use, and these are mostly to do with interface index treatment.
739 * The literature available discusses the use of either the delta / ASM
740 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
741 using setsourcefilter(3)/getsourcefilter(3). For more information
742 please refer to RFC 3768, 'Socket Interface Extensions for
743 Multicast Source Filters'.
745 * Applications which use the published RFC 3678 APIs should be fine.
747 For systems administrators:
749 * The mtest(8) utility has been refactored to support IPv6, in
750 addition to IPv4. Interface addresses are no longer accepted
751 as arguments, their names must be used instead. The utility
752 will map the interface name to its first IPv4 address as
753 returned by getifaddrs(3).
755 * The ifmcstat(8) utility has also been updated to print the MLDv2
756 endpoint state and source filter lists via sysctl(3).
758 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
759 loopback of IPv6 multicast datagrams by default; it defaults to 1
760 to preserve the existing behaviour. Disabling multicast loopback is
761 recommended for optimal system performance.
763 * The IPv6 MROUTING code has been changed to examine this sysctl
764 instead of attempting to perform a group lookup before looping
765 back forwarded datagrams.
767 Bump __FreeBSD_version to 800084.
770 Implement low-level Bluetooth HCI API.
771 Bump __FreeBSD_version to 800083.
774 The layout of struct malloc_type, used by modules to register new
775 memory allocation types, has changed. Most modules will need to
776 be rebuilt or panics may be experienced.
777 Bump __FreeBSD_version to 800081.
780 Anticipate overflowing inp_flags - add inp_flags2.
781 This changes most offsets in inpcb, so checking v4 connection
782 state will require a world rebuild.
783 Bump __FreeBSD_version to 800080.
786 Add an llentry to struct route and struct route_in6. Modules
787 embedding a struct route will need to be recompiled.
788 Bump __FreeBSD_version to 800079.
791 The size of rt_metrics_lite and by extension rtentry has changed.
792 Networking administration apps will need to be recompiled.
793 The route command now supports show as an alias for get, weighting
794 of routes, sticky and nostick flags to alter the behavior of stateful
796 Bump __FreeBSD_version to 800078.
799 Do not use Giant for kbdmux(4) locking. This is wrong and
800 apparently causing more problems than it solves. This will
801 re-open the issue where interrupt handlers may race with
802 kbdmux(4) in polling mode. Typical symptoms include (but
803 not limited to) duplicated and/or missing characters when
804 low level console functions (such as gets) are used while
805 interrupts are enabled (for example geli password prompt,
806 mountroot prompt etc.). Disabling kbdmux(4) may help.
809 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
810 kernel modules referencing any of the above need to be recompiled.
811 Bump __FreeBSD_version to 800075.
814 GEOM_PART has become the default partition slicer for storage devices,
815 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
816 introduces some changes:
818 MSDOS/EBR: the devices created from MSDOS extended partition entries
819 (EBR) can be named differently than with GEOM_MBR and are now symlinks
820 to devices with offset-based names. fstabs may need to be modified.
822 BSD: the "geometry does not match label" warning is harmless in most
823 cases but it points to problems in file system misalignment with
824 disk geometry. The "c" partition is now implicit, covers the whole
825 top-level drive and cannot be (mis)used by users.
827 General: Kernel dumps are now not allowed to be written to devices
828 whose partition types indicate they are meant to be used for file
829 systems (or, in case of MSDOS partitions, as something else than
832 Most of these changes date approximately from 200812.
835 The uscanner(4) driver has been removed from the kernel. This follows
836 Linux removing theirs in 2.6 and making libusb the default interface
840 The multicast forwarding code has been cleaned up. netstat(1)
841 only relies on KVM now for printing bandwidth upcall meters.
842 The IPv4 and IPv6 modules are split into ip_mroute_mod and
843 ip6_mroute_mod respectively. The config(5) options for statically
844 compiling this code remain the same, i.e. 'options MROUTING'.
847 Support for the IFF_NEEDSGIANT network interface flag has been
848 removed, which means that non-MPSAFE network device drivers are no
849 longer supported. In particular, if_ar, if_sr, and network device
850 drivers from the old (legacy) USB stack can no longer be built or
854 POSIX.1 Native Language Support (NLS) has been enabled in libc and
855 a bunch of new language catalog files have also been added.
856 This means that some common libc messages are now localized and
857 they depend on the LC_MESSAGES environmental variable.
860 The k8temp(4) driver has been renamed to amdtemp(4) since
861 support for Family 10 and Family 11 CPU families was added.
864 IGMPv3 and Source-Specific Multicast (SSM) have been merged
865 to the IPv4 stack. VIMAGE hooks are in but not yet used.
867 For kernel developers, the most important changes are that the
868 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
869 and this lock has been downgraded to a non-recursive mutex.
871 Transport protocols (UDP, Raw IP) are now responsible for filtering
872 inbound multicast traffic according to group membership and source
873 filters. The imo_multicast_filter() KPI exists for this purpose.
874 Transports which do not use multicast (SCTP, TCP) already reject
875 multicast by default. Forwarding and receive performance may improve
876 as a mutex acquisition is no longer needed in the ip_input()
877 low-level input path. in_addmulti() and in_delmulti() are shimmed
878 to new KPIs which exist to support SSM in-kernel.
880 For application developers, it is recommended that loopback of
881 multicast datagrams be disabled for best performance, as this
882 will still cause the lock to be taken for each looped-back
883 datagram transmission. The net.inet.ip.mcast.loop sysctl may
884 be tuned to 0 to disable loopback by default; it defaults to 1
885 to preserve the existing behaviour.
887 For systems administrators, to obtain best performance with
888 multicast reception and multiple groups, it is always recommended
889 that a card with a suitably precise hash filter is used. Hash
890 collisions will still result in the lock being taken within the
891 transport protocol input path to check group membership.
893 If deploying FreeBSD in an environment with IGMP snooping switches,
894 it is recommended that the net.inet.igmp.sendlocal sysctl remain
895 enabled; this forces 224.0.0.0/24 group membership to be announced
898 The size of 'struct igmpstat' has changed; netstat needs to be
899 recompiled to reflect this.
900 Bump __FreeBSD_version to 800070.
903 libusb20.so.1 is now installed as libusb.so.1 and the ports system
904 updated to use it. This requires a buildworld/installworld in order to
905 update the library and dependencies (usbconfig, etc). Its advisable to
906 rebuild all ports which uses libusb. More specific directions are given
907 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
908 libusb are no longer required and can be removed.
911 A workaround is committed to allow the creation of System V shared
912 memory segment of size > 2 GB on the 64-bit architectures.
913 Due to a limitation of the existing ABI, the shm_segsz member
914 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
915 wrong for large segments. Note that limits must be explicitly
916 raised to allow such segments to be created.
919 The layout of struct ifnet has changed, requiring a rebuild of all
920 network device driver modules.
923 The /dev handling for the new USB stack has changed, a
924 buildworld/installworld is required for libusb20.
927 The new USB2 stack has now been permanently moved in and all kernel and
928 module names reverted to their previous values (eg, usb, ehci, ohci,
929 ums, ...). The old usb stack can be compiled in by prefixing the name
930 with the letter 'o', the old usb modules have been removed.
931 Updating entry 20090216 for xorg and 20090215 for libmap may still
935 The rc.conf(5) option if_up_delay has been renamed to
936 defaultroute_delay to better reflect its purpose. If you have
937 customized this setting in /etc/rc.conf you need to update it to
941 xorg 7.4 wants to configure its input devices via hald which does not
942 yet work with USB2. If the keyboard/mouse does not work in xorg then
944 Option "AllowEmptyInput" "off"
945 to your ServerLayout section. This will cause X to use the configured
946 kbd and mouse sections from your xorg.conf.
949 The GENERIC kernels for all architectures now default to the new USB2
950 stack. No kernel config options or code have been removed so if a
951 problem arises please report it and optionally revert to the old USB
952 stack. If you are loading USB kernel modules or have a custom kernel
953 that includes GENERIC then ensure that usb names are also changed over,
954 eg uftdi -> usb2_serial_ftdi.
956 Older programs linked against the ports libusb 0.1 need to be
957 redirected to the new stack's libusb20. /etc/libmap.conf can
959 # Map old usb library to new one for usb2 stack
960 libusb-0.1.so.8 libusb20.so.1
963 All USB ethernet devices now attach as interfaces under the name ueN
964 (eg. ue0). This is to provide a predictable name as vendors often
965 change usb chipsets in a product without notice.
968 The ichsmb(4) driver has been changed to require SMBus slave
969 addresses be left-justified (xxxxxxx0b) rather than right-justified.
970 All of the other SMBus controller drivers require left-justified
971 slave addresses, so this change makes all the drivers provide the
975 INET6 statistics (struct ip6stat) was updated.
976 netstat(1) needs to be recompiled.
979 NTFS has been removed from GENERIC kernel on amd64 to match
980 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
981 will load ntfs.ko module automatically when NTFS support is
982 actually needed, unless ntfs.ko is not installed or security
983 level prohibits loading kernel modules. If either is the case,
984 "options NTFS" has to be added into kernel config.
987 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
988 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
989 800061. User space tools that rely on the size of struct tcpcb in
990 tcp_var.h (e.g. sockstat) need to be recompiled.
993 ng_tty(4) module updated to match the new TTY subsystem.
994 Due to API change, user-level applications must be updated.
995 New API support added to mpd5 CVS and expected to be present
996 in next mpd5.3 release.
999 With __FreeBSD_version 800060 the makefs tool is part of
1000 the base system (it was a port).
1003 The afdata and ifnet locks have been changed from mutexes to
1004 rwlocks, network modules will need to be re-compiled.
1007 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1008 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1009 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1010 architecture and 40 bytes on 64-bit architecture. The userland
1011 applications "arp" and "ndp" have been updated accordingly.
1012 The output from "netstat -r" shows only routing entries and
1013 none of the L2 information.
1016 __FreeBSD_version 800057 marks the switchover from the
1017 binary ath hal to source code. Users must add the line:
1019 options AH_SUPPORT_AR5416
1021 to their kernel config files when specifying:
1025 The ath_hal module no longer exists; the code is now compiled
1026 together with the driver in the ath module. It is now
1027 possible to tailor chip support (i.e. reduce the set of chips
1028 and thereby the code size); consult ath_hal(4) for details.
1031 __FreeBSD_version 800054 adds memory barriers to
1032 <machine/atomic.h>, new interfaces to ifnet to facilitate
1033 multiple hardware transmit queues for cards that support
1034 them, and a lock-less ring-buffer implementation to
1035 enable drivers to more efficiently manage queueing of
1039 A new version of ZFS (version 13) has been merged to -HEAD.
1040 This version has zpool attribute "listsnapshots" off by
1041 default, which means "zfs list" does not show snapshots,
1042 and is the same as Solaris behavior.
1045 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1048 The uhci, ohci, ehci and slhci USB Host controller drivers have
1049 been put into separate modules. If you load the usb module
1050 separately through loader.conf you will need to load the
1051 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1052 controller add the following to loader.conf:
1058 The ABI used by the PMC toolset has changed. Please keep
1059 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1063 atapci kernel module now includes only generic PCI ATA
1064 driver. AHCI driver moved to ataahci kernel module.
1065 All vendor-specific code moved into separate kernel modules:
1066 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1067 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1068 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1069 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1072 The TTY subsystem of the kernel has been replaced by a new
1073 implementation, which provides better scalability and an
1074 improved driver model. Most common drivers have been migrated to
1075 the new TTY subsystem, while others have not. The following
1076 drivers have not yet been ported to the new TTY layer:
1079 cy, digi, rc, rp, sio
1085 ng_h4, ng_tty, ppp, sl, snp
1087 Adding these drivers to your kernel configuration file shall
1088 cause compilation to fail.
1091 ntpd has been upgraded to 4.2.4p5.
1094 OpenSSH has been upgraded to 5.1p1.
1096 For many years, FreeBSD's version of OpenSSH preferred DSA
1097 over RSA for host and user authentication keys. With this
1098 upgrade, we've switched to the vendor's default of RSA over
1099 DSA. This may cause upgraded clients to warn about unknown
1100 host keys even for previously known hosts. Users should
1101 follow the usual procedure for verifying host keys before
1102 accepting the RSA key.
1104 This can be circumvented by setting the "HostKeyAlgorithms"
1105 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1108 Please note that the sequence of keys offered for
1109 authentication has been changed as well. You may want to
1110 specify IdentityFile in a different order to revert this
1114 The sio(4) driver has been removed from the i386 and amd64
1115 kernel configuration files. This means uart(4) is now the
1116 default serial port driver on those platforms as well.
1118 To prevent collisions with the sio(4) driver, the uart(4) driver
1119 uses different names for its device nodes. This means the
1120 onboard serial port will now most likely be called "ttyu0"
1121 instead of "ttyd0". You may need to reconfigure applications to
1122 use the new device names.
1124 When using the serial port as a boot console, be sure to update
1125 /boot/device.hints and /etc/ttys before booting the new kernel.
1126 If you forget to do so, you can still manually specify the hints
1127 at the loader prompt:
1129 set hint.uart.0.at="isa"
1130 set hint.uart.0.port="0x3F8"
1131 set hint.uart.0.flags="0x10"
1132 set hint.uart.0.irq="4"
1136 The gpt(8) utility has been removed. Use gpart(8) to partition
1140 The version that Linuxulator emulates was changed from 2.4.2
1141 to 2.6.16. If you experience any problems with Linux binaries
1142 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1143 if it fixes the problem contact emulation mailing list.
1146 ISDN4BSD (I4B) was removed from the src tree. You may need to
1147 update a your kernel configuration and remove relevant entries.
1150 I have checked in code to support multiple routing tables.
1151 See the man pages setfib(1) and setfib(2).
1152 This is a hopefully backwards compatible version,
1153 but to make use of it you need to compile your kernel
1154 with options ROUTETABLES=2 (or more up to 16).
1157 The 802.11 wireless support was redone to enable multi-bss
1158 operation on devices that are capable. The underlying device
1159 is no longer used directly but instead wlanX devices are
1160 cloned with ifconfig. This requires changes to rc.conf files.
1161 For example, change:
1162 ifconfig_ath0="WPA DHCP"
1165 ifconfig_wlan0="WPA DHCP"
1166 see rc.conf(5) for more details. In addition, mergemaster of
1167 /etc/rc.d is highly recommended. Simultaneous update of userland
1168 and kernel wouldn't hurt either.
1170 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1171 modules were merged into the base wlan module. All references
1172 to these modules (e.g. in kernel config files) must be removed.
1175 psm(4) has gained write(2) support in native operation level.
1176 Arbitrary commands can be written to /dev/psm%d and status can
1177 be read back from it. Therefore, an application is responsible
1178 for status validation and error recovery. It is a no-op in
1179 other operation levels.
1182 Support for KSE threading has been removed from the kernel. To
1183 run legacy applications linked against KSE libmap.conf may
1184 be used. The following libmap.conf may be used to ensure
1185 compatibility with any prior release:
1187 libpthread.so.1 libthr.so.1
1188 libpthread.so.2 libthr.so.2
1189 libkse.so.3 libthr.so.3
1192 The layout of struct vmspace has changed. This affects libkvm
1193 and any executables that link against libkvm and use the
1194 kvm_getprocs() function. In particular, but not exclusively,
1195 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1196 The effects are minimal, but it's advisable to upgrade world
1200 The latest em driver no longer has support in it for the
1201 82575 adapter, this is now moved to the igb driver. The
1202 split was done to make new features that are incompatible
1203 with older hardware easier to do.
1206 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1207 likewise the kernel option is now GEOM_LINUX_LVM.
1210 The default NFS mount mode has changed from UDP to TCP for
1211 increased reliability. If you rely on (insecurely) NFS
1212 mounting across a firewall you may need to update your
1216 Belatedly note the addition of m_collapse for compacting
1220 The fts(3) structures have been changed to use adequate
1221 integer types for their members and so to be able to cope
1222 with huge file trees. The old fts(3) ABI is preserved
1223 through symbol versioning in libc, so third-party binaries
1224 using fts(3) should still work, although they will not take
1225 advantage of the extended types. At the same time, some
1226 third-party software might fail to build after this change
1227 due to unportable assumptions made in its source code about
1228 fts(3) structure members. Such software should be fixed
1229 by its vendor or, in the worst case, in the ports tree.
1230 FreeBSD_version 800015 marks this change for the unlikely
1231 case that a portable fix is impossible.
1234 To upgrade to -current after this date, you must be running
1235 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1236 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1239 The ADAPTIVE_GIANT kernel option has been retired because its
1240 functionality is the default now.
1243 The AT keyboard emulation of sunkbd(4) has been turned on
1244 by default. In order to make the special symbols of the Sun
1245 keyboards driven by sunkbd(4) work under X these now have
1246 to be configured the same way as Sun USB keyboards driven
1247 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1249 Option "XkbLayout" "us"
1250 Option "XkbRules" "xorg"
1251 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1254 It has been decided that it is desirable to provide ABI
1255 backwards compatibility to the FreeBSD 4/5/6 versions of the
1256 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1257 broken with the introduction of PCI domain support (see the
1258 20070930 entry). Unfortunately, this required the ABI of
1259 PCIOCGETCONF to be broken again in order to be able to
1260 provide backwards compatibility to the old version of that
1261 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1262 again. As for prominent ports this affects neither pciutils
1263 nor xorg-server this time, the hal port needs to be rebuilt
1267 The misnamed kthread_create() and friends have been renamed
1268 to kproc_create() etc. Many of the callers already
1269 used kproc_start()..
1270 I will return kthread_create() and friends in a while
1271 with implementations that actually create threads, not procs.
1272 Renaming corresponds with version 800002.
1281 Avoid using make -j when upgrading. While generally safe, there are
1282 sometimes problems using -j to upgrade. If your upgrade fails with
1283 -j, please try again without -j. From time to time in the past there
1284 have been problems using -j with buildworld and/or installworld. This
1285 is especially true when upgrading between "distant" versions (eg one
1286 that cross a major release boundary or several minor releases, or when
1287 several months have passed on the -current branch).
1289 Sometimes, obscure build problems are the result of environment
1290 poisoning. This can happen because the make utility reads its
1291 environment when searching for values for global variables. To run
1292 your build attempts in an "environmental clean room", prefix all make
1293 commands with 'env -i '. See the env(1) manual page for more details.
1295 When upgrading from one major version to another it is generally best
1296 to upgrade to the latest code in the currently installed branch first,
1297 then do an upgrade to the new branch. This is the best-tested upgrade
1298 path, and has the highest probability of being successful. Please try
1299 this approach before reporting problems with a major version upgrade.
1303 When upgrading the boot ZFS pool to a new version, always follow
1306 1.) recompile and reinstall the ZFS boot loader and boot block
1307 (this is part of "make buildworld" and "make installworld")
1309 2.) update the ZFS boot block on your boot drive
1311 The following example updates the ZFS boot block on the first
1312 partition (freebsd-boot) of a GPT partitioned drive ad0:
1313 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1315 Non-boot pools do not need these updates.
1319 If you are updating from a prior version of FreeBSD (even one just
1320 a few days old), you should follow this procedure. It is the most
1321 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1323 make kernel-toolchain
1324 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1325 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1327 To test a kernel once
1328 ---------------------
1329 If you just want to boot a kernel once (because you are not sure
1330 if it works, or if you want to boot a known bad kernel to provide
1331 debugging information) run
1332 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1333 nextboot -k testkernel
1335 To just build a kernel when you know that it won't mess you up
1336 --------------------------------------------------------------
1337 This assumes you are already running a CURRENT system. Replace
1338 ${arch} with the architecture of your machine (e.g. "i386",
1339 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1341 cd src/sys/${arch}/conf
1342 config KERNEL_NAME_HERE
1343 cd ../compile/KERNEL_NAME_HERE
1348 If this fails, go to the "To build a kernel" section.
1350 To rebuild everything and install it on the current system.
1351 -----------------------------------------------------------
1352 # Note: sometimes if you are running current you gotta do more than
1353 # is listed here if you are upgrading from a really old current.
1355 <make sure you have good level 0 dumps>
1357 make kernel KERNCONF=YOUR_KERNEL_HERE
1359 <reboot in single user> [3]
1367 To cross-install current onto a separate partition
1368 --------------------------------------------------
1369 # In this approach we use a separate partition to hold
1370 # current's root, 'usr', and 'var' directories. A partition
1371 # holding "/", "/usr" and "/var" should be about 2GB in
1374 <make sure you have good level 0 dumps>
1377 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1378 <maybe newfs current's root partition>
1379 <mount current's root partition on directory ${CURRENT_ROOT}>
1380 make installworld DESTDIR=${CURRENT_ROOT}
1381 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1382 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1383 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1384 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1385 <reboot into current>
1386 <do a "native" rebuild/install as described in the previous section>
1387 <maybe install compatibility libraries from ports/misc/compat*>
1391 To upgrade in-place from 8.x-stable to current
1392 ----------------------------------------------
1393 <make sure you have good level 0 dumps>
1395 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1397 <reboot in single user> [3]
1404 Make sure that you've read the UPDATING file to understand the
1405 tweaks to various things you need. At this point in the life
1406 cycle of current, things change often and you are on your own
1407 to cope. The defaults can also change, so please read ALL of
1408 the UPDATING entries.
1410 Also, if you are tracking -current, you must be subscribed to
1411 freebsd-current@freebsd.org. Make sure that before you update
1412 your sources that you have read and understood all the recent
1413 messages there. If in doubt, please track -stable which has
1414 much fewer pitfalls.
1416 [1] If you have third party modules, such as vmware, you
1417 should disable them at this point so they don't crash your
1420 [3] From the bootblocks, boot -s, and then do
1425 adjkerntz -i # if CMOS is wall time
1426 Also, when doing a major release upgrade, it is required that
1427 you boot into single user mode to do the installworld.
1429 [4] Note: This step is non-optional. Failure to do this step
1430 can result in a significant reduction in the functionality of the
1431 system. Attempting to do it by hand is not recommended and those
1432 that pursue this avenue should read this file carefully, as well
1433 as the archives of freebsd-current and freebsd-hackers mailing lists
1434 for potential gotchas. The -U option is also useful to consider.
1435 See mergemaster(8) for more information.
1437 [5] Usually this step is a noop. However, from time to time
1438 you may need to do this if you get unknown user in the following
1439 step. It never hurts to do it all the time. You may need to
1440 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1441 install) after the buildworld before this step if you last updated
1442 from current before 20020224 or from -stable before 20020408.
1444 [6] This only deletes old files and directories. Old libraries
1445 can be deleted by "make delete-old-libs", but you have to make
1446 sure that no program is using those libraries anymore.
1448 [8] In order to have a kernel that can run the 4.x binaries needed to
1449 do an installworld, you must include the COMPAT_FREEBSD4 option in
1450 your kernel. Failure to do so may leave you with a system that is
1451 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1452 required to run the 5.x binaries on more recent kernels. And so on
1453 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1455 Make sure that you merge any new devices from GENERIC since the
1456 last time you updated your kernel config file.
1458 [9] When checking out sources, you must include the -P flag to have
1459 cvs prune empty directories.
1461 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1462 "?=" instead of the "=" assignment operator, so that buildworld can
1463 override the CPUTYPE if it needs to.
1465 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1466 not on the command line, or in /etc/make.conf. buildworld will
1467 warn if it is improperly defined.
1470 This file contains a list, in reverse chronological order, of major
1471 breakages in tracking -current. It is not guaranteed to be a complete
1472 list of such breakages, and only contains entries since October 10, 2007.
1473 If you need to see UPDATING entries from before that date, you will need
1474 to fetch an UPDATING file from an older FreeBSD release.
1476 Copyright information:
1478 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1480 Redistribution, publication, translation and use, with or without
1481 modification, in full or in part, in any form or format of this
1482 document are permitted without further permission from the author.
1484 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1485 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1486 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1487 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1488 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1489 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1490 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1491 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1492 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1493 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1494 POSSIBILITY OF SUCH DAMAGE.
1496 Contact Warner Losh if you have any questions about your use of