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 http://www.freebsd.org/doc/en_US.ISO8859-1/books/handbook/makeworld.html
11 Items affecting the ports and packages system can be found in
12 /usr/ports/UPDATING. Please read that file before running portupgrade.
15 The VFS KBI was changed with the merge of several nullfs
16 optimizations and fixes. All filesystem modules must be
20 With the addition of auditdistd(8), a new auditdistd user is now
21 depended on during installworld. "mergemaster -p" can be used to add
22 the user prior to installworld, as documented in the handbook.
28 A new version of ZFS (pool version 5000) has been merged to 9-STABLE.
29 Starting with this version the old system of ZFS pool versioning
30 is superseded by "feature flags". This concept enables forward
31 compatibility against certain future changes in functionality of ZFS
32 pools. The first two read-only compatible "feature flags" for ZFS
33 pools are "com.delphix:async_destroy" and "com.delphix:empty_bpobj".
34 For more information read the new zpool-features(7) manual page.
35 Please refer to the "ZFS notes" section of this file for information
36 on upgrading boot ZFS pools.
39 The IPFIREWALL_FORWARD kernel option has been removed. Its
40 functionality now turned on by default.
43 The random(4) support for the VIA hardware random number
44 generator (`PADLOCK') is no longer enabled unconditionally.
45 Add the PADLOCK_RNG option in the custom kernel config if
46 needed. The GENERIC kernels on i386 and amd64 do include the
47 option, so the change only affects the custom kernel
51 The sparc64 ZFS loader has been changed to no longer try to auto-
52 detect ZFS providers based on diskN aliases but now requires these
53 to be explicitly listed in the OFW boot-device environment variable.
56 Now unix domain sockets behave "as expected" on nullfs(5). Previously
57 nullfs(5) did not pass through all behaviours to the underlying layer,
58 as a result if we bound to a socket on the lower layer we could connect
59 only to the lower path; if we bound to the upper layer we could connect
60 only to the upper path. The new behavior is one can connect to both the
61 lower and the upper paths regardless what layer path one binds to.
64 The acpi_wmi(4) status device /dev/wmistat has been renamed to
68 A new VOP_ADVISE() was added to support posix_fadvise(2). All
69 filesystem modules must be recompiled.
72 The interface of the VOP_VPTOCNP(9) changed, now the returned
73 vnode shall be referenced, previously it was required to be
74 only held. All in-tree filesystems are converted.
80 The broken amd(4) driver has been replaced with esp(4) in the amd64,
81 i386 and pc98 GENERIC kernel configuration files.
84 This commit modifies vfs_register() so that it uses a hash
85 calculation to set vfc_typenum, which is enabled by default.
86 The first time a system is booted after this change, the
87 vfc_typenum values will change for all file systems. The
88 main effect of this is a change to the NFS server file handles
89 for file systems that use vfc_typenum in their fsid, such as ZFS.
90 It will, however, prevent vfc_typenum from changing when file
91 systems are loaded in a different order for subsequent reboots.
92 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
93 until you are ready to remount all NFS clients after a reboot.
96 Bump the shared library version numbers for libraries that
97 do not use symbol versioning, have changed the ABI compared
98 to stable/8 and which shared library version was not bumped.
99 Done as part of 9.0-RELEASE cycle.
102 During the merge of Capsicum features, the fget(9) KPI was modified.
103 This may require the rebuilding of out-of-tree device drivers --
104 issues have been reported specifically with the nVidia device driver.
105 __FreeBSD_version is bumped to 900041.
107 Also, there is a period between 20110811 and 20110814 where the
108 special devices /dev/{stdin,stdout,stderr} did not work correctly.
109 Building world from a kernel during that window may not work.
112 The packet filter (pf) code has been updated to OpenBSD 4.5.
113 You need to update userland tools to be in sync with kernel.
114 This update breaks backward compatibility with earlier pfsync(4)
115 versions. Care must be taken when updating redundant firewall setups.
118 The following sysctls and tunables are retired on x86 platforms:
120 machdep.hlt_logical_cpus
121 The following sysctl is retired:
122 machdep.hyperthreading_allowed
123 The sysctls were supposed to provide a way to dynamically offline and
124 online selected CPUs on x86 platforms, but the implementation has not
125 been reliable especially with SCHED_ULE scheduler.
126 machdep.hyperthreading_allowed tunable is still available to ignore
127 hyperthreading CPUs at OS level.
128 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
129 where X is an APIC ID of a CPU. Be advised, though, that disabling
130 CPUs in non-uniform fashion will result in non-uniform topology and
131 may lead to sub-optimal system performance with SCHED_ULE, which is
135 cpumask_t type is retired and cpuset_t is used in order to describe
139 Changes to ifconfig(8) for dynamic address family detection mandate
140 that you are running a kernel of 20110525 or later. Make sure to
141 follow the update procedure to boot a new kernel before installing
145 Support for sun4v architecture is officially dropped
148 Several KPI breaking changes have been committed to the mii(4) layer,
149 the PHY drivers and consequently some Ethernet drivers using mii(4).
150 This means that miibus.ko and the modules of the affected Ethernet
151 drivers need to be recompiled.
153 Note to kernel developers: Given that the OUI bit reversion problem
154 was fixed as part of these changes all mii(4) commits related to OUIs,
155 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
156 handling, no longer can be merged verbatim to stable/8 and previous
160 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
161 into their kernel configurations along with 'device pci'.
164 The default NFS client is now the new NFS client, so fstype "newnfs"
165 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
166 Although mounts via fstype "nfs" will usually work without userland
167 changes, it is recommended that the mount(8) and mount_nfs(8)
168 commands be rebuilt from sources and that a link to mount_nfs called
169 mount_oldnfs be created. The new client is compiled into the
170 kernel with "options NFSCL" and this is needed for diskless root
171 file systems. The GENERIC kernel configs have been changed to use
172 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
173 To use the regular/old client, you can "mount -t oldnfs ...". For
174 a diskless root file system, you must also include a line like:
176 vfs.root.mountfrom="oldnfs:"
178 in the boot/loader.conf on the root fs on the NFS server to make
179 a diskless root fs use the old client.
182 The GENERIC kernels for all architectures now default to the new
183 CAM-based ATA stack. It means that all legacy ATA drivers were
184 removed and replaced by respective CAM drivers. If you are using
185 ATA device names in /etc/fstab or other places, make sure to update
186 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
187 where 'Y's are the sequential numbers starting from zero for each type
188 in order of detection, unless configured otherwise with tunables,
189 see cam(4)). There will be symbolic links created in /dev/ to map
190 old adX devices to the respective adaY. They should provide basic
191 compatibility for file systems mounting in most cases, but they do
192 not support old user-level APIs and do not have respective providers
193 in GEOM. Consider using updated management tools with new device names.
195 It is possible to load devices ahci, ata, siis and mvs as modules,
196 but option ATA_CAM should remain in kernel configuration to make ata
197 module work as CAM driver supporting legacy ATA controllers. Device ata
198 still can be used in modular fashion (atacore + ...). Modules atadisk
199 and atapi* are not used and won't affect operation in ATA_CAM mode.
200 Note that to use CAM-based ATA kernel should include CAM devices
201 scbus, pass, da (or explicitly ada), cd and optionally others. All of
202 them are parts of the cam module.
204 ataraid(4) functionality is now supported by the RAID GEOM class.
205 To use it you can load geom_raid kernel module and use graid(8) tool
206 for management. Instead of /dev/arX device names, use /dev/raid/rX.
208 No kernel config options or code have been removed, so if a problem
209 arises, please report it and optionally revert to the old ATA stack.
210 In order to do it you can remove from the kernel config:
215 , and instead add back:
216 device atadisk # ATA disk drives
217 device ataraid # ATA RAID drives
218 device atapicd # ATAPI CDROM drives
219 device atapifd # ATAPI floppy drives
220 device atapist # ATAPI tape drives
223 The default NFS server has been changed to the new server, which
224 was referred to as the experimental server. If you need to switch
225 back to the old NFS server, you must now put the "-o" option on
226 both the mountd and nfsd commands. This can be done using the
227 mountd_flags and nfs_server_flags rc.conf variables until an
228 update to the rc scripts is committed, which is coming soon.
231 The GNU Objective-C runtime library (libobjc), and other Objective-C
232 related components have been removed from the base system. If you
233 require an Objective-C library, please use one of the available ports.
236 ath(4) has been split into bus- and device- modules. if_ath contains
237 the HAL, the TX rate control and the network device code. if_ath_pci
238 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
239 contains the AHB glue. Users need to load both if_ath_pci and if_ath
240 in order to use ath on everything else.
242 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
243 need to load if_ath and if_ath_pci for ath(4) operation.
246 As part of the replacement of sysinstall, the process of building
247 release media has changed significantly. For details, please re-read
248 release(7), which has been updated to reflect the new build process.
251 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
252 is the last available version under GPLv2. It brings a number of new
253 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
254 4.1 and SSE 4.2), better support for powerpc64, a number of new
255 directives, and lots of other small improvements. See the ChangeLog
256 file in contrib/binutils for the full details.
259 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
260 compliant, and will now use half of hash for authentication.
261 This will break interoperability with all stacks (including all
262 actual FreeBSD versions) who implement
263 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
265 The only workaround with such peers is to use another HMAC
266 algorithm for IPsec ("phase 2") authentication.
269 Remove the uio_yield prototype and symbol. This function has
270 been misnamed since it was introduced and should not be
271 globally exposed with this name. The equivalent functionality
272 is now available using kern_yield(curthread->td_user_pri).
273 The function remains undocumented.
276 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
277 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
278 for scalar sysctls is defined but disabled. Code that needs
279 UQUAD to pass the type checking that must compile on older
280 systems where the define is not present can check against
281 __FreeBSD_version >= 900030.
283 The system dialog(1) has been replaced with a new version previously
284 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
285 with the previous version, but the libdialog associated with it has
286 a largely incompatible API. As such, the original version of libdialog
287 will be kept temporarily as libodialog, until its base system consumers
288 are replaced or updated. Bump __FreeBSD_version to 900030.
291 If you are trying to run make universe on a -stable system, and you get
292 the following warning:
293 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
294 unknown. config(8) likely too old."
295 or something similar to it, then you must upgrade your -stable system
296 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
297 stable/8) or build the config from the latest stable/8 branch and
298 install it on your system.
300 Prior to this date, building a current universe on 8-stable system from
301 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
302 error in the first kernel build phase. A new config on those old
303 systems will fix that problem for older versions of -current.
306 The TCP stack has been modified to allow Khelp modules to interact with
307 it via helper hook points and store per-connection data in the TCP
308 control block. Bump __FreeBSD_version to 900029. User space tools that
309 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
313 Generic IEEE 802.3 annex 31B full duplex flow control support has been
314 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
315 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
316 to take advantage of it instead of using custom implementations. This
317 means that these drivers now no longer unconditionally advertise
318 support for flow control but only do so if flow control is a selected
319 media option. This was implemented in the generic support that way in
320 order to allow flow control to be switched on and off via ifconfig(8)
321 with the PHY specific default to typically off in order to protect
322 from unwanted effects. Consequently, if you used flow control with
323 one of the above mentioned drivers you now need to explicitly enable
325 ifconfig bge0 media auto mediaopt flowcontrol
327 Along with the above mentioned changes generic support for setting
328 1000baseT master mode also has been added and brgphy(4), ciphy(4),
329 e1000phy(4) as well as ip1000phy(4) have been converted to take
330 advantage of it. This means that these drivers now no longer take the
331 link0 parameter for selecting master mode but the master media option
332 has to be used instead, for example like in the following:
333 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
335 Selection of master mode now is also available with all other PHY
336 drivers supporting 1000baseT.
339 The TCP stack has received a significant update to add support for
340 modularised congestion control and generally improve the clarity of
341 congestion control decisions. Bump __FreeBSD_version to 900025. User
342 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
343 sockstat) need to be recompiled.
346 The man(1) utility has been replaced by a new version that no longer
347 uses /etc/manpath.config. Please consult man.conf(5) for how to
348 migrate local entries to the new format.
351 The copyright strings printed by login(1) and sshd(8) at the time of a
352 new connection have been removed to follow other operating systems and
356 A workaround for a fixed ld bug has been removed in kernel code,
357 so make sure that your system ld is built from sources after
358 revision 210245 from 2010-07-19 (r211583 if building head kernel
359 on stable/8, r211584 for stable/7; both from 2010-08-21).
360 A symptom of incorrect ld version is different addresses for
361 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
364 The $ipv6_prefer variable in rc.conf(5) has been split into
365 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
367 The $ip6addrctl_policy is a variable to choose a pre-defined
368 address selection policy set by ip6addrctl(8). A value
369 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
372 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
373 flag (see an entry of 20090926) is set on an interface with no
374 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
375 security reason. If you want IPv6 link-local address on all
376 interfaces by default, set this to "YES".
378 The old ipv6_prefer="YES" is equivalent to
379 ipv6_activate_all_interfaces="YES" and
380 ip6addrctl_policy="ipv6_prefer".
383 DTrace has grown support for userland tracing. Due to this, DTrace is
384 now i386 and amd64 only.
385 dtruss(1) is now installed by default on those systems and a new
386 kernel module is needed for userland tracing: fasttrap.
387 No changes to your kernel config file are necessary to enable
388 userland tracing, but you might consider adding 'STRIP=' and
389 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
390 to have informative userland stack traces in DTrace (ustack).
393 The acpi_aiboost(4) driver has been removed in favor of the new
394 aibs(4) driver. You should update your kernel configuration file.
397 BSD grep has been imported to the base system and it is built by
398 default. It is completely BSD licensed, highly GNU-compatible, uses
399 less memory than its GNU counterpart and has a small codebase.
400 However, it is slower than its GNU counterpart, which is mostly
401 noticeable for larger searches, for smaller ones it is measurable
402 but not significant. The reason is complex, the most important factor
403 is that we lack a modern and efficient regex library and GNU
404 overcomes this by optimizing the searches internally. Future work
405 on improving the regex performance is planned, for the meantime,
406 users that need better performance, can build GNU grep instead by
407 setting the WITH_GNU_GREP knob.
410 Due to the import of powerpc64 support, all existing powerpc kernel
411 configuration files must be updated with a machine directive like this:
412 machine powerpc powerpc
414 In addition, an updated config(8) is required to build powerpc kernels
418 A new version of ZFS (version 15) has been merged to -HEAD.
419 This version uses a python library for the following subcommands:
420 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
421 For full functionality of these commands the following port must
422 be installed: sysutils/py-zfs
425 'vm_page's are now hashed by physical address to an array of mutexes.
426 Currently this is only used to serialize access to hold_count. Over
427 time the page queue mutex will be peeled away. This changes the size
428 of pmap on every architecture. And requires all callers of vm_page_hold
429 and vm_page_unhold to be updated.
432 WITH_CTF can now be specified in src.conf (not recommended, there
433 are some problems with static executables), make.conf (would also
434 affect ports which do not use GNU make and do not override the
435 compile targets) or in the kernel config (via "makeoptions
437 When WITH_CTF was specified there before this was silently ignored,
438 so make sure that WITH_CTF is not used in places which could lead
439 to unwanted behavior.
442 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
443 to allow 32-bit compatibility on non-x86 platforms. All kernel
444 configurations on amd64 and ia64 platforms using these options must
445 be modified accordingly.
448 The utmp user accounting database has been replaced with utmpx,
449 the user accounting interface standardized by POSIX.
450 Unfortunately the semantics of utmp and utmpx don't match,
451 making it practically impossible to support both interfaces.
452 The user accounting database is used by tools like finger(1),
453 last(1), talk(1), w(1) and ac(8).
455 All applications in the base system use utmpx. This means only
456 local binaries (e.g. from the ports tree) may still use these
457 utmp database files. These applications must be rebuilt to make
460 After the system has been upgraded, it is safe to remove the old
461 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
462 assuming their contents is of no importance anymore. Old wtmp
463 databases can only be used by last(1) and ac(8) after they have
464 been converted to the new format using wtmpcvt(1).
467 Introduce the kernel thread "deadlock resolver" (which can be enabled
468 via the DEADLKRES option, see NOTES for more details) and the
469 sleepq_type() function for sleepqueues.
472 The rc.firewall and rc.firewall6 were unified, and
473 rc.firewall6 and rc.d/ip6fw were removed.
474 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
475 variables are obsoleted. Instead, the following new rc
476 variables are added to rc.d/ipfw:
478 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
479 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
480 firewall_simple_onet_ipv6, firewall_trusted_ipv6
482 The meanings correspond to the relevant IPv4 variables.
488 The default terminal emulation for syscons(4) has been changed
489 from cons25 to xterm on all platforms except pc98. This means
490 that the /etc/ttys file needs to be updated to ensure correct
491 operation of applications on the console.
493 The terminal emulation style can be toggled per window by using
494 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
495 options can be used to change the compile-time default back to
498 To prevent graphical artifacts, make sure the TERM environment
499 variable is set to match the terminal emulation that is being
500 performed by syscons(4).
503 The layout of the structure ieee80211req_scan_result has changed.
504 Applications that require wireless scan results (e.g. ifconfig(8))
505 from net80211 need to be recompiled.
507 Applications such as wpa_supplicant(8) may require a full world
508 build without using NO_CLEAN in order to get synchronized with the
512 The iwn(4) driver has been updated to support the 5000 and 5150 series.
513 There's one kernel module for each firmware. Adding "device iwnfw"
514 to the kernel configuration file means including all three firmware
515 images inside the kernel. If you want to include just the one for
516 your wireless card, use the devices iwn4965fw, iwn5000fw or
520 The rc.d/network_ipv6, IPv6 configuration script has been integrated
521 into rc.d/netif. The changes are the following:
523 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
524 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
525 Note that both variables need the "inet6" keyword at the head.
527 Do not set $ipv6_network_interfaces manually if you do not
528 understand what you are doing. It is not needed in most cases.
530 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
533 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
534 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
536 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
537 all configured interfaces have "inet6 accept_rtadv" in the
538 $ifconfig_IF_ipv6. These are for backward compatibility.
540 3. A new variable $ipv6_prefer has been added. If NO, IPv6
541 functionality of interfaces with no corresponding
542 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
543 and the default address selection policy of ip6addrctl(8)
544 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
545 Note that if you want to configure IPv6 functionality on the
546 disabled interfaces after boot, first you need to clear the flag by
547 using ifconfig(8) like:
549 ifconfig em0 inet6 -ifdisabled
551 If YES, the default address selection policy is set as
554 The default value of $ipv6_prefer is NO.
556 4. If your system need to receive Router Advertisement messages,
557 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
558 scripts automatically invoke rtsol(8) when the interface becomes
559 UP. The Router Advertisement messages are used for SLAAC
560 (State-Less Address AutoConfiguration).
563 802.11s D3.03 support was committed. This is incompatible with the
564 previous code, which was based on D3.0.
567 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
568 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
569 control whether accepting Router Advertisement messages or not.
570 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
571 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
572 The ifconfig(8) utility now supports these flags.
575 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
576 mount(8) and -a option for df(1) to see them.
579 The old tunable hw.bus.devctl_disable has been superseded by
580 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
581 replaced by hw.bus.devctl_queue=0. The default for this new tunable
585 Remove the option STOP_NMI. The default action is now to use NMI only
586 for KDB via the newly introduced function stop_cpus_hard() and
587 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
590 The stable/8 branch created in subversion. This corresponds to the
591 RELENG_8 branch in CVS.
594 Bump the shared library version numbers for all libraries that do not
595 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
596 __FreeBSD_version to 800105.
599 Due to changes in the implementation of virtual network stack support,
600 all network-related kernel modules must be recompiled. As this change
601 breaks the ABI, bump __FreeBSD_version to 800104.
604 The TOE interface to the TCP syncache has been modified to remove
605 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
606 The cxgb driver is the only TOE consumer affected by this change, and
607 needs to be recompiled along with the kernel. As this change breaks
608 the ABI, bump __FreeBSD_version to 800103.
611 Padding has been added to struct tcpcb, sackhint and tcpstat in
612 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
613 maintaining the ABI. However, this change breaks the ABI, so bump
614 __FreeBSD_version to 800102. User space tools that rely on the size of
615 any of these structs (e.g. sockstat) need to be recompiled.
618 The NFS_LEGACYRPC option has been removed along with the old kernel
619 RPC implementation that this option selected. Kernel configurations
620 may need to be adjusted.
623 The network interface device nodes at /dev/net/<interface> have been
624 removed. All ioctl operations can be performed the normal way using
625 routing sockets. The kqueue functionality can generally be replaced
626 with routing sockets.
629 The documentation from the FreeBSD Documentation Project (Handbook,
630 FAQ, etc.) is now installed via packages by sysinstall(8) and under
631 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
634 The ABI of various structures related to the SYSV IPC API have been
635 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
636 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
640 Layout of struct vnet has changed as routing related variables were
641 moved to their own Vimage module. Modules need to be recompiled. Bump
642 __FreeBSD_version to 800099.
645 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
646 respectively. As long as no more than 16 groups per process are used,
647 no changes should be visible. When more than 16 groups are used, old
648 binaries may fail if they call getgroups() or getgrouplist() with
649 statically sized storage. Recompiling will work around this, but
650 applications should be modified to use dynamically allocated storage
651 for group arrays as POSIX.1-2008 does not cap an implementation's
652 number of supported groups at NGROUPS_MAX+1 as previous versions did.
654 NFS and portalfs mounts may also be affected as the list of groups is
655 truncated to 16. Users of NFS who use more than 16 groups, should
656 take care that negative group permissions are not used on the exported
657 file systems as they will not be reliable unless a GSSAPI based
658 authentication method is used.
661 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
662 option compiles in the support for adaptive spinning for lockmgrs
663 which want to enable it. The lockinit() function now accepts the flag
664 LK_ADAPTIVE in order to make the lock object subject to adaptive
665 spinning when both held in write and read mode.
668 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
669 changed. User applications that use this ioctl need to be rebuilt.
672 The layout of struct thread has changed. Kernel and modules need to
676 The layout of structs ifnet, domain, protosw and vnet_net has changed.
677 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
680 window(1) has been removed from the base system. It can now be
681 installed from ports. The port is called misc/window.
684 The way we are storing and accessing `routing table' entries has
685 changed. Programs reading the FIB, like netstat, need to be
689 A new netisr implementation has been added for FreeBSD 8. Network
690 file system modules, such as igmp, ipdivert, and others, should be
692 Bump __FreeBSD_version to 800096.
695 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
699 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
700 Bump __FreeBSD_version to 800094.
703 Add mnt_xflag field to 'struct mount'. File system modules need to be
705 Bump __FreeBSD_version to 800093.
708 The compiling option ADAPTIVE_SX has been retired while it has been
709 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
710 The KPI for sx_init_flags() changes as accepting flags:
711 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
712 been introduced in order to handle the reversed logic.
713 Bump __FreeBSD_version to 800092.
716 Add support for hierarchical jails. Remove global securelevel.
717 Bump __FreeBSD_version to 800091.
720 The layout of struct vnet_net has changed, therefore modules
722 Bump __FreeBSD_version to 800090.
725 The newly imported zic(8) produces a new format in the output. Please
726 run tzsetup(8) to install the newly created data to /etc/localtime.
729 The sysctl tree for the usb stack has renamed from hw.usb2.* to
730 hw.usb.* and is now consistent again with previous releases.
733 802.11 monitor mode support was revised and driver api's were changed.
734 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
735 of DLT_IEEE802_11. No user-visible data structures were changed but
736 applications that use DLT_IEEE802_11 may require changes.
737 Bump __FreeBSD_version to 800088.
740 The layout of the following structs has changed: sysctl_oid,
741 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
742 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
743 panics may be experienced. World rebuild is required for
744 correctly checking networking state from userland.
745 Bump __FreeBSD_version to 800085.
748 MLDv2 and Source-Specific Multicast (SSM) have been merged
749 to the IPv6 stack. VIMAGE hooks are in but not yet used.
750 The implementation of SSM within FreeBSD's IPv6 stack closely
751 follows the IPv4 implementation.
753 For kernel developers:
755 * The most important changes are that the ip6_output() and
756 ip6_input() paths no longer take the IN6_MULTI_LOCK,
757 and this lock has been downgraded to a non-recursive mutex.
759 * As with the changes to the IPv4 stack to support SSM, filtering
760 of inbound multicast traffic must now be performed by transport
761 protocols within the IPv6 stack. This does not apply to TCP and
762 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
764 * The KPIs used by IPv6 multicast are similar to those used by
765 the IPv4 stack, with the following differences:
766 * im6o_mc_filter() is analogous to imo_multicast_filter().
767 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
768 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
769 * IN6_LOOKUP_MULTI() has been deprecated and removed.
770 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
771 for MLDv1 have an additional 'timer' argument which is used to
772 jitter the initial membership report for the solicited-node
773 multicast membership on-link.
774 * This is not strictly needed for MLDv2, which already jitters
775 its report transmissions. However, the 'timer' argument is
776 preserved in case MLDv1 is active on the interface.
778 * The KAME linked-list based IPv6 membership implementation has
779 been refactored to use a vector similar to that used by the IPv4
781 Code which maintains a list of its own multicast memberships
782 internally, e.g. carp, has been updated to reflect the new
785 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
786 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
787 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
788 implementation constraint which needs to be addressed in HEAD.
790 For application developers:
792 * The changes are broadly similar to those made for the IPv4
795 * The use of IPv4 and IPv6 multicast socket options on the same
796 socket, using mapped addresses, HAS NOT been tested or supported.
798 * There are a number of issues with the implementation of various
799 IPv6 multicast APIs which need to be resolved in the API surface
800 before the implementation is fully compatible with KAME userland
801 use, and these are mostly to do with interface index treatment.
803 * The literature available discusses the use of either the delta / ASM
804 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
805 using setsourcefilter(3)/getsourcefilter(3). For more information
806 please refer to RFC 3768, 'Socket Interface Extensions for
807 Multicast Source Filters'.
809 * Applications which use the published RFC 3678 APIs should be fine.
811 For systems administrators:
813 * The mtest(8) utility has been refactored to support IPv6, in
814 addition to IPv4. Interface addresses are no longer accepted
815 as arguments, their names must be used instead. The utility
816 will map the interface name to its first IPv4 address as
817 returned by getifaddrs(3).
819 * The ifmcstat(8) utility has also been updated to print the MLDv2
820 endpoint state and source filter lists via sysctl(3).
822 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
823 loopback of IPv6 multicast datagrams by default; it defaults to 1
824 to preserve the existing behaviour. Disabling multicast loopback is
825 recommended for optimal system performance.
827 * The IPv6 MROUTING code has been changed to examine this sysctl
828 instead of attempting to perform a group lookup before looping
829 back forwarded datagrams.
831 Bump __FreeBSD_version to 800084.
834 Implement low-level Bluetooth HCI API.
835 Bump __FreeBSD_version to 800083.
838 The layout of struct malloc_type, used by modules to register new
839 memory allocation types, has changed. Most modules will need to
840 be rebuilt or panics may be experienced.
841 Bump __FreeBSD_version to 800081.
844 Anticipate overflowing inp_flags - add inp_flags2.
845 This changes most offsets in inpcb, so checking v4 connection
846 state will require a world rebuild.
847 Bump __FreeBSD_version to 800080.
850 Add an llentry to struct route and struct route_in6. Modules
851 embedding a struct route will need to be recompiled.
852 Bump __FreeBSD_version to 800079.
855 The size of rt_metrics_lite and by extension rtentry has changed.
856 Networking administration apps will need to be recompiled.
857 The route command now supports show as an alias for get, weighting
858 of routes, sticky and nostick flags to alter the behavior of stateful
860 Bump __FreeBSD_version to 800078.
863 Do not use Giant for kbdmux(4) locking. This is wrong and
864 apparently causing more problems than it solves. This will
865 re-open the issue where interrupt handlers may race with
866 kbdmux(4) in polling mode. Typical symptoms include (but
867 not limited to) duplicated and/or missing characters when
868 low level console functions (such as gets) are used while
869 interrupts are enabled (for example geli password prompt,
870 mountroot prompt etc.). Disabling kbdmux(4) may help.
873 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
874 kernel modules referencing any of the above need to be recompiled.
875 Bump __FreeBSD_version to 800075.
878 GEOM_PART has become the default partition slicer for storage devices,
879 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
880 introduces some changes:
882 MSDOS/EBR: the devices created from MSDOS extended partition entries
883 (EBR) can be named differently than with GEOM_MBR and are now symlinks
884 to devices with offset-based names. fstabs may need to be modified.
886 BSD: the "geometry does not match label" warning is harmless in most
887 cases but it points to problems in file system misalignment with
888 disk geometry. The "c" partition is now implicit, covers the whole
889 top-level drive and cannot be (mis)used by users.
891 General: Kernel dumps are now not allowed to be written to devices
892 whose partition types indicate they are meant to be used for file
893 systems (or, in case of MSDOS partitions, as something else than
896 Most of these changes date approximately from 200812.
899 The uscanner(4) driver has been removed from the kernel. This follows
900 Linux removing theirs in 2.6 and making libusb the default interface
904 The multicast forwarding code has been cleaned up. netstat(1)
905 only relies on KVM now for printing bandwidth upcall meters.
906 The IPv4 and IPv6 modules are split into ip_mroute_mod and
907 ip6_mroute_mod respectively. The config(5) options for statically
908 compiling this code remain the same, i.e. 'options MROUTING'.
911 Support for the IFF_NEEDSGIANT network interface flag has been
912 removed, which means that non-MPSAFE network device drivers are no
913 longer supported. In particular, if_ar, if_sr, and network device
914 drivers from the old (legacy) USB stack can no longer be built or
918 POSIX.1 Native Language Support (NLS) has been enabled in libc and
919 a bunch of new language catalog files have also been added.
920 This means that some common libc messages are now localized and
921 they depend on the LC_MESSAGES environmental variable.
924 The k8temp(4) driver has been renamed to amdtemp(4) since
925 support for Family 10 and Family 11 CPU families was added.
928 IGMPv3 and Source-Specific Multicast (SSM) have been merged
929 to the IPv4 stack. VIMAGE hooks are in but not yet used.
931 For kernel developers, the most important changes are that the
932 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
933 and this lock has been downgraded to a non-recursive mutex.
935 Transport protocols (UDP, Raw IP) are now responsible for filtering
936 inbound multicast traffic according to group membership and source
937 filters. The imo_multicast_filter() KPI exists for this purpose.
938 Transports which do not use multicast (SCTP, TCP) already reject
939 multicast by default. Forwarding and receive performance may improve
940 as a mutex acquisition is no longer needed in the ip_input()
941 low-level input path. in_addmulti() and in_delmulti() are shimmed
942 to new KPIs which exist to support SSM in-kernel.
944 For application developers, it is recommended that loopback of
945 multicast datagrams be disabled for best performance, as this
946 will still cause the lock to be taken for each looped-back
947 datagram transmission. The net.inet.ip.mcast.loop sysctl may
948 be tuned to 0 to disable loopback by default; it defaults to 1
949 to preserve the existing behaviour.
951 For systems administrators, to obtain best performance with
952 multicast reception and multiple groups, it is always recommended
953 that a card with a suitably precise hash filter is used. Hash
954 collisions will still result in the lock being taken within the
955 transport protocol input path to check group membership.
957 If deploying FreeBSD in an environment with IGMP snooping switches,
958 it is recommended that the net.inet.igmp.sendlocal sysctl remain
959 enabled; this forces 224.0.0.0/24 group membership to be announced
962 The size of 'struct igmpstat' has changed; netstat needs to be
963 recompiled to reflect this.
964 Bump __FreeBSD_version to 800070.
967 libusb20.so.1 is now installed as libusb.so.1 and the ports system
968 updated to use it. This requires a buildworld/installworld in order to
969 update the library and dependencies (usbconfig, etc). Its advisable to
970 rebuild all ports which uses libusb. More specific directions are given
971 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
972 libusb are no longer required and can be removed.
975 A workaround is committed to allow the creation of System V shared
976 memory segment of size > 2 GB on the 64-bit architectures.
977 Due to a limitation of the existing ABI, the shm_segsz member
978 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
979 wrong for large segments. Note that limits must be explicitly
980 raised to allow such segments to be created.
983 The layout of struct ifnet has changed, requiring a rebuild of all
984 network device driver modules.
987 The /dev handling for the new USB stack has changed, a
988 buildworld/installworld is required for libusb20.
991 The new USB2 stack has now been permanently moved in and all kernel and
992 module names reverted to their previous values (eg, usb, ehci, ohci,
993 ums, ...). The old usb stack can be compiled in by prefixing the name
994 with the letter 'o', the old usb modules have been removed.
995 Updating entry 20090216 for xorg and 20090215 for libmap may still
999 The rc.conf(5) option if_up_delay has been renamed to
1000 defaultroute_delay to better reflect its purpose. If you have
1001 customized this setting in /etc/rc.conf you need to update it to
1005 xorg 7.4 wants to configure its input devices via hald which does not
1006 yet work with USB2. If the keyboard/mouse does not work in xorg then
1008 Option "AllowEmptyInput" "off"
1009 to your ServerLayout section. This will cause X to use the configured
1010 kbd and mouse sections from your xorg.conf.
1013 The GENERIC kernels for all architectures now default to the new USB2
1014 stack. No kernel config options or code have been removed so if a
1015 problem arises please report it and optionally revert to the old USB
1016 stack. If you are loading USB kernel modules or have a custom kernel
1017 that includes GENERIC then ensure that usb names are also changed over,
1018 eg uftdi -> usb2_serial_ftdi.
1020 Older programs linked against the ports libusb 0.1 need to be
1021 redirected to the new stack's libusb20. /etc/libmap.conf can
1023 # Map old usb library to new one for usb2 stack
1024 libusb-0.1.so.8 libusb20.so.1
1027 All USB ethernet devices now attach as interfaces under the name ueN
1028 (eg. ue0). This is to provide a predictable name as vendors often
1029 change usb chipsets in a product without notice.
1032 The ichsmb(4) driver has been changed to require SMBus slave
1033 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1034 All of the other SMBus controller drivers require left-justified
1035 slave addresses, so this change makes all the drivers provide the
1039 INET6 statistics (struct ip6stat) was updated.
1040 netstat(1) needs to be recompiled.
1043 NTFS has been removed from GENERIC kernel on amd64 to match
1044 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1045 will load ntfs.ko module automatically when NTFS support is
1046 actually needed, unless ntfs.ko is not installed or security
1047 level prohibits loading kernel modules. If either is the case,
1048 "options NTFS" has to be added into kernel config.
1051 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1052 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1053 800061. User space tools that rely on the size of struct tcpcb in
1054 tcp_var.h (e.g. sockstat) need to be recompiled.
1057 ng_tty(4) module updated to match the new TTY subsystem.
1058 Due to API change, user-level applications must be updated.
1059 New API support added to mpd5 CVS and expected to be present
1060 in next mpd5.3 release.
1063 With __FreeBSD_version 800060 the makefs tool is part of
1064 the base system (it was a port).
1067 The afdata and ifnet locks have been changed from mutexes to
1068 rwlocks, network modules will need to be re-compiled.
1071 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1072 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1073 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1074 architecture and 40 bytes on 64-bit architecture. The userland
1075 applications "arp" and "ndp" have been updated accordingly.
1076 The output from "netstat -r" shows only routing entries and
1077 none of the L2 information.
1080 __FreeBSD_version 800057 marks the switchover from the
1081 binary ath hal to source code. Users must add the line:
1083 options AH_SUPPORT_AR5416
1085 to their kernel config files when specifying:
1089 The ath_hal module no longer exists; the code is now compiled
1090 together with the driver in the ath module. It is now
1091 possible to tailor chip support (i.e. reduce the set of chips
1092 and thereby the code size); consult ath_hal(4) for details.
1095 __FreeBSD_version 800054 adds memory barriers to
1096 <machine/atomic.h>, new interfaces to ifnet to facilitate
1097 multiple hardware transmit queues for cards that support
1098 them, and a lock-less ring-buffer implementation to
1099 enable drivers to more efficiently manage queueing of
1103 A new version of ZFS (version 13) has been merged to -HEAD.
1104 This version has zpool attribute "listsnapshots" off by
1105 default, which means "zfs list" does not show snapshots,
1106 and is the same as Solaris behavior.
1109 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1112 The uhci, ohci, ehci and slhci USB Host controller drivers have
1113 been put into separate modules. If you load the usb module
1114 separately through loader.conf you will need to load the
1115 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1116 controller add the following to loader.conf:
1122 The ABI used by the PMC toolset has changed. Please keep
1123 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1127 atapci kernel module now includes only generic PCI ATA
1128 driver. AHCI driver moved to ataahci kernel module.
1129 All vendor-specific code moved into separate kernel modules:
1130 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1131 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1132 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1133 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1136 The TTY subsystem of the kernel has been replaced by a new
1137 implementation, which provides better scalability and an
1138 improved driver model. Most common drivers have been migrated to
1139 the new TTY subsystem, while others have not. The following
1140 drivers have not yet been ported to the new TTY layer:
1143 cy, digi, rc, rp, sio
1149 ng_h4, ng_tty, ppp, sl, snp
1151 Adding these drivers to your kernel configuration file shall
1152 cause compilation to fail.
1155 ntpd has been upgraded to 4.2.4p5.
1158 OpenSSH has been upgraded to 5.1p1.
1160 For many years, FreeBSD's version of OpenSSH preferred DSA
1161 over RSA for host and user authentication keys. With this
1162 upgrade, we've switched to the vendor's default of RSA over
1163 DSA. This may cause upgraded clients to warn about unknown
1164 host keys even for previously known hosts. Users should
1165 follow the usual procedure for verifying host keys before
1166 accepting the RSA key.
1168 This can be circumvented by setting the "HostKeyAlgorithms"
1169 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1172 Please note that the sequence of keys offered for
1173 authentication has been changed as well. You may want to
1174 specify IdentityFile in a different order to revert this
1178 The sio(4) driver has been removed from the i386 and amd64
1179 kernel configuration files. This means uart(4) is now the
1180 default serial port driver on those platforms as well.
1182 To prevent collisions with the sio(4) driver, the uart(4) driver
1183 uses different names for its device nodes. This means the
1184 onboard serial port will now most likely be called "ttyu0"
1185 instead of "ttyd0". You may need to reconfigure applications to
1186 use the new device names.
1188 When using the serial port as a boot console, be sure to update
1189 /boot/device.hints and /etc/ttys before booting the new kernel.
1190 If you forget to do so, you can still manually specify the hints
1191 at the loader prompt:
1193 set hint.uart.0.at="isa"
1194 set hint.uart.0.port="0x3F8"
1195 set hint.uart.0.flags="0x10"
1196 set hint.uart.0.irq="4"
1200 The gpt(8) utility has been removed. Use gpart(8) to partition
1204 The version that Linuxulator emulates was changed from 2.4.2
1205 to 2.6.16. If you experience any problems with Linux binaries
1206 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1207 if it fixes the problem contact emulation mailing list.
1210 ISDN4BSD (I4B) was removed from the src tree. You may need to
1211 update a your kernel configuration and remove relevant entries.
1214 I have checked in code to support multiple routing tables.
1215 See the man pages setfib(1) and setfib(2).
1216 This is a hopefully backwards compatible version,
1217 but to make use of it you need to compile your kernel
1218 with options ROUTETABLES=2 (or more up to 16).
1221 The 802.11 wireless support was redone to enable multi-bss
1222 operation on devices that are capable. The underlying device
1223 is no longer used directly but instead wlanX devices are
1224 cloned with ifconfig. This requires changes to rc.conf files.
1225 For example, change:
1226 ifconfig_ath0="WPA DHCP"
1229 ifconfig_wlan0="WPA DHCP"
1230 see rc.conf(5) for more details. In addition, mergemaster of
1231 /etc/rc.d is highly recommended. Simultaneous update of userland
1232 and kernel wouldn't hurt either.
1234 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1235 modules were merged into the base wlan module. All references
1236 to these modules (e.g. in kernel config files) must be removed.
1239 psm(4) has gained write(2) support in native operation level.
1240 Arbitrary commands can be written to /dev/psm%d and status can
1241 be read back from it. Therefore, an application is responsible
1242 for status validation and error recovery. It is a no-op in
1243 other operation levels.
1246 Support for KSE threading has been removed from the kernel. To
1247 run legacy applications linked against KSE libmap.conf may
1248 be used. The following libmap.conf may be used to ensure
1249 compatibility with any prior release:
1251 libpthread.so.1 libthr.so.1
1252 libpthread.so.2 libthr.so.2
1253 libkse.so.3 libthr.so.3
1256 The layout of struct vmspace has changed. This affects libkvm
1257 and any executables that link against libkvm and use the
1258 kvm_getprocs() function. In particular, but not exclusively,
1259 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1260 The effects are minimal, but it's advisable to upgrade world
1264 The latest em driver no longer has support in it for the
1265 82575 adapter, this is now moved to the igb driver. The
1266 split was done to make new features that are incompatible
1267 with older hardware easier to do.
1270 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1271 likewise the kernel option is now GEOM_LINUX_LVM.
1274 The default NFS mount mode has changed from UDP to TCP for
1275 increased reliability. If you rely on (insecurely) NFS
1276 mounting across a firewall you may need to update your
1280 Belatedly note the addition of m_collapse for compacting
1284 The fts(3) structures have been changed to use adequate
1285 integer types for their members and so to be able to cope
1286 with huge file trees. The old fts(3) ABI is preserved
1287 through symbol versioning in libc, so third-party binaries
1288 using fts(3) should still work, although they will not take
1289 advantage of the extended types. At the same time, some
1290 third-party software might fail to build after this change
1291 due to unportable assumptions made in its source code about
1292 fts(3) structure members. Such software should be fixed
1293 by its vendor or, in the worst case, in the ports tree.
1294 FreeBSD_version 800015 marks this change for the unlikely
1295 case that a portable fix is impossible.
1298 To upgrade to -current after this date, you must be running
1299 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1300 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1303 The ADAPTIVE_GIANT kernel option has been retired because its
1304 functionality is the default now.
1307 The AT keyboard emulation of sunkbd(4) has been turned on
1308 by default. In order to make the special symbols of the Sun
1309 keyboards driven by sunkbd(4) work under X these now have
1310 to be configured the same way as Sun USB keyboards driven
1311 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1313 Option "XkbLayout" "us"
1314 Option "XkbRules" "xorg"
1315 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1318 It has been decided that it is desirable to provide ABI
1319 backwards compatibility to the FreeBSD 4/5/6 versions of the
1320 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1321 broken with the introduction of PCI domain support (see the
1322 20070930 entry). Unfortunately, this required the ABI of
1323 PCIOCGETCONF to be broken again in order to be able to
1324 provide backwards compatibility to the old version of that
1325 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1326 again. As for prominent ports this affects neither pciutils
1327 nor xorg-server this time, the hal port needs to be rebuilt
1331 The misnamed kthread_create() and friends have been renamed
1332 to kproc_create() etc. Many of the callers already
1333 used kproc_start()..
1334 I will return kthread_create() and friends in a while
1335 with implementations that actually create threads, not procs.
1336 Renaming corresponds with version 800002.
1345 Avoid using make -j when upgrading. While generally safe, there are
1346 sometimes problems using -j to upgrade. If your upgrade fails with
1347 -j, please try again without -j. From time to time in the past there
1348 have been problems using -j with buildworld and/or installworld. This
1349 is especially true when upgrading between "distant" versions (eg one
1350 that cross a major release boundary or several minor releases, or when
1351 several months have passed on the -current branch).
1353 Sometimes, obscure build problems are the result of environment
1354 poisoning. This can happen because the make utility reads its
1355 environment when searching for values for global variables. To run
1356 your build attempts in an "environmental clean room", prefix all make
1357 commands with 'env -i '. See the env(1) manual page for more details.
1359 When upgrading from one major version to another it is generally best
1360 to upgrade to the latest code in the currently installed branch first,
1361 then do an upgrade to the new branch. This is the best-tested upgrade
1362 path, and has the highest probability of being successful. Please try
1363 this approach before reporting problems with a major version upgrade.
1367 When upgrading the boot ZFS pool to a new version, always follow
1370 1.) recompile and reinstall the ZFS boot loader and boot block
1371 (this is part of "make buildworld" and "make installworld")
1373 2.) update the ZFS boot block on your boot drive
1375 The following example updates the ZFS boot block on the first
1376 partition (freebsd-boot) of a GPT partitioned drive ad0:
1377 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1379 Non-boot pools do not need these updates.
1383 If you are updating from a prior version of FreeBSD (even one just
1384 a few days old), you should follow this procedure. It is the most
1385 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1387 make kernel-toolchain
1388 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1389 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1391 To test a kernel once
1392 ---------------------
1393 If you just want to boot a kernel once (because you are not sure
1394 if it works, or if you want to boot a known bad kernel to provide
1395 debugging information) run
1396 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1397 nextboot -k testkernel
1399 To just build a kernel when you know that it won't mess you up
1400 --------------------------------------------------------------
1401 This assumes you are already running a CURRENT system. Replace
1402 ${arch} with the architecture of your machine (e.g. "i386",
1403 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1405 cd src/sys/${arch}/conf
1406 config KERNEL_NAME_HERE
1407 cd ../compile/KERNEL_NAME_HERE
1412 If this fails, go to the "To build a kernel" section.
1414 To rebuild everything and install it on the current system.
1415 -----------------------------------------------------------
1416 # Note: sometimes if you are running current you gotta do more than
1417 # is listed here if you are upgrading from a really old current.
1419 <make sure you have good level 0 dumps>
1421 make kernel KERNCONF=YOUR_KERNEL_HERE
1423 <reboot in single user> [3]
1431 To cross-install current onto a separate partition
1432 --------------------------------------------------
1433 # In this approach we use a separate partition to hold
1434 # current's root, 'usr', and 'var' directories. A partition
1435 # holding "/", "/usr" and "/var" should be about 2GB in
1438 <make sure you have good level 0 dumps>
1441 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1442 <maybe newfs current's root partition>
1443 <mount current's root partition on directory ${CURRENT_ROOT}>
1444 make installworld DESTDIR=${CURRENT_ROOT}
1445 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1446 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1447 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1448 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1449 <reboot into current>
1450 <do a "native" rebuild/install as described in the previous section>
1451 <maybe install compatibility libraries from ports/misc/compat*>
1455 To upgrade in-place from stable to current
1456 ----------------------------------------------
1457 <make sure you have good level 0 dumps>
1459 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1461 <reboot in single user> [3]
1468 Make sure that you've read the UPDATING file to understand the
1469 tweaks to various things you need. At this point in the life
1470 cycle of current, things change often and you are on your own
1471 to cope. The defaults can also change, so please read ALL of
1472 the UPDATING entries.
1474 Also, if you are tracking -current, you must be subscribed to
1475 freebsd-current@freebsd.org. Make sure that before you update
1476 your sources that you have read and understood all the recent
1477 messages there. If in doubt, please track -stable which has
1478 much fewer pitfalls.
1480 [1] If you have third party modules, such as vmware, you
1481 should disable them at this point so they don't crash your
1484 [3] From the bootblocks, boot -s, and then do
1489 adjkerntz -i # if CMOS is wall time
1490 Also, when doing a major release upgrade, it is required that
1491 you boot into single user mode to do the installworld.
1493 [4] Note: This step is non-optional. Failure to do this step
1494 can result in a significant reduction in the functionality of the
1495 system. Attempting to do it by hand is not recommended and those
1496 that pursue this avenue should read this file carefully, as well
1497 as the archives of freebsd-current and freebsd-hackers mailing lists
1498 for potential gotchas. The -U option is also useful to consider.
1499 See mergemaster(8) for more information.
1501 [5] Usually this step is a noop. However, from time to time
1502 you may need to do this if you get unknown user in the following
1503 step. It never hurts to do it all the time. You may need to
1504 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1505 install) after the buildworld before this step if you last updated
1506 from current before 20020224 or from -stable before 20020408.
1508 [6] This only deletes old files and directories. Old libraries
1509 can be deleted by "make delete-old-libs", but you have to make
1510 sure that no program is using those libraries anymore.
1512 [8] In order to have a kernel that can run the 4.x binaries needed to
1513 do an installworld, you must include the COMPAT_FREEBSD4 option in
1514 your kernel. Failure to do so may leave you with a system that is
1515 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1516 required to run the 5.x binaries on more recent kernels. And so on
1517 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1519 Make sure that you merge any new devices from GENERIC since the
1520 last time you updated your kernel config file.
1522 [9] When checking out sources, you must include the -P flag to have
1523 cvs prune empty directories.
1525 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1526 "?=" instead of the "=" assignment operator, so that buildworld can
1527 override the CPUTYPE if it needs to.
1529 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1530 not on the command line, or in /etc/make.conf. buildworld will
1531 warn if it is improperly defined.
1534 This file contains a list, in reverse chronological order, of major
1535 breakages in tracking -current. It is not guaranteed to be a complete
1536 list of such breakages, and only contains entries since October 10, 2007.
1537 If you need to see UPDATING entries from before that date, you will need
1538 to fetch an UPDATING file from an older FreeBSD release.
1540 Copyright information:
1542 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1544 Redistribution, publication, translation and use, with or without
1545 modification, in full or in part, in any form or format of this
1546 document are permitted without further permission from the author.
1548 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1549 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1550 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1551 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1552 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1553 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1554 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1555 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1556 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1557 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1558 POSSIBILITY OF SUCH DAMAGE.
1560 Contact Warner Losh if you have any questions about your use of