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 completely disable malloc
23 debugging, define MALLOC_PRODUCTION in /etc/make.conf, or to merely
24 disable the most expensive debugging functionality run
25 "ln -s 'abort:false,junk:false' /etc/malloc.conf".)
28 The malloc(3) implementation embedded in libc now uses sources imported
29 as contrib/jemalloc. The most disruptive API change is to
30 /etc/malloc.conf. If your system has an old-style /etc/malloc.conf,
31 delete it prior to installworld, and optionally re-create it using the
32 new format after rebooting. See malloc.conf(5) for details
33 (specifically the TUNING section and the "opt.*" entries in the MALLCTL
37 Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb
38 is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is
39 now spelled mips. This is to aid compatibility with third-party
40 software that expects this naming scheme in uname(3). Little-endian
41 settings are unchanged.
44 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
48 Now unix domain sockets behave "as expected" on nullfs(5). Previously
49 nullfs(5) did not pass through all behaviours to the underlying layer,
50 as a result if we bound to a socket on the lower layer we could connect
51 only to the lower path; if we bound to the upper layer we could connect
52 only to the upper path. The new behavior is one can connect to both the
53 lower and the upper paths regardless what layer path one binds to.
56 The getifaddrs upgrade path broken with 20111215 has been restored.
57 If you have upgraded in between 20111215 and 20120209 you need to
58 recompile libc again with your kernel. You still need to recompile
59 world to be able to configure CARP but this restriction already
63 The set_rcvar() function has been removed from /etc/rc.subr. All
64 base and ports rc.d scripts have been updated, so if you have a
65 port installed with a script in /usr/local/etc/rc.d you can either
66 hand-edit the rcvar= line, or reinstall the port.
68 An easy way to handle the mass-update of /etc/rc.d:
69 rm /etc/rc.d/* && mergemaster -i
72 panic(9) now stops other CPUs in the SMP systems, disables interrupts
73 on the current CPU and prevents other threads from running.
74 This behavior can be reverted using the kern.stop_scheduler_on_panic
76 The new behavior can be incompatible with kern.sync_on_panic.
79 The carp(4) facility has been changed significantly. Configuration
80 of the CARP protocol via ifconfig(8) has changed, as well as format
81 of CARP events submitted to devd(8) has changed. See manual pages
82 for more information. The arpbalance feature of carp(4) is currently
83 not supported anymore.
85 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
86 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
87 need to be recompiled.
90 The acpi_wmi(4) status device /dev/wmistat has been renamed to
94 The option VFS_ALLOW_NONMPSAFE option has been added in order to
95 explicitely support non-MPSAFE filesystems.
96 It is on by default for all supported platform at this present
100 The broken amd(4) driver has been replaced with esp(4) in the amd64,
101 i386 and pc98 GENERIC kernel configuration files.
104 sysinstall has been removed
107 The stable/9 branch created in subversion. This corresponds to the
108 RELENG_9 branch in CVS.
111 This commit modifies vfs_register() so that it uses a hash
112 calculation to set vfc_typenum, which is enabled by default.
113 The first time a system is booted after this change, the
114 vfc_typenum values will change for all file systems. The
115 main effect of this is a change to the NFS server file handles
116 for file systems that use vfc_typenum in their fsid, such as ZFS.
117 It will, however, prevent vfc_typenum from changing when file
118 systems are loaded in a different order for subsequent reboots.
119 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
120 until you are ready to remount all NFS clients after a reboot.
123 Bump the shared library version numbers for libraries that
124 do not use symbol versioning, have changed the ABI compared
125 to stable/8 and which shared library version was not bumped.
126 Done as part of 9.0-RELEASE cycle.
129 During the merge of Capsicum features, the fget(9) KPI was modified.
130 This may require the rebuilding of out-of-tree device drivers --
131 issues have been reported specifically with the nVidia device driver.
132 __FreeBSD_version is bumped to 900041.
134 Also, there is a period between 20110811 and 20110814 where the
135 special devices /dev/{stdin,stdout,stderr} did not work correctly.
136 Building world from a kernel during that window may not work.
139 The packet filter (pf) code has been updated to OpenBSD 4.5.
140 You need to update userland tools to be in sync with kernel.
141 This update breaks backward compatibility with earlier pfsync(4)
142 versions. Care must be taken when updating redundant firewall setups.
145 The following sysctls and tunables are retired on x86 platforms:
147 machdep.hlt_logical_cpus
148 The following sysctl is retired:
149 machdep.hyperthreading_allowed
150 The sysctls were supposed to provide a way to dynamically offline and
151 online selected CPUs on x86 platforms, but the implementation has not
152 been reliable especially with SCHED_ULE scheduler.
153 machdep.hyperthreading_allowed tunable is still available to ignore
154 hyperthreading CPUs at OS level.
155 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
156 where X is an APIC ID of a CPU. Be advised, though, that disabling
157 CPUs in non-uniform fashion will result in non-uniform topology and
158 may lead to sub-optimal system performance with SCHED_ULE, which is
162 cpumask_t type is retired and cpuset_t is used in order to describe
166 Changes to ifconfig(8) for dynamic address family detection mandate
167 that you are running a kernel of 20110525 or later. Make sure to
168 follow the update procedure to boot a new kernel before installing
172 Support for sun4v architecture is officially dropped
175 Several KPI breaking changes have been committed to the mii(4) layer,
176 the PHY drivers and consequently some Ethernet drivers using mii(4).
177 This means that miibus.ko and the modules of the affected Ethernet
178 drivers need to be recompiled.
180 Note to kernel developers: Given that the OUI bit reversion problem
181 was fixed as part of these changes all mii(4) commits related to OUIs,
182 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
183 handling, no longer can be merged verbatim to stable/8 and previous
187 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
188 into their kernel configurations along with 'device pci'.
191 The default NFS client is now the new NFS client, so fstype "newnfs"
192 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
193 Although mounts via fstype "nfs" will usually work without userland
194 changes, it is recommended that the mount(8) and mount_nfs(8)
195 commands be rebuilt from sources and that a link to mount_nfs called
196 mount_oldnfs be created. The new client is compiled into the
197 kernel with "options NFSCL" and this is needed for diskless root
198 file systems. The GENERIC kernel configs have been changed to use
199 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
200 To use the regular/old client, you can "mount -t oldnfs ...". For
201 a diskless root file system, you must also include a line like:
203 vfs.root.mountfrom="oldnfs:"
205 in the boot/loader.conf on the root fs on the NFS server to make
206 a diskless root fs use the old client.
209 The GENERIC kernels for all architectures now default to the new
210 CAM-based ATA stack. It means that all legacy ATA drivers were
211 removed and replaced by respective CAM drivers. If you are using
212 ATA device names in /etc/fstab or other places, make sure to update
213 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
214 where 'Y's are the sequential numbers starting from zero for each type
215 in order of detection, unless configured otherwise with tunables,
216 see cam(4)). There will be symbolic links created in /dev/ to map
217 old adX devices to the respective adaY. They should provide basic
218 compatibility for file systems mounting in most cases, but they do
219 not support old user-level APIs and do not have respective providers
220 in GEOM. Consider using updated management tools with new device names.
222 It is possible to load devices ahci, ata, siis and mvs as modules,
223 but option ATA_CAM should remain in kernel configuration to make ata
224 module work as CAM driver supporting legacy ATA controllers. Device ata
225 still can be used in modular fashion (atacore + ...). Modules atadisk
226 and atapi* are not used and won't affect operation in ATA_CAM mode.
227 Note that to use CAM-based ATA kernel should include CAM devices
228 scbus, pass, da (or explicitly ada), cd and optionally others. All of
229 them are parts of the cam module.
231 ataraid(4) functionality is now supported by the RAID GEOM class.
232 To use it you can load geom_raid kernel module and use graid(8) tool
233 for management. Instead of /dev/arX device names, use /dev/raid/rX.
235 No kernel config options or code have been removed, so if a problem
236 arises, please report it and optionally revert to the old ATA stack.
237 In order to do it you can remove from the kernel config:
242 , and instead add back:
243 device atadisk # ATA disk drives
244 device ataraid # ATA RAID drives
245 device atapicd # ATAPI CDROM drives
246 device atapifd # ATAPI floppy drives
247 device atapist # ATAPI tape drives
250 The default NFS server has been changed to the new server, which
251 was referred to as the experimental server. If you need to switch
252 back to the old NFS server, you must now put the "-o" option on
253 both the mountd and nfsd commands. This can be done using the
254 mountd_flags and nfs_server_flags rc.conf variables until an
255 update to the rc scripts is committed, which is coming soon.
258 The GNU Objective-C runtime library (libobjc), and other Objective-C
259 related components have been removed from the base system. If you
260 require an Objective-C library, please use one of the available ports.
263 ath(4) has been split into bus- and device- modules. if_ath contains
264 the HAL, the TX rate control and the network device code. if_ath_pci
265 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
266 contains the AHB glue. Users need to load both if_ath_pci and if_ath
267 in order to use ath on everything else.
269 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
270 need to load if_ath and if_ath_pci for ath(4) operation.
273 As part of the replacement of sysinstall, the process of building
274 release media has changed significantly. For details, please re-read
275 release(7), which has been updated to reflect the new build process.
278 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
279 is the last available version under GPLv2. It brings a number of new
280 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
281 4.1 and SSE 4.2), better support for powerpc64, a number of new
282 directives, and lots of other small improvements. See the ChangeLog
283 file in contrib/binutils for the full details.
286 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
287 compliant, and will now use half of hash for authentication.
288 This will break interoperability with all stacks (including all
289 actual FreeBSD versions) who implement
290 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
292 The only workaround with such peers is to use another HMAC
293 algorithm for IPsec ("phase 2") authentication.
296 Remove the uio_yield prototype and symbol. This function has
297 been misnamed since it was introduced and should not be
298 globally exposed with this name. The equivalent functionality
299 is now available using kern_yield(curthread->td_user_pri).
300 The function remains undocumented.
303 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
304 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
305 for scalar sysctls is defined but disabled. Code that needs
306 UQUAD to pass the type checking that must compile on older
307 systems where the define is not present can check against
308 __FreeBSD_version >= 900030.
310 The system dialog(1) has been replaced with a new version previously
311 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
312 with the previous version, but the libdialog associated with it has
313 a largely incompatible API. As such, the original version of libdialog
314 will be kept temporarily as libodialog, until its base system consumers
315 are replaced or updated. Bump __FreeBSD_version to 900030.
318 If you are trying to run make universe on a -stable system, and you get
319 the following warning:
320 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
321 unknown. config(8) likely too old."
322 or something similar to it, then you must upgrade your -stable system
323 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
324 stable/8) or build the config from the latest stable/8 branch and
325 install it on your system.
327 Prior to this date, building a current universe on 8-stable system from
328 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
329 error in the first kernel build phase. A new config on those old
330 systems will fix that problem for older versions of -current.
333 The TCP stack has been modified to allow Khelp modules to interact with
334 it via helper hook points and store per-connection data in the TCP
335 control block. Bump __FreeBSD_version to 900029. User space tools that
336 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
340 Generic IEEE 802.3 annex 31B full duplex flow control support has been
341 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
342 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
343 to take advantage of it instead of using custom implementations. This
344 means that these drivers now no longer unconditionally advertise
345 support for flow control but only do so if flow control is a selected
346 media option. This was implemented in the generic support that way in
347 order to allow flow control to be switched on and off via ifconfig(8)
348 with the PHY specific default to typically off in order to protect
349 from unwanted effects. Consequently, if you used flow control with
350 one of the above mentioned drivers you now need to explicitly enable
352 ifconfig bge0 media auto mediaopt flowcontrol
354 Along with the above mentioned changes generic support for setting
355 1000baseT master mode also has been added and brgphy(4), ciphy(4),
356 e1000phy(4) as well as ip1000phy(4) have been converted to take
357 advantage of it. This means that these drivers now no longer take the
358 link0 parameter for selecting master mode but the master media option
359 has to be used instead, for example like in the following:
360 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
362 Selection of master mode now is also available with all other PHY
363 drivers supporting 1000baseT.
366 The TCP stack has received a significant update to add support for
367 modularised congestion control and generally improve the clarity of
368 congestion control decisions. Bump __FreeBSD_version to 900025. User
369 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
370 sockstat) need to be recompiled.
373 The man(1) utility has been replaced by a new version that no longer
374 uses /etc/manpath.config. Please consult man.conf(5) for how to
375 migrate local entries to the new format.
378 The copyright strings printed by login(1) and sshd(8) at the time of a
379 new connection have been removed to follow other operating systems and
383 A workaround for a fixed ld bug has been removed in kernel code,
384 so make sure that your system ld is built from sources after
385 revision 210245 from 2010-07-19 (r211583 if building head kernel
386 on stable/8, r211584 for stable/7; both from 2010-08-21).
387 A symptom of incorrect ld version is different addresses for
388 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
391 The $ipv6_prefer variable in rc.conf(5) has been split into
392 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
394 The $ip6addrctl_policy is a variable to choose a pre-defined
395 address selection policy set by ip6addrctl(8). A value
396 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
399 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
400 flag (see an entry of 20090926) is set on an interface with no
401 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
402 security reason. If you want IPv6 link-local address on all
403 interfaces by default, set this to "YES".
405 The old ipv6_prefer="YES" is equivalent to
406 ipv6_activate_all_interfaces="YES" and
407 ip6addrctl_policy="ipv6_prefer".
410 DTrace has grown support for userland tracing. Due to this, DTrace is
411 now i386 and amd64 only.
412 dtruss(1) is now installed by default on those systems and a new
413 kernel module is needed for userland tracing: fasttrap.
414 No changes to your kernel config file are necessary to enable
415 userland tracing, but you might consider adding 'STRIP=' and
416 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
417 to have informative userland stack traces in DTrace (ustack).
420 The acpi_aiboost(4) driver has been removed in favor of the new
421 aibs(4) driver. You should update your kernel configuration file.
424 BSD grep has been imported to the base system and it is built by
425 default. It is completely BSD licensed, highly GNU-compatible, uses
426 less memory than its GNU counterpart and has a small codebase.
427 However, it is slower than its GNU counterpart, which is mostly
428 noticeable for larger searches, for smaller ones it is measurable
429 but not significant. The reason is complex, the most important factor
430 is that we lack a modern and efficient regex library and GNU
431 overcomes this by optimizing the searches internally. Future work
432 on improving the regex performance is planned, for the meantime,
433 users that need better performance, can build GNU grep instead by
434 setting the WITH_GNU_GREP knob.
437 Due to the import of powerpc64 support, all existing powerpc kernel
438 configuration files must be updated with a machine directive like this:
439 machine powerpc powerpc
441 In addition, an updated config(8) is required to build powerpc kernels
445 A new version of ZFS (version 15) has been merged to -HEAD.
446 This version uses a python library for the following subcommands:
447 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
448 For full functionality of these commands the following port must
449 be installed: sysutils/py-zfs
452 'vm_page's are now hashed by physical address to an array of mutexes.
453 Currently this is only used to serialize access to hold_count. Over
454 time the page queue mutex will be peeled away. This changes the size
455 of pmap on every architecture. And requires all callers of vm_page_hold
456 and vm_page_unhold to be updated.
459 WITH_CTF can now be specified in src.conf (not recommended, there
460 are some problems with static executables), make.conf (would also
461 affect ports which do not use GNU make and do not override the
462 compile targets) or in the kernel config (via "makeoptions
464 When WITH_CTF was specified there before this was silently ignored,
465 so make sure that WITH_CTF is not used in places which could lead
466 to unwanted behavior.
469 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
470 to allow 32-bit compatibility on non-x86 platforms. All kernel
471 configurations on amd64 and ia64 platforms using these options must
472 be modified accordingly.
475 The utmp user accounting database has been replaced with utmpx,
476 the user accounting interface standardized by POSIX.
477 Unfortunately the semantics of utmp and utmpx don't match,
478 making it practically impossible to support both interfaces.
479 The user accounting database is used by tools like finger(1),
480 last(1), talk(1), w(1) and ac(8).
482 All applications in the base system use utmpx. This means only
483 local binaries (e.g. from the ports tree) may still use these
484 utmp database files. These applications must be rebuilt to make
487 After the system has been upgraded, it is safe to remove the old
488 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
489 assuming their contents is of no importance anymore. Old wtmp
490 databases can only be used by last(1) and ac(8) after they have
491 been converted to the new format using wtmpcvt(1).
494 Introduce the kernel thread "deadlock resolver" (which can be enabled
495 via the DEADLKRES option, see NOTES for more details) and the
496 sleepq_type() function for sleepqueues.
499 The rc.firewall and rc.firewall6 were unified, and
500 rc.firewall6 and rc.d/ip6fw were removed.
501 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
502 variables are obsoleted. Instead, the following new rc
503 variables are added to rc.d/ipfw:
505 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
506 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
507 firewall_simple_onet_ipv6, firewall_trusted_ipv6
509 The meanings correspond to the relevant IPv4 variables.
515 The default terminal emulation for syscons(4) has been changed
516 from cons25 to xterm on all platforms except pc98. This means
517 that the /etc/ttys file needs to be updated to ensure correct
518 operation of applications on the console.
520 The terminal emulation style can be toggled per window by using
521 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
522 options can be used to change the compile-time default back to
525 To prevent graphical artifacts, make sure the TERM environment
526 variable is set to match the terminal emulation that is being
527 performed by syscons(4).
530 The layout of the structure ieee80211req_scan_result has changed.
531 Applications that require wireless scan results (e.g. ifconfig(8))
532 from net80211 need to be recompiled.
534 Applications such as wpa_supplicant(8) may require a full world
535 build without using NO_CLEAN in order to get synchronized with the
539 The iwn(4) driver has been updated to support the 5000 and 5150 series.
540 There's one kernel module for each firmware. Adding "device iwnfw"
541 to the kernel configuration file means including all three firmware
542 images inside the kernel. If you want to include just the one for
543 your wireless card, use the devices iwn4965fw, iwn5000fw or
547 The rc.d/network_ipv6, IPv6 configuration script has been integrated
548 into rc.d/netif. The changes are the following:
550 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
551 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
552 Note that both variables need the "inet6" keyword at the head.
554 Do not set $ipv6_network_interfaces manually if you do not
555 understand what you are doing. It is not needed in most cases.
557 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
560 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
561 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
563 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
564 all configured interfaces have "inet6 accept_rtadv" in the
565 $ifconfig_IF_ipv6. These are for backward compatibility.
567 3. A new variable $ipv6_prefer has been added. If NO, IPv6
568 functionality of interfaces with no corresponding
569 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
570 and the default address selection policy of ip6addrctl(8)
571 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
572 Note that if you want to configure IPv6 functionality on the
573 disabled interfaces after boot, first you need to clear the flag by
574 using ifconfig(8) like:
576 ifconfig em0 inet6 -ifdisabled
578 If YES, the default address selection policy is set as
581 The default value of $ipv6_prefer is NO.
583 4. If your system need to receive Router Advertisement messages,
584 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
585 scripts automatically invoke rtsol(8) when the interface becomes
586 UP. The Router Advertisement messages are used for SLAAC
587 (State-Less Address AutoConfiguration).
590 802.11s D3.03 support was committed. This is incompatible with the
591 previous code, which was based on D3.0.
594 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
595 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
596 control whether accepting Router Advertisement messages or not.
597 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
598 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
599 The ifconfig(8) utility now supports these flags.
602 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
603 mount(8) and -a option for df(1) to see them.
606 The old tunable hw.bus.devctl_disable has been superseded by
607 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
608 replaced by hw.bus.devctl_queue=0. The default for this new tunable
612 Remove the option STOP_NMI. The default action is now to use NMI only
613 for KDB via the newly introduced function stop_cpus_hard() and
614 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
617 The stable/8 branch created in subversion. This corresponds to the
618 RELENG_8 branch in CVS.
621 Bump the shared library version numbers for all libraries that do not
622 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
623 __FreeBSD_version to 800105.
626 Due to changes in the implementation of virtual network stack support,
627 all network-related kernel modules must be recompiled. As this change
628 breaks the ABI, bump __FreeBSD_version to 800104.
631 The TOE interface to the TCP syncache has been modified to remove
632 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
633 The cxgb driver is the only TOE consumer affected by this change, and
634 needs to be recompiled along with the kernel. As this change breaks
635 the ABI, bump __FreeBSD_version to 800103.
638 Padding has been added to struct tcpcb, sackhint and tcpstat in
639 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
640 maintaining the ABI. However, this change breaks the ABI, so bump
641 __FreeBSD_version to 800102. User space tools that rely on the size of
642 any of these structs (e.g. sockstat) need to be recompiled.
645 The NFS_LEGACYRPC option has been removed along with the old kernel
646 RPC implementation that this option selected. Kernel configurations
647 may need to be adjusted.
650 The network interface device nodes at /dev/net/<interface> have been
651 removed. All ioctl operations can be performed the normal way using
652 routing sockets. The kqueue functionality can generally be replaced
653 with routing sockets.
656 The documentation from the FreeBSD Documentation Project (Handbook,
657 FAQ, etc.) is now installed via packages by sysinstall(8) and under
658 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
661 The ABI of various structures related to the SYSV IPC API have been
662 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
663 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
667 Layout of struct vnet has changed as routing related variables were
668 moved to their own Vimage module. Modules need to be recompiled. Bump
669 __FreeBSD_version to 800099.
672 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
673 respectively. As long as no more than 16 groups per process are used,
674 no changes should be visible. When more than 16 groups are used, old
675 binaries may fail if they call getgroups() or getgrouplist() with
676 statically sized storage. Recompiling will work around this, but
677 applications should be modified to use dynamically allocated storage
678 for group arrays as POSIX.1-2008 does not cap an implementation's
679 number of supported groups at NGROUPS_MAX+1 as previous versions did.
681 NFS and portalfs mounts may also be affected as the list of groups is
682 truncated to 16. Users of NFS who use more than 16 groups, should
683 take care that negative group permissions are not used on the exported
684 file systems as they will not be reliable unless a GSSAPI based
685 authentication method is used.
688 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
689 option compiles in the support for adaptive spinning for lockmgrs
690 which want to enable it. The lockinit() function now accepts the flag
691 LK_ADAPTIVE in order to make the lock object subject to adaptive
692 spinning when both held in write and read mode.
695 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
696 changed. User applications that use this ioctl need to be rebuilt.
699 The layout of struct thread has changed. Kernel and modules need to
703 The layout of structs ifnet, domain, protosw and vnet_net has changed.
704 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
707 window(1) has been removed from the base system. It can now be
708 installed from ports. The port is called misc/window.
711 The way we are storing and accessing `routing table' entries has
712 changed. Programs reading the FIB, like netstat, need to be
716 A new netisr implementation has been added for FreeBSD 8. Network
717 file system modules, such as igmp, ipdivert, and others, should be
719 Bump __FreeBSD_version to 800096.
722 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
726 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
727 Bump __FreeBSD_version to 800094.
730 Add mnt_xflag field to 'struct mount'. File system modules need to be
732 Bump __FreeBSD_version to 800093.
735 The compiling option ADAPTIVE_SX has been retired while it has been
736 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
737 The KPI for sx_init_flags() changes as accepting flags:
738 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
739 been introduced in order to handle the reversed logic.
740 Bump __FreeBSD_version to 800092.
743 Add support for hierarchical jails. Remove global securelevel.
744 Bump __FreeBSD_version to 800091.
747 The layout of struct vnet_net has changed, therefore modules
749 Bump __FreeBSD_version to 800090.
752 The newly imported zic(8) produces a new format in the output. Please
753 run tzsetup(8) to install the newly created data to /etc/localtime.
756 The sysctl tree for the usb stack has renamed from hw.usb2.* to
757 hw.usb.* and is now consistent again with previous releases.
760 802.11 monitor mode support was revised and driver api's were changed.
761 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
762 of DLT_IEEE802_11. No user-visible data structures were changed but
763 applications that use DLT_IEEE802_11 may require changes.
764 Bump __FreeBSD_version to 800088.
767 The layout of the following structs has changed: sysctl_oid,
768 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
769 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
770 panics may be experienced. World rebuild is required for
771 correctly checking networking state from userland.
772 Bump __FreeBSD_version to 800085.
775 MLDv2 and Source-Specific Multicast (SSM) have been merged
776 to the IPv6 stack. VIMAGE hooks are in but not yet used.
777 The implementation of SSM within FreeBSD's IPv6 stack closely
778 follows the IPv4 implementation.
780 For kernel developers:
782 * The most important changes are that the ip6_output() and
783 ip6_input() paths no longer take the IN6_MULTI_LOCK,
784 and this lock has been downgraded to a non-recursive mutex.
786 * As with the changes to the IPv4 stack to support SSM, filtering
787 of inbound multicast traffic must now be performed by transport
788 protocols within the IPv6 stack. This does not apply to TCP and
789 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
791 * The KPIs used by IPv6 multicast are similar to those used by
792 the IPv4 stack, with the following differences:
793 * im6o_mc_filter() is analogous to imo_multicast_filter().
794 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
795 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
796 * IN6_LOOKUP_MULTI() has been deprecated and removed.
797 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
798 for MLDv1 have an additional 'timer' argument which is used to
799 jitter the initial membership report for the solicited-node
800 multicast membership on-link.
801 * This is not strictly needed for MLDv2, which already jitters
802 its report transmissions. However, the 'timer' argument is
803 preserved in case MLDv1 is active on the interface.
805 * The KAME linked-list based IPv6 membership implementation has
806 been refactored to use a vector similar to that used by the IPv4
808 Code which maintains a list of its own multicast memberships
809 internally, e.g. carp, has been updated to reflect the new
812 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
813 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
814 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
815 implementation constraint which needs to be addressed in HEAD.
817 For application developers:
819 * The changes are broadly similar to those made for the IPv4
822 * The use of IPv4 and IPv6 multicast socket options on the same
823 socket, using mapped addresses, HAS NOT been tested or supported.
825 * There are a number of issues with the implementation of various
826 IPv6 multicast APIs which need to be resolved in the API surface
827 before the implementation is fully compatible with KAME userland
828 use, and these are mostly to do with interface index treatment.
830 * The literature available discusses the use of either the delta / ASM
831 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
832 using setsourcefilter(3)/getsourcefilter(3). For more information
833 please refer to RFC 3768, 'Socket Interface Extensions for
834 Multicast Source Filters'.
836 * Applications which use the published RFC 3678 APIs should be fine.
838 For systems administrators:
840 * The mtest(8) utility has been refactored to support IPv6, in
841 addition to IPv4. Interface addresses are no longer accepted
842 as arguments, their names must be used instead. The utility
843 will map the interface name to its first IPv4 address as
844 returned by getifaddrs(3).
846 * The ifmcstat(8) utility has also been updated to print the MLDv2
847 endpoint state and source filter lists via sysctl(3).
849 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
850 loopback of IPv6 multicast datagrams by default; it defaults to 1
851 to preserve the existing behaviour. Disabling multicast loopback is
852 recommended for optimal system performance.
854 * The IPv6 MROUTING code has been changed to examine this sysctl
855 instead of attempting to perform a group lookup before looping
856 back forwarded datagrams.
858 Bump __FreeBSD_version to 800084.
861 Implement low-level Bluetooth HCI API.
862 Bump __FreeBSD_version to 800083.
865 The layout of struct malloc_type, used by modules to register new
866 memory allocation types, has changed. Most modules will need to
867 be rebuilt or panics may be experienced.
868 Bump __FreeBSD_version to 800081.
871 Anticipate overflowing inp_flags - add inp_flags2.
872 This changes most offsets in inpcb, so checking v4 connection
873 state will require a world rebuild.
874 Bump __FreeBSD_version to 800080.
877 Add an llentry to struct route and struct route_in6. Modules
878 embedding a struct route will need to be recompiled.
879 Bump __FreeBSD_version to 800079.
882 The size of rt_metrics_lite and by extension rtentry has changed.
883 Networking administration apps will need to be recompiled.
884 The route command now supports show as an alias for get, weighting
885 of routes, sticky and nostick flags to alter the behavior of stateful
887 Bump __FreeBSD_version to 800078.
890 Do not use Giant for kbdmux(4) locking. This is wrong and
891 apparently causing more problems than it solves. This will
892 re-open the issue where interrupt handlers may race with
893 kbdmux(4) in polling mode. Typical symptoms include (but
894 not limited to) duplicated and/or missing characters when
895 low level console functions (such as gets) are used while
896 interrupts are enabled (for example geli password prompt,
897 mountroot prompt etc.). Disabling kbdmux(4) may help.
900 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
901 kernel modules referencing any of the above need to be recompiled.
902 Bump __FreeBSD_version to 800075.
905 GEOM_PART has become the default partition slicer for storage devices,
906 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
907 introduces some changes:
909 MSDOS/EBR: the devices created from MSDOS extended partition entries
910 (EBR) can be named differently than with GEOM_MBR and are now symlinks
911 to devices with offset-based names. fstabs may need to be modified.
913 BSD: the "geometry does not match label" warning is harmless in most
914 cases but it points to problems in file system misalignment with
915 disk geometry. The "c" partition is now implicit, covers the whole
916 top-level drive and cannot be (mis)used by users.
918 General: Kernel dumps are now not allowed to be written to devices
919 whose partition types indicate they are meant to be used for file
920 systems (or, in case of MSDOS partitions, as something else than
923 Most of these changes date approximately from 200812.
926 The uscanner(4) driver has been removed from the kernel. This follows
927 Linux removing theirs in 2.6 and making libusb the default interface
931 The multicast forwarding code has been cleaned up. netstat(1)
932 only relies on KVM now for printing bandwidth upcall meters.
933 The IPv4 and IPv6 modules are split into ip_mroute_mod and
934 ip6_mroute_mod respectively. The config(5) options for statically
935 compiling this code remain the same, i.e. 'options MROUTING'.
938 Support for the IFF_NEEDSGIANT network interface flag has been
939 removed, which means that non-MPSAFE network device drivers are no
940 longer supported. In particular, if_ar, if_sr, and network device
941 drivers from the old (legacy) USB stack can no longer be built or
945 POSIX.1 Native Language Support (NLS) has been enabled in libc and
946 a bunch of new language catalog files have also been added.
947 This means that some common libc messages are now localized and
948 they depend on the LC_MESSAGES environmental variable.
951 The k8temp(4) driver has been renamed to amdtemp(4) since
952 support for Family 10 and Family 11 CPU families was added.
955 IGMPv3 and Source-Specific Multicast (SSM) have been merged
956 to the IPv4 stack. VIMAGE hooks are in but not yet used.
958 For kernel developers, the most important changes are that the
959 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
960 and this lock has been downgraded to a non-recursive mutex.
962 Transport protocols (UDP, Raw IP) are now responsible for filtering
963 inbound multicast traffic according to group membership and source
964 filters. The imo_multicast_filter() KPI exists for this purpose.
965 Transports which do not use multicast (SCTP, TCP) already reject
966 multicast by default. Forwarding and receive performance may improve
967 as a mutex acquisition is no longer needed in the ip_input()
968 low-level input path. in_addmulti() and in_delmulti() are shimmed
969 to new KPIs which exist to support SSM in-kernel.
971 For application developers, it is recommended that loopback of
972 multicast datagrams be disabled for best performance, as this
973 will still cause the lock to be taken for each looped-back
974 datagram transmission. The net.inet.ip.mcast.loop sysctl may
975 be tuned to 0 to disable loopback by default; it defaults to 1
976 to preserve the existing behaviour.
978 For systems administrators, to obtain best performance with
979 multicast reception and multiple groups, it is always recommended
980 that a card with a suitably precise hash filter is used. Hash
981 collisions will still result in the lock being taken within the
982 transport protocol input path to check group membership.
984 If deploying FreeBSD in an environment with IGMP snooping switches,
985 it is recommended that the net.inet.igmp.sendlocal sysctl remain
986 enabled; this forces 224.0.0.0/24 group membership to be announced
989 The size of 'struct igmpstat' has changed; netstat needs to be
990 recompiled to reflect this.
991 Bump __FreeBSD_version to 800070.
994 libusb20.so.1 is now installed as libusb.so.1 and the ports system
995 updated to use it. This requires a buildworld/installworld in order to
996 update the library and dependencies (usbconfig, etc). Its advisable to
997 rebuild all ports which uses libusb. More specific directions are given
998 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
999 libusb are no longer required and can be removed.
1002 A workaround is committed to allow the creation of System V shared
1003 memory segment of size > 2 GB on the 64-bit architectures.
1004 Due to a limitation of the existing ABI, the shm_segsz member
1005 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
1006 wrong for large segments. Note that limits must be explicitly
1007 raised to allow such segments to be created.
1010 The layout of struct ifnet has changed, requiring a rebuild of all
1011 network device driver modules.
1014 The /dev handling for the new USB stack has changed, a
1015 buildworld/installworld is required for libusb20.
1018 The new USB2 stack has now been permanently moved in and all kernel and
1019 module names reverted to their previous values (eg, usb, ehci, ohci,
1020 ums, ...). The old usb stack can be compiled in by prefixing the name
1021 with the letter 'o', the old usb modules have been removed.
1022 Updating entry 20090216 for xorg and 20090215 for libmap may still
1026 The rc.conf(5) option if_up_delay has been renamed to
1027 defaultroute_delay to better reflect its purpose. If you have
1028 customized this setting in /etc/rc.conf you need to update it to
1032 xorg 7.4 wants to configure its input devices via hald which does not
1033 yet work with USB2. If the keyboard/mouse does not work in xorg then
1035 Option "AllowEmptyInput" "off"
1036 to your ServerLayout section. This will cause X to use the configured
1037 kbd and mouse sections from your xorg.conf.
1040 The GENERIC kernels for all architectures now default to the new USB2
1041 stack. No kernel config options or code have been removed so if a
1042 problem arises please report it and optionally revert to the old USB
1043 stack. If you are loading USB kernel modules or have a custom kernel
1044 that includes GENERIC then ensure that usb names are also changed over,
1045 eg uftdi -> usb2_serial_ftdi.
1047 Older programs linked against the ports libusb 0.1 need to be
1048 redirected to the new stack's libusb20. /etc/libmap.conf can
1050 # Map old usb library to new one for usb2 stack
1051 libusb-0.1.so.8 libusb20.so.1
1054 All USB ethernet devices now attach as interfaces under the name ueN
1055 (eg. ue0). This is to provide a predictable name as vendors often
1056 change usb chipsets in a product without notice.
1059 The ichsmb(4) driver has been changed to require SMBus slave
1060 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1061 All of the other SMBus controller drivers require left-justified
1062 slave addresses, so this change makes all the drivers provide the
1066 INET6 statistics (struct ip6stat) was updated.
1067 netstat(1) needs to be recompiled.
1070 NTFS has been removed from GENERIC kernel on amd64 to match
1071 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1072 will load ntfs.ko module automatically when NTFS support is
1073 actually needed, unless ntfs.ko is not installed or security
1074 level prohibits loading kernel modules. If either is the case,
1075 "options NTFS" has to be added into kernel config.
1078 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1079 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1080 800061. User space tools that rely on the size of struct tcpcb in
1081 tcp_var.h (e.g. sockstat) need to be recompiled.
1084 ng_tty(4) module updated to match the new TTY subsystem.
1085 Due to API change, user-level applications must be updated.
1086 New API support added to mpd5 CVS and expected to be present
1087 in next mpd5.3 release.
1090 With __FreeBSD_version 800060 the makefs tool is part of
1091 the base system (it was a port).
1094 The afdata and ifnet locks have been changed from mutexes to
1095 rwlocks, network modules will need to be re-compiled.
1098 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1099 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1100 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1101 architecture and 40 bytes on 64-bit architecture. The userland
1102 applications "arp" and "ndp" have been updated accordingly.
1103 The output from "netstat -r" shows only routing entries and
1104 none of the L2 information.
1107 __FreeBSD_version 800057 marks the switchover from the
1108 binary ath hal to source code. Users must add the line:
1110 options AH_SUPPORT_AR5416
1112 to their kernel config files when specifying:
1116 The ath_hal module no longer exists; the code is now compiled
1117 together with the driver in the ath module. It is now
1118 possible to tailor chip support (i.e. reduce the set of chips
1119 and thereby the code size); consult ath_hal(4) for details.
1122 __FreeBSD_version 800054 adds memory barriers to
1123 <machine/atomic.h>, new interfaces to ifnet to facilitate
1124 multiple hardware transmit queues for cards that support
1125 them, and a lock-less ring-buffer implementation to
1126 enable drivers to more efficiently manage queueing of
1130 A new version of ZFS (version 13) has been merged to -HEAD.
1131 This version has zpool attribute "listsnapshots" off by
1132 default, which means "zfs list" does not show snapshots,
1133 and is the same as Solaris behavior.
1136 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1139 The uhci, ohci, ehci and slhci USB Host controller drivers have
1140 been put into separate modules. If you load the usb module
1141 separately through loader.conf you will need to load the
1142 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1143 controller add the following to loader.conf:
1149 The ABI used by the PMC toolset has changed. Please keep
1150 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1154 atapci kernel module now includes only generic PCI ATA
1155 driver. AHCI driver moved to ataahci kernel module.
1156 All vendor-specific code moved into separate kernel modules:
1157 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1158 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1159 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1160 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1163 The TTY subsystem of the kernel has been replaced by a new
1164 implementation, which provides better scalability and an
1165 improved driver model. Most common drivers have been migrated to
1166 the new TTY subsystem, while others have not. The following
1167 drivers have not yet been ported to the new TTY layer:
1170 cy, digi, rc, rp, sio
1176 ng_h4, ng_tty, ppp, sl, snp
1178 Adding these drivers to your kernel configuration file shall
1179 cause compilation to fail.
1182 ntpd has been upgraded to 4.2.4p5.
1185 OpenSSH has been upgraded to 5.1p1.
1187 For many years, FreeBSD's version of OpenSSH preferred DSA
1188 over RSA for host and user authentication keys. With this
1189 upgrade, we've switched to the vendor's default of RSA over
1190 DSA. This may cause upgraded clients to warn about unknown
1191 host keys even for previously known hosts. Users should
1192 follow the usual procedure for verifying host keys before
1193 accepting the RSA key.
1195 This can be circumvented by setting the "HostKeyAlgorithms"
1196 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1199 Please note that the sequence of keys offered for
1200 authentication has been changed as well. You may want to
1201 specify IdentityFile in a different order to revert this
1205 The sio(4) driver has been removed from the i386 and amd64
1206 kernel configuration files. This means uart(4) is now the
1207 default serial port driver on those platforms as well.
1209 To prevent collisions with the sio(4) driver, the uart(4) driver
1210 uses different names for its device nodes. This means the
1211 onboard serial port will now most likely be called "ttyu0"
1212 instead of "ttyd0". You may need to reconfigure applications to
1213 use the new device names.
1215 When using the serial port as a boot console, be sure to update
1216 /boot/device.hints and /etc/ttys before booting the new kernel.
1217 If you forget to do so, you can still manually specify the hints
1218 at the loader prompt:
1220 set hint.uart.0.at="isa"
1221 set hint.uart.0.port="0x3F8"
1222 set hint.uart.0.flags="0x10"
1223 set hint.uart.0.irq="4"
1227 The gpt(8) utility has been removed. Use gpart(8) to partition
1231 The version that Linuxulator emulates was changed from 2.4.2
1232 to 2.6.16. If you experience any problems with Linux binaries
1233 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1234 if it fixes the problem contact emulation mailing list.
1237 ISDN4BSD (I4B) was removed from the src tree. You may need to
1238 update a your kernel configuration and remove relevant entries.
1241 I have checked in code to support multiple routing tables.
1242 See the man pages setfib(1) and setfib(2).
1243 This is a hopefully backwards compatible version,
1244 but to make use of it you need to compile your kernel
1245 with options ROUTETABLES=2 (or more up to 16).
1248 The 802.11 wireless support was redone to enable multi-bss
1249 operation on devices that are capable. The underlying device
1250 is no longer used directly but instead wlanX devices are
1251 cloned with ifconfig. This requires changes to rc.conf files.
1252 For example, change:
1253 ifconfig_ath0="WPA DHCP"
1256 ifconfig_wlan0="WPA DHCP"
1257 see rc.conf(5) for more details. In addition, mergemaster of
1258 /etc/rc.d is highly recommended. Simultaneous update of userland
1259 and kernel wouldn't hurt either.
1261 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1262 modules were merged into the base wlan module. All references
1263 to these modules (e.g. in kernel config files) must be removed.
1266 psm(4) has gained write(2) support in native operation level.
1267 Arbitrary commands can be written to /dev/psm%d and status can
1268 be read back from it. Therefore, an application is responsible
1269 for status validation and error recovery. It is a no-op in
1270 other operation levels.
1273 Support for KSE threading has been removed from the kernel. To
1274 run legacy applications linked against KSE libmap.conf may
1275 be used. The following libmap.conf may be used to ensure
1276 compatibility with any prior release:
1278 libpthread.so.1 libthr.so.1
1279 libpthread.so.2 libthr.so.2
1280 libkse.so.3 libthr.so.3
1283 The layout of struct vmspace has changed. This affects libkvm
1284 and any executables that link against libkvm and use the
1285 kvm_getprocs() function. In particular, but not exclusively,
1286 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1287 The effects are minimal, but it's advisable to upgrade world
1291 The latest em driver no longer has support in it for the
1292 82575 adapter, this is now moved to the igb driver. The
1293 split was done to make new features that are incompatible
1294 with older hardware easier to do.
1297 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1298 likewise the kernel option is now GEOM_LINUX_LVM.
1301 The default NFS mount mode has changed from UDP to TCP for
1302 increased reliability. If you rely on (insecurely) NFS
1303 mounting across a firewall you may need to update your
1307 Belatedly note the addition of m_collapse for compacting
1311 The fts(3) structures have been changed to use adequate
1312 integer types for their members and so to be able to cope
1313 with huge file trees. The old fts(3) ABI is preserved
1314 through symbol versioning in libc, so third-party binaries
1315 using fts(3) should still work, although they will not take
1316 advantage of the extended types. At the same time, some
1317 third-party software might fail to build after this change
1318 due to unportable assumptions made in its source code about
1319 fts(3) structure members. Such software should be fixed
1320 by its vendor or, in the worst case, in the ports tree.
1321 FreeBSD_version 800015 marks this change for the unlikely
1322 case that a portable fix is impossible.
1325 To upgrade to -current after this date, you must be running
1326 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1327 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1330 The ADAPTIVE_GIANT kernel option has been retired because its
1331 functionality is the default now.
1334 The AT keyboard emulation of sunkbd(4) has been turned on
1335 by default. In order to make the special symbols of the Sun
1336 keyboards driven by sunkbd(4) work under X these now have
1337 to be configured the same way as Sun USB keyboards driven
1338 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1340 Option "XkbLayout" "us"
1341 Option "XkbRules" "xorg"
1342 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1345 It has been decided that it is desirable to provide ABI
1346 backwards compatibility to the FreeBSD 4/5/6 versions of the
1347 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1348 broken with the introduction of PCI domain support (see the
1349 20070930 entry). Unfortunately, this required the ABI of
1350 PCIOCGETCONF to be broken again in order to be able to
1351 provide backwards compatibility to the old version of that
1352 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1353 again. As for prominent ports this affects neither pciutils
1354 nor xorg-server this time, the hal port needs to be rebuilt
1358 The misnamed kthread_create() and friends have been renamed
1359 to kproc_create() etc. Many of the callers already
1360 used kproc_start()..
1361 I will return kthread_create() and friends in a while
1362 with implementations that actually create threads, not procs.
1363 Renaming corresponds with version 800002.
1372 Avoid using make -j when upgrading. While generally safe, there are
1373 sometimes problems using -j to upgrade. If your upgrade fails with
1374 -j, please try again without -j. From time to time in the past there
1375 have been problems using -j with buildworld and/or installworld. This
1376 is especially true when upgrading between "distant" versions (eg one
1377 that cross a major release boundary or several minor releases, or when
1378 several months have passed on the -current branch).
1380 Sometimes, obscure build problems are the result of environment
1381 poisoning. This can happen because the make utility reads its
1382 environment when searching for values for global variables. To run
1383 your build attempts in an "environmental clean room", prefix all make
1384 commands with 'env -i '. See the env(1) manual page for more details.
1386 When upgrading from one major version to another it is generally best
1387 to upgrade to the latest code in the currently installed branch first,
1388 then do an upgrade to the new branch. This is the best-tested upgrade
1389 path, and has the highest probability of being successful. Please try
1390 this approach before reporting problems with a major version upgrade.
1394 When upgrading the boot ZFS pool to a new version, always follow
1397 1.) recompile and reinstall the ZFS boot loader and boot block
1398 (this is part of "make buildworld" and "make installworld")
1400 2.) update the ZFS boot block on your boot drive
1402 The following example updates the ZFS boot block on the first
1403 partition (freebsd-boot) of a GPT partitioned drive ad0:
1404 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1406 Non-boot pools do not need these updates.
1410 If you are updating from a prior version of FreeBSD (even one just
1411 a few days old), you should follow this procedure. It is the most
1412 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1414 make kernel-toolchain
1415 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1416 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1418 To test a kernel once
1419 ---------------------
1420 If you just want to boot a kernel once (because you are not sure
1421 if it works, or if you want to boot a known bad kernel to provide
1422 debugging information) run
1423 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1424 nextboot -k testkernel
1426 To just build a kernel when you know that it won't mess you up
1427 --------------------------------------------------------------
1428 This assumes you are already running a CURRENT system. Replace
1429 ${arch} with the architecture of your machine (e.g. "i386",
1430 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1432 cd src/sys/${arch}/conf
1433 config KERNEL_NAME_HERE
1434 cd ../compile/KERNEL_NAME_HERE
1439 If this fails, go to the "To build a kernel" section.
1441 To rebuild everything and install it on the current system.
1442 -----------------------------------------------------------
1443 # Note: sometimes if you are running current you gotta do more than
1444 # is listed here if you are upgrading from a really old current.
1446 <make sure you have good level 0 dumps>
1448 make kernel KERNCONF=YOUR_KERNEL_HERE
1450 <reboot in single user> [3]
1458 To cross-install current onto a separate partition
1459 --------------------------------------------------
1460 # In this approach we use a separate partition to hold
1461 # current's root, 'usr', and 'var' directories. A partition
1462 # holding "/", "/usr" and "/var" should be about 2GB in
1465 <make sure you have good level 0 dumps>
1468 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1469 <maybe newfs current's root partition>
1470 <mount current's root partition on directory ${CURRENT_ROOT}>
1471 make installworld DESTDIR=${CURRENT_ROOT}
1472 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1473 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1474 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1475 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1476 <reboot into current>
1477 <do a "native" rebuild/install as described in the previous section>
1478 <maybe install compatibility libraries from ports/misc/compat*>
1482 To upgrade in-place from 8.x-stable to current
1483 ----------------------------------------------
1484 <make sure you have good level 0 dumps>
1486 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1488 <reboot in single user> [3]
1495 Make sure that you've read the UPDATING file to understand the
1496 tweaks to various things you need. At this point in the life
1497 cycle of current, things change often and you are on your own
1498 to cope. The defaults can also change, so please read ALL of
1499 the UPDATING entries.
1501 Also, if you are tracking -current, you must be subscribed to
1502 freebsd-current@freebsd.org. Make sure that before you update
1503 your sources that you have read and understood all the recent
1504 messages there. If in doubt, please track -stable which has
1505 much fewer pitfalls.
1507 [1] If you have third party modules, such as vmware, you
1508 should disable them at this point so they don't crash your
1511 [3] From the bootblocks, boot -s, and then do
1516 adjkerntz -i # if CMOS is wall time
1517 Also, when doing a major release upgrade, it is required that
1518 you boot into single user mode to do the installworld.
1520 [4] Note: This step is non-optional. Failure to do this step
1521 can result in a significant reduction in the functionality of the
1522 system. Attempting to do it by hand is not recommended and those
1523 that pursue this avenue should read this file carefully, as well
1524 as the archives of freebsd-current and freebsd-hackers mailing lists
1525 for potential gotchas. The -U option is also useful to consider.
1526 See mergemaster(8) for more information.
1528 [5] Usually this step is a noop. However, from time to time
1529 you may need to do this if you get unknown user in the following
1530 step. It never hurts to do it all the time. You may need to
1531 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1532 install) after the buildworld before this step if you last updated
1533 from current before 20020224 or from -stable before 20020408.
1535 [6] This only deletes old files and directories. Old libraries
1536 can be deleted by "make delete-old-libs", but you have to make
1537 sure that no program is using those libraries anymore.
1539 [8] In order to have a kernel that can run the 4.x binaries needed to
1540 do an installworld, you must include the COMPAT_FREEBSD4 option in
1541 your kernel. Failure to do so may leave you with a system that is
1542 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1543 required to run the 5.x binaries on more recent kernels. And so on
1544 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1546 Make sure that you merge any new devices from GENERIC since the
1547 last time you updated your kernel config file.
1549 [9] When checking out sources, you must include the -P flag to have
1550 cvs prune empty directories.
1552 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1553 "?=" instead of the "=" assignment operator, so that buildworld can
1554 override the CPUTYPE if it needs to.
1556 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1557 not on the command line, or in /etc/make.conf. buildworld will
1558 warn if it is improperly defined.
1561 This file contains a list, in reverse chronological order, of major
1562 breakages in tracking -current. It is not guaranteed to be a complete
1563 list of such breakages, and only contains entries since October 10, 2007.
1564 If you need to see UPDATING entries from before that date, you will need
1565 to fetch an UPDATING file from an older FreeBSD release.
1567 Copyright information:
1569 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1571 Redistribution, publication, translation and use, with or without
1572 modification, in full or in part, in any form or format of this
1573 document are permitted without further permission from the author.
1575 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1576 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1577 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1578 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1579 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1580 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1581 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1582 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1583 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1584 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1585 POSSIBILITY OF SUCH DAMAGE.
1587 Contact Warner Losh if you have any questions about your use of