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 sparc64 ZFS loader has been changed to no longer try to auto-
14 detect ZFS providers based on diskN aliases but now requires these
15 to be explicitly listed in the OFW boot-device environment variable.
18 Now unix domain sockets behave "as expected" on nullfs(5). Previously
19 nullfs(5) did not pass through all behaviours to the underlying layer,
20 as a result if we bound to a socket on the lower layer we could connect
21 only to the lower path; if we bound to the upper layer we could connect
22 only to the upper path. The new behavior is one can connect to both the
23 lower and the upper paths regardless what layer path one binds to.
26 The acpi_wmi(4) status device /dev/wmistat has been renamed to
30 A new VOP_ADVISE() was added to support posix_fadvise(2). All
31 filesystem modules must be recompiled.
34 The interface of the VOP_VPTOCNP(9) changed, now the returned
35 vnode shall be referenced, previously it was required to be
36 only held. All in-tree filesystems are converted.
42 The broken amd(4) driver has been replaced with esp(4) in the amd64,
43 i386 and pc98 GENERIC kernel configuration files.
46 This commit modifies vfs_register() so that it uses a hash
47 calculation to set vfc_typenum, which is enabled by default.
48 The first time a system is booted after this change, the
49 vfc_typenum values will change for all file systems. The
50 main effect of this is a change to the NFS server file handles
51 for file systems that use vfc_typenum in their fsid, such as ZFS.
52 It will, however, prevent vfc_typenum from changing when file
53 systems are loaded in a different order for subsequent reboots.
54 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
55 until you are ready to remount all NFS clients after a reboot.
58 Bump the shared library version numbers for libraries that
59 do not use symbol versioning, have changed the ABI compared
60 to stable/8 and which shared library version was not bumped.
61 Done as part of 9.0-RELEASE cycle.
64 During the merge of Capsicum features, the fget(9) KPI was modified.
65 This may require the rebuilding of out-of-tree device drivers --
66 issues have been reported specifically with the nVidia device driver.
67 __FreeBSD_version is bumped to 900041.
69 Also, there is a period between 20110811 and 20110814 where the
70 special devices /dev/{stdin,stdout,stderr} did not work correctly.
71 Building world from a kernel during that window may not work.
74 The packet filter (pf) code has been updated to OpenBSD 4.5.
75 You need to update userland tools to be in sync with kernel.
76 This update breaks backward compatibility with earlier pfsync(4)
77 versions. Care must be taken when updating redundant firewall setups.
80 The following sysctls and tunables are retired on x86 platforms:
82 machdep.hlt_logical_cpus
83 The following sysctl is retired:
84 machdep.hyperthreading_allowed
85 The sysctls were supposed to provide a way to dynamically offline and
86 online selected CPUs on x86 platforms, but the implementation has not
87 been reliable especially with SCHED_ULE scheduler.
88 machdep.hyperthreading_allowed tunable is still available to ignore
89 hyperthreading CPUs at OS level.
90 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
91 where X is an APIC ID of a CPU. Be advised, though, that disabling
92 CPUs in non-uniform fashion will result in non-uniform topology and
93 may lead to sub-optimal system performance with SCHED_ULE, which is
97 cpumask_t type is retired and cpuset_t is used in order to describe
101 Changes to ifconfig(8) for dynamic address family detection mandate
102 that you are running a kernel of 20110525 or later. Make sure to
103 follow the update procedure to boot a new kernel before installing
107 Support for sun4v architecture is officially dropped
110 Several KPI breaking changes have been committed to the mii(4) layer,
111 the PHY drivers and consequently some Ethernet drivers using mii(4).
112 This means that miibus.ko and the modules of the affected Ethernet
113 drivers need to be recompiled.
115 Note to kernel developers: Given that the OUI bit reversion problem
116 was fixed as part of these changes all mii(4) commits related to OUIs,
117 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
118 handling, no longer can be merged verbatim to stable/8 and previous
122 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
123 into their kernel configurations along with 'device pci'.
126 The default NFS client is now the new NFS client, so fstype "newnfs"
127 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
128 Although mounts via fstype "nfs" will usually work without userland
129 changes, it is recommended that the mount(8) and mount_nfs(8)
130 commands be rebuilt from sources and that a link to mount_nfs called
131 mount_oldnfs be created. The new client is compiled into the
132 kernel with "options NFSCL" and this is needed for diskless root
133 file systems. The GENERIC kernel configs have been changed to use
134 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
135 To use the regular/old client, you can "mount -t oldnfs ...". For
136 a diskless root file system, you must also include a line like:
138 vfs.root.mountfrom="oldnfs:"
140 in the boot/loader.conf on the root fs on the NFS server to make
141 a diskless root fs use the old client.
144 The GENERIC kernels for all architectures now default to the new
145 CAM-based ATA stack. It means that all legacy ATA drivers were
146 removed and replaced by respective CAM drivers. If you are using
147 ATA device names in /etc/fstab or other places, make sure to update
148 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
149 where 'Y's are the sequential numbers starting from zero for each type
150 in order of detection, unless configured otherwise with tunables,
151 see cam(4)). There will be symbolic links created in /dev/ to map
152 old adX devices to the respective adaY. They should provide basic
153 compatibility for file systems mounting in most cases, but they do
154 not support old user-level APIs and do not have respective providers
155 in GEOM. Consider using updated management tools with new device names.
157 It is possible to load devices ahci, ata, siis and mvs as modules,
158 but option ATA_CAM should remain in kernel configuration to make ata
159 module work as CAM driver supporting legacy ATA controllers. Device ata
160 still can be used in modular fashion (atacore + ...). Modules atadisk
161 and atapi* are not used and won't affect operation in ATA_CAM mode.
162 Note that to use CAM-based ATA kernel should include CAM devices
163 scbus, pass, da (or explicitly ada), cd and optionally others. All of
164 them are parts of the cam module.
166 ataraid(4) functionality is now supported by the RAID GEOM class.
167 To use it you can load geom_raid kernel module and use graid(8) tool
168 for management. Instead of /dev/arX device names, use /dev/raid/rX.
170 No kernel config options or code have been removed, so if a problem
171 arises, please report it and optionally revert to the old ATA stack.
172 In order to do it you can remove from the kernel config:
177 , and instead add back:
178 device atadisk # ATA disk drives
179 device ataraid # ATA RAID drives
180 device atapicd # ATAPI CDROM drives
181 device atapifd # ATAPI floppy drives
182 device atapist # ATAPI tape drives
185 The default NFS server has been changed to the new server, which
186 was referred to as the experimental server. If you need to switch
187 back to the old NFS server, you must now put the "-o" option on
188 both the mountd and nfsd commands. This can be done using the
189 mountd_flags and nfs_server_flags rc.conf variables until an
190 update to the rc scripts is committed, which is coming soon.
193 The GNU Objective-C runtime library (libobjc), and other Objective-C
194 related components have been removed from the base system. If you
195 require an Objective-C library, please use one of the available ports.
198 ath(4) has been split into bus- and device- modules. if_ath contains
199 the HAL, the TX rate control and the network device code. if_ath_pci
200 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
201 contains the AHB glue. Users need to load both if_ath_pci and if_ath
202 in order to use ath on everything else.
204 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
205 need to load if_ath and if_ath_pci for ath(4) operation.
208 As part of the replacement of sysinstall, the process of building
209 release media has changed significantly. For details, please re-read
210 release(7), which has been updated to reflect the new build process.
213 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
214 is the last available version under GPLv2. It brings a number of new
215 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
216 4.1 and SSE 4.2), better support for powerpc64, a number of new
217 directives, and lots of other small improvements. See the ChangeLog
218 file in contrib/binutils for the full details.
221 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
222 compliant, and will now use half of hash for authentication.
223 This will break interoperability with all stacks (including all
224 actual FreeBSD versions) who implement
225 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
227 The only workaround with such peers is to use another HMAC
228 algorithm for IPsec ("phase 2") authentication.
231 Remove the uio_yield prototype and symbol. This function has
232 been misnamed since it was introduced and should not be
233 globally exposed with this name. The equivalent functionality
234 is now available using kern_yield(curthread->td_user_pri).
235 The function remains undocumented.
238 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
239 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
240 for scalar sysctls is defined but disabled. Code that needs
241 UQUAD to pass the type checking that must compile on older
242 systems where the define is not present can check against
243 __FreeBSD_version >= 900030.
245 The system dialog(1) has been replaced with a new version previously
246 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
247 with the previous version, but the libdialog associated with it has
248 a largely incompatible API. As such, the original version of libdialog
249 will be kept temporarily as libodialog, until its base system consumers
250 are replaced or updated. Bump __FreeBSD_version to 900030.
253 If you are trying to run make universe on a -stable system, and you get
254 the following warning:
255 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
256 unknown. config(8) likely too old."
257 or something similar to it, then you must upgrade your -stable system
258 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
259 stable/8) or build the config from the latest stable/8 branch and
260 install it on your system.
262 Prior to this date, building a current universe on 8-stable system from
263 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
264 error in the first kernel build phase. A new config on those old
265 systems will fix that problem for older versions of -current.
268 The TCP stack has been modified to allow Khelp modules to interact with
269 it via helper hook points and store per-connection data in the TCP
270 control block. Bump __FreeBSD_version to 900029. User space tools that
271 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
275 Generic IEEE 802.3 annex 31B full duplex flow control support has been
276 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
277 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
278 to take advantage of it instead of using custom implementations. This
279 means that these drivers now no longer unconditionally advertise
280 support for flow control but only do so if flow control is a selected
281 media option. This was implemented in the generic support that way in
282 order to allow flow control to be switched on and off via ifconfig(8)
283 with the PHY specific default to typically off in order to protect
284 from unwanted effects. Consequently, if you used flow control with
285 one of the above mentioned drivers you now need to explicitly enable
287 ifconfig bge0 media auto mediaopt flowcontrol
289 Along with the above mentioned changes generic support for setting
290 1000baseT master mode also has been added and brgphy(4), ciphy(4),
291 e1000phy(4) as well as ip1000phy(4) have been converted to take
292 advantage of it. This means that these drivers now no longer take the
293 link0 parameter for selecting master mode but the master media option
294 has to be used instead, for example like in the following:
295 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
297 Selection of master mode now is also available with all other PHY
298 drivers supporting 1000baseT.
301 The TCP stack has received a significant update to add support for
302 modularised congestion control and generally improve the clarity of
303 congestion control decisions. Bump __FreeBSD_version to 900025. User
304 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
305 sockstat) need to be recompiled.
308 The man(1) utility has been replaced by a new version that no longer
309 uses /etc/manpath.config. Please consult man.conf(5) for how to
310 migrate local entries to the new format.
313 The copyright strings printed by login(1) and sshd(8) at the time of a
314 new connection have been removed to follow other operating systems and
318 A workaround for a fixed ld bug has been removed in kernel code,
319 so make sure that your system ld is built from sources after
320 revision 210245 from 2010-07-19 (r211583 if building head kernel
321 on stable/8, r211584 for stable/7; both from 2010-08-21).
322 A symptom of incorrect ld version is different addresses for
323 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
326 The $ipv6_prefer variable in rc.conf(5) has been split into
327 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
329 The $ip6addrctl_policy is a variable to choose a pre-defined
330 address selection policy set by ip6addrctl(8). A value
331 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
334 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
335 flag (see an entry of 20090926) is set on an interface with no
336 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
337 security reason. If you want IPv6 link-local address on all
338 interfaces by default, set this to "YES".
340 The old ipv6_prefer="YES" is equivalent to
341 ipv6_activate_all_interfaces="YES" and
342 ip6addrctl_policy="ipv6_prefer".
345 DTrace has grown support for userland tracing. Due to this, DTrace is
346 now i386 and amd64 only.
347 dtruss(1) is now installed by default on those systems and a new
348 kernel module is needed for userland tracing: fasttrap.
349 No changes to your kernel config file are necessary to enable
350 userland tracing, but you might consider adding 'STRIP=' and
351 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
352 to have informative userland stack traces in DTrace (ustack).
355 The acpi_aiboost(4) driver has been removed in favor of the new
356 aibs(4) driver. You should update your kernel configuration file.
359 BSD grep has been imported to the base system and it is built by
360 default. It is completely BSD licensed, highly GNU-compatible, uses
361 less memory than its GNU counterpart and has a small codebase.
362 However, it is slower than its GNU counterpart, which is mostly
363 noticeable for larger searches, for smaller ones it is measurable
364 but not significant. The reason is complex, the most important factor
365 is that we lack a modern and efficient regex library and GNU
366 overcomes this by optimizing the searches internally. Future work
367 on improving the regex performance is planned, for the meantime,
368 users that need better performance, can build GNU grep instead by
369 setting the WITH_GNU_GREP knob.
372 Due to the import of powerpc64 support, all existing powerpc kernel
373 configuration files must be updated with a machine directive like this:
374 machine powerpc powerpc
376 In addition, an updated config(8) is required to build powerpc kernels
380 A new version of ZFS (version 15) has been merged to -HEAD.
381 This version uses a python library for the following subcommands:
382 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
383 For full functionality of these commands the following port must
384 be installed: sysutils/py-zfs
387 'vm_page's are now hashed by physical address to an array of mutexes.
388 Currently this is only used to serialize access to hold_count. Over
389 time the page queue mutex will be peeled away. This changes the size
390 of pmap on every architecture. And requires all callers of vm_page_hold
391 and vm_page_unhold to be updated.
394 WITH_CTF can now be specified in src.conf (not recommended, there
395 are some problems with static executables), make.conf (would also
396 affect ports which do not use GNU make and do not override the
397 compile targets) or in the kernel config (via "makeoptions
399 When WITH_CTF was specified there before this was silently ignored,
400 so make sure that WITH_CTF is not used in places which could lead
401 to unwanted behavior.
404 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
405 to allow 32-bit compatibility on non-x86 platforms. All kernel
406 configurations on amd64 and ia64 platforms using these options must
407 be modified accordingly.
410 The utmp user accounting database has been replaced with utmpx,
411 the user accounting interface standardized by POSIX.
412 Unfortunately the semantics of utmp and utmpx don't match,
413 making it practically impossible to support both interfaces.
414 The user accounting database is used by tools like finger(1),
415 last(1), talk(1), w(1) and ac(8).
417 All applications in the base system use utmpx. This means only
418 local binaries (e.g. from the ports tree) may still use these
419 utmp database files. These applications must be rebuilt to make
422 After the system has been upgraded, it is safe to remove the old
423 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
424 assuming their contents is of no importance anymore. Old wtmp
425 databases can only be used by last(1) and ac(8) after they have
426 been converted to the new format using wtmpcvt(1).
429 Introduce the kernel thread "deadlock resolver" (which can be enabled
430 via the DEADLKRES option, see NOTES for more details) and the
431 sleepq_type() function for sleepqueues.
434 The rc.firewall and rc.firewall6 were unified, and
435 rc.firewall6 and rc.d/ip6fw were removed.
436 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
437 variables are obsoleted. Instead, the following new rc
438 variables are added to rc.d/ipfw:
440 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
441 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
442 firewall_simple_onet_ipv6, firewall_trusted_ipv6
444 The meanings correspond to the relevant IPv4 variables.
450 The default terminal emulation for syscons(4) has been changed
451 from cons25 to xterm on all platforms except pc98. This means
452 that the /etc/ttys file needs to be updated to ensure correct
453 operation of applications on the console.
455 The terminal emulation style can be toggled per window by using
456 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
457 options can be used to change the compile-time default back to
460 To prevent graphical artifacts, make sure the TERM environment
461 variable is set to match the terminal emulation that is being
462 performed by syscons(4).
465 The layout of the structure ieee80211req_scan_result has changed.
466 Applications that require wireless scan results (e.g. ifconfig(8))
467 from net80211 need to be recompiled.
469 Applications such as wpa_supplicant(8) may require a full world
470 build without using NO_CLEAN in order to get synchronized with the
474 The iwn(4) driver has been updated to support the 5000 and 5150 series.
475 There's one kernel module for each firmware. Adding "device iwnfw"
476 to the kernel configuration file means including all three firmware
477 images inside the kernel. If you want to include just the one for
478 your wireless card, use the devices iwn4965fw, iwn5000fw or
482 The rc.d/network_ipv6, IPv6 configuration script has been integrated
483 into rc.d/netif. The changes are the following:
485 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
486 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
487 Note that both variables need the "inet6" keyword at the head.
489 Do not set $ipv6_network_interfaces manually if you do not
490 understand what you are doing. It is not needed in most cases.
492 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
495 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
496 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
498 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
499 all configured interfaces have "inet6 accept_rtadv" in the
500 $ifconfig_IF_ipv6. These are for backward compatibility.
502 3. A new variable $ipv6_prefer has been added. If NO, IPv6
503 functionality of interfaces with no corresponding
504 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
505 and the default address selection policy of ip6addrctl(8)
506 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
507 Note that if you want to configure IPv6 functionality on the
508 disabled interfaces after boot, first you need to clear the flag by
509 using ifconfig(8) like:
511 ifconfig em0 inet6 -ifdisabled
513 If YES, the default address selection policy is set as
516 The default value of $ipv6_prefer is NO.
518 4. If your system need to receive Router Advertisement messages,
519 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
520 scripts automatically invoke rtsol(8) when the interface becomes
521 UP. The Router Advertisement messages are used for SLAAC
522 (State-Less Address AutoConfiguration).
525 802.11s D3.03 support was committed. This is incompatible with the
526 previous code, which was based on D3.0.
529 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
530 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
531 control whether accepting Router Advertisement messages or not.
532 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
533 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
534 The ifconfig(8) utility now supports these flags.
537 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
538 mount(8) and -a option for df(1) to see them.
541 The old tunable hw.bus.devctl_disable has been superseded by
542 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
543 replaced by hw.bus.devctl_queue=0. The default for this new tunable
547 Remove the option STOP_NMI. The default action is now to use NMI only
548 for KDB via the newly introduced function stop_cpus_hard() and
549 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
552 The stable/8 branch created in subversion. This corresponds to the
553 RELENG_8 branch in CVS.
556 Bump the shared library version numbers for all libraries that do not
557 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
558 __FreeBSD_version to 800105.
561 Due to changes in the implementation of virtual network stack support,
562 all network-related kernel modules must be recompiled. As this change
563 breaks the ABI, bump __FreeBSD_version to 800104.
566 The TOE interface to the TCP syncache has been modified to remove
567 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
568 The cxgb driver is the only TOE consumer affected by this change, and
569 needs to be recompiled along with the kernel. As this change breaks
570 the ABI, bump __FreeBSD_version to 800103.
573 Padding has been added to struct tcpcb, sackhint and tcpstat in
574 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
575 maintaining the ABI. However, this change breaks the ABI, so bump
576 __FreeBSD_version to 800102. User space tools that rely on the size of
577 any of these structs (e.g. sockstat) need to be recompiled.
580 The NFS_LEGACYRPC option has been removed along with the old kernel
581 RPC implementation that this option selected. Kernel configurations
582 may need to be adjusted.
585 The network interface device nodes at /dev/net/<interface> have been
586 removed. All ioctl operations can be performed the normal way using
587 routing sockets. The kqueue functionality can generally be replaced
588 with routing sockets.
591 The documentation from the FreeBSD Documentation Project (Handbook,
592 FAQ, etc.) is now installed via packages by sysinstall(8) and under
593 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
596 The ABI of various structures related to the SYSV IPC API have been
597 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
598 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
602 Layout of struct vnet has changed as routing related variables were
603 moved to their own Vimage module. Modules need to be recompiled. Bump
604 __FreeBSD_version to 800099.
607 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
608 respectively. As long as no more than 16 groups per process are used,
609 no changes should be visible. When more than 16 groups are used, old
610 binaries may fail if they call getgroups() or getgrouplist() with
611 statically sized storage. Recompiling will work around this, but
612 applications should be modified to use dynamically allocated storage
613 for group arrays as POSIX.1-2008 does not cap an implementation's
614 number of supported groups at NGROUPS_MAX+1 as previous versions did.
616 NFS and portalfs mounts may also be affected as the list of groups is
617 truncated to 16. Users of NFS who use more than 16 groups, should
618 take care that negative group permissions are not used on the exported
619 file systems as they will not be reliable unless a GSSAPI based
620 authentication method is used.
623 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
624 option compiles in the support for adaptive spinning for lockmgrs
625 which want to enable it. The lockinit() function now accepts the flag
626 LK_ADAPTIVE in order to make the lock object subject to adaptive
627 spinning when both held in write and read mode.
630 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
631 changed. User applications that use this ioctl need to be rebuilt.
634 The layout of struct thread has changed. Kernel and modules need to
638 The layout of structs ifnet, domain, protosw and vnet_net has changed.
639 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
642 window(1) has been removed from the base system. It can now be
643 installed from ports. The port is called misc/window.
646 The way we are storing and accessing `routing table' entries has
647 changed. Programs reading the FIB, like netstat, need to be
651 A new netisr implementation has been added for FreeBSD 8. Network
652 file system modules, such as igmp, ipdivert, and others, should be
654 Bump __FreeBSD_version to 800096.
657 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
661 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
662 Bump __FreeBSD_version to 800094.
665 Add mnt_xflag field to 'struct mount'. File system modules need to be
667 Bump __FreeBSD_version to 800093.
670 The compiling option ADAPTIVE_SX has been retired while it has been
671 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
672 The KPI for sx_init_flags() changes as accepting flags:
673 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
674 been introduced in order to handle the reversed logic.
675 Bump __FreeBSD_version to 800092.
678 Add support for hierarchical jails. Remove global securelevel.
679 Bump __FreeBSD_version to 800091.
682 The layout of struct vnet_net has changed, therefore modules
684 Bump __FreeBSD_version to 800090.
687 The newly imported zic(8) produces a new format in the output. Please
688 run tzsetup(8) to install the newly created data to /etc/localtime.
691 The sysctl tree for the usb stack has renamed from hw.usb2.* to
692 hw.usb.* and is now consistent again with previous releases.
695 802.11 monitor mode support was revised and driver api's were changed.
696 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
697 of DLT_IEEE802_11. No user-visible data structures were changed but
698 applications that use DLT_IEEE802_11 may require changes.
699 Bump __FreeBSD_version to 800088.
702 The layout of the following structs has changed: sysctl_oid,
703 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
704 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
705 panics may be experienced. World rebuild is required for
706 correctly checking networking state from userland.
707 Bump __FreeBSD_version to 800085.
710 MLDv2 and Source-Specific Multicast (SSM) have been merged
711 to the IPv6 stack. VIMAGE hooks are in but not yet used.
712 The implementation of SSM within FreeBSD's IPv6 stack closely
713 follows the IPv4 implementation.
715 For kernel developers:
717 * The most important changes are that the ip6_output() and
718 ip6_input() paths no longer take the IN6_MULTI_LOCK,
719 and this lock has been downgraded to a non-recursive mutex.
721 * As with the changes to the IPv4 stack to support SSM, filtering
722 of inbound multicast traffic must now be performed by transport
723 protocols within the IPv6 stack. This does not apply to TCP and
724 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
726 * The KPIs used by IPv6 multicast are similar to those used by
727 the IPv4 stack, with the following differences:
728 * im6o_mc_filter() is analogous to imo_multicast_filter().
729 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
730 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
731 * IN6_LOOKUP_MULTI() has been deprecated and removed.
732 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
733 for MLDv1 have an additional 'timer' argument which is used to
734 jitter the initial membership report for the solicited-node
735 multicast membership on-link.
736 * This is not strictly needed for MLDv2, which already jitters
737 its report transmissions. However, the 'timer' argument is
738 preserved in case MLDv1 is active on the interface.
740 * The KAME linked-list based IPv6 membership implementation has
741 been refactored to use a vector similar to that used by the IPv4
743 Code which maintains a list of its own multicast memberships
744 internally, e.g. carp, has been updated to reflect the new
747 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
748 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
749 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
750 implementation constraint which needs to be addressed in HEAD.
752 For application developers:
754 * The changes are broadly similar to those made for the IPv4
757 * The use of IPv4 and IPv6 multicast socket options on the same
758 socket, using mapped addresses, HAS NOT been tested or supported.
760 * There are a number of issues with the implementation of various
761 IPv6 multicast APIs which need to be resolved in the API surface
762 before the implementation is fully compatible with KAME userland
763 use, and these are mostly to do with interface index treatment.
765 * The literature available discusses the use of either the delta / ASM
766 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
767 using setsourcefilter(3)/getsourcefilter(3). For more information
768 please refer to RFC 3768, 'Socket Interface Extensions for
769 Multicast Source Filters'.
771 * Applications which use the published RFC 3678 APIs should be fine.
773 For systems administrators:
775 * The mtest(8) utility has been refactored to support IPv6, in
776 addition to IPv4. Interface addresses are no longer accepted
777 as arguments, their names must be used instead. The utility
778 will map the interface name to its first IPv4 address as
779 returned by getifaddrs(3).
781 * The ifmcstat(8) utility has also been updated to print the MLDv2
782 endpoint state and source filter lists via sysctl(3).
784 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
785 loopback of IPv6 multicast datagrams by default; it defaults to 1
786 to preserve the existing behaviour. Disabling multicast loopback is
787 recommended for optimal system performance.
789 * The IPv6 MROUTING code has been changed to examine this sysctl
790 instead of attempting to perform a group lookup before looping
791 back forwarded datagrams.
793 Bump __FreeBSD_version to 800084.
796 Implement low-level Bluetooth HCI API.
797 Bump __FreeBSD_version to 800083.
800 The layout of struct malloc_type, used by modules to register new
801 memory allocation types, has changed. Most modules will need to
802 be rebuilt or panics may be experienced.
803 Bump __FreeBSD_version to 800081.
806 Anticipate overflowing inp_flags - add inp_flags2.
807 This changes most offsets in inpcb, so checking v4 connection
808 state will require a world rebuild.
809 Bump __FreeBSD_version to 800080.
812 Add an llentry to struct route and struct route_in6. Modules
813 embedding a struct route will need to be recompiled.
814 Bump __FreeBSD_version to 800079.
817 The size of rt_metrics_lite and by extension rtentry has changed.
818 Networking administration apps will need to be recompiled.
819 The route command now supports show as an alias for get, weighting
820 of routes, sticky and nostick flags to alter the behavior of stateful
822 Bump __FreeBSD_version to 800078.
825 Do not use Giant for kbdmux(4) locking. This is wrong and
826 apparently causing more problems than it solves. This will
827 re-open the issue where interrupt handlers may race with
828 kbdmux(4) in polling mode. Typical symptoms include (but
829 not limited to) duplicated and/or missing characters when
830 low level console functions (such as gets) are used while
831 interrupts are enabled (for example geli password prompt,
832 mountroot prompt etc.). Disabling kbdmux(4) may help.
835 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
836 kernel modules referencing any of the above need to be recompiled.
837 Bump __FreeBSD_version to 800075.
840 GEOM_PART has become the default partition slicer for storage devices,
841 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
842 introduces some changes:
844 MSDOS/EBR: the devices created from MSDOS extended partition entries
845 (EBR) can be named differently than with GEOM_MBR and are now symlinks
846 to devices with offset-based names. fstabs may need to be modified.
848 BSD: the "geometry does not match label" warning is harmless in most
849 cases but it points to problems in file system misalignment with
850 disk geometry. The "c" partition is now implicit, covers the whole
851 top-level drive and cannot be (mis)used by users.
853 General: Kernel dumps are now not allowed to be written to devices
854 whose partition types indicate they are meant to be used for file
855 systems (or, in case of MSDOS partitions, as something else than
858 Most of these changes date approximately from 200812.
861 The uscanner(4) driver has been removed from the kernel. This follows
862 Linux removing theirs in 2.6 and making libusb the default interface
866 The multicast forwarding code has been cleaned up. netstat(1)
867 only relies on KVM now for printing bandwidth upcall meters.
868 The IPv4 and IPv6 modules are split into ip_mroute_mod and
869 ip6_mroute_mod respectively. The config(5) options for statically
870 compiling this code remain the same, i.e. 'options MROUTING'.
873 Support for the IFF_NEEDSGIANT network interface flag has been
874 removed, which means that non-MPSAFE network device drivers are no
875 longer supported. In particular, if_ar, if_sr, and network device
876 drivers from the old (legacy) USB stack can no longer be built or
880 POSIX.1 Native Language Support (NLS) has been enabled in libc and
881 a bunch of new language catalog files have also been added.
882 This means that some common libc messages are now localized and
883 they depend on the LC_MESSAGES environmental variable.
886 The k8temp(4) driver has been renamed to amdtemp(4) since
887 support for Family 10 and Family 11 CPU families was added.
890 IGMPv3 and Source-Specific Multicast (SSM) have been merged
891 to the IPv4 stack. VIMAGE hooks are in but not yet used.
893 For kernel developers, the most important changes are that the
894 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
895 and this lock has been downgraded to a non-recursive mutex.
897 Transport protocols (UDP, Raw IP) are now responsible for filtering
898 inbound multicast traffic according to group membership and source
899 filters. The imo_multicast_filter() KPI exists for this purpose.
900 Transports which do not use multicast (SCTP, TCP) already reject
901 multicast by default. Forwarding and receive performance may improve
902 as a mutex acquisition is no longer needed in the ip_input()
903 low-level input path. in_addmulti() and in_delmulti() are shimmed
904 to new KPIs which exist to support SSM in-kernel.
906 For application developers, it is recommended that loopback of
907 multicast datagrams be disabled for best performance, as this
908 will still cause the lock to be taken for each looped-back
909 datagram transmission. The net.inet.ip.mcast.loop sysctl may
910 be tuned to 0 to disable loopback by default; it defaults to 1
911 to preserve the existing behaviour.
913 For systems administrators, to obtain best performance with
914 multicast reception and multiple groups, it is always recommended
915 that a card with a suitably precise hash filter is used. Hash
916 collisions will still result in the lock being taken within the
917 transport protocol input path to check group membership.
919 If deploying FreeBSD in an environment with IGMP snooping switches,
920 it is recommended that the net.inet.igmp.sendlocal sysctl remain
921 enabled; this forces 224.0.0.0/24 group membership to be announced
924 The size of 'struct igmpstat' has changed; netstat needs to be
925 recompiled to reflect this.
926 Bump __FreeBSD_version to 800070.
929 libusb20.so.1 is now installed as libusb.so.1 and the ports system
930 updated to use it. This requires a buildworld/installworld in order to
931 update the library and dependencies (usbconfig, etc). Its advisable to
932 rebuild all ports which uses libusb. More specific directions are given
933 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
934 libusb are no longer required and can be removed.
937 A workaround is committed to allow the creation of System V shared
938 memory segment of size > 2 GB on the 64-bit architectures.
939 Due to a limitation of the existing ABI, the shm_segsz member
940 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
941 wrong for large segments. Note that limits must be explicitly
942 raised to allow such segments to be created.
945 The layout of struct ifnet has changed, requiring a rebuild of all
946 network device driver modules.
949 The /dev handling for the new USB stack has changed, a
950 buildworld/installworld is required for libusb20.
953 The new USB2 stack has now been permanently moved in and all kernel and
954 module names reverted to their previous values (eg, usb, ehci, ohci,
955 ums, ...). The old usb stack can be compiled in by prefixing the name
956 with the letter 'o', the old usb modules have been removed.
957 Updating entry 20090216 for xorg and 20090215 for libmap may still
961 The rc.conf(5) option if_up_delay has been renamed to
962 defaultroute_delay to better reflect its purpose. If you have
963 customized this setting in /etc/rc.conf you need to update it to
967 xorg 7.4 wants to configure its input devices via hald which does not
968 yet work with USB2. If the keyboard/mouse does not work in xorg then
970 Option "AllowEmptyInput" "off"
971 to your ServerLayout section. This will cause X to use the configured
972 kbd and mouse sections from your xorg.conf.
975 The GENERIC kernels for all architectures now default to the new USB2
976 stack. No kernel config options or code have been removed so if a
977 problem arises please report it and optionally revert to the old USB
978 stack. If you are loading USB kernel modules or have a custom kernel
979 that includes GENERIC then ensure that usb names are also changed over,
980 eg uftdi -> usb2_serial_ftdi.
982 Older programs linked against the ports libusb 0.1 need to be
983 redirected to the new stack's libusb20. /etc/libmap.conf can
985 # Map old usb library to new one for usb2 stack
986 libusb-0.1.so.8 libusb20.so.1
989 All USB ethernet devices now attach as interfaces under the name ueN
990 (eg. ue0). This is to provide a predictable name as vendors often
991 change usb chipsets in a product without notice.
994 The ichsmb(4) driver has been changed to require SMBus slave
995 addresses be left-justified (xxxxxxx0b) rather than right-justified.
996 All of the other SMBus controller drivers require left-justified
997 slave addresses, so this change makes all the drivers provide the
1001 INET6 statistics (struct ip6stat) was updated.
1002 netstat(1) needs to be recompiled.
1005 NTFS has been removed from GENERIC kernel on amd64 to match
1006 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1007 will load ntfs.ko module automatically when NTFS support is
1008 actually needed, unless ntfs.ko is not installed or security
1009 level prohibits loading kernel modules. If either is the case,
1010 "options NTFS" has to be added into kernel config.
1013 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1014 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1015 800061. User space tools that rely on the size of struct tcpcb in
1016 tcp_var.h (e.g. sockstat) need to be recompiled.
1019 ng_tty(4) module updated to match the new TTY subsystem.
1020 Due to API change, user-level applications must be updated.
1021 New API support added to mpd5 CVS and expected to be present
1022 in next mpd5.3 release.
1025 With __FreeBSD_version 800060 the makefs tool is part of
1026 the base system (it was a port).
1029 The afdata and ifnet locks have been changed from mutexes to
1030 rwlocks, network modules will need to be re-compiled.
1033 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1034 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1035 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1036 architecture and 40 bytes on 64-bit architecture. The userland
1037 applications "arp" and "ndp" have been updated accordingly.
1038 The output from "netstat -r" shows only routing entries and
1039 none of the L2 information.
1042 __FreeBSD_version 800057 marks the switchover from the
1043 binary ath hal to source code. Users must add the line:
1045 options AH_SUPPORT_AR5416
1047 to their kernel config files when specifying:
1051 The ath_hal module no longer exists; the code is now compiled
1052 together with the driver in the ath module. It is now
1053 possible to tailor chip support (i.e. reduce the set of chips
1054 and thereby the code size); consult ath_hal(4) for details.
1057 __FreeBSD_version 800054 adds memory barriers to
1058 <machine/atomic.h>, new interfaces to ifnet to facilitate
1059 multiple hardware transmit queues for cards that support
1060 them, and a lock-less ring-buffer implementation to
1061 enable drivers to more efficiently manage queueing of
1065 A new version of ZFS (version 13) has been merged to -HEAD.
1066 This version has zpool attribute "listsnapshots" off by
1067 default, which means "zfs list" does not show snapshots,
1068 and is the same as Solaris behavior.
1071 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1074 The uhci, ohci, ehci and slhci USB Host controller drivers have
1075 been put into separate modules. If you load the usb module
1076 separately through loader.conf you will need to load the
1077 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1078 controller add the following to loader.conf:
1084 The ABI used by the PMC toolset has changed. Please keep
1085 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1089 atapci kernel module now includes only generic PCI ATA
1090 driver. AHCI driver moved to ataahci kernel module.
1091 All vendor-specific code moved into separate kernel modules:
1092 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1093 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1094 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1095 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1098 The TTY subsystem of the kernel has been replaced by a new
1099 implementation, which provides better scalability and an
1100 improved driver model. Most common drivers have been migrated to
1101 the new TTY subsystem, while others have not. The following
1102 drivers have not yet been ported to the new TTY layer:
1105 cy, digi, rc, rp, sio
1111 ng_h4, ng_tty, ppp, sl, snp
1113 Adding these drivers to your kernel configuration file shall
1114 cause compilation to fail.
1117 ntpd has been upgraded to 4.2.4p5.
1120 OpenSSH has been upgraded to 5.1p1.
1122 For many years, FreeBSD's version of OpenSSH preferred DSA
1123 over RSA for host and user authentication keys. With this
1124 upgrade, we've switched to the vendor's default of RSA over
1125 DSA. This may cause upgraded clients to warn about unknown
1126 host keys even for previously known hosts. Users should
1127 follow the usual procedure for verifying host keys before
1128 accepting the RSA key.
1130 This can be circumvented by setting the "HostKeyAlgorithms"
1131 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1134 Please note that the sequence of keys offered for
1135 authentication has been changed as well. You may want to
1136 specify IdentityFile in a different order to revert this
1140 The sio(4) driver has been removed from the i386 and amd64
1141 kernel configuration files. This means uart(4) is now the
1142 default serial port driver on those platforms as well.
1144 To prevent collisions with the sio(4) driver, the uart(4) driver
1145 uses different names for its device nodes. This means the
1146 onboard serial port will now most likely be called "ttyu0"
1147 instead of "ttyd0". You may need to reconfigure applications to
1148 use the new device names.
1150 When using the serial port as a boot console, be sure to update
1151 /boot/device.hints and /etc/ttys before booting the new kernel.
1152 If you forget to do so, you can still manually specify the hints
1153 at the loader prompt:
1155 set hint.uart.0.at="isa"
1156 set hint.uart.0.port="0x3F8"
1157 set hint.uart.0.flags="0x10"
1158 set hint.uart.0.irq="4"
1162 The gpt(8) utility has been removed. Use gpart(8) to partition
1166 The version that Linuxulator emulates was changed from 2.4.2
1167 to 2.6.16. If you experience any problems with Linux binaries
1168 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1169 if it fixes the problem contact emulation mailing list.
1172 ISDN4BSD (I4B) was removed from the src tree. You may need to
1173 update a your kernel configuration and remove relevant entries.
1176 I have checked in code to support multiple routing tables.
1177 See the man pages setfib(1) and setfib(2).
1178 This is a hopefully backwards compatible version,
1179 but to make use of it you need to compile your kernel
1180 with options ROUTETABLES=2 (or more up to 16).
1183 The 802.11 wireless support was redone to enable multi-bss
1184 operation on devices that are capable. The underlying device
1185 is no longer used directly but instead wlanX devices are
1186 cloned with ifconfig. This requires changes to rc.conf files.
1187 For example, change:
1188 ifconfig_ath0="WPA DHCP"
1191 ifconfig_wlan0="WPA DHCP"
1192 see rc.conf(5) for more details. In addition, mergemaster of
1193 /etc/rc.d is highly recommended. Simultaneous update of userland
1194 and kernel wouldn't hurt either.
1196 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1197 modules were merged into the base wlan module. All references
1198 to these modules (e.g. in kernel config files) must be removed.
1201 psm(4) has gained write(2) support in native operation level.
1202 Arbitrary commands can be written to /dev/psm%d and status can
1203 be read back from it. Therefore, an application is responsible
1204 for status validation and error recovery. It is a no-op in
1205 other operation levels.
1208 Support for KSE threading has been removed from the kernel. To
1209 run legacy applications linked against KSE libmap.conf may
1210 be used. The following libmap.conf may be used to ensure
1211 compatibility with any prior release:
1213 libpthread.so.1 libthr.so.1
1214 libpthread.so.2 libthr.so.2
1215 libkse.so.3 libthr.so.3
1218 The layout of struct vmspace has changed. This affects libkvm
1219 and any executables that link against libkvm and use the
1220 kvm_getprocs() function. In particular, but not exclusively,
1221 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1222 The effects are minimal, but it's advisable to upgrade world
1226 The latest em driver no longer has support in it for the
1227 82575 adapter, this is now moved to the igb driver. The
1228 split was done to make new features that are incompatible
1229 with older hardware easier to do.
1232 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1233 likewise the kernel option is now GEOM_LINUX_LVM.
1236 The default NFS mount mode has changed from UDP to TCP for
1237 increased reliability. If you rely on (insecurely) NFS
1238 mounting across a firewall you may need to update your
1242 Belatedly note the addition of m_collapse for compacting
1246 The fts(3) structures have been changed to use adequate
1247 integer types for their members and so to be able to cope
1248 with huge file trees. The old fts(3) ABI is preserved
1249 through symbol versioning in libc, so third-party binaries
1250 using fts(3) should still work, although they will not take
1251 advantage of the extended types. At the same time, some
1252 third-party software might fail to build after this change
1253 due to unportable assumptions made in its source code about
1254 fts(3) structure members. Such software should be fixed
1255 by its vendor or, in the worst case, in the ports tree.
1256 FreeBSD_version 800015 marks this change for the unlikely
1257 case that a portable fix is impossible.
1260 To upgrade to -current after this date, you must be running
1261 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1262 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1265 The ADAPTIVE_GIANT kernel option has been retired because its
1266 functionality is the default now.
1269 The AT keyboard emulation of sunkbd(4) has been turned on
1270 by default. In order to make the special symbols of the Sun
1271 keyboards driven by sunkbd(4) work under X these now have
1272 to be configured the same way as Sun USB keyboards driven
1273 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1275 Option "XkbLayout" "us"
1276 Option "XkbRules" "xorg"
1277 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1280 It has been decided that it is desirable to provide ABI
1281 backwards compatibility to the FreeBSD 4/5/6 versions of the
1282 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1283 broken with the introduction of PCI domain support (see the
1284 20070930 entry). Unfortunately, this required the ABI of
1285 PCIOCGETCONF to be broken again in order to be able to
1286 provide backwards compatibility to the old version of that
1287 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1288 again. As for prominent ports this affects neither pciutils
1289 nor xorg-server this time, the hal port needs to be rebuilt
1293 The misnamed kthread_create() and friends have been renamed
1294 to kproc_create() etc. Many of the callers already
1295 used kproc_start()..
1296 I will return kthread_create() and friends in a while
1297 with implementations that actually create threads, not procs.
1298 Renaming corresponds with version 800002.
1307 Avoid using make -j when upgrading. While generally safe, there are
1308 sometimes problems using -j to upgrade. If your upgrade fails with
1309 -j, please try again without -j. From time to time in the past there
1310 have been problems using -j with buildworld and/or installworld. This
1311 is especially true when upgrading between "distant" versions (eg one
1312 that cross a major release boundary or several minor releases, or when
1313 several months have passed on the -current branch).
1315 Sometimes, obscure build problems are the result of environment
1316 poisoning. This can happen because the make utility reads its
1317 environment when searching for values for global variables. To run
1318 your build attempts in an "environmental clean room", prefix all make
1319 commands with 'env -i '. See the env(1) manual page for more details.
1321 When upgrading from one major version to another it is generally best
1322 to upgrade to the latest code in the currently installed branch first,
1323 then do an upgrade to the new branch. This is the best-tested upgrade
1324 path, and has the highest probability of being successful. Please try
1325 this approach before reporting problems with a major version upgrade.
1329 When upgrading the boot ZFS pool to a new version, always follow
1332 1.) recompile and reinstall the ZFS boot loader and boot block
1333 (this is part of "make buildworld" and "make installworld")
1335 2.) update the ZFS boot block on your boot drive
1337 The following example updates the ZFS boot block on the first
1338 partition (freebsd-boot) of a GPT partitioned drive ad0:
1339 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1341 Non-boot pools do not need these updates.
1345 If you are updating from a prior version of FreeBSD (even one just
1346 a few days old), you should follow this procedure. It is the most
1347 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1349 make kernel-toolchain
1350 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1351 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1353 To test a kernel once
1354 ---------------------
1355 If you just want to boot a kernel once (because you are not sure
1356 if it works, or if you want to boot a known bad kernel to provide
1357 debugging information) run
1358 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1359 nextboot -k testkernel
1361 To just build a kernel when you know that it won't mess you up
1362 --------------------------------------------------------------
1363 This assumes you are already running a CURRENT system. Replace
1364 ${arch} with the architecture of your machine (e.g. "i386",
1365 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1367 cd src/sys/${arch}/conf
1368 config KERNEL_NAME_HERE
1369 cd ../compile/KERNEL_NAME_HERE
1374 If this fails, go to the "To build a kernel" section.
1376 To rebuild everything and install it on the current system.
1377 -----------------------------------------------------------
1378 # Note: sometimes if you are running current you gotta do more than
1379 # is listed here if you are upgrading from a really old current.
1381 <make sure you have good level 0 dumps>
1383 make kernel KERNCONF=YOUR_KERNEL_HERE
1385 <reboot in single user> [3]
1393 To cross-install current onto a separate partition
1394 --------------------------------------------------
1395 # In this approach we use a separate partition to hold
1396 # current's root, 'usr', and 'var' directories. A partition
1397 # holding "/", "/usr" and "/var" should be about 2GB in
1400 <make sure you have good level 0 dumps>
1403 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1404 <maybe newfs current's root partition>
1405 <mount current's root partition on directory ${CURRENT_ROOT}>
1406 make installworld DESTDIR=${CURRENT_ROOT}
1407 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1408 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1409 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1410 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1411 <reboot into current>
1412 <do a "native" rebuild/install as described in the previous section>
1413 <maybe install compatibility libraries from ports/misc/compat*>
1417 To upgrade in-place from 8.x-stable to current
1418 ----------------------------------------------
1419 <make sure you have good level 0 dumps>
1421 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1423 <reboot in single user> [3]
1430 Make sure that you've read the UPDATING file to understand the
1431 tweaks to various things you need. At this point in the life
1432 cycle of current, things change often and you are on your own
1433 to cope. The defaults can also change, so please read ALL of
1434 the UPDATING entries.
1436 Also, if you are tracking -current, you must be subscribed to
1437 freebsd-current@freebsd.org. Make sure that before you update
1438 your sources that you have read and understood all the recent
1439 messages there. If in doubt, please track -stable which has
1440 much fewer pitfalls.
1442 [1] If you have third party modules, such as vmware, you
1443 should disable them at this point so they don't crash your
1446 [3] From the bootblocks, boot -s, and then do
1451 adjkerntz -i # if CMOS is wall time
1452 Also, when doing a major release upgrade, it is required that
1453 you boot into single user mode to do the installworld.
1455 [4] Note: This step is non-optional. Failure to do this step
1456 can result in a significant reduction in the functionality of the
1457 system. Attempting to do it by hand is not recommended and those
1458 that pursue this avenue should read this file carefully, as well
1459 as the archives of freebsd-current and freebsd-hackers mailing lists
1460 for potential gotchas. The -U option is also useful to consider.
1461 See mergemaster(8) for more information.
1463 [5] Usually this step is a noop. However, from time to time
1464 you may need to do this if you get unknown user in the following
1465 step. It never hurts to do it all the time. You may need to
1466 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1467 install) after the buildworld before this step if you last updated
1468 from current before 20020224 or from -stable before 20020408.
1470 [6] This only deletes old files and directories. Old libraries
1471 can be deleted by "make delete-old-libs", but you have to make
1472 sure that no program is using those libraries anymore.
1474 [8] In order to have a kernel that can run the 4.x binaries needed to
1475 do an installworld, you must include the COMPAT_FREEBSD4 option in
1476 your kernel. Failure to do so may leave you with a system that is
1477 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1478 required to run the 5.x binaries on more recent kernels. And so on
1479 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1481 Make sure that you merge any new devices from GENERIC since the
1482 last time you updated your kernel config file.
1484 [9] When checking out sources, you must include the -P flag to have
1485 cvs prune empty directories.
1487 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1488 "?=" instead of the "=" assignment operator, so that buildworld can
1489 override the CPUTYPE if it needs to.
1491 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1492 not on the command line, or in /etc/make.conf. buildworld will
1493 warn if it is improperly defined.
1496 This file contains a list, in reverse chronological order, of major
1497 breakages in tracking -current. It is not guaranteed to be a complete
1498 list of such breakages, and only contains entries since October 10, 2007.
1499 If you need to see UPDATING entries from before that date, you will need
1500 to fetch an UPDATING file from an older FreeBSD release.
1502 Copyright information:
1504 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1506 Redistribution, publication, translation and use, with or without
1507 modification, in full or in part, in any form or format of this
1508 document are permitted without further permission from the author.
1510 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1511 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1512 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1513 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1514 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1515 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1516 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1517 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1518 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1519 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1520 POSSIBILITY OF SUCH DAMAGE.
1522 Contact Warner Losh if you have any questions about your use of