1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
12 NOTE TO PEOPLE WHO THINK THAT FreeBSD 10.x IS SLOW:
13 FreeBSD 10.x has many debugging features turned on, in both the kernel
14 and userland. These features attempt to detect incorrect use of
15 system primitives, and encourage loud failure through extra sanity
16 checking and fail stop semantics. They also substantially impact
17 system performance. If you want to do performance measurement,
18 benchmarking, and optimization, you'll want to turn them off. This
19 includes various WITNESS- related kernel options, INVARIANTS, malloc
20 debugging flags in userland, and various verbose features in the
21 kernel. Many developers choose to disable these features on build
22 machines to maximize performance. (To disable malloc debugging, run
23 ln -s aj /etc/malloc.conf.)
26 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
30 Now unix domain sockets behave "as expected" on nullfs(5). Previously
31 nullfs(5) did not pass through all behaviours to the underlying layer,
32 as a result if we bound to a socket on the lower layer we could connect
33 only to the lower path; if we bound to the upper layer we could connect
34 only to the upper path. The new behavior is one can connect to both the
35 lower and the upper paths regardless what layer path one binds to.
38 The getifaddrs upgrade path broken with 20111215 has been restored.
39 If you have upgraded in between 20111215 and 20120209 you need to
40 recompile libc again with your kernel. You still need to recompile
41 world to be able to configure CARP but this restriction already
45 The set_rcvar() function has been removed from /etc/rc.subr. All
46 base and ports rc.d scripts have been updated, so if you have a
47 port installed with a script in /usr/local/etc/rc.d you can either
48 hand-edit the rcvar= line, or reinstall the port.
50 An easy way to handle the mass-update of /etc/rc.d:
51 rm /etc/rc.d/* && mergemaster -i
54 panic(9) now stops other CPUs in the SMP systems, disables interrupts
55 on the current CPU and prevents other threads from running.
56 This behavior can be reverted using the kern.stop_scheduler_on_panic
58 The new behavior can be incompatible with kern.sync_on_panic.
61 The carp(4) facility has been changed significantly. Configuration
62 of the CARP protocol via ifconfig(8) has changed, as well as format
63 of CARP events submitted to devd(8) has changed. See manual pages
64 for more information. The arpbalance feature of carp(4) is currently
65 not supported anymore.
67 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
68 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
69 need to be recompiled.
72 The acpi_wmi(4) status device /dev/wmistat has been renamed to
76 The option VFS_ALLOW_NONMPSAFE option has been added in order to
77 explicitely support non-MPSAFE filesystems.
78 It is on by default for all supported platform at this present
82 The broken amd(4) driver has been replaced with esp(4) in the amd64,
83 i386 and pc98 GENERIC kernel configuration files.
86 sysinstall has been removed
89 The stable/9 branch created in subversion. This corresponds to the
90 RELENG_9 branch in CVS.
93 This commit modifies vfs_register() so that it uses a hash
94 calculation to set vfc_typenum, which is enabled by default.
95 The first time a system is booted after this change, the
96 vfc_typenum values will change for all file systems. The
97 main effect of this is a change to the NFS server file handles
98 for file systems that use vfc_typenum in their fsid, such as ZFS.
99 It will, however, prevent vfc_typenum from changing when file
100 systems are loaded in a different order for subsequent reboots.
101 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
102 until you are ready to remount all NFS clients after a reboot.
105 Bump the shared library version numbers for libraries that
106 do not use symbol versioning, have changed the ABI compared
107 to stable/8 and which shared library version was not bumped.
108 Done as part of 9.0-RELEASE cycle.
111 During the merge of Capsicum features, the fget(9) KPI was modified.
112 This may require the rebuilding of out-of-tree device drivers --
113 issues have been reported specifically with the nVidia device driver.
114 __FreeBSD_version is bumped to 900041.
116 Also, there is a period between 20110811 and 20110814 where the
117 special devices /dev/{stdin,stdout,stderr} did not work correctly.
118 Building world from a kernel during that window may not work.
121 The packet filter (pf) code has been updated to OpenBSD 4.5.
122 You need to update userland tools to be in sync with kernel.
123 This update breaks backward compatibility with earlier pfsync(4)
124 versions. Care must be taken when updating redundant firewall setups.
127 The following sysctls and tunables are retired on x86 platforms:
129 machdep.hlt_logical_cpus
130 The following sysctl is retired:
131 machdep.hyperthreading_allowed
132 The sysctls were supposed to provide a way to dynamically offline and
133 online selected CPUs on x86 platforms, but the implementation has not
134 been reliable especially with SCHED_ULE scheduler.
135 machdep.hyperthreading_allowed tunable is still available to ignore
136 hyperthreading CPUs at OS level.
137 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
138 where X is an APIC ID of a CPU. Be advised, though, that disabling
139 CPUs in non-uniform fashion will result in non-uniform topology and
140 may lead to sub-optimal system performance with SCHED_ULE, which is
144 cpumask_t type is retired and cpuset_t is used in order to describe
148 Changes to ifconfig(8) for dynamic address family detection mandate
149 that you are running a kernel of 20110525 or later. Make sure to
150 follow the update procedure to boot a new kernel before installing
154 Support for sun4v architecture is officially dropped
157 Several KPI breaking changes have been committed to the mii(4) layer,
158 the PHY drivers and consequently some Ethernet drivers using mii(4).
159 This means that miibus.ko and the modules of the affected Ethernet
160 drivers need to be recompiled.
162 Note to kernel developers: Given that the OUI bit reversion problem
163 was fixed as part of these changes all mii(4) commits related to OUIs,
164 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
165 handling, no longer can be merged verbatim to stable/8 and previous
169 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
170 into their kernel configurations along with 'device pci'.
173 The default NFS client is now the new NFS client, so fstype "newnfs"
174 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
175 Although mounts via fstype "nfs" will usually work without userland
176 changes, it is recommended that the mount(8) and mount_nfs(8)
177 commands be rebuilt from sources and that a link to mount_nfs called
178 mount_oldnfs be created. The new client is compiled into the
179 kernel with "options NFSCL" and this is needed for diskless root
180 file systems. The GENERIC kernel configs have been changed to use
181 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
182 To use the regular/old client, you can "mount -t oldnfs ...". For
183 a diskless root file system, you must also include a line like:
185 vfs.root.mountfrom="oldnfs:"
187 in the boot/loader.conf on the root fs on the NFS server to make
188 a diskless root fs use the old client.
191 The GENERIC kernels for all architectures now default to the new
192 CAM-based ATA stack. It means that all legacy ATA drivers were
193 removed and replaced by respective CAM drivers. If you are using
194 ATA device names in /etc/fstab or other places, make sure to update
195 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
196 where 'Y's are the sequential numbers starting from zero for each type
197 in order of detection, unless configured otherwise with tunables,
198 see cam(4)). There will be symbolic links created in /dev/ to map
199 old adX devices to the respective adaY. They should provide basic
200 compatibility for file systems mounting in most cases, but they do
201 not support old user-level APIs and do not have respective providers
202 in GEOM. Consider using updated management tools with new device names.
204 It is possible to load devices ahci, ata, siis and mvs as modules,
205 but option ATA_CAM should remain in kernel configuration to make ata
206 module work as CAM driver supporting legacy ATA controllers. Device ata
207 still can be used in modular fashion (atacore + ...). Modules atadisk
208 and atapi* are not used and won't affect operation in ATA_CAM mode.
209 Note that to use CAM-based ATA kernel should include CAM devices
210 scbus, pass, da (or explicitly ada), cd and optionally others. All of
211 them are parts of the cam module.
213 ataraid(4) functionality is now supported by the RAID GEOM class.
214 To use it you can load geom_raid kernel module and use graid(8) tool
215 for management. Instead of /dev/arX device names, use /dev/raid/rX.
217 No kernel config options or code have been removed, so if a problem
218 arises, please report it and optionally revert to the old ATA stack.
219 In order to do it you can remove from the kernel config:
224 , and instead add back:
225 device atadisk # ATA disk drives
226 device ataraid # ATA RAID drives
227 device atapicd # ATAPI CDROM drives
228 device atapifd # ATAPI floppy drives
229 device atapist # ATAPI tape drives
232 The default NFS server has been changed to the new server, which
233 was referred to as the experimental server. If you need to switch
234 back to the old NFS server, you must now put the "-o" option on
235 both the mountd and nfsd commands. This can be done using the
236 mountd_flags and nfs_server_flags rc.conf variables until an
237 update to the rc scripts is committed, which is coming soon.
240 The GNU Objective-C runtime library (libobjc), and other Objective-C
241 related components have been removed from the base system. If you
242 require an Objective-C library, please use one of the available ports.
245 ath(4) has been split into bus- and device- modules. if_ath contains
246 the HAL, the TX rate control and the network device code. if_ath_pci
247 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
248 contains the AHB glue. Users need to load both if_ath_pci and if_ath
249 in order to use ath on everything else.
251 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
252 need to load if_ath and if_ath_pci for ath(4) operation.
255 As part of the replacement of sysinstall, the process of building
256 release media has changed significantly. For details, please re-read
257 release(7), which has been updated to reflect the new build process.
260 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
261 is the last available version under GPLv2. It brings a number of new
262 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
263 4.1 and SSE 4.2), better support for powerpc64, a number of new
264 directives, and lots of other small improvements. See the ChangeLog
265 file in contrib/binutils for the full details.
268 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
269 compliant, and will now use half of hash for authentication.
270 This will break interoperability with all stacks (including all
271 actual FreeBSD versions) who implement
272 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
274 The only workaround with such peers is to use another HMAC
275 algorithm for IPsec ("phase 2") authentication.
278 Remove the uio_yield prototype and symbol. This function has
279 been misnamed since it was introduced and should not be
280 globally exposed with this name. The equivalent functionality
281 is now available using kern_yield(curthread->td_user_pri).
282 The function remains undocumented.
285 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
286 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
287 for scalar sysctls is defined but disabled. Code that needs
288 UQUAD to pass the type checking that must compile on older
289 systems where the define is not present can check against
290 __FreeBSD_version >= 900030.
292 The system dialog(1) has been replaced with a new version previously
293 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
294 with the previous version, but the libdialog associated with it has
295 a largely incompatible API. As such, the original version of libdialog
296 will be kept temporarily as libodialog, until its base system consumers
297 are replaced or updated. Bump __FreeBSD_version to 900030.
300 If you are trying to run make universe on a -stable system, and you get
301 the following warning:
302 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
303 unknown. config(8) likely too old."
304 or something similar to it, then you must upgrade your -stable system
305 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
306 stable/8) or build the config from the latest stable/8 branch and
307 install it on your system.
309 Prior to this date, building a current universe on 8-stable system from
310 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
311 error in the first kernel build phase. A new config on those old
312 systems will fix that problem for older versions of -current.
315 The TCP stack has been modified to allow Khelp modules to interact with
316 it via helper hook points and store per-connection data in the TCP
317 control block. Bump __FreeBSD_version to 900029. User space tools that
318 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
322 Generic IEEE 802.3 annex 31B full duplex flow control support has been
323 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
324 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
325 to take advantage of it instead of using custom implementations. This
326 means that these drivers now no longer unconditionally advertise
327 support for flow control but only do so if flow control is a selected
328 media option. This was implemented in the generic support that way in
329 order to allow flow control to be switched on and off via ifconfig(8)
330 with the PHY specific default to typically off in order to protect
331 from unwanted effects. Consequently, if you used flow control with
332 one of the above mentioned drivers you now need to explicitly enable
334 ifconfig bge0 media auto mediaopt flowcontrol
336 Along with the above mentioned changes generic support for setting
337 1000baseT master mode also has been added and brgphy(4), ciphy(4),
338 e1000phy(4) as well as ip1000phy(4) have been converted to take
339 advantage of it. This means that these drivers now no longer take the
340 link0 parameter for selecting master mode but the master media option
341 has to be used instead, for example like in the following:
342 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
344 Selection of master mode now is also available with all other PHY
345 drivers supporting 1000baseT.
348 The TCP stack has received a significant update to add support for
349 modularised congestion control and generally improve the clarity of
350 congestion control decisions. Bump __FreeBSD_version to 900025. User
351 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
352 sockstat) need to be recompiled.
355 The man(1) utility has been replaced by a new version that no longer
356 uses /etc/manpath.config. Please consult man.conf(5) for how to
357 migrate local entries to the new format.
360 The copyright strings printed by login(1) and sshd(8) at the time of a
361 new connection have been removed to follow other operating systems and
365 A workaround for a fixed ld bug has been removed in kernel code,
366 so make sure that your system ld is built from sources after
367 revision 210245 from 2010-07-19 (r211583 if building head kernel
368 on stable/8, r211584 for stable/7; both from 2010-08-21).
369 A symptom of incorrect ld version is different addresses for
370 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
373 The $ipv6_prefer variable in rc.conf(5) has been split into
374 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
376 The $ip6addrctl_policy is a variable to choose a pre-defined
377 address selection policy set by ip6addrctl(8). A value
378 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
381 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
382 flag (see an entry of 20090926) is set on an interface with no
383 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
384 security reason. If you want IPv6 link-local address on all
385 interfaces by default, set this to "YES".
387 The old ipv6_prefer="YES" is equivalent to
388 ipv6_activate_all_interfaces="YES" and
389 ip6addrctl_policy="ipv6_prefer".
392 DTrace has grown support for userland tracing. Due to this, DTrace is
393 now i386 and amd64 only.
394 dtruss(1) is now installed by default on those systems and a new
395 kernel module is needed for userland tracing: fasttrap.
396 No changes to your kernel config file are necessary to enable
397 userland tracing, but you might consider adding 'STRIP=' and
398 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
399 to have informative userland stack traces in DTrace (ustack).
402 The acpi_aiboost(4) driver has been removed in favor of the new
403 aibs(4) driver. You should update your kernel configuration file.
406 BSD grep has been imported to the base system and it is built by
407 default. It is completely BSD licensed, highly GNU-compatible, uses
408 less memory than its GNU counterpart and has a small codebase.
409 However, it is slower than its GNU counterpart, which is mostly
410 noticeable for larger searches, for smaller ones it is measurable
411 but not significant. The reason is complex, the most important factor
412 is that we lack a modern and efficient regex library and GNU
413 overcomes this by optimizing the searches internally. Future work
414 on improving the regex performance is planned, for the meantime,
415 users that need better performance, can build GNU grep instead by
416 setting the WITH_GNU_GREP knob.
419 Due to the import of powerpc64 support, all existing powerpc kernel
420 configuration files must be updated with a machine directive like this:
421 machine powerpc powerpc
423 In addition, an updated config(8) is required to build powerpc kernels
427 A new version of ZFS (version 15) has been merged to -HEAD.
428 This version uses a python library for the following subcommands:
429 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
430 For full functionality of these commands the following port must
431 be installed: sysutils/py-zfs
434 'vm_page's are now hashed by physical address to an array of mutexes.
435 Currently this is only used to serialize access to hold_count. Over
436 time the page queue mutex will be peeled away. This changes the size
437 of pmap on every architecture. And requires all callers of vm_page_hold
438 and vm_page_unhold to be updated.
441 WITH_CTF can now be specified in src.conf (not recommended, there
442 are some problems with static executables), make.conf (would also
443 affect ports which do not use GNU make and do not override the
444 compile targets) or in the kernel config (via "makeoptions
446 When WITH_CTF was specified there before this was silently ignored,
447 so make sure that WITH_CTF is not used in places which could lead
448 to unwanted behavior.
451 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
452 to allow 32-bit compatibility on non-x86 platforms. All kernel
453 configurations on amd64 and ia64 platforms using these options must
454 be modified accordingly.
457 The utmp user accounting database has been replaced with utmpx,
458 the user accounting interface standardized by POSIX.
459 Unfortunately the semantics of utmp and utmpx don't match,
460 making it practically impossible to support both interfaces.
461 The user accounting database is used by tools like finger(1),
462 last(1), talk(1), w(1) and ac(8).
464 All applications in the base system use utmpx. This means only
465 local binaries (e.g. from the ports tree) may still use these
466 utmp database files. These applications must be rebuilt to make
469 After the system has been upgraded, it is safe to remove the old
470 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
471 assuming their contents is of no importance anymore. Old wtmp
472 databases can only be used by last(1) and ac(8) after they have
473 been converted to the new format using wtmpcvt(1).
476 Introduce the kernel thread "deadlock resolver" (which can be enabled
477 via the DEADLKRES option, see NOTES for more details) and the
478 sleepq_type() function for sleepqueues.
481 The rc.firewall and rc.firewall6 were unified, and
482 rc.firewall6 and rc.d/ip6fw were removed.
483 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
484 variables are obsoleted. Instead, the following new rc
485 variables are added to rc.d/ipfw:
487 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
488 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
489 firewall_simple_onet_ipv6, firewall_trusted_ipv6
491 The meanings correspond to the relevant IPv4 variables.
497 The default terminal emulation for syscons(4) has been changed
498 from cons25 to xterm on all platforms except pc98. This means
499 that the /etc/ttys file needs to be updated to ensure correct
500 operation of applications on the console.
502 The terminal emulation style can be toggled per window by using
503 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
504 options can be used to change the compile-time default back to
507 To prevent graphical artifacts, make sure the TERM environment
508 variable is set to match the terminal emulation that is being
509 performed by syscons(4).
512 The layout of the structure ieee80211req_scan_result has changed.
513 Applications that require wireless scan results (e.g. ifconfig(8))
514 from net80211 need to be recompiled.
516 Applications such as wpa_supplicant(8) may require a full world
517 build without using NO_CLEAN in order to get synchronized with the
521 The iwn(4) driver has been updated to support the 5000 and 5150 series.
522 There's one kernel module for each firmware. Adding "device iwnfw"
523 to the kernel configuration file means including all three firmware
524 images inside the kernel. If you want to include just the one for
525 your wireless card, use the devices iwn4965fw, iwn5000fw or
529 The rc.d/network_ipv6, IPv6 configuration script has been integrated
530 into rc.d/netif. The changes are the following:
532 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
533 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
534 Note that both variables need the "inet6" keyword at the head.
536 Do not set $ipv6_network_interfaces manually if you do not
537 understand what you are doing. It is not needed in most cases.
539 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
542 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
543 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
545 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
546 all configured interfaces have "inet6 accept_rtadv" in the
547 $ifconfig_IF_ipv6. These are for backward compatibility.
549 3. A new variable $ipv6_prefer has been added. If NO, IPv6
550 functionality of interfaces with no corresponding
551 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
552 and the default address selection policy of ip6addrctl(8)
553 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
554 Note that if you want to configure IPv6 functionality on the
555 disabled interfaces after boot, first you need to clear the flag by
556 using ifconfig(8) like:
558 ifconfig em0 inet6 -ifdisabled
560 If YES, the default address selection policy is set as
563 The default value of $ipv6_prefer is NO.
565 4. If your system need to receive Router Advertisement messages,
566 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
567 scripts automatically invoke rtsol(8) when the interface becomes
568 UP. The Router Advertisement messages are used for SLAAC
569 (State-Less Address AutoConfiguration).
572 802.11s D3.03 support was committed. This is incompatible with the
573 previous code, which was based on D3.0.
576 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
577 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
578 control whether accepting Router Advertisement messages or not.
579 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
580 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
581 The ifconfig(8) utility now supports these flags.
584 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
585 mount(8) and -a option for df(1) to see them.
588 The old tunable hw.bus.devctl_disable has been superseded by
589 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
590 replaced by hw.bus.devctl_queue=0. The default for this new tunable
594 Remove the option STOP_NMI. The default action is now to use NMI only
595 for KDB via the newly introduced function stop_cpus_hard() and
596 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
599 The stable/8 branch created in subversion. This corresponds to the
600 RELENG_8 branch in CVS.
603 Bump the shared library version numbers for all libraries that do not
604 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
605 __FreeBSD_version to 800105.
608 Due to changes in the implementation of virtual network stack support,
609 all network-related kernel modules must be recompiled. As this change
610 breaks the ABI, bump __FreeBSD_version to 800104.
613 The TOE interface to the TCP syncache has been modified to remove
614 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
615 The cxgb driver is the only TOE consumer affected by this change, and
616 needs to be recompiled along with the kernel. As this change breaks
617 the ABI, bump __FreeBSD_version to 800103.
620 Padding has been added to struct tcpcb, sackhint and tcpstat in
621 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
622 maintaining the ABI. However, this change breaks the ABI, so bump
623 __FreeBSD_version to 800102. User space tools that rely on the size of
624 any of these structs (e.g. sockstat) need to be recompiled.
627 The NFS_LEGACYRPC option has been removed along with the old kernel
628 RPC implementation that this option selected. Kernel configurations
629 may need to be adjusted.
632 The network interface device nodes at /dev/net/<interface> have been
633 removed. All ioctl operations can be performed the normal way using
634 routing sockets. The kqueue functionality can generally be replaced
635 with routing sockets.
638 The documentation from the FreeBSD Documentation Project (Handbook,
639 FAQ, etc.) is now installed via packages by sysinstall(8) and under
640 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
643 The ABI of various structures related to the SYSV IPC API have been
644 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
645 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
649 Layout of struct vnet has changed as routing related variables were
650 moved to their own Vimage module. Modules need to be recompiled. Bump
651 __FreeBSD_version to 800099.
654 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
655 respectively. As long as no more than 16 groups per process are used,
656 no changes should be visible. When more than 16 groups are used, old
657 binaries may fail if they call getgroups() or getgrouplist() with
658 statically sized storage. Recompiling will work around this, but
659 applications should be modified to use dynamically allocated storage
660 for group arrays as POSIX.1-2008 does not cap an implementation's
661 number of supported groups at NGROUPS_MAX+1 as previous versions did.
663 NFS and portalfs mounts may also be affected as the list of groups is
664 truncated to 16. Users of NFS who use more than 16 groups, should
665 take care that negative group permissions are not used on the exported
666 file systems as they will not be reliable unless a GSSAPI based
667 authentication method is used.
670 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
671 option compiles in the support for adaptive spinning for lockmgrs
672 which want to enable it. The lockinit() function now accepts the flag
673 LK_ADAPTIVE in order to make the lock object subject to adaptive
674 spinning when both held in write and read mode.
677 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
678 changed. User applications that use this ioctl need to be rebuilt.
681 The layout of struct thread has changed. Kernel and modules need to
685 The layout of structs ifnet, domain, protosw and vnet_net has changed.
686 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
689 window(1) has been removed from the base system. It can now be
690 installed from ports. The port is called misc/window.
693 The way we are storing and accessing `routing table' entries has
694 changed. Programs reading the FIB, like netstat, need to be
698 A new netisr implementation has been added for FreeBSD 8. Network
699 file system modules, such as igmp, ipdivert, and others, should be
701 Bump __FreeBSD_version to 800096.
704 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
708 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
709 Bump __FreeBSD_version to 800094.
712 Add mnt_xflag field to 'struct mount'. File system modules need to be
714 Bump __FreeBSD_version to 800093.
717 The compiling option ADAPTIVE_SX has been retired while it has been
718 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
719 The KPI for sx_init_flags() changes as accepting flags:
720 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
721 been introduced in order to handle the reversed logic.
722 Bump __FreeBSD_version to 800092.
725 Add support for hierarchical jails. Remove global securelevel.
726 Bump __FreeBSD_version to 800091.
729 The layout of struct vnet_net has changed, therefore modules
731 Bump __FreeBSD_version to 800090.
734 The newly imported zic(8) produces a new format in the output. Please
735 run tzsetup(8) to install the newly created data to /etc/localtime.
738 The sysctl tree for the usb stack has renamed from hw.usb2.* to
739 hw.usb.* and is now consistent again with previous releases.
742 802.11 monitor mode support was revised and driver api's were changed.
743 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
744 of DLT_IEEE802_11. No user-visible data structures were changed but
745 applications that use DLT_IEEE802_11 may require changes.
746 Bump __FreeBSD_version to 800088.
749 The layout of the following structs has changed: sysctl_oid,
750 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
751 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
752 panics may be experienced. World rebuild is required for
753 correctly checking networking state from userland.
754 Bump __FreeBSD_version to 800085.
757 MLDv2 and Source-Specific Multicast (SSM) have been merged
758 to the IPv6 stack. VIMAGE hooks are in but not yet used.
759 The implementation of SSM within FreeBSD's IPv6 stack closely
760 follows the IPv4 implementation.
762 For kernel developers:
764 * The most important changes are that the ip6_output() and
765 ip6_input() paths no longer take the IN6_MULTI_LOCK,
766 and this lock has been downgraded to a non-recursive mutex.
768 * As with the changes to the IPv4 stack to support SSM, filtering
769 of inbound multicast traffic must now be performed by transport
770 protocols within the IPv6 stack. This does not apply to TCP and
771 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
773 * The KPIs used by IPv6 multicast are similar to those used by
774 the IPv4 stack, with the following differences:
775 * im6o_mc_filter() is analogous to imo_multicast_filter().
776 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
777 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
778 * IN6_LOOKUP_MULTI() has been deprecated and removed.
779 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
780 for MLDv1 have an additional 'timer' argument which is used to
781 jitter the initial membership report for the solicited-node
782 multicast membership on-link.
783 * This is not strictly needed for MLDv2, which already jitters
784 its report transmissions. However, the 'timer' argument is
785 preserved in case MLDv1 is active on the interface.
787 * The KAME linked-list based IPv6 membership implementation has
788 been refactored to use a vector similar to that used by the IPv4
790 Code which maintains a list of its own multicast memberships
791 internally, e.g. carp, has been updated to reflect the new
794 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
795 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
796 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
797 implementation constraint which needs to be addressed in HEAD.
799 For application developers:
801 * The changes are broadly similar to those made for the IPv4
804 * The use of IPv4 and IPv6 multicast socket options on the same
805 socket, using mapped addresses, HAS NOT been tested or supported.
807 * There are a number of issues with the implementation of various
808 IPv6 multicast APIs which need to be resolved in the API surface
809 before the implementation is fully compatible with KAME userland
810 use, and these are mostly to do with interface index treatment.
812 * The literature available discusses the use of either the delta / ASM
813 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
814 using setsourcefilter(3)/getsourcefilter(3). For more information
815 please refer to RFC 3768, 'Socket Interface Extensions for
816 Multicast Source Filters'.
818 * Applications which use the published RFC 3678 APIs should be fine.
820 For systems administrators:
822 * The mtest(8) utility has been refactored to support IPv6, in
823 addition to IPv4. Interface addresses are no longer accepted
824 as arguments, their names must be used instead. The utility
825 will map the interface name to its first IPv4 address as
826 returned by getifaddrs(3).
828 * The ifmcstat(8) utility has also been updated to print the MLDv2
829 endpoint state and source filter lists via sysctl(3).
831 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
832 loopback of IPv6 multicast datagrams by default; it defaults to 1
833 to preserve the existing behaviour. Disabling multicast loopback is
834 recommended for optimal system performance.
836 * The IPv6 MROUTING code has been changed to examine this sysctl
837 instead of attempting to perform a group lookup before looping
838 back forwarded datagrams.
840 Bump __FreeBSD_version to 800084.
843 Implement low-level Bluetooth HCI API.
844 Bump __FreeBSD_version to 800083.
847 The layout of struct malloc_type, used by modules to register new
848 memory allocation types, has changed. Most modules will need to
849 be rebuilt or panics may be experienced.
850 Bump __FreeBSD_version to 800081.
853 Anticipate overflowing inp_flags - add inp_flags2.
854 This changes most offsets in inpcb, so checking v4 connection
855 state will require a world rebuild.
856 Bump __FreeBSD_version to 800080.
859 Add an llentry to struct route and struct route_in6. Modules
860 embedding a struct route will need to be recompiled.
861 Bump __FreeBSD_version to 800079.
864 The size of rt_metrics_lite and by extension rtentry has changed.
865 Networking administration apps will need to be recompiled.
866 The route command now supports show as an alias for get, weighting
867 of routes, sticky and nostick flags to alter the behavior of stateful
869 Bump __FreeBSD_version to 800078.
872 Do not use Giant for kbdmux(4) locking. This is wrong and
873 apparently causing more problems than it solves. This will
874 re-open the issue where interrupt handlers may race with
875 kbdmux(4) in polling mode. Typical symptoms include (but
876 not limited to) duplicated and/or missing characters when
877 low level console functions (such as gets) are used while
878 interrupts are enabled (for example geli password prompt,
879 mountroot prompt etc.). Disabling kbdmux(4) may help.
882 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
883 kernel modules referencing any of the above need to be recompiled.
884 Bump __FreeBSD_version to 800075.
887 GEOM_PART has become the default partition slicer for storage devices,
888 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
889 introduces some changes:
891 MSDOS/EBR: the devices created from MSDOS extended partition entries
892 (EBR) can be named differently than with GEOM_MBR and are now symlinks
893 to devices with offset-based names. fstabs may need to be modified.
895 BSD: the "geometry does not match label" warning is harmless in most
896 cases but it points to problems in file system misalignment with
897 disk geometry. The "c" partition is now implicit, covers the whole
898 top-level drive and cannot be (mis)used by users.
900 General: Kernel dumps are now not allowed to be written to devices
901 whose partition types indicate they are meant to be used for file
902 systems (or, in case of MSDOS partitions, as something else than
905 Most of these changes date approximately from 200812.
908 The uscanner(4) driver has been removed from the kernel. This follows
909 Linux removing theirs in 2.6 and making libusb the default interface
913 The multicast forwarding code has been cleaned up. netstat(1)
914 only relies on KVM now for printing bandwidth upcall meters.
915 The IPv4 and IPv6 modules are split into ip_mroute_mod and
916 ip6_mroute_mod respectively. The config(5) options for statically
917 compiling this code remain the same, i.e. 'options MROUTING'.
920 Support for the IFF_NEEDSGIANT network interface flag has been
921 removed, which means that non-MPSAFE network device drivers are no
922 longer supported. In particular, if_ar, if_sr, and network device
923 drivers from the old (legacy) USB stack can no longer be built or
927 POSIX.1 Native Language Support (NLS) has been enabled in libc and
928 a bunch of new language catalog files have also been added.
929 This means that some common libc messages are now localized and
930 they depend on the LC_MESSAGES environmental variable.
933 The k8temp(4) driver has been renamed to amdtemp(4) since
934 support for Family 10 and Family 11 CPU families was added.
937 IGMPv3 and Source-Specific Multicast (SSM) have been merged
938 to the IPv4 stack. VIMAGE hooks are in but not yet used.
940 For kernel developers, the most important changes are that the
941 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
942 and this lock has been downgraded to a non-recursive mutex.
944 Transport protocols (UDP, Raw IP) are now responsible for filtering
945 inbound multicast traffic according to group membership and source
946 filters. The imo_multicast_filter() KPI exists for this purpose.
947 Transports which do not use multicast (SCTP, TCP) already reject
948 multicast by default. Forwarding and receive performance may improve
949 as a mutex acquisition is no longer needed in the ip_input()
950 low-level input path. in_addmulti() and in_delmulti() are shimmed
951 to new KPIs which exist to support SSM in-kernel.
953 For application developers, it is recommended that loopback of
954 multicast datagrams be disabled for best performance, as this
955 will still cause the lock to be taken for each looped-back
956 datagram transmission. The net.inet.ip.mcast.loop sysctl may
957 be tuned to 0 to disable loopback by default; it defaults to 1
958 to preserve the existing behaviour.
960 For systems administrators, to obtain best performance with
961 multicast reception and multiple groups, it is always recommended
962 that a card with a suitably precise hash filter is used. Hash
963 collisions will still result in the lock being taken within the
964 transport protocol input path to check group membership.
966 If deploying FreeBSD in an environment with IGMP snooping switches,
967 it is recommended that the net.inet.igmp.sendlocal sysctl remain
968 enabled; this forces 224.0.0.0/24 group membership to be announced
971 The size of 'struct igmpstat' has changed; netstat needs to be
972 recompiled to reflect this.
973 Bump __FreeBSD_version to 800070.
976 libusb20.so.1 is now installed as libusb.so.1 and the ports system
977 updated to use it. This requires a buildworld/installworld in order to
978 update the library and dependencies (usbconfig, etc). Its advisable to
979 rebuild all ports which uses libusb. More specific directions are given
980 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
981 libusb are no longer required and can be removed.
984 A workaround is committed to allow the creation of System V shared
985 memory segment of size > 2 GB on the 64-bit architectures.
986 Due to a limitation of the existing ABI, the shm_segsz member
987 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
988 wrong for large segments. Note that limits must be explicitly
989 raised to allow such segments to be created.
992 The layout of struct ifnet has changed, requiring a rebuild of all
993 network device driver modules.
996 The /dev handling for the new USB stack has changed, a
997 buildworld/installworld is required for libusb20.
1000 The new USB2 stack has now been permanently moved in and all kernel and
1001 module names reverted to their previous values (eg, usb, ehci, ohci,
1002 ums, ...). The old usb stack can be compiled in by prefixing the name
1003 with the letter 'o', the old usb modules have been removed.
1004 Updating entry 20090216 for xorg and 20090215 for libmap may still
1008 The rc.conf(5) option if_up_delay has been renamed to
1009 defaultroute_delay to better reflect its purpose. If you have
1010 customized this setting in /etc/rc.conf you need to update it to
1014 xorg 7.4 wants to configure its input devices via hald which does not
1015 yet work with USB2. If the keyboard/mouse does not work in xorg then
1017 Option "AllowEmptyInput" "off"
1018 to your ServerLayout section. This will cause X to use the configured
1019 kbd and mouse sections from your xorg.conf.
1022 The GENERIC kernels for all architectures now default to the new USB2
1023 stack. No kernel config options or code have been removed so if a
1024 problem arises please report it and optionally revert to the old USB
1025 stack. If you are loading USB kernel modules or have a custom kernel
1026 that includes GENERIC then ensure that usb names are also changed over,
1027 eg uftdi -> usb2_serial_ftdi.
1029 Older programs linked against the ports libusb 0.1 need to be
1030 redirected to the new stack's libusb20. /etc/libmap.conf can
1032 # Map old usb library to new one for usb2 stack
1033 libusb-0.1.so.8 libusb20.so.1
1036 All USB ethernet devices now attach as interfaces under the name ueN
1037 (eg. ue0). This is to provide a predictable name as vendors often
1038 change usb chipsets in a product without notice.
1041 The ichsmb(4) driver has been changed to require SMBus slave
1042 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1043 All of the other SMBus controller drivers require left-justified
1044 slave addresses, so this change makes all the drivers provide the
1048 INET6 statistics (struct ip6stat) was updated.
1049 netstat(1) needs to be recompiled.
1052 NTFS has been removed from GENERIC kernel on amd64 to match
1053 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1054 will load ntfs.ko module automatically when NTFS support is
1055 actually needed, unless ntfs.ko is not installed or security
1056 level prohibits loading kernel modules. If either is the case,
1057 "options NTFS" has to be added into kernel config.
1060 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1061 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1062 800061. User space tools that rely on the size of struct tcpcb in
1063 tcp_var.h (e.g. sockstat) need to be recompiled.
1066 ng_tty(4) module updated to match the new TTY subsystem.
1067 Due to API change, user-level applications must be updated.
1068 New API support added to mpd5 CVS and expected to be present
1069 in next mpd5.3 release.
1072 With __FreeBSD_version 800060 the makefs tool is part of
1073 the base system (it was a port).
1076 The afdata and ifnet locks have been changed from mutexes to
1077 rwlocks, network modules will need to be re-compiled.
1080 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1081 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1082 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1083 architecture and 40 bytes on 64-bit architecture. The userland
1084 applications "arp" and "ndp" have been updated accordingly.
1085 The output from "netstat -r" shows only routing entries and
1086 none of the L2 information.
1089 __FreeBSD_version 800057 marks the switchover from the
1090 binary ath hal to source code. Users must add the line:
1092 options AH_SUPPORT_AR5416
1094 to their kernel config files when specifying:
1098 The ath_hal module no longer exists; the code is now compiled
1099 together with the driver in the ath module. It is now
1100 possible to tailor chip support (i.e. reduce the set of chips
1101 and thereby the code size); consult ath_hal(4) for details.
1104 __FreeBSD_version 800054 adds memory barriers to
1105 <machine/atomic.h>, new interfaces to ifnet to facilitate
1106 multiple hardware transmit queues for cards that support
1107 them, and a lock-less ring-buffer implementation to
1108 enable drivers to more efficiently manage queueing of
1112 A new version of ZFS (version 13) has been merged to -HEAD.
1113 This version has zpool attribute "listsnapshots" off by
1114 default, which means "zfs list" does not show snapshots,
1115 and is the same as Solaris behavior.
1118 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1121 The uhci, ohci, ehci and slhci USB Host controller drivers have
1122 been put into separate modules. If you load the usb module
1123 separately through loader.conf you will need to load the
1124 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1125 controller add the following to loader.conf:
1131 The ABI used by the PMC toolset has changed. Please keep
1132 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1136 atapci kernel module now includes only generic PCI ATA
1137 driver. AHCI driver moved to ataahci kernel module.
1138 All vendor-specific code moved into separate kernel modules:
1139 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1140 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1141 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1142 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1145 The TTY subsystem of the kernel has been replaced by a new
1146 implementation, which provides better scalability and an
1147 improved driver model. Most common drivers have been migrated to
1148 the new TTY subsystem, while others have not. The following
1149 drivers have not yet been ported to the new TTY layer:
1152 cy, digi, rc, rp, sio
1158 ng_h4, ng_tty, ppp, sl, snp
1160 Adding these drivers to your kernel configuration file shall
1161 cause compilation to fail.
1164 ntpd has been upgraded to 4.2.4p5.
1167 OpenSSH has been upgraded to 5.1p1.
1169 For many years, FreeBSD's version of OpenSSH preferred DSA
1170 over RSA for host and user authentication keys. With this
1171 upgrade, we've switched to the vendor's default of RSA over
1172 DSA. This may cause upgraded clients to warn about unknown
1173 host keys even for previously known hosts. Users should
1174 follow the usual procedure for verifying host keys before
1175 accepting the RSA key.
1177 This can be circumvented by setting the "HostKeyAlgorithms"
1178 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1181 Please note that the sequence of keys offered for
1182 authentication has been changed as well. You may want to
1183 specify IdentityFile in a different order to revert this
1187 The sio(4) driver has been removed from the i386 and amd64
1188 kernel configuration files. This means uart(4) is now the
1189 default serial port driver on those platforms as well.
1191 To prevent collisions with the sio(4) driver, the uart(4) driver
1192 uses different names for its device nodes. This means the
1193 onboard serial port will now most likely be called "ttyu0"
1194 instead of "ttyd0". You may need to reconfigure applications to
1195 use the new device names.
1197 When using the serial port as a boot console, be sure to update
1198 /boot/device.hints and /etc/ttys before booting the new kernel.
1199 If you forget to do so, you can still manually specify the hints
1200 at the loader prompt:
1202 set hint.uart.0.at="isa"
1203 set hint.uart.0.port="0x3F8"
1204 set hint.uart.0.flags="0x10"
1205 set hint.uart.0.irq="4"
1209 The gpt(8) utility has been removed. Use gpart(8) to partition
1213 The version that Linuxulator emulates was changed from 2.4.2
1214 to 2.6.16. If you experience any problems with Linux binaries
1215 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1216 if it fixes the problem contact emulation mailing list.
1219 ISDN4BSD (I4B) was removed from the src tree. You may need to
1220 update a your kernel configuration and remove relevant entries.
1223 I have checked in code to support multiple routing tables.
1224 See the man pages setfib(1) and setfib(2).
1225 This is a hopefully backwards compatible version,
1226 but to make use of it you need to compile your kernel
1227 with options ROUTETABLES=2 (or more up to 16).
1230 The 802.11 wireless support was redone to enable multi-bss
1231 operation on devices that are capable. The underlying device
1232 is no longer used directly but instead wlanX devices are
1233 cloned with ifconfig. This requires changes to rc.conf files.
1234 For example, change:
1235 ifconfig_ath0="WPA DHCP"
1238 ifconfig_wlan0="WPA DHCP"
1239 see rc.conf(5) for more details. In addition, mergemaster of
1240 /etc/rc.d is highly recommended. Simultaneous update of userland
1241 and kernel wouldn't hurt either.
1243 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1244 modules were merged into the base wlan module. All references
1245 to these modules (e.g. in kernel config files) must be removed.
1248 psm(4) has gained write(2) support in native operation level.
1249 Arbitrary commands can be written to /dev/psm%d and status can
1250 be read back from it. Therefore, an application is responsible
1251 for status validation and error recovery. It is a no-op in
1252 other operation levels.
1255 Support for KSE threading has been removed from the kernel. To
1256 run legacy applications linked against KSE libmap.conf may
1257 be used. The following libmap.conf may be used to ensure
1258 compatibility with any prior release:
1260 libpthread.so.1 libthr.so.1
1261 libpthread.so.2 libthr.so.2
1262 libkse.so.3 libthr.so.3
1265 The layout of struct vmspace has changed. This affects libkvm
1266 and any executables that link against libkvm and use the
1267 kvm_getprocs() function. In particular, but not exclusively,
1268 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1269 The effects are minimal, but it's advisable to upgrade world
1273 The latest em driver no longer has support in it for the
1274 82575 adapter, this is now moved to the igb driver. The
1275 split was done to make new features that are incompatible
1276 with older hardware easier to do.
1279 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1280 likewise the kernel option is now GEOM_LINUX_LVM.
1283 The default NFS mount mode has changed from UDP to TCP for
1284 increased reliability. If you rely on (insecurely) NFS
1285 mounting across a firewall you may need to update your
1289 Belatedly note the addition of m_collapse for compacting
1293 The fts(3) structures have been changed to use adequate
1294 integer types for their members and so to be able to cope
1295 with huge file trees. The old fts(3) ABI is preserved
1296 through symbol versioning in libc, so third-party binaries
1297 using fts(3) should still work, although they will not take
1298 advantage of the extended types. At the same time, some
1299 third-party software might fail to build after this change
1300 due to unportable assumptions made in its source code about
1301 fts(3) structure members. Such software should be fixed
1302 by its vendor or, in the worst case, in the ports tree.
1303 FreeBSD_version 800015 marks this change for the unlikely
1304 case that a portable fix is impossible.
1307 To upgrade to -current after this date, you must be running
1308 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1309 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1312 The ADAPTIVE_GIANT kernel option has been retired because its
1313 functionality is the default now.
1316 The AT keyboard emulation of sunkbd(4) has been turned on
1317 by default. In order to make the special symbols of the Sun
1318 keyboards driven by sunkbd(4) work under X these now have
1319 to be configured the same way as Sun USB keyboards driven
1320 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1322 Option "XkbLayout" "us"
1323 Option "XkbRules" "xorg"
1324 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1327 It has been decided that it is desirable to provide ABI
1328 backwards compatibility to the FreeBSD 4/5/6 versions of the
1329 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1330 broken with the introduction of PCI domain support (see the
1331 20070930 entry). Unfortunately, this required the ABI of
1332 PCIOCGETCONF to be broken again in order to be able to
1333 provide backwards compatibility to the old version of that
1334 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1335 again. As for prominent ports this affects neither pciutils
1336 nor xorg-server this time, the hal port needs to be rebuilt
1340 The misnamed kthread_create() and friends have been renamed
1341 to kproc_create() etc. Many of the callers already
1342 used kproc_start()..
1343 I will return kthread_create() and friends in a while
1344 with implementations that actually create threads, not procs.
1345 Renaming corresponds with version 800002.
1354 Avoid using make -j when upgrading. While generally safe, there are
1355 sometimes problems using -j to upgrade. If your upgrade fails with
1356 -j, please try again without -j. From time to time in the past there
1357 have been problems using -j with buildworld and/or installworld. This
1358 is especially true when upgrading between "distant" versions (eg one
1359 that cross a major release boundary or several minor releases, or when
1360 several months have passed on the -current branch).
1362 Sometimes, obscure build problems are the result of environment
1363 poisoning. This can happen because the make utility reads its
1364 environment when searching for values for global variables. To run
1365 your build attempts in an "environmental clean room", prefix all make
1366 commands with 'env -i '. See the env(1) manual page for more details.
1368 When upgrading from one major version to another it is generally best
1369 to upgrade to the latest code in the currently installed branch first,
1370 then do an upgrade to the new branch. This is the best-tested upgrade
1371 path, and has the highest probability of being successful. Please try
1372 this approach before reporting problems with a major version upgrade.
1376 When upgrading the boot ZFS pool to a new version, always follow
1379 1.) recompile and reinstall the ZFS boot loader and boot block
1380 (this is part of "make buildworld" and "make installworld")
1382 2.) update the ZFS boot block on your boot drive
1384 The following example updates the ZFS boot block on the first
1385 partition (freebsd-boot) of a GPT partitioned drive ad0:
1386 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1388 Non-boot pools do not need these updates.
1392 If you are updating from a prior version of FreeBSD (even one just
1393 a few days old), you should follow this procedure. It is the most
1394 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1396 make kernel-toolchain
1397 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1398 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1400 To test a kernel once
1401 ---------------------
1402 If you just want to boot a kernel once (because you are not sure
1403 if it works, or if you want to boot a known bad kernel to provide
1404 debugging information) run
1405 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1406 nextboot -k testkernel
1408 To just build a kernel when you know that it won't mess you up
1409 --------------------------------------------------------------
1410 This assumes you are already running a CURRENT system. Replace
1411 ${arch} with the architecture of your machine (e.g. "i386",
1412 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1414 cd src/sys/${arch}/conf
1415 config KERNEL_NAME_HERE
1416 cd ../compile/KERNEL_NAME_HERE
1421 If this fails, go to the "To build a kernel" section.
1423 To rebuild everything and install it on the current system.
1424 -----------------------------------------------------------
1425 # Note: sometimes if you are running current you gotta do more than
1426 # is listed here if you are upgrading from a really old current.
1428 <make sure you have good level 0 dumps>
1430 make kernel KERNCONF=YOUR_KERNEL_HERE
1432 <reboot in single user> [3]
1440 To cross-install current onto a separate partition
1441 --------------------------------------------------
1442 # In this approach we use a separate partition to hold
1443 # current's root, 'usr', and 'var' directories. A partition
1444 # holding "/", "/usr" and "/var" should be about 2GB in
1447 <make sure you have good level 0 dumps>
1450 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1451 <maybe newfs current's root partition>
1452 <mount current's root partition on directory ${CURRENT_ROOT}>
1453 make installworld DESTDIR=${CURRENT_ROOT}
1454 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1455 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1456 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1457 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1458 <reboot into current>
1459 <do a "native" rebuild/install as described in the previous section>
1460 <maybe install compatibility libraries from ports/misc/compat*>
1464 To upgrade in-place from 8.x-stable to current
1465 ----------------------------------------------
1466 <make sure you have good level 0 dumps>
1468 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1470 <reboot in single user> [3]
1477 Make sure that you've read the UPDATING file to understand the
1478 tweaks to various things you need. At this point in the life
1479 cycle of current, things change often and you are on your own
1480 to cope. The defaults can also change, so please read ALL of
1481 the UPDATING entries.
1483 Also, if you are tracking -current, you must be subscribed to
1484 freebsd-current@freebsd.org. Make sure that before you update
1485 your sources that you have read and understood all the recent
1486 messages there. If in doubt, please track -stable which has
1487 much fewer pitfalls.
1489 [1] If you have third party modules, such as vmware, you
1490 should disable them at this point so they don't crash your
1493 [3] From the bootblocks, boot -s, and then do
1498 adjkerntz -i # if CMOS is wall time
1499 Also, when doing a major release upgrade, it is required that
1500 you boot into single user mode to do the installworld.
1502 [4] Note: This step is non-optional. Failure to do this step
1503 can result in a significant reduction in the functionality of the
1504 system. Attempting to do it by hand is not recommended and those
1505 that pursue this avenue should read this file carefully, as well
1506 as the archives of freebsd-current and freebsd-hackers mailing lists
1507 for potential gotchas. The -U option is also useful to consider.
1508 See mergemaster(8) for more information.
1510 [5] Usually this step is a noop. However, from time to time
1511 you may need to do this if you get unknown user in the following
1512 step. It never hurts to do it all the time. You may need to
1513 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1514 install) after the buildworld before this step if you last updated
1515 from current before 20020224 or from -stable before 20020408.
1517 [6] This only deletes old files and directories. Old libraries
1518 can be deleted by "make delete-old-libs", but you have to make
1519 sure that no program is using those libraries anymore.
1521 [8] In order to have a kernel that can run the 4.x binaries needed to
1522 do an installworld, you must include the COMPAT_FREEBSD4 option in
1523 your kernel. Failure to do so may leave you with a system that is
1524 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1525 required to run the 5.x binaries on more recent kernels. And so on
1526 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1528 Make sure that you merge any new devices from GENERIC since the
1529 last time you updated your kernel config file.
1531 [9] When checking out sources, you must include the -P flag to have
1532 cvs prune empty directories.
1534 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1535 "?=" instead of the "=" assignment operator, so that buildworld can
1536 override the CPUTYPE if it needs to.
1538 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1539 not on the command line, or in /etc/make.conf. buildworld will
1540 warn if it is improperly defined.
1543 This file contains a list, in reverse chronological order, of major
1544 breakages in tracking -current. It is not guaranteed to be a complete
1545 list of such breakages, and only contains entries since October 10, 2007.
1546 If you need to see UPDATING entries from before that date, you will need
1547 to fetch an UPDATING file from an older FreeBSD release.
1549 Copyright information:
1551 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1553 Redistribution, publication, translation and use, with or without
1554 modification, in full or in part, in any form or format of this
1555 document are permitted without further permission from the author.
1557 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1558 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1559 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1560 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1561 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1562 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1563 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1564 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1565 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1566 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1567 POSSIBILITY OF SUCH DAMAGE.
1569 Contact Warner Losh if you have any questions about your use of