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 sparc64 ZFS loader has been changed to no longer try to auto-
29 detect ZFS providers based on diskN aliases but now requires these
30 to be explicitly listed in the OFW boot-device environment variable.
33 The OpenSSL has been upgraded to 1.0.1c. Any binaries requiring
34 libcrypto.so.6 or libssl.so.6 must be recompiled. Also, there are
35 configuration changes. Make sure to merge /etc/ssl/openssl.cnf.
38 The following sysctls and tunables have been renamed for consistency
40 kern.cam.da.da_send_ordered -> kern.cam.da.send_ordered
41 kern.cam.ada.ada_send_ordered -> kern.cam.ada.send_ordered
44 The sort utility has been replaced with BSD sort. For now, GNU sort
45 is also available as "gnusort" or the default can be set back to
46 GNU sort by setting WITH_GNU_SORT. In this case, BSD sort will be
47 installed as "bsdsort".
50 A new version of ZFS (pool version 5000) has been merged to -HEAD.
51 Starting with this version the old system of ZFS pool versioning
52 is superseded by "feature flags". This concept enables forward
53 compatibility against certain future changes in functionality of ZFS
54 pools. The first read-only compatible "feature flag" for ZFS pools
55 is named "com.delphix:async_destroy". For more information
56 read the new zpool-features(5) manual page.
57 Please refer to the "ZFS notes" section of this file for information
58 on upgrading boot ZFS pools.
61 The malloc(3) implementation embedded in libc now uses sources imported
62 as contrib/jemalloc. The most disruptive API change is to
63 /etc/malloc.conf. If your system has an old-style /etc/malloc.conf,
64 delete it prior to installworld, and optionally re-create it using the
65 new format after rebooting. See malloc.conf(5) for details
66 (specifically the TUNING section and the "opt.*" entries in the MALLCTL
70 Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb
71 is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is
72 now spelled mips. This is to aid compatibility with third-party
73 software that expects this naming scheme in uname(3). Little-endian
74 settings are unchanged.
77 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
81 Now unix domain sockets behave "as expected" on nullfs(5). Previously
82 nullfs(5) did not pass through all behaviours to the underlying layer,
83 as a result if we bound to a socket on the lower layer we could connect
84 only to the lower path; if we bound to the upper layer we could connect
85 only to the upper path. The new behavior is one can connect to both the
86 lower and the upper paths regardless what layer path one binds to.
89 The getifaddrs upgrade path broken with 20111215 has been restored.
90 If you have upgraded in between 20111215 and 20120209 you need to
91 recompile libc again with your kernel. You still need to recompile
92 world to be able to configure CARP but this restriction already
96 The set_rcvar() function has been removed from /etc/rc.subr. All
97 base and ports rc.d scripts have been updated, so if you have a
98 port installed with a script in /usr/local/etc/rc.d you can either
99 hand-edit the rcvar= line, or reinstall the port.
101 An easy way to handle the mass-update of /etc/rc.d:
102 rm /etc/rc.d/* && mergemaster -i
105 panic(9) now stops other CPUs in the SMP systems, disables interrupts
106 on the current CPU and prevents other threads from running.
107 This behavior can be reverted using the kern.stop_scheduler_on_panic
109 The new behavior can be incompatible with kern.sync_on_panic.
112 The carp(4) facility has been changed significantly. Configuration
113 of the CARP protocol via ifconfig(8) has changed, as well as format
114 of CARP events submitted to devd(8) has changed. See manual pages
115 for more information. The arpbalance feature of carp(4) is currently
116 not supported anymore.
118 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
119 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
120 need to be recompiled.
123 The acpi_wmi(4) status device /dev/wmistat has been renamed to
127 The option VFS_ALLOW_NONMPSAFE option has been added in order to
128 explicitely support non-MPSAFE filesystems.
129 It is on by default for all supported platform at this present
133 The broken amd(4) driver has been replaced with esp(4) in the amd64,
134 i386 and pc98 GENERIC kernel configuration files.
137 sysinstall has been removed
140 The stable/9 branch created in subversion. This corresponds to the
141 RELENG_9 branch in CVS.
144 This commit modifies vfs_register() so that it uses a hash
145 calculation to set vfc_typenum, which is enabled by default.
146 The first time a system is booted after this change, the
147 vfc_typenum values will change for all file systems. The
148 main effect of this is a change to the NFS server file handles
149 for file systems that use vfc_typenum in their fsid, such as ZFS.
150 It will, however, prevent vfc_typenum from changing when file
151 systems are loaded in a different order for subsequent reboots.
152 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
153 until you are ready to remount all NFS clients after a reboot.
156 Bump the shared library version numbers for libraries that
157 do not use symbol versioning, have changed the ABI compared
158 to stable/8 and which shared library version was not bumped.
159 Done as part of 9.0-RELEASE cycle.
162 During the merge of Capsicum features, the fget(9) KPI was modified.
163 This may require the rebuilding of out-of-tree device drivers --
164 issues have been reported specifically with the nVidia device driver.
165 __FreeBSD_version is bumped to 900041.
167 Also, there is a period between 20110811 and 20110814 where the
168 special devices /dev/{stdin,stdout,stderr} did not work correctly.
169 Building world from a kernel during that window may not work.
172 The packet filter (pf) code has been updated to OpenBSD 4.5.
173 You need to update userland tools to be in sync with kernel.
174 This update breaks backward compatibility with earlier pfsync(4)
175 versions. Care must be taken when updating redundant firewall setups.
178 The following sysctls and tunables are retired on x86 platforms:
180 machdep.hlt_logical_cpus
181 The following sysctl is retired:
182 machdep.hyperthreading_allowed
183 The sysctls were supposed to provide a way to dynamically offline and
184 online selected CPUs on x86 platforms, but the implementation has not
185 been reliable especially with SCHED_ULE scheduler.
186 machdep.hyperthreading_allowed tunable is still available to ignore
187 hyperthreading CPUs at OS level.
188 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
189 where X is an APIC ID of a CPU. Be advised, though, that disabling
190 CPUs in non-uniform fashion will result in non-uniform topology and
191 may lead to sub-optimal system performance with SCHED_ULE, which is
195 cpumask_t type is retired and cpuset_t is used in order to describe
199 Changes to ifconfig(8) for dynamic address family detection mandate
200 that you are running a kernel of 20110525 or later. Make sure to
201 follow the update procedure to boot a new kernel before installing
205 Support for sun4v architecture is officially dropped
208 Several KPI breaking changes have been committed to the mii(4) layer,
209 the PHY drivers and consequently some Ethernet drivers using mii(4).
210 This means that miibus.ko and the modules of the affected Ethernet
211 drivers need to be recompiled.
213 Note to kernel developers: Given that the OUI bit reversion problem
214 was fixed as part of these changes all mii(4) commits related to OUIs,
215 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
216 handling, no longer can be merged verbatim to stable/8 and previous
220 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
221 into their kernel configurations along with 'device pci'.
224 The default NFS client is now the new NFS client, so fstype "newnfs"
225 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
226 Although mounts via fstype "nfs" will usually work without userland
227 changes, it is recommended that the mount(8) and mount_nfs(8)
228 commands be rebuilt from sources and that a link to mount_nfs called
229 mount_oldnfs be created. The new client is compiled into the
230 kernel with "options NFSCL" and this is needed for diskless root
231 file systems. The GENERIC kernel configs have been changed to use
232 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
233 To use the regular/old client, you can "mount -t oldnfs ...". For
234 a diskless root file system, you must also include a line like:
236 vfs.root.mountfrom="oldnfs:"
238 in the boot/loader.conf on the root fs on the NFS server to make
239 a diskless root fs use the old client.
242 The GENERIC kernels for all architectures now default to the new
243 CAM-based ATA stack. It means that all legacy ATA drivers were
244 removed and replaced by respective CAM drivers. If you are using
245 ATA device names in /etc/fstab or other places, make sure to update
246 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
247 where 'Y's are the sequential numbers starting from zero for each type
248 in order of detection, unless configured otherwise with tunables,
249 see cam(4)). There will be symbolic links created in /dev/ to map
250 old adX devices to the respective adaY. They should provide basic
251 compatibility for file systems mounting in most cases, but they do
252 not support old user-level APIs and do not have respective providers
253 in GEOM. Consider using updated management tools with new device names.
255 It is possible to load devices ahci, ata, siis and mvs as modules,
256 but option ATA_CAM should remain in kernel configuration to make ata
257 module work as CAM driver supporting legacy ATA controllers. Device ata
258 still can be used in modular fashion (atacore + ...). Modules atadisk
259 and atapi* are not used and won't affect operation in ATA_CAM mode.
260 Note that to use CAM-based ATA kernel should include CAM devices
261 scbus, pass, da (or explicitly ada), cd and optionally others. All of
262 them are parts of the cam module.
264 ataraid(4) functionality is now supported by the RAID GEOM class.
265 To use it you can load geom_raid kernel module and use graid(8) tool
266 for management. Instead of /dev/arX device names, use /dev/raid/rX.
268 No kernel config options or code have been removed, so if a problem
269 arises, please report it and optionally revert to the old ATA stack.
270 In order to do it you can remove from the kernel config:
275 , and instead add back:
276 device atadisk # ATA disk drives
277 device ataraid # ATA RAID drives
278 device atapicd # ATAPI CDROM drives
279 device atapifd # ATAPI floppy drives
280 device atapist # ATAPI tape drives
283 The default NFS server has been changed to the new server, which
284 was referred to as the experimental server. If you need to switch
285 back to the old NFS server, you must now put the "-o" option on
286 both the mountd and nfsd commands. This can be done using the
287 mountd_flags and nfs_server_flags rc.conf variables until an
288 update to the rc scripts is committed, which is coming soon.
291 The GNU Objective-C runtime library (libobjc), and other Objective-C
292 related components have been removed from the base system. If you
293 require an Objective-C library, please use one of the available ports.
296 ath(4) has been split into bus- and device- modules. if_ath contains
297 the HAL, the TX rate control and the network device code. if_ath_pci
298 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
299 contains the AHB glue. Users need to load both if_ath_pci and if_ath
300 in order to use ath on everything else.
302 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
303 need to load if_ath and if_ath_pci for ath(4) operation.
306 As part of the replacement of sysinstall, the process of building
307 release media has changed significantly. For details, please re-read
308 release(7), which has been updated to reflect the new build process.
311 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
312 is the last available version under GPLv2. It brings a number of new
313 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
314 4.1 and SSE 4.2), better support for powerpc64, a number of new
315 directives, and lots of other small improvements. See the ChangeLog
316 file in contrib/binutils for the full details.
319 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
320 compliant, and will now use half of hash for authentication.
321 This will break interoperability with all stacks (including all
322 actual FreeBSD versions) who implement
323 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
325 The only workaround with such peers is to use another HMAC
326 algorithm for IPsec ("phase 2") authentication.
329 Remove the uio_yield prototype and symbol. This function has
330 been misnamed since it was introduced and should not be
331 globally exposed with this name. The equivalent functionality
332 is now available using kern_yield(curthread->td_user_pri).
333 The function remains undocumented.
336 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
337 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
338 for scalar sysctls is defined but disabled. Code that needs
339 UQUAD to pass the type checking that must compile on older
340 systems where the define is not present can check against
341 __FreeBSD_version >= 900030.
343 The system dialog(1) has been replaced with a new version previously
344 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
345 with the previous version, but the libdialog associated with it has
346 a largely incompatible API. As such, the original version of libdialog
347 will be kept temporarily as libodialog, until its base system consumers
348 are replaced or updated. Bump __FreeBSD_version to 900030.
351 If you are trying to run make universe on a -stable system, and you get
352 the following warning:
353 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
354 unknown. config(8) likely too old."
355 or something similar to it, then you must upgrade your -stable system
356 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
357 stable/8) or build the config from the latest stable/8 branch and
358 install it on your system.
360 Prior to this date, building a current universe on 8-stable system from
361 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
362 error in the first kernel build phase. A new config on those old
363 systems will fix that problem for older versions of -current.
366 The TCP stack has been modified to allow Khelp modules to interact with
367 it via helper hook points and store per-connection data in the TCP
368 control block. Bump __FreeBSD_version to 900029. User space tools that
369 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
373 Generic IEEE 802.3 annex 31B full duplex flow control support has been
374 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
375 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
376 to take advantage of it instead of using custom implementations. This
377 means that these drivers now no longer unconditionally advertise
378 support for flow control but only do so if flow control is a selected
379 media option. This was implemented in the generic support that way in
380 order to allow flow control to be switched on and off via ifconfig(8)
381 with the PHY specific default to typically off in order to protect
382 from unwanted effects. Consequently, if you used flow control with
383 one of the above mentioned drivers you now need to explicitly enable
385 ifconfig bge0 media auto mediaopt flowcontrol
387 Along with the above mentioned changes generic support for setting
388 1000baseT master mode also has been added and brgphy(4), ciphy(4),
389 e1000phy(4) as well as ip1000phy(4) have been converted to take
390 advantage of it. This means that these drivers now no longer take the
391 link0 parameter for selecting master mode but the master media option
392 has to be used instead, for example like in the following:
393 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
395 Selection of master mode now is also available with all other PHY
396 drivers supporting 1000baseT.
399 The TCP stack has received a significant update to add support for
400 modularised congestion control and generally improve the clarity of
401 congestion control decisions. Bump __FreeBSD_version to 900025. User
402 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
403 sockstat) need to be recompiled.
406 The man(1) utility has been replaced by a new version that no longer
407 uses /etc/manpath.config. Please consult man.conf(5) for how to
408 migrate local entries to the new format.
411 The copyright strings printed by login(1) and sshd(8) at the time of a
412 new connection have been removed to follow other operating systems and
416 A workaround for a fixed ld bug has been removed in kernel code,
417 so make sure that your system ld is built from sources after
418 revision 210245 from 2010-07-19 (r211583 if building head kernel
419 on stable/8, r211584 for stable/7; both from 2010-08-21).
420 A symptom of incorrect ld version is different addresses for
421 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
424 The $ipv6_prefer variable in rc.conf(5) has been split into
425 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
427 The $ip6addrctl_policy is a variable to choose a pre-defined
428 address selection policy set by ip6addrctl(8). A value
429 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
432 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
433 flag (see an entry of 20090926) is set on an interface with no
434 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
435 security reason. If you want IPv6 link-local address on all
436 interfaces by default, set this to "YES".
438 The old ipv6_prefer="YES" is equivalent to
439 ipv6_activate_all_interfaces="YES" and
440 ip6addrctl_policy="ipv6_prefer".
443 DTrace has grown support for userland tracing. Due to this, DTrace is
444 now i386 and amd64 only.
445 dtruss(1) is now installed by default on those systems and a new
446 kernel module is needed for userland tracing: fasttrap.
447 No changes to your kernel config file are necessary to enable
448 userland tracing, but you might consider adding 'STRIP=' and
449 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
450 to have informative userland stack traces in DTrace (ustack).
453 The acpi_aiboost(4) driver has been removed in favor of the new
454 aibs(4) driver. You should update your kernel configuration file.
457 BSD grep has been imported to the base system and it is built by
458 default. It is completely BSD licensed, highly GNU-compatible, uses
459 less memory than its GNU counterpart and has a small codebase.
460 However, it is slower than its GNU counterpart, which is mostly
461 noticeable for larger searches, for smaller ones it is measurable
462 but not significant. The reason is complex, the most important factor
463 is that we lack a modern and efficient regex library and GNU
464 overcomes this by optimizing the searches internally. Future work
465 on improving the regex performance is planned, for the meantime,
466 users that need better performance, can build GNU grep instead by
467 setting the WITH_GNU_GREP knob.
470 Due to the import of powerpc64 support, all existing powerpc kernel
471 configuration files must be updated with a machine directive like this:
472 machine powerpc powerpc
474 In addition, an updated config(8) is required to build powerpc kernels
478 A new version of ZFS (version 15) has been merged to -HEAD.
479 This version uses a python library for the following subcommands:
480 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
481 For full functionality of these commands the following port must
482 be installed: sysutils/py-zfs
485 'vm_page's are now hashed by physical address to an array of mutexes.
486 Currently this is only used to serialize access to hold_count. Over
487 time the page queue mutex will be peeled away. This changes the size
488 of pmap on every architecture. And requires all callers of vm_page_hold
489 and vm_page_unhold to be updated.
492 WITH_CTF can now be specified in src.conf (not recommended, there
493 are some problems with static executables), make.conf (would also
494 affect ports which do not use GNU make and do not override the
495 compile targets) or in the kernel config (via "makeoptions
497 When WITH_CTF was specified there before this was silently ignored,
498 so make sure that WITH_CTF is not used in places which could lead
499 to unwanted behavior.
502 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
503 to allow 32-bit compatibility on non-x86 platforms. All kernel
504 configurations on amd64 and ia64 platforms using these options must
505 be modified accordingly.
508 The utmp user accounting database has been replaced with utmpx,
509 the user accounting interface standardized by POSIX.
510 Unfortunately the semantics of utmp and utmpx don't match,
511 making it practically impossible to support both interfaces.
512 The user accounting database is used by tools like finger(1),
513 last(1), talk(1), w(1) and ac(8).
515 All applications in the base system use utmpx. This means only
516 local binaries (e.g. from the ports tree) may still use these
517 utmp database files. These applications must be rebuilt to make
520 After the system has been upgraded, it is safe to remove the old
521 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
522 assuming their contents is of no importance anymore. Old wtmp
523 databases can only be used by last(1) and ac(8) after they have
524 been converted to the new format using wtmpcvt(1).
527 Introduce the kernel thread "deadlock resolver" (which can be enabled
528 via the DEADLKRES option, see NOTES for more details) and the
529 sleepq_type() function for sleepqueues.
532 The rc.firewall and rc.firewall6 were unified, and
533 rc.firewall6 and rc.d/ip6fw were removed.
534 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
535 variables are obsoleted. Instead, the following new rc
536 variables are added to rc.d/ipfw:
538 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
539 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
540 firewall_simple_onet_ipv6, firewall_trusted_ipv6
542 The meanings correspond to the relevant IPv4 variables.
548 The default terminal emulation for syscons(4) has been changed
549 from cons25 to xterm on all platforms except pc98. This means
550 that the /etc/ttys file needs to be updated to ensure correct
551 operation of applications on the console.
553 The terminal emulation style can be toggled per window by using
554 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
555 options can be used to change the compile-time default back to
558 To prevent graphical artifacts, make sure the TERM environment
559 variable is set to match the terminal emulation that is being
560 performed by syscons(4).
563 The layout of the structure ieee80211req_scan_result has changed.
564 Applications that require wireless scan results (e.g. ifconfig(8))
565 from net80211 need to be recompiled.
567 Applications such as wpa_supplicant(8) may require a full world
568 build without using NO_CLEAN in order to get synchronized with the
572 The iwn(4) driver has been updated to support the 5000 and 5150 series.
573 There's one kernel module for each firmware. Adding "device iwnfw"
574 to the kernel configuration file means including all three firmware
575 images inside the kernel. If you want to include just the one for
576 your wireless card, use the devices iwn4965fw, iwn5000fw or
580 The rc.d/network_ipv6, IPv6 configuration script has been integrated
581 into rc.d/netif. The changes are the following:
583 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
584 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
585 Note that both variables need the "inet6" keyword at the head.
587 Do not set $ipv6_network_interfaces manually if you do not
588 understand what you are doing. It is not needed in most cases.
590 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
593 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
594 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
596 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
597 all configured interfaces have "inet6 accept_rtadv" in the
598 $ifconfig_IF_ipv6. These are for backward compatibility.
600 3. A new variable $ipv6_prefer has been added. If NO, IPv6
601 functionality of interfaces with no corresponding
602 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
603 and the default address selection policy of ip6addrctl(8)
604 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
605 Note that if you want to configure IPv6 functionality on the
606 disabled interfaces after boot, first you need to clear the flag by
607 using ifconfig(8) like:
609 ifconfig em0 inet6 -ifdisabled
611 If YES, the default address selection policy is set as
614 The default value of $ipv6_prefer is NO.
616 4. If your system need to receive Router Advertisement messages,
617 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
618 scripts automatically invoke rtsol(8) when the interface becomes
619 UP. The Router Advertisement messages are used for SLAAC
620 (State-Less Address AutoConfiguration).
623 802.11s D3.03 support was committed. This is incompatible with the
624 previous code, which was based on D3.0.
627 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
628 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
629 control whether accepting Router Advertisement messages or not.
630 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
631 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
632 The ifconfig(8) utility now supports these flags.
635 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
636 mount(8) and -a option for df(1) to see them.
639 The old tunable hw.bus.devctl_disable has been superseded by
640 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
641 replaced by hw.bus.devctl_queue=0. The default for this new tunable
645 Remove the option STOP_NMI. The default action is now to use NMI only
646 for KDB via the newly introduced function stop_cpus_hard() and
647 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
650 The stable/8 branch created in subversion. This corresponds to the
651 RELENG_8 branch in CVS.
654 Bump the shared library version numbers for all libraries that do not
655 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
656 __FreeBSD_version to 800105.
659 Due to changes in the implementation of virtual network stack support,
660 all network-related kernel modules must be recompiled. As this change
661 breaks the ABI, bump __FreeBSD_version to 800104.
664 The TOE interface to the TCP syncache has been modified to remove
665 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
666 The cxgb driver is the only TOE consumer affected by this change, and
667 needs to be recompiled along with the kernel. As this change breaks
668 the ABI, bump __FreeBSD_version to 800103.
671 Padding has been added to struct tcpcb, sackhint and tcpstat in
672 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
673 maintaining the ABI. However, this change breaks the ABI, so bump
674 __FreeBSD_version to 800102. User space tools that rely on the size of
675 any of these structs (e.g. sockstat) need to be recompiled.
678 The NFS_LEGACYRPC option has been removed along with the old kernel
679 RPC implementation that this option selected. Kernel configurations
680 may need to be adjusted.
683 The network interface device nodes at /dev/net/<interface> have been
684 removed. All ioctl operations can be performed the normal way using
685 routing sockets. The kqueue functionality can generally be replaced
686 with routing sockets.
689 The documentation from the FreeBSD Documentation Project (Handbook,
690 FAQ, etc.) is now installed via packages by sysinstall(8) and under
691 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
694 The ABI of various structures related to the SYSV IPC API have been
695 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
696 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
700 Layout of struct vnet has changed as routing related variables were
701 moved to their own Vimage module. Modules need to be recompiled. Bump
702 __FreeBSD_version to 800099.
705 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
706 respectively. As long as no more than 16 groups per process are used,
707 no changes should be visible. When more than 16 groups are used, old
708 binaries may fail if they call getgroups() or getgrouplist() with
709 statically sized storage. Recompiling will work around this, but
710 applications should be modified to use dynamically allocated storage
711 for group arrays as POSIX.1-2008 does not cap an implementation's
712 number of supported groups at NGROUPS_MAX+1 as previous versions did.
714 NFS and portalfs mounts may also be affected as the list of groups is
715 truncated to 16. Users of NFS who use more than 16 groups, should
716 take care that negative group permissions are not used on the exported
717 file systems as they will not be reliable unless a GSSAPI based
718 authentication method is used.
721 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
722 option compiles in the support for adaptive spinning for lockmgrs
723 which want to enable it. The lockinit() function now accepts the flag
724 LK_ADAPTIVE in order to make the lock object subject to adaptive
725 spinning when both held in write and read mode.
728 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
729 changed. User applications that use this ioctl need to be rebuilt.
732 The layout of struct thread has changed. Kernel and modules need to
736 The layout of structs ifnet, domain, protosw and vnet_net has changed.
737 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
740 window(1) has been removed from the base system. It can now be
741 installed from ports. The port is called misc/window.
744 The way we are storing and accessing `routing table' entries has
745 changed. Programs reading the FIB, like netstat, need to be
749 A new netisr implementation has been added for FreeBSD 8. Network
750 file system modules, such as igmp, ipdivert, and others, should be
752 Bump __FreeBSD_version to 800096.
755 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
759 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
760 Bump __FreeBSD_version to 800094.
763 Add mnt_xflag field to 'struct mount'. File system modules need to be
765 Bump __FreeBSD_version to 800093.
768 The compiling option ADAPTIVE_SX has been retired while it has been
769 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
770 The KPI for sx_init_flags() changes as accepting flags:
771 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
772 been introduced in order to handle the reversed logic.
773 Bump __FreeBSD_version to 800092.
776 Add support for hierarchical jails. Remove global securelevel.
777 Bump __FreeBSD_version to 800091.
780 The layout of struct vnet_net has changed, therefore modules
782 Bump __FreeBSD_version to 800090.
785 The newly imported zic(8) produces a new format in the output. Please
786 run tzsetup(8) to install the newly created data to /etc/localtime.
789 The sysctl tree for the usb stack has renamed from hw.usb2.* to
790 hw.usb.* and is now consistent again with previous releases.
793 802.11 monitor mode support was revised and driver api's were changed.
794 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
795 of DLT_IEEE802_11. No user-visible data structures were changed but
796 applications that use DLT_IEEE802_11 may require changes.
797 Bump __FreeBSD_version to 800088.
800 The layout of the following structs has changed: sysctl_oid,
801 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
802 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
803 panics may be experienced. World rebuild is required for
804 correctly checking networking state from userland.
805 Bump __FreeBSD_version to 800085.
808 MLDv2 and Source-Specific Multicast (SSM) have been merged
809 to the IPv6 stack. VIMAGE hooks are in but not yet used.
810 The implementation of SSM within FreeBSD's IPv6 stack closely
811 follows the IPv4 implementation.
813 For kernel developers:
815 * The most important changes are that the ip6_output() and
816 ip6_input() paths no longer take the IN6_MULTI_LOCK,
817 and this lock has been downgraded to a non-recursive mutex.
819 * As with the changes to the IPv4 stack to support SSM, filtering
820 of inbound multicast traffic must now be performed by transport
821 protocols within the IPv6 stack. This does not apply to TCP and
822 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
824 * The KPIs used by IPv6 multicast are similar to those used by
825 the IPv4 stack, with the following differences:
826 * im6o_mc_filter() is analogous to imo_multicast_filter().
827 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
828 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
829 * IN6_LOOKUP_MULTI() has been deprecated and removed.
830 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
831 for MLDv1 have an additional 'timer' argument which is used to
832 jitter the initial membership report for the solicited-node
833 multicast membership on-link.
834 * This is not strictly needed for MLDv2, which already jitters
835 its report transmissions. However, the 'timer' argument is
836 preserved in case MLDv1 is active on the interface.
838 * The KAME linked-list based IPv6 membership implementation has
839 been refactored to use a vector similar to that used by the IPv4
841 Code which maintains a list of its own multicast memberships
842 internally, e.g. carp, has been updated to reflect the new
845 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
846 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
847 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
848 implementation constraint which needs to be addressed in HEAD.
850 For application developers:
852 * The changes are broadly similar to those made for the IPv4
855 * The use of IPv4 and IPv6 multicast socket options on the same
856 socket, using mapped addresses, HAS NOT been tested or supported.
858 * There are a number of issues with the implementation of various
859 IPv6 multicast APIs which need to be resolved in the API surface
860 before the implementation is fully compatible with KAME userland
861 use, and these are mostly to do with interface index treatment.
863 * The literature available discusses the use of either the delta / ASM
864 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
865 using setsourcefilter(3)/getsourcefilter(3). For more information
866 please refer to RFC 3768, 'Socket Interface Extensions for
867 Multicast Source Filters'.
869 * Applications which use the published RFC 3678 APIs should be fine.
871 For systems administrators:
873 * The mtest(8) utility has been refactored to support IPv6, in
874 addition to IPv4. Interface addresses are no longer accepted
875 as arguments, their names must be used instead. The utility
876 will map the interface name to its first IPv4 address as
877 returned by getifaddrs(3).
879 * The ifmcstat(8) utility has also been updated to print the MLDv2
880 endpoint state and source filter lists via sysctl(3).
882 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
883 loopback of IPv6 multicast datagrams by default; it defaults to 1
884 to preserve the existing behaviour. Disabling multicast loopback is
885 recommended for optimal system performance.
887 * The IPv6 MROUTING code has been changed to examine this sysctl
888 instead of attempting to perform a group lookup before looping
889 back forwarded datagrams.
891 Bump __FreeBSD_version to 800084.
894 Implement low-level Bluetooth HCI API.
895 Bump __FreeBSD_version to 800083.
898 The layout of struct malloc_type, used by modules to register new
899 memory allocation types, has changed. Most modules will need to
900 be rebuilt or panics may be experienced.
901 Bump __FreeBSD_version to 800081.
904 Anticipate overflowing inp_flags - add inp_flags2.
905 This changes most offsets in inpcb, so checking v4 connection
906 state will require a world rebuild.
907 Bump __FreeBSD_version to 800080.
910 Add an llentry to struct route and struct route_in6. Modules
911 embedding a struct route will need to be recompiled.
912 Bump __FreeBSD_version to 800079.
915 The size of rt_metrics_lite and by extension rtentry has changed.
916 Networking administration apps will need to be recompiled.
917 The route command now supports show as an alias for get, weighting
918 of routes, sticky and nostick flags to alter the behavior of stateful
920 Bump __FreeBSD_version to 800078.
923 Do not use Giant for kbdmux(4) locking. This is wrong and
924 apparently causing more problems than it solves. This will
925 re-open the issue where interrupt handlers may race with
926 kbdmux(4) in polling mode. Typical symptoms include (but
927 not limited to) duplicated and/or missing characters when
928 low level console functions (such as gets) are used while
929 interrupts are enabled (for example geli password prompt,
930 mountroot prompt etc.). Disabling kbdmux(4) may help.
933 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
934 kernel modules referencing any of the above need to be recompiled.
935 Bump __FreeBSD_version to 800075.
938 GEOM_PART has become the default partition slicer for storage devices,
939 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
940 introduces some changes:
942 MSDOS/EBR: the devices created from MSDOS extended partition entries
943 (EBR) can be named differently than with GEOM_MBR and are now symlinks
944 to devices with offset-based names. fstabs may need to be modified.
946 BSD: the "geometry does not match label" warning is harmless in most
947 cases but it points to problems in file system misalignment with
948 disk geometry. The "c" partition is now implicit, covers the whole
949 top-level drive and cannot be (mis)used by users.
951 General: Kernel dumps are now not allowed to be written to devices
952 whose partition types indicate they are meant to be used for file
953 systems (or, in case of MSDOS partitions, as something else than
956 Most of these changes date approximately from 200812.
959 The uscanner(4) driver has been removed from the kernel. This follows
960 Linux removing theirs in 2.6 and making libusb the default interface
964 The multicast forwarding code has been cleaned up. netstat(1)
965 only relies on KVM now for printing bandwidth upcall meters.
966 The IPv4 and IPv6 modules are split into ip_mroute_mod and
967 ip6_mroute_mod respectively. The config(5) options for statically
968 compiling this code remain the same, i.e. 'options MROUTING'.
971 Support for the IFF_NEEDSGIANT network interface flag has been
972 removed, which means that non-MPSAFE network device drivers are no
973 longer supported. In particular, if_ar, if_sr, and network device
974 drivers from the old (legacy) USB stack can no longer be built or
978 POSIX.1 Native Language Support (NLS) has been enabled in libc and
979 a bunch of new language catalog files have also been added.
980 This means that some common libc messages are now localized and
981 they depend on the LC_MESSAGES environmental variable.
984 The k8temp(4) driver has been renamed to amdtemp(4) since
985 support for Family 10 and Family 11 CPU families was added.
988 IGMPv3 and Source-Specific Multicast (SSM) have been merged
989 to the IPv4 stack. VIMAGE hooks are in but not yet used.
991 For kernel developers, the most important changes are that the
992 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
993 and this lock has been downgraded to a non-recursive mutex.
995 Transport protocols (UDP, Raw IP) are now responsible for filtering
996 inbound multicast traffic according to group membership and source
997 filters. The imo_multicast_filter() KPI exists for this purpose.
998 Transports which do not use multicast (SCTP, TCP) already reject
999 multicast by default. Forwarding and receive performance may improve
1000 as a mutex acquisition is no longer needed in the ip_input()
1001 low-level input path. in_addmulti() and in_delmulti() are shimmed
1002 to new KPIs which exist to support SSM in-kernel.
1004 For application developers, it is recommended that loopback of
1005 multicast datagrams be disabled for best performance, as this
1006 will still cause the lock to be taken for each looped-back
1007 datagram transmission. The net.inet.ip.mcast.loop sysctl may
1008 be tuned to 0 to disable loopback by default; it defaults to 1
1009 to preserve the existing behaviour.
1011 For systems administrators, to obtain best performance with
1012 multicast reception and multiple groups, it is always recommended
1013 that a card with a suitably precise hash filter is used. Hash
1014 collisions will still result in the lock being taken within the
1015 transport protocol input path to check group membership.
1017 If deploying FreeBSD in an environment with IGMP snooping switches,
1018 it is recommended that the net.inet.igmp.sendlocal sysctl remain
1019 enabled; this forces 224.0.0.0/24 group membership to be announced
1022 The size of 'struct igmpstat' has changed; netstat needs to be
1023 recompiled to reflect this.
1024 Bump __FreeBSD_version to 800070.
1027 libusb20.so.1 is now installed as libusb.so.1 and the ports system
1028 updated to use it. This requires a buildworld/installworld in order to
1029 update the library and dependencies (usbconfig, etc). Its advisable to
1030 rebuild all ports which uses libusb. More specific directions are given
1031 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
1032 libusb are no longer required and can be removed.
1035 A workaround is committed to allow the creation of System V shared
1036 memory segment of size > 2 GB on the 64-bit architectures.
1037 Due to a limitation of the existing ABI, the shm_segsz member
1038 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
1039 wrong for large segments. Note that limits must be explicitly
1040 raised to allow such segments to be created.
1043 The layout of struct ifnet has changed, requiring a rebuild of all
1044 network device driver modules.
1047 The /dev handling for the new USB stack has changed, a
1048 buildworld/installworld is required for libusb20.
1051 The new USB2 stack has now been permanently moved in and all kernel and
1052 module names reverted to their previous values (eg, usb, ehci, ohci,
1053 ums, ...). The old usb stack can be compiled in by prefixing the name
1054 with the letter 'o', the old usb modules have been removed.
1055 Updating entry 20090216 for xorg and 20090215 for libmap may still
1059 The rc.conf(5) option if_up_delay has been renamed to
1060 defaultroute_delay to better reflect its purpose. If you have
1061 customized this setting in /etc/rc.conf you need to update it to
1065 xorg 7.4 wants to configure its input devices via hald which does not
1066 yet work with USB2. If the keyboard/mouse does not work in xorg then
1068 Option "AllowEmptyInput" "off"
1069 to your ServerLayout section. This will cause X to use the configured
1070 kbd and mouse sections from your xorg.conf.
1073 The GENERIC kernels for all architectures now default to the new USB2
1074 stack. No kernel config options or code have been removed so if a
1075 problem arises please report it and optionally revert to the old USB
1076 stack. If you are loading USB kernel modules or have a custom kernel
1077 that includes GENERIC then ensure that usb names are also changed over,
1078 eg uftdi -> usb2_serial_ftdi.
1080 Older programs linked against the ports libusb 0.1 need to be
1081 redirected to the new stack's libusb20. /etc/libmap.conf can
1083 # Map old usb library to new one for usb2 stack
1084 libusb-0.1.so.8 libusb20.so.1
1087 All USB ethernet devices now attach as interfaces under the name ueN
1088 (eg. ue0). This is to provide a predictable name as vendors often
1089 change usb chipsets in a product without notice.
1092 The ichsmb(4) driver has been changed to require SMBus slave
1093 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1094 All of the other SMBus controller drivers require left-justified
1095 slave addresses, so this change makes all the drivers provide the
1099 INET6 statistics (struct ip6stat) was updated.
1100 netstat(1) needs to be recompiled.
1103 NTFS has been removed from GENERIC kernel on amd64 to match
1104 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1105 will load ntfs.ko module automatically when NTFS support is
1106 actually needed, unless ntfs.ko is not installed or security
1107 level prohibits loading kernel modules. If either is the case,
1108 "options NTFS" has to be added into kernel config.
1111 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1112 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1113 800061. User space tools that rely on the size of struct tcpcb in
1114 tcp_var.h (e.g. sockstat) need to be recompiled.
1117 ng_tty(4) module updated to match the new TTY subsystem.
1118 Due to API change, user-level applications must be updated.
1119 New API support added to mpd5 CVS and expected to be present
1120 in next mpd5.3 release.
1123 With __FreeBSD_version 800060 the makefs tool is part of
1124 the base system (it was a port).
1127 The afdata and ifnet locks have been changed from mutexes to
1128 rwlocks, network modules will need to be re-compiled.
1131 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1132 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1133 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1134 architecture and 40 bytes on 64-bit architecture. The userland
1135 applications "arp" and "ndp" have been updated accordingly.
1136 The output from "netstat -r" shows only routing entries and
1137 none of the L2 information.
1140 __FreeBSD_version 800057 marks the switchover from the
1141 binary ath hal to source code. Users must add the line:
1143 options AH_SUPPORT_AR5416
1145 to their kernel config files when specifying:
1149 The ath_hal module no longer exists; the code is now compiled
1150 together with the driver in the ath module. It is now
1151 possible to tailor chip support (i.e. reduce the set of chips
1152 and thereby the code size); consult ath_hal(4) for details.
1155 __FreeBSD_version 800054 adds memory barriers to
1156 <machine/atomic.h>, new interfaces to ifnet to facilitate
1157 multiple hardware transmit queues for cards that support
1158 them, and a lock-less ring-buffer implementation to
1159 enable drivers to more efficiently manage queueing of
1163 A new version of ZFS (version 13) has been merged to -HEAD.
1164 This version has zpool attribute "listsnapshots" off by
1165 default, which means "zfs list" does not show snapshots,
1166 and is the same as Solaris behavior.
1169 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1172 The uhci, ohci, ehci and slhci USB Host controller drivers have
1173 been put into separate modules. If you load the usb module
1174 separately through loader.conf you will need to load the
1175 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1176 controller add the following to loader.conf:
1182 The ABI used by the PMC toolset has changed. Please keep
1183 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1187 atapci kernel module now includes only generic PCI ATA
1188 driver. AHCI driver moved to ataahci kernel module.
1189 All vendor-specific code moved into separate kernel modules:
1190 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1191 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1192 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1193 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1196 The TTY subsystem of the kernel has been replaced by a new
1197 implementation, which provides better scalability and an
1198 improved driver model. Most common drivers have been migrated to
1199 the new TTY subsystem, while others have not. The following
1200 drivers have not yet been ported to the new TTY layer:
1203 cy, digi, rc, rp, sio
1209 ng_h4, ng_tty, ppp, sl, snp
1211 Adding these drivers to your kernel configuration file shall
1212 cause compilation to fail.
1215 ntpd has been upgraded to 4.2.4p5.
1218 OpenSSH has been upgraded to 5.1p1.
1220 For many years, FreeBSD's version of OpenSSH preferred DSA
1221 over RSA for host and user authentication keys. With this
1222 upgrade, we've switched to the vendor's default of RSA over
1223 DSA. This may cause upgraded clients to warn about unknown
1224 host keys even for previously known hosts. Users should
1225 follow the usual procedure for verifying host keys before
1226 accepting the RSA key.
1228 This can be circumvented by setting the "HostKeyAlgorithms"
1229 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1232 Please note that the sequence of keys offered for
1233 authentication has been changed as well. You may want to
1234 specify IdentityFile in a different order to revert this
1238 The sio(4) driver has been removed from the i386 and amd64
1239 kernel configuration files. This means uart(4) is now the
1240 default serial port driver on those platforms as well.
1242 To prevent collisions with the sio(4) driver, the uart(4) driver
1243 uses different names for its device nodes. This means the
1244 onboard serial port will now most likely be called "ttyu0"
1245 instead of "ttyd0". You may need to reconfigure applications to
1246 use the new device names.
1248 When using the serial port as a boot console, be sure to update
1249 /boot/device.hints and /etc/ttys before booting the new kernel.
1250 If you forget to do so, you can still manually specify the hints
1251 at the loader prompt:
1253 set hint.uart.0.at="isa"
1254 set hint.uart.0.port="0x3F8"
1255 set hint.uart.0.flags="0x10"
1256 set hint.uart.0.irq="4"
1260 The gpt(8) utility has been removed. Use gpart(8) to partition
1264 The version that Linuxulator emulates was changed from 2.4.2
1265 to 2.6.16. If you experience any problems with Linux binaries
1266 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1267 if it fixes the problem contact emulation mailing list.
1270 ISDN4BSD (I4B) was removed from the src tree. You may need to
1271 update a your kernel configuration and remove relevant entries.
1274 I have checked in code to support multiple routing tables.
1275 See the man pages setfib(1) and setfib(2).
1276 This is a hopefully backwards compatible version,
1277 but to make use of it you need to compile your kernel
1278 with options ROUTETABLES=2 (or more up to 16).
1281 The 802.11 wireless support was redone to enable multi-bss
1282 operation on devices that are capable. The underlying device
1283 is no longer used directly but instead wlanX devices are
1284 cloned with ifconfig. This requires changes to rc.conf files.
1285 For example, change:
1286 ifconfig_ath0="WPA DHCP"
1289 ifconfig_wlan0="WPA DHCP"
1290 see rc.conf(5) for more details. In addition, mergemaster of
1291 /etc/rc.d is highly recommended. Simultaneous update of userland
1292 and kernel wouldn't hurt either.
1294 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1295 modules were merged into the base wlan module. All references
1296 to these modules (e.g. in kernel config files) must be removed.
1299 psm(4) has gained write(2) support in native operation level.
1300 Arbitrary commands can be written to /dev/psm%d and status can
1301 be read back from it. Therefore, an application is responsible
1302 for status validation and error recovery. It is a no-op in
1303 other operation levels.
1306 Support for KSE threading has been removed from the kernel. To
1307 run legacy applications linked against KSE libmap.conf may
1308 be used. The following libmap.conf may be used to ensure
1309 compatibility with any prior release:
1311 libpthread.so.1 libthr.so.1
1312 libpthread.so.2 libthr.so.2
1313 libkse.so.3 libthr.so.3
1316 The layout of struct vmspace has changed. This affects libkvm
1317 and any executables that link against libkvm and use the
1318 kvm_getprocs() function. In particular, but not exclusively,
1319 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1320 The effects are minimal, but it's advisable to upgrade world
1324 The latest em driver no longer has support in it for the
1325 82575 adapter, this is now moved to the igb driver. The
1326 split was done to make new features that are incompatible
1327 with older hardware easier to do.
1330 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1331 likewise the kernel option is now GEOM_LINUX_LVM.
1334 The default NFS mount mode has changed from UDP to TCP for
1335 increased reliability. If you rely on (insecurely) NFS
1336 mounting across a firewall you may need to update your
1340 Belatedly note the addition of m_collapse for compacting
1344 The fts(3) structures have been changed to use adequate
1345 integer types for their members and so to be able to cope
1346 with huge file trees. The old fts(3) ABI is preserved
1347 through symbol versioning in libc, so third-party binaries
1348 using fts(3) should still work, although they will not take
1349 advantage of the extended types. At the same time, some
1350 third-party software might fail to build after this change
1351 due to unportable assumptions made in its source code about
1352 fts(3) structure members. Such software should be fixed
1353 by its vendor or, in the worst case, in the ports tree.
1354 FreeBSD_version 800015 marks this change for the unlikely
1355 case that a portable fix is impossible.
1358 To upgrade to -current after this date, you must be running
1359 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1360 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1363 The ADAPTIVE_GIANT kernel option has been retired because its
1364 functionality is the default now.
1367 The AT keyboard emulation of sunkbd(4) has been turned on
1368 by default. In order to make the special symbols of the Sun
1369 keyboards driven by sunkbd(4) work under X these now have
1370 to be configured the same way as Sun USB keyboards driven
1371 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1373 Option "XkbLayout" "us"
1374 Option "XkbRules" "xorg"
1375 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1378 It has been decided that it is desirable to provide ABI
1379 backwards compatibility to the FreeBSD 4/5/6 versions of the
1380 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1381 broken with the introduction of PCI domain support (see the
1382 20070930 entry). Unfortunately, this required the ABI of
1383 PCIOCGETCONF to be broken again in order to be able to
1384 provide backwards compatibility to the old version of that
1385 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1386 again. As for prominent ports this affects neither pciutils
1387 nor xorg-server this time, the hal port needs to be rebuilt
1391 The misnamed kthread_create() and friends have been renamed
1392 to kproc_create() etc. Many of the callers already
1393 used kproc_start()..
1394 I will return kthread_create() and friends in a while
1395 with implementations that actually create threads, not procs.
1396 Renaming corresponds with version 800002.
1405 Avoid using make -j when upgrading. While generally safe, there are
1406 sometimes problems using -j to upgrade. If your upgrade fails with
1407 -j, please try again without -j. From time to time in the past there
1408 have been problems using -j with buildworld and/or installworld. This
1409 is especially true when upgrading between "distant" versions (eg one
1410 that cross a major release boundary or several minor releases, or when
1411 several months have passed on the -current branch).
1413 Sometimes, obscure build problems are the result of environment
1414 poisoning. This can happen because the make utility reads its
1415 environment when searching for values for global variables. To run
1416 your build attempts in an "environmental clean room", prefix all make
1417 commands with 'env -i '. See the env(1) manual page for more details.
1419 When upgrading from one major version to another it is generally best
1420 to upgrade to the latest code in the currently installed branch first,
1421 then do an upgrade to the new branch. This is the best-tested upgrade
1422 path, and has the highest probability of being successful. Please try
1423 this approach before reporting problems with a major version upgrade.
1427 When upgrading the boot ZFS pool to a new version, always follow
1430 1.) recompile and reinstall the ZFS boot loader and boot block
1431 (this is part of "make buildworld" and "make installworld")
1433 2.) update the ZFS boot block on your boot drive
1435 The following example updates the ZFS boot block on the first
1436 partition (freebsd-boot) of a GPT partitioned drive ad0:
1437 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1439 Non-boot pools do not need these updates.
1443 If you are updating from a prior version of FreeBSD (even one just
1444 a few days old), you should follow this procedure. It is the most
1445 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1447 make kernel-toolchain
1448 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1449 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1451 To test a kernel once
1452 ---------------------
1453 If you just want to boot a kernel once (because you are not sure
1454 if it works, or if you want to boot a known bad kernel to provide
1455 debugging information) run
1456 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1457 nextboot -k testkernel
1459 To just build a kernel when you know that it won't mess you up
1460 --------------------------------------------------------------
1461 This assumes you are already running a CURRENT system. Replace
1462 ${arch} with the architecture of your machine (e.g. "i386",
1463 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1465 cd src/sys/${arch}/conf
1466 config KERNEL_NAME_HERE
1467 cd ../compile/KERNEL_NAME_HERE
1472 If this fails, go to the "To build a kernel" section.
1474 To rebuild everything and install it on the current system.
1475 -----------------------------------------------------------
1476 # Note: sometimes if you are running current you gotta do more than
1477 # is listed here if you are upgrading from a really old current.
1479 <make sure you have good level 0 dumps>
1481 make kernel KERNCONF=YOUR_KERNEL_HERE
1483 <reboot in single user> [3]
1491 To cross-install current onto a separate partition
1492 --------------------------------------------------
1493 # In this approach we use a separate partition to hold
1494 # current's root, 'usr', and 'var' directories. A partition
1495 # holding "/", "/usr" and "/var" should be about 2GB in
1498 <make sure you have good level 0 dumps>
1501 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1502 <maybe newfs current's root partition>
1503 <mount current's root partition on directory ${CURRENT_ROOT}>
1504 make installworld DESTDIR=${CURRENT_ROOT}
1505 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1506 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1507 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1508 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1509 <reboot into current>
1510 <do a "native" rebuild/install as described in the previous section>
1511 <maybe install compatibility libraries from ports/misc/compat*>
1515 To upgrade in-place from 8.x-stable to current
1516 ----------------------------------------------
1517 <make sure you have good level 0 dumps>
1519 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1521 <reboot in single user> [3]
1528 Make sure that you've read the UPDATING file to understand the
1529 tweaks to various things you need. At this point in the life
1530 cycle of current, things change often and you are on your own
1531 to cope. The defaults can also change, so please read ALL of
1532 the UPDATING entries.
1534 Also, if you are tracking -current, you must be subscribed to
1535 freebsd-current@freebsd.org. Make sure that before you update
1536 your sources that you have read and understood all the recent
1537 messages there. If in doubt, please track -stable which has
1538 much fewer pitfalls.
1540 [1] If you have third party modules, such as vmware, you
1541 should disable them at this point so they don't crash your
1544 [3] From the bootblocks, boot -s, and then do
1549 adjkerntz -i # if CMOS is wall time
1550 Also, when doing a major release upgrade, it is required that
1551 you boot into single user mode to do the installworld.
1553 [4] Note: This step is non-optional. Failure to do this step
1554 can result in a significant reduction in the functionality of the
1555 system. Attempting to do it by hand is not recommended and those
1556 that pursue this avenue should read this file carefully, as well
1557 as the archives of freebsd-current and freebsd-hackers mailing lists
1558 for potential gotchas. The -U option is also useful to consider.
1559 See mergemaster(8) for more information.
1561 [5] Usually this step is a noop. However, from time to time
1562 you may need to do this if you get unknown user in the following
1563 step. It never hurts to do it all the time. You may need to
1564 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1565 install) after the buildworld before this step if you last updated
1566 from current before 20020224 or from -stable before 20020408.
1568 [6] This only deletes old files and directories. Old libraries
1569 can be deleted by "make delete-old-libs", but you have to make
1570 sure that no program is using those libraries anymore.
1572 [8] In order to have a kernel that can run the 4.x binaries needed to
1573 do an installworld, you must include the COMPAT_FREEBSD4 option in
1574 your kernel. Failure to do so may leave you with a system that is
1575 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1576 required to run the 5.x binaries on more recent kernels. And so on
1577 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1579 Make sure that you merge any new devices from GENERIC since the
1580 last time you updated your kernel config file.
1582 [9] When checking out sources, you must include the -P flag to have
1583 cvs prune empty directories.
1585 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1586 "?=" instead of the "=" assignment operator, so that buildworld can
1587 override the CPUTYPE if it needs to.
1589 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1590 not on the command line, or in /etc/make.conf. buildworld will
1591 warn if it is improperly defined.
1594 This file contains a list, in reverse chronological order, of major
1595 breakages in tracking -current. It is not guaranteed to be a complete
1596 list of such breakages, and only contains entries since October 10, 2007.
1597 If you need to see UPDATING entries from before that date, you will need
1598 to fetch an UPDATING file from an older FreeBSD release.
1600 Copyright information:
1602 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1604 Redistribution, publication, translation and use, with or without
1605 modification, in full or in part, in any form or format of this
1606 document are permitted without further permission from the author.
1608 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1609 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1610 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1611 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1612 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1613 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1614 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1615 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1616 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1617 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1618 POSSIBILITY OF SUCH DAMAGE.
1620 Contact Warner Losh if you have any questions about your use of