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 OpenSSL has been upgraded to 1.0.1c. Any binaries requiring
29 libcrypto.so.6 or libssl.so.6 must be recompiled. Also, there are
30 configuration changes. Make sure to merge /etc/ssl/openssl.cnf.
33 The following sysctls and tunables have been renamed for consistency
35 kern.cam.da.da_send_ordered -> kern.cam.da.send_ordered
36 kern.cam.ada.ada_send_ordered -> kern.cam.ada.send_ordered
39 The sort utility has been replaced with BSD sort. For now, GNU sort
40 is also available as "gnusort" or the default can be set back to
41 GNU sort by setting WITH_GNU_SORT. In this case, BSD sort will be
42 installed as "bsdsort".
45 A new version of ZFS (pool version 5000) has been merged to -HEAD.
46 Starting with this version the old system of ZFS pool versioning
47 is superseded by "feature flags". This concept enables forward
48 compatibility against certain future changes in functionality of ZFS
49 pools. The first read-only compatible "feature flag" for ZFS pools
50 is named "com.delphix:async_destroy". For more information
51 read the new zpool-features(5) manual page.
52 Please refer to the "ZFS notes" section of this file for information
53 on upgrading boot ZFS pools.
56 The malloc(3) implementation embedded in libc now uses sources imported
57 as contrib/jemalloc. The most disruptive API change is to
58 /etc/malloc.conf. If your system has an old-style /etc/malloc.conf,
59 delete it prior to installworld, and optionally re-create it using the
60 new format after rebooting. See malloc.conf(5) for details
61 (specifically the TUNING section and the "opt.*" entries in the MALLCTL
65 Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb
66 is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is
67 now spelled mips. This is to aid compatibility with third-party
68 software that expects this naming scheme in uname(3). Little-endian
69 settings are unchanged.
72 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
76 Now unix domain sockets behave "as expected" on nullfs(5). Previously
77 nullfs(5) did not pass through all behaviours to the underlying layer,
78 as a result if we bound to a socket on the lower layer we could connect
79 only to the lower path; if we bound to the upper layer we could connect
80 only to the upper path. The new behavior is one can connect to both the
81 lower and the upper paths regardless what layer path one binds to.
84 The getifaddrs upgrade path broken with 20111215 has been restored.
85 If you have upgraded in between 20111215 and 20120209 you need to
86 recompile libc again with your kernel. You still need to recompile
87 world to be able to configure CARP but this restriction already
91 The set_rcvar() function has been removed from /etc/rc.subr. All
92 base and ports rc.d scripts have been updated, so if you have a
93 port installed with a script in /usr/local/etc/rc.d you can either
94 hand-edit the rcvar= line, or reinstall the port.
96 An easy way to handle the mass-update of /etc/rc.d:
97 rm /etc/rc.d/* && mergemaster -i
100 panic(9) now stops other CPUs in the SMP systems, disables interrupts
101 on the current CPU and prevents other threads from running.
102 This behavior can be reverted using the kern.stop_scheduler_on_panic
104 The new behavior can be incompatible with kern.sync_on_panic.
107 The carp(4) facility has been changed significantly. Configuration
108 of the CARP protocol via ifconfig(8) has changed, as well as format
109 of CARP events submitted to devd(8) has changed. See manual pages
110 for more information. The arpbalance feature of carp(4) is currently
111 not supported anymore.
113 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
114 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
115 need to be recompiled.
118 The acpi_wmi(4) status device /dev/wmistat has been renamed to
122 The option VFS_ALLOW_NONMPSAFE option has been added in order to
123 explicitely support non-MPSAFE filesystems.
124 It is on by default for all supported platform at this present
128 The broken amd(4) driver has been replaced with esp(4) in the amd64,
129 i386 and pc98 GENERIC kernel configuration files.
132 sysinstall has been removed
135 The stable/9 branch created in subversion. This corresponds to the
136 RELENG_9 branch in CVS.
139 This commit modifies vfs_register() so that it uses a hash
140 calculation to set vfc_typenum, which is enabled by default.
141 The first time a system is booted after this change, the
142 vfc_typenum values will change for all file systems. The
143 main effect of this is a change to the NFS server file handles
144 for file systems that use vfc_typenum in their fsid, such as ZFS.
145 It will, however, prevent vfc_typenum from changing when file
146 systems are loaded in a different order for subsequent reboots.
147 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
148 until you are ready to remount all NFS clients after a reboot.
151 Bump the shared library version numbers for libraries that
152 do not use symbol versioning, have changed the ABI compared
153 to stable/8 and which shared library version was not bumped.
154 Done as part of 9.0-RELEASE cycle.
157 During the merge of Capsicum features, the fget(9) KPI was modified.
158 This may require the rebuilding of out-of-tree device drivers --
159 issues have been reported specifically with the nVidia device driver.
160 __FreeBSD_version is bumped to 900041.
162 Also, there is a period between 20110811 and 20110814 where the
163 special devices /dev/{stdin,stdout,stderr} did not work correctly.
164 Building world from a kernel during that window may not work.
167 The packet filter (pf) code has been updated to OpenBSD 4.5.
168 You need to update userland tools to be in sync with kernel.
169 This update breaks backward compatibility with earlier pfsync(4)
170 versions. Care must be taken when updating redundant firewall setups.
173 The following sysctls and tunables are retired on x86 platforms:
175 machdep.hlt_logical_cpus
176 The following sysctl is retired:
177 machdep.hyperthreading_allowed
178 The sysctls were supposed to provide a way to dynamically offline and
179 online selected CPUs on x86 platforms, but the implementation has not
180 been reliable especially with SCHED_ULE scheduler.
181 machdep.hyperthreading_allowed tunable is still available to ignore
182 hyperthreading CPUs at OS level.
183 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
184 where X is an APIC ID of a CPU. Be advised, though, that disabling
185 CPUs in non-uniform fashion will result in non-uniform topology and
186 may lead to sub-optimal system performance with SCHED_ULE, which is
190 cpumask_t type is retired and cpuset_t is used in order to describe
194 Changes to ifconfig(8) for dynamic address family detection mandate
195 that you are running a kernel of 20110525 or later. Make sure to
196 follow the update procedure to boot a new kernel before installing
200 Support for sun4v architecture is officially dropped
203 Several KPI breaking changes have been committed to the mii(4) layer,
204 the PHY drivers and consequently some Ethernet drivers using mii(4).
205 This means that miibus.ko and the modules of the affected Ethernet
206 drivers need to be recompiled.
208 Note to kernel developers: Given that the OUI bit reversion problem
209 was fixed as part of these changes all mii(4) commits related to OUIs,
210 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
211 handling, no longer can be merged verbatim to stable/8 and previous
215 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
216 into their kernel configurations along with 'device pci'.
219 The default NFS client is now the new NFS client, so fstype "newnfs"
220 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
221 Although mounts via fstype "nfs" will usually work without userland
222 changes, it is recommended that the mount(8) and mount_nfs(8)
223 commands be rebuilt from sources and that a link to mount_nfs called
224 mount_oldnfs be created. The new client is compiled into the
225 kernel with "options NFSCL" and this is needed for diskless root
226 file systems. The GENERIC kernel configs have been changed to use
227 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
228 To use the regular/old client, you can "mount -t oldnfs ...". For
229 a diskless root file system, you must also include a line like:
231 vfs.root.mountfrom="oldnfs:"
233 in the boot/loader.conf on the root fs on the NFS server to make
234 a diskless root fs use the old client.
237 The GENERIC kernels for all architectures now default to the new
238 CAM-based ATA stack. It means that all legacy ATA drivers were
239 removed and replaced by respective CAM drivers. If you are using
240 ATA device names in /etc/fstab or other places, make sure to update
241 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
242 where 'Y's are the sequential numbers starting from zero for each type
243 in order of detection, unless configured otherwise with tunables,
244 see cam(4)). There will be symbolic links created in /dev/ to map
245 old adX devices to the respective adaY. They should provide basic
246 compatibility for file systems mounting in most cases, but they do
247 not support old user-level APIs and do not have respective providers
248 in GEOM. Consider using updated management tools with new device names.
250 It is possible to load devices ahci, ata, siis and mvs as modules,
251 but option ATA_CAM should remain in kernel configuration to make ata
252 module work as CAM driver supporting legacy ATA controllers. Device ata
253 still can be used in modular fashion (atacore + ...). Modules atadisk
254 and atapi* are not used and won't affect operation in ATA_CAM mode.
255 Note that to use CAM-based ATA kernel should include CAM devices
256 scbus, pass, da (or explicitly ada), cd and optionally others. All of
257 them are parts of the cam module.
259 ataraid(4) functionality is now supported by the RAID GEOM class.
260 To use it you can load geom_raid kernel module and use graid(8) tool
261 for management. Instead of /dev/arX device names, use /dev/raid/rX.
263 No kernel config options or code have been removed, so if a problem
264 arises, please report it and optionally revert to the old ATA stack.
265 In order to do it you can remove from the kernel config:
270 , and instead add back:
271 device atadisk # ATA disk drives
272 device ataraid # ATA RAID drives
273 device atapicd # ATAPI CDROM drives
274 device atapifd # ATAPI floppy drives
275 device atapist # ATAPI tape drives
278 The default NFS server has been changed to the new server, which
279 was referred to as the experimental server. If you need to switch
280 back to the old NFS server, you must now put the "-o" option on
281 both the mountd and nfsd commands. This can be done using the
282 mountd_flags and nfs_server_flags rc.conf variables until an
283 update to the rc scripts is committed, which is coming soon.
286 The GNU Objective-C runtime library (libobjc), and other Objective-C
287 related components have been removed from the base system. If you
288 require an Objective-C library, please use one of the available ports.
291 ath(4) has been split into bus- and device- modules. if_ath contains
292 the HAL, the TX rate control and the network device code. if_ath_pci
293 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
294 contains the AHB glue. Users need to load both if_ath_pci and if_ath
295 in order to use ath on everything else.
297 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
298 need to load if_ath and if_ath_pci for ath(4) operation.
301 As part of the replacement of sysinstall, the process of building
302 release media has changed significantly. For details, please re-read
303 release(7), which has been updated to reflect the new build process.
306 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
307 is the last available version under GPLv2. It brings a number of new
308 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
309 4.1 and SSE 4.2), better support for powerpc64, a number of new
310 directives, and lots of other small improvements. See the ChangeLog
311 file in contrib/binutils for the full details.
314 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
315 compliant, and will now use half of hash for authentication.
316 This will break interoperability with all stacks (including all
317 actual FreeBSD versions) who implement
318 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
320 The only workaround with such peers is to use another HMAC
321 algorithm for IPsec ("phase 2") authentication.
324 Remove the uio_yield prototype and symbol. This function has
325 been misnamed since it was introduced and should not be
326 globally exposed with this name. The equivalent functionality
327 is now available using kern_yield(curthread->td_user_pri).
328 The function remains undocumented.
331 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
332 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
333 for scalar sysctls is defined but disabled. Code that needs
334 UQUAD to pass the type checking that must compile on older
335 systems where the define is not present can check against
336 __FreeBSD_version >= 900030.
338 The system dialog(1) has been replaced with a new version previously
339 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
340 with the previous version, but the libdialog associated with it has
341 a largely incompatible API. As such, the original version of libdialog
342 will be kept temporarily as libodialog, until its base system consumers
343 are replaced or updated. Bump __FreeBSD_version to 900030.
346 If you are trying to run make universe on a -stable system, and you get
347 the following warning:
348 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
349 unknown. config(8) likely too old."
350 or something similar to it, then you must upgrade your -stable system
351 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
352 stable/8) or build the config from the latest stable/8 branch and
353 install it on your system.
355 Prior to this date, building a current universe on 8-stable system from
356 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
357 error in the first kernel build phase. A new config on those old
358 systems will fix that problem for older versions of -current.
361 The TCP stack has been modified to allow Khelp modules to interact with
362 it via helper hook points and store per-connection data in the TCP
363 control block. Bump __FreeBSD_version to 900029. User space tools that
364 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
368 Generic IEEE 802.3 annex 31B full duplex flow control support has been
369 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
370 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
371 to take advantage of it instead of using custom implementations. This
372 means that these drivers now no longer unconditionally advertise
373 support for flow control but only do so if flow control is a selected
374 media option. This was implemented in the generic support that way in
375 order to allow flow control to be switched on and off via ifconfig(8)
376 with the PHY specific default to typically off in order to protect
377 from unwanted effects. Consequently, if you used flow control with
378 one of the above mentioned drivers you now need to explicitly enable
380 ifconfig bge0 media auto mediaopt flowcontrol
382 Along with the above mentioned changes generic support for setting
383 1000baseT master mode also has been added and brgphy(4), ciphy(4),
384 e1000phy(4) as well as ip1000phy(4) have been converted to take
385 advantage of it. This means that these drivers now no longer take the
386 link0 parameter for selecting master mode but the master media option
387 has to be used instead, for example like in the following:
388 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
390 Selection of master mode now is also available with all other PHY
391 drivers supporting 1000baseT.
394 The TCP stack has received a significant update to add support for
395 modularised congestion control and generally improve the clarity of
396 congestion control decisions. Bump __FreeBSD_version to 900025. User
397 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
398 sockstat) need to be recompiled.
401 The man(1) utility has been replaced by a new version that no longer
402 uses /etc/manpath.config. Please consult man.conf(5) for how to
403 migrate local entries to the new format.
406 The copyright strings printed by login(1) and sshd(8) at the time of a
407 new connection have been removed to follow other operating systems and
411 A workaround for a fixed ld bug has been removed in kernel code,
412 so make sure that your system ld is built from sources after
413 revision 210245 from 2010-07-19 (r211583 if building head kernel
414 on stable/8, r211584 for stable/7; both from 2010-08-21).
415 A symptom of incorrect ld version is different addresses for
416 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
419 The $ipv6_prefer variable in rc.conf(5) has been split into
420 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
422 The $ip6addrctl_policy is a variable to choose a pre-defined
423 address selection policy set by ip6addrctl(8). A value
424 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
427 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
428 flag (see an entry of 20090926) is set on an interface with no
429 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
430 security reason. If you want IPv6 link-local address on all
431 interfaces by default, set this to "YES".
433 The old ipv6_prefer="YES" is equivalent to
434 ipv6_activate_all_interfaces="YES" and
435 ip6addrctl_policy="ipv6_prefer".
438 DTrace has grown support for userland tracing. Due to this, DTrace is
439 now i386 and amd64 only.
440 dtruss(1) is now installed by default on those systems and a new
441 kernel module is needed for userland tracing: fasttrap.
442 No changes to your kernel config file are necessary to enable
443 userland tracing, but you might consider adding 'STRIP=' and
444 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
445 to have informative userland stack traces in DTrace (ustack).
448 The acpi_aiboost(4) driver has been removed in favor of the new
449 aibs(4) driver. You should update your kernel configuration file.
452 BSD grep has been imported to the base system and it is built by
453 default. It is completely BSD licensed, highly GNU-compatible, uses
454 less memory than its GNU counterpart and has a small codebase.
455 However, it is slower than its GNU counterpart, which is mostly
456 noticeable for larger searches, for smaller ones it is measurable
457 but not significant. The reason is complex, the most important factor
458 is that we lack a modern and efficient regex library and GNU
459 overcomes this by optimizing the searches internally. Future work
460 on improving the regex performance is planned, for the meantime,
461 users that need better performance, can build GNU grep instead by
462 setting the WITH_GNU_GREP knob.
465 Due to the import of powerpc64 support, all existing powerpc kernel
466 configuration files must be updated with a machine directive like this:
467 machine powerpc powerpc
469 In addition, an updated config(8) is required to build powerpc kernels
473 A new version of ZFS (version 15) has been merged to -HEAD.
474 This version uses a python library for the following subcommands:
475 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
476 For full functionality of these commands the following port must
477 be installed: sysutils/py-zfs
480 'vm_page's are now hashed by physical address to an array of mutexes.
481 Currently this is only used to serialize access to hold_count. Over
482 time the page queue mutex will be peeled away. This changes the size
483 of pmap on every architecture. And requires all callers of vm_page_hold
484 and vm_page_unhold to be updated.
487 WITH_CTF can now be specified in src.conf (not recommended, there
488 are some problems with static executables), make.conf (would also
489 affect ports which do not use GNU make and do not override the
490 compile targets) or in the kernel config (via "makeoptions
492 When WITH_CTF was specified there before this was silently ignored,
493 so make sure that WITH_CTF is not used in places which could lead
494 to unwanted behavior.
497 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
498 to allow 32-bit compatibility on non-x86 platforms. All kernel
499 configurations on amd64 and ia64 platforms using these options must
500 be modified accordingly.
503 The utmp user accounting database has been replaced with utmpx,
504 the user accounting interface standardized by POSIX.
505 Unfortunately the semantics of utmp and utmpx don't match,
506 making it practically impossible to support both interfaces.
507 The user accounting database is used by tools like finger(1),
508 last(1), talk(1), w(1) and ac(8).
510 All applications in the base system use utmpx. This means only
511 local binaries (e.g. from the ports tree) may still use these
512 utmp database files. These applications must be rebuilt to make
515 After the system has been upgraded, it is safe to remove the old
516 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
517 assuming their contents is of no importance anymore. Old wtmp
518 databases can only be used by last(1) and ac(8) after they have
519 been converted to the new format using wtmpcvt(1).
522 Introduce the kernel thread "deadlock resolver" (which can be enabled
523 via the DEADLKRES option, see NOTES for more details) and the
524 sleepq_type() function for sleepqueues.
527 The rc.firewall and rc.firewall6 were unified, and
528 rc.firewall6 and rc.d/ip6fw were removed.
529 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
530 variables are obsoleted. Instead, the following new rc
531 variables are added to rc.d/ipfw:
533 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
534 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
535 firewall_simple_onet_ipv6, firewall_trusted_ipv6
537 The meanings correspond to the relevant IPv4 variables.
543 The default terminal emulation for syscons(4) has been changed
544 from cons25 to xterm on all platforms except pc98. This means
545 that the /etc/ttys file needs to be updated to ensure correct
546 operation of applications on the console.
548 The terminal emulation style can be toggled per window by using
549 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
550 options can be used to change the compile-time default back to
553 To prevent graphical artifacts, make sure the TERM environment
554 variable is set to match the terminal emulation that is being
555 performed by syscons(4).
558 The layout of the structure ieee80211req_scan_result has changed.
559 Applications that require wireless scan results (e.g. ifconfig(8))
560 from net80211 need to be recompiled.
562 Applications such as wpa_supplicant(8) may require a full world
563 build without using NO_CLEAN in order to get synchronized with the
567 The iwn(4) driver has been updated to support the 5000 and 5150 series.
568 There's one kernel module for each firmware. Adding "device iwnfw"
569 to the kernel configuration file means including all three firmware
570 images inside the kernel. If you want to include just the one for
571 your wireless card, use the devices iwn4965fw, iwn5000fw or
575 The rc.d/network_ipv6, IPv6 configuration script has been integrated
576 into rc.d/netif. The changes are the following:
578 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
579 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
580 Note that both variables need the "inet6" keyword at the head.
582 Do not set $ipv6_network_interfaces manually if you do not
583 understand what you are doing. It is not needed in most cases.
585 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
588 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
589 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
591 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
592 all configured interfaces have "inet6 accept_rtadv" in the
593 $ifconfig_IF_ipv6. These are for backward compatibility.
595 3. A new variable $ipv6_prefer has been added. If NO, IPv6
596 functionality of interfaces with no corresponding
597 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
598 and the default address selection policy of ip6addrctl(8)
599 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
600 Note that if you want to configure IPv6 functionality on the
601 disabled interfaces after boot, first you need to clear the flag by
602 using ifconfig(8) like:
604 ifconfig em0 inet6 -ifdisabled
606 If YES, the default address selection policy is set as
609 The default value of $ipv6_prefer is NO.
611 4. If your system need to receive Router Advertisement messages,
612 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
613 scripts automatically invoke rtsol(8) when the interface becomes
614 UP. The Router Advertisement messages are used for SLAAC
615 (State-Less Address AutoConfiguration).
618 802.11s D3.03 support was committed. This is incompatible with the
619 previous code, which was based on D3.0.
622 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
623 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
624 control whether accepting Router Advertisement messages or not.
625 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
626 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
627 The ifconfig(8) utility now supports these flags.
630 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
631 mount(8) and -a option for df(1) to see them.
634 The old tunable hw.bus.devctl_disable has been superseded by
635 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
636 replaced by hw.bus.devctl_queue=0. The default for this new tunable
640 Remove the option STOP_NMI. The default action is now to use NMI only
641 for KDB via the newly introduced function stop_cpus_hard() and
642 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
645 The stable/8 branch created in subversion. This corresponds to the
646 RELENG_8 branch in CVS.
649 Bump the shared library version numbers for all libraries that do not
650 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
651 __FreeBSD_version to 800105.
654 Due to changes in the implementation of virtual network stack support,
655 all network-related kernel modules must be recompiled. As this change
656 breaks the ABI, bump __FreeBSD_version to 800104.
659 The TOE interface to the TCP syncache has been modified to remove
660 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
661 The cxgb driver is the only TOE consumer affected by this change, and
662 needs to be recompiled along with the kernel. As this change breaks
663 the ABI, bump __FreeBSD_version to 800103.
666 Padding has been added to struct tcpcb, sackhint and tcpstat in
667 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
668 maintaining the ABI. However, this change breaks the ABI, so bump
669 __FreeBSD_version to 800102. User space tools that rely on the size of
670 any of these structs (e.g. sockstat) need to be recompiled.
673 The NFS_LEGACYRPC option has been removed along with the old kernel
674 RPC implementation that this option selected. Kernel configurations
675 may need to be adjusted.
678 The network interface device nodes at /dev/net/<interface> have been
679 removed. All ioctl operations can be performed the normal way using
680 routing sockets. The kqueue functionality can generally be replaced
681 with routing sockets.
684 The documentation from the FreeBSD Documentation Project (Handbook,
685 FAQ, etc.) is now installed via packages by sysinstall(8) and under
686 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
689 The ABI of various structures related to the SYSV IPC API have been
690 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
691 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
695 Layout of struct vnet has changed as routing related variables were
696 moved to their own Vimage module. Modules need to be recompiled. Bump
697 __FreeBSD_version to 800099.
700 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
701 respectively. As long as no more than 16 groups per process are used,
702 no changes should be visible. When more than 16 groups are used, old
703 binaries may fail if they call getgroups() or getgrouplist() with
704 statically sized storage. Recompiling will work around this, but
705 applications should be modified to use dynamically allocated storage
706 for group arrays as POSIX.1-2008 does not cap an implementation's
707 number of supported groups at NGROUPS_MAX+1 as previous versions did.
709 NFS and portalfs mounts may also be affected as the list of groups is
710 truncated to 16. Users of NFS who use more than 16 groups, should
711 take care that negative group permissions are not used on the exported
712 file systems as they will not be reliable unless a GSSAPI based
713 authentication method is used.
716 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
717 option compiles in the support for adaptive spinning for lockmgrs
718 which want to enable it. The lockinit() function now accepts the flag
719 LK_ADAPTIVE in order to make the lock object subject to adaptive
720 spinning when both held in write and read mode.
723 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
724 changed. User applications that use this ioctl need to be rebuilt.
727 The layout of struct thread has changed. Kernel and modules need to
731 The layout of structs ifnet, domain, protosw and vnet_net has changed.
732 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
735 window(1) has been removed from the base system. It can now be
736 installed from ports. The port is called misc/window.
739 The way we are storing and accessing `routing table' entries has
740 changed. Programs reading the FIB, like netstat, need to be
744 A new netisr implementation has been added for FreeBSD 8. Network
745 file system modules, such as igmp, ipdivert, and others, should be
747 Bump __FreeBSD_version to 800096.
750 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
754 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
755 Bump __FreeBSD_version to 800094.
758 Add mnt_xflag field to 'struct mount'. File system modules need to be
760 Bump __FreeBSD_version to 800093.
763 The compiling option ADAPTIVE_SX has been retired while it has been
764 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
765 The KPI for sx_init_flags() changes as accepting flags:
766 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
767 been introduced in order to handle the reversed logic.
768 Bump __FreeBSD_version to 800092.
771 Add support for hierarchical jails. Remove global securelevel.
772 Bump __FreeBSD_version to 800091.
775 The layout of struct vnet_net has changed, therefore modules
777 Bump __FreeBSD_version to 800090.
780 The newly imported zic(8) produces a new format in the output. Please
781 run tzsetup(8) to install the newly created data to /etc/localtime.
784 The sysctl tree for the usb stack has renamed from hw.usb2.* to
785 hw.usb.* and is now consistent again with previous releases.
788 802.11 monitor mode support was revised and driver api's were changed.
789 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
790 of DLT_IEEE802_11. No user-visible data structures were changed but
791 applications that use DLT_IEEE802_11 may require changes.
792 Bump __FreeBSD_version to 800088.
795 The layout of the following structs has changed: sysctl_oid,
796 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
797 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
798 panics may be experienced. World rebuild is required for
799 correctly checking networking state from userland.
800 Bump __FreeBSD_version to 800085.
803 MLDv2 and Source-Specific Multicast (SSM) have been merged
804 to the IPv6 stack. VIMAGE hooks are in but not yet used.
805 The implementation of SSM within FreeBSD's IPv6 stack closely
806 follows the IPv4 implementation.
808 For kernel developers:
810 * The most important changes are that the ip6_output() and
811 ip6_input() paths no longer take the IN6_MULTI_LOCK,
812 and this lock has been downgraded to a non-recursive mutex.
814 * As with the changes to the IPv4 stack to support SSM, filtering
815 of inbound multicast traffic must now be performed by transport
816 protocols within the IPv6 stack. This does not apply to TCP and
817 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
819 * The KPIs used by IPv6 multicast are similar to those used by
820 the IPv4 stack, with the following differences:
821 * im6o_mc_filter() is analogous to imo_multicast_filter().
822 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
823 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
824 * IN6_LOOKUP_MULTI() has been deprecated and removed.
825 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
826 for MLDv1 have an additional 'timer' argument which is used to
827 jitter the initial membership report for the solicited-node
828 multicast membership on-link.
829 * This is not strictly needed for MLDv2, which already jitters
830 its report transmissions. However, the 'timer' argument is
831 preserved in case MLDv1 is active on the interface.
833 * The KAME linked-list based IPv6 membership implementation has
834 been refactored to use a vector similar to that used by the IPv4
836 Code which maintains a list of its own multicast memberships
837 internally, e.g. carp, has been updated to reflect the new
840 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
841 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
842 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
843 implementation constraint which needs to be addressed in HEAD.
845 For application developers:
847 * The changes are broadly similar to those made for the IPv4
850 * The use of IPv4 and IPv6 multicast socket options on the same
851 socket, using mapped addresses, HAS NOT been tested or supported.
853 * There are a number of issues with the implementation of various
854 IPv6 multicast APIs which need to be resolved in the API surface
855 before the implementation is fully compatible with KAME userland
856 use, and these are mostly to do with interface index treatment.
858 * The literature available discusses the use of either the delta / ASM
859 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
860 using setsourcefilter(3)/getsourcefilter(3). For more information
861 please refer to RFC 3768, 'Socket Interface Extensions for
862 Multicast Source Filters'.
864 * Applications which use the published RFC 3678 APIs should be fine.
866 For systems administrators:
868 * The mtest(8) utility has been refactored to support IPv6, in
869 addition to IPv4. Interface addresses are no longer accepted
870 as arguments, their names must be used instead. The utility
871 will map the interface name to its first IPv4 address as
872 returned by getifaddrs(3).
874 * The ifmcstat(8) utility has also been updated to print the MLDv2
875 endpoint state and source filter lists via sysctl(3).
877 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
878 loopback of IPv6 multicast datagrams by default; it defaults to 1
879 to preserve the existing behaviour. Disabling multicast loopback is
880 recommended for optimal system performance.
882 * The IPv6 MROUTING code has been changed to examine this sysctl
883 instead of attempting to perform a group lookup before looping
884 back forwarded datagrams.
886 Bump __FreeBSD_version to 800084.
889 Implement low-level Bluetooth HCI API.
890 Bump __FreeBSD_version to 800083.
893 The layout of struct malloc_type, used by modules to register new
894 memory allocation types, has changed. Most modules will need to
895 be rebuilt or panics may be experienced.
896 Bump __FreeBSD_version to 800081.
899 Anticipate overflowing inp_flags - add inp_flags2.
900 This changes most offsets in inpcb, so checking v4 connection
901 state will require a world rebuild.
902 Bump __FreeBSD_version to 800080.
905 Add an llentry to struct route and struct route_in6. Modules
906 embedding a struct route will need to be recompiled.
907 Bump __FreeBSD_version to 800079.
910 The size of rt_metrics_lite and by extension rtentry has changed.
911 Networking administration apps will need to be recompiled.
912 The route command now supports show as an alias for get, weighting
913 of routes, sticky and nostick flags to alter the behavior of stateful
915 Bump __FreeBSD_version to 800078.
918 Do not use Giant for kbdmux(4) locking. This is wrong and
919 apparently causing more problems than it solves. This will
920 re-open the issue where interrupt handlers may race with
921 kbdmux(4) in polling mode. Typical symptoms include (but
922 not limited to) duplicated and/or missing characters when
923 low level console functions (such as gets) are used while
924 interrupts are enabled (for example geli password prompt,
925 mountroot prompt etc.). Disabling kbdmux(4) may help.
928 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
929 kernel modules referencing any of the above need to be recompiled.
930 Bump __FreeBSD_version to 800075.
933 GEOM_PART has become the default partition slicer for storage devices,
934 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
935 introduces some changes:
937 MSDOS/EBR: the devices created from MSDOS extended partition entries
938 (EBR) can be named differently than with GEOM_MBR and are now symlinks
939 to devices with offset-based names. fstabs may need to be modified.
941 BSD: the "geometry does not match label" warning is harmless in most
942 cases but it points to problems in file system misalignment with
943 disk geometry. The "c" partition is now implicit, covers the whole
944 top-level drive and cannot be (mis)used by users.
946 General: Kernel dumps are now not allowed to be written to devices
947 whose partition types indicate they are meant to be used for file
948 systems (or, in case of MSDOS partitions, as something else than
951 Most of these changes date approximately from 200812.
954 The uscanner(4) driver has been removed from the kernel. This follows
955 Linux removing theirs in 2.6 and making libusb the default interface
959 The multicast forwarding code has been cleaned up. netstat(1)
960 only relies on KVM now for printing bandwidth upcall meters.
961 The IPv4 and IPv6 modules are split into ip_mroute_mod and
962 ip6_mroute_mod respectively. The config(5) options for statically
963 compiling this code remain the same, i.e. 'options MROUTING'.
966 Support for the IFF_NEEDSGIANT network interface flag has been
967 removed, which means that non-MPSAFE network device drivers are no
968 longer supported. In particular, if_ar, if_sr, and network device
969 drivers from the old (legacy) USB stack can no longer be built or
973 POSIX.1 Native Language Support (NLS) has been enabled in libc and
974 a bunch of new language catalog files have also been added.
975 This means that some common libc messages are now localized and
976 they depend on the LC_MESSAGES environmental variable.
979 The k8temp(4) driver has been renamed to amdtemp(4) since
980 support for Family 10 and Family 11 CPU families was added.
983 IGMPv3 and Source-Specific Multicast (SSM) have been merged
984 to the IPv4 stack. VIMAGE hooks are in but not yet used.
986 For kernel developers, the most important changes are that the
987 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
988 and this lock has been downgraded to a non-recursive mutex.
990 Transport protocols (UDP, Raw IP) are now responsible for filtering
991 inbound multicast traffic according to group membership and source
992 filters. The imo_multicast_filter() KPI exists for this purpose.
993 Transports which do not use multicast (SCTP, TCP) already reject
994 multicast by default. Forwarding and receive performance may improve
995 as a mutex acquisition is no longer needed in the ip_input()
996 low-level input path. in_addmulti() and in_delmulti() are shimmed
997 to new KPIs which exist to support SSM in-kernel.
999 For application developers, it is recommended that loopback of
1000 multicast datagrams be disabled for best performance, as this
1001 will still cause the lock to be taken for each looped-back
1002 datagram transmission. The net.inet.ip.mcast.loop sysctl may
1003 be tuned to 0 to disable loopback by default; it defaults to 1
1004 to preserve the existing behaviour.
1006 For systems administrators, to obtain best performance with
1007 multicast reception and multiple groups, it is always recommended
1008 that a card with a suitably precise hash filter is used. Hash
1009 collisions will still result in the lock being taken within the
1010 transport protocol input path to check group membership.
1012 If deploying FreeBSD in an environment with IGMP snooping switches,
1013 it is recommended that the net.inet.igmp.sendlocal sysctl remain
1014 enabled; this forces 224.0.0.0/24 group membership to be announced
1017 The size of 'struct igmpstat' has changed; netstat needs to be
1018 recompiled to reflect this.
1019 Bump __FreeBSD_version to 800070.
1022 libusb20.so.1 is now installed as libusb.so.1 and the ports system
1023 updated to use it. This requires a buildworld/installworld in order to
1024 update the library and dependencies (usbconfig, etc). Its advisable to
1025 rebuild all ports which uses libusb. More specific directions are given
1026 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
1027 libusb are no longer required and can be removed.
1030 A workaround is committed to allow the creation of System V shared
1031 memory segment of size > 2 GB on the 64-bit architectures.
1032 Due to a limitation of the existing ABI, the shm_segsz member
1033 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
1034 wrong for large segments. Note that limits must be explicitly
1035 raised to allow such segments to be created.
1038 The layout of struct ifnet has changed, requiring a rebuild of all
1039 network device driver modules.
1042 The /dev handling for the new USB stack has changed, a
1043 buildworld/installworld is required for libusb20.
1046 The new USB2 stack has now been permanently moved in and all kernel and
1047 module names reverted to their previous values (eg, usb, ehci, ohci,
1048 ums, ...). The old usb stack can be compiled in by prefixing the name
1049 with the letter 'o', the old usb modules have been removed.
1050 Updating entry 20090216 for xorg and 20090215 for libmap may still
1054 The rc.conf(5) option if_up_delay has been renamed to
1055 defaultroute_delay to better reflect its purpose. If you have
1056 customized this setting in /etc/rc.conf you need to update it to
1060 xorg 7.4 wants to configure its input devices via hald which does not
1061 yet work with USB2. If the keyboard/mouse does not work in xorg then
1063 Option "AllowEmptyInput" "off"
1064 to your ServerLayout section. This will cause X to use the configured
1065 kbd and mouse sections from your xorg.conf.
1068 The GENERIC kernels for all architectures now default to the new USB2
1069 stack. No kernel config options or code have been removed so if a
1070 problem arises please report it and optionally revert to the old USB
1071 stack. If you are loading USB kernel modules or have a custom kernel
1072 that includes GENERIC then ensure that usb names are also changed over,
1073 eg uftdi -> usb2_serial_ftdi.
1075 Older programs linked against the ports libusb 0.1 need to be
1076 redirected to the new stack's libusb20. /etc/libmap.conf can
1078 # Map old usb library to new one for usb2 stack
1079 libusb-0.1.so.8 libusb20.so.1
1082 All USB ethernet devices now attach as interfaces under the name ueN
1083 (eg. ue0). This is to provide a predictable name as vendors often
1084 change usb chipsets in a product without notice.
1087 The ichsmb(4) driver has been changed to require SMBus slave
1088 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1089 All of the other SMBus controller drivers require left-justified
1090 slave addresses, so this change makes all the drivers provide the
1094 INET6 statistics (struct ip6stat) was updated.
1095 netstat(1) needs to be recompiled.
1098 NTFS has been removed from GENERIC kernel on amd64 to match
1099 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1100 will load ntfs.ko module automatically when NTFS support is
1101 actually needed, unless ntfs.ko is not installed or security
1102 level prohibits loading kernel modules. If either is the case,
1103 "options NTFS" has to be added into kernel config.
1106 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1107 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1108 800061. User space tools that rely on the size of struct tcpcb in
1109 tcp_var.h (e.g. sockstat) need to be recompiled.
1112 ng_tty(4) module updated to match the new TTY subsystem.
1113 Due to API change, user-level applications must be updated.
1114 New API support added to mpd5 CVS and expected to be present
1115 in next mpd5.3 release.
1118 With __FreeBSD_version 800060 the makefs tool is part of
1119 the base system (it was a port).
1122 The afdata and ifnet locks have been changed from mutexes to
1123 rwlocks, network modules will need to be re-compiled.
1126 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1127 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1128 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1129 architecture and 40 bytes on 64-bit architecture. The userland
1130 applications "arp" and "ndp" have been updated accordingly.
1131 The output from "netstat -r" shows only routing entries and
1132 none of the L2 information.
1135 __FreeBSD_version 800057 marks the switchover from the
1136 binary ath hal to source code. Users must add the line:
1138 options AH_SUPPORT_AR5416
1140 to their kernel config files when specifying:
1144 The ath_hal module no longer exists; the code is now compiled
1145 together with the driver in the ath module. It is now
1146 possible to tailor chip support (i.e. reduce the set of chips
1147 and thereby the code size); consult ath_hal(4) for details.
1150 __FreeBSD_version 800054 adds memory barriers to
1151 <machine/atomic.h>, new interfaces to ifnet to facilitate
1152 multiple hardware transmit queues for cards that support
1153 them, and a lock-less ring-buffer implementation to
1154 enable drivers to more efficiently manage queueing of
1158 A new version of ZFS (version 13) has been merged to -HEAD.
1159 This version has zpool attribute "listsnapshots" off by
1160 default, which means "zfs list" does not show snapshots,
1161 and is the same as Solaris behavior.
1164 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1167 The uhci, ohci, ehci and slhci USB Host controller drivers have
1168 been put into separate modules. If you load the usb module
1169 separately through loader.conf you will need to load the
1170 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1171 controller add the following to loader.conf:
1177 The ABI used by the PMC toolset has changed. Please keep
1178 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1182 atapci kernel module now includes only generic PCI ATA
1183 driver. AHCI driver moved to ataahci kernel module.
1184 All vendor-specific code moved into separate kernel modules:
1185 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1186 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1187 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1188 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1191 The TTY subsystem of the kernel has been replaced by a new
1192 implementation, which provides better scalability and an
1193 improved driver model. Most common drivers have been migrated to
1194 the new TTY subsystem, while others have not. The following
1195 drivers have not yet been ported to the new TTY layer:
1198 cy, digi, rc, rp, sio
1204 ng_h4, ng_tty, ppp, sl, snp
1206 Adding these drivers to your kernel configuration file shall
1207 cause compilation to fail.
1210 ntpd has been upgraded to 4.2.4p5.
1213 OpenSSH has been upgraded to 5.1p1.
1215 For many years, FreeBSD's version of OpenSSH preferred DSA
1216 over RSA for host and user authentication keys. With this
1217 upgrade, we've switched to the vendor's default of RSA over
1218 DSA. This may cause upgraded clients to warn about unknown
1219 host keys even for previously known hosts. Users should
1220 follow the usual procedure for verifying host keys before
1221 accepting the RSA key.
1223 This can be circumvented by setting the "HostKeyAlgorithms"
1224 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1227 Please note that the sequence of keys offered for
1228 authentication has been changed as well. You may want to
1229 specify IdentityFile in a different order to revert this
1233 The sio(4) driver has been removed from the i386 and amd64
1234 kernel configuration files. This means uart(4) is now the
1235 default serial port driver on those platforms as well.
1237 To prevent collisions with the sio(4) driver, the uart(4) driver
1238 uses different names for its device nodes. This means the
1239 onboard serial port will now most likely be called "ttyu0"
1240 instead of "ttyd0". You may need to reconfigure applications to
1241 use the new device names.
1243 When using the serial port as a boot console, be sure to update
1244 /boot/device.hints and /etc/ttys before booting the new kernel.
1245 If you forget to do so, you can still manually specify the hints
1246 at the loader prompt:
1248 set hint.uart.0.at="isa"
1249 set hint.uart.0.port="0x3F8"
1250 set hint.uart.0.flags="0x10"
1251 set hint.uart.0.irq="4"
1255 The gpt(8) utility has been removed. Use gpart(8) to partition
1259 The version that Linuxulator emulates was changed from 2.4.2
1260 to 2.6.16. If you experience any problems with Linux binaries
1261 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1262 if it fixes the problem contact emulation mailing list.
1265 ISDN4BSD (I4B) was removed from the src tree. You may need to
1266 update a your kernel configuration and remove relevant entries.
1269 I have checked in code to support multiple routing tables.
1270 See the man pages setfib(1) and setfib(2).
1271 This is a hopefully backwards compatible version,
1272 but to make use of it you need to compile your kernel
1273 with options ROUTETABLES=2 (or more up to 16).
1276 The 802.11 wireless support was redone to enable multi-bss
1277 operation on devices that are capable. The underlying device
1278 is no longer used directly but instead wlanX devices are
1279 cloned with ifconfig. This requires changes to rc.conf files.
1280 For example, change:
1281 ifconfig_ath0="WPA DHCP"
1284 ifconfig_wlan0="WPA DHCP"
1285 see rc.conf(5) for more details. In addition, mergemaster of
1286 /etc/rc.d is highly recommended. Simultaneous update of userland
1287 and kernel wouldn't hurt either.
1289 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1290 modules were merged into the base wlan module. All references
1291 to these modules (e.g. in kernel config files) must be removed.
1294 psm(4) has gained write(2) support in native operation level.
1295 Arbitrary commands can be written to /dev/psm%d and status can
1296 be read back from it. Therefore, an application is responsible
1297 for status validation and error recovery. It is a no-op in
1298 other operation levels.
1301 Support for KSE threading has been removed from the kernel. To
1302 run legacy applications linked against KSE libmap.conf may
1303 be used. The following libmap.conf may be used to ensure
1304 compatibility with any prior release:
1306 libpthread.so.1 libthr.so.1
1307 libpthread.so.2 libthr.so.2
1308 libkse.so.3 libthr.so.3
1311 The layout of struct vmspace has changed. This affects libkvm
1312 and any executables that link against libkvm and use the
1313 kvm_getprocs() function. In particular, but not exclusively,
1314 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1315 The effects are minimal, but it's advisable to upgrade world
1319 The latest em driver no longer has support in it for the
1320 82575 adapter, this is now moved to the igb driver. The
1321 split was done to make new features that are incompatible
1322 with older hardware easier to do.
1325 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1326 likewise the kernel option is now GEOM_LINUX_LVM.
1329 The default NFS mount mode has changed from UDP to TCP for
1330 increased reliability. If you rely on (insecurely) NFS
1331 mounting across a firewall you may need to update your
1335 Belatedly note the addition of m_collapse for compacting
1339 The fts(3) structures have been changed to use adequate
1340 integer types for their members and so to be able to cope
1341 with huge file trees. The old fts(3) ABI is preserved
1342 through symbol versioning in libc, so third-party binaries
1343 using fts(3) should still work, although they will not take
1344 advantage of the extended types. At the same time, some
1345 third-party software might fail to build after this change
1346 due to unportable assumptions made in its source code about
1347 fts(3) structure members. Such software should be fixed
1348 by its vendor or, in the worst case, in the ports tree.
1349 FreeBSD_version 800015 marks this change for the unlikely
1350 case that a portable fix is impossible.
1353 To upgrade to -current after this date, you must be running
1354 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1355 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1358 The ADAPTIVE_GIANT kernel option has been retired because its
1359 functionality is the default now.
1362 The AT keyboard emulation of sunkbd(4) has been turned on
1363 by default. In order to make the special symbols of the Sun
1364 keyboards driven by sunkbd(4) work under X these now have
1365 to be configured the same way as Sun USB keyboards driven
1366 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1368 Option "XkbLayout" "us"
1369 Option "XkbRules" "xorg"
1370 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1373 It has been decided that it is desirable to provide ABI
1374 backwards compatibility to the FreeBSD 4/5/6 versions of the
1375 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1376 broken with the introduction of PCI domain support (see the
1377 20070930 entry). Unfortunately, this required the ABI of
1378 PCIOCGETCONF to be broken again in order to be able to
1379 provide backwards compatibility to the old version of that
1380 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1381 again. As for prominent ports this affects neither pciutils
1382 nor xorg-server this time, the hal port needs to be rebuilt
1386 The misnamed kthread_create() and friends have been renamed
1387 to kproc_create() etc. Many of the callers already
1388 used kproc_start()..
1389 I will return kthread_create() and friends in a while
1390 with implementations that actually create threads, not procs.
1391 Renaming corresponds with version 800002.
1400 Avoid using make -j when upgrading. While generally safe, there are
1401 sometimes problems using -j to upgrade. If your upgrade fails with
1402 -j, please try again without -j. From time to time in the past there
1403 have been problems using -j with buildworld and/or installworld. This
1404 is especially true when upgrading between "distant" versions (eg one
1405 that cross a major release boundary or several minor releases, or when
1406 several months have passed on the -current branch).
1408 Sometimes, obscure build problems are the result of environment
1409 poisoning. This can happen because the make utility reads its
1410 environment when searching for values for global variables. To run
1411 your build attempts in an "environmental clean room", prefix all make
1412 commands with 'env -i '. See the env(1) manual page for more details.
1414 When upgrading from one major version to another it is generally best
1415 to upgrade to the latest code in the currently installed branch first,
1416 then do an upgrade to the new branch. This is the best-tested upgrade
1417 path, and has the highest probability of being successful. Please try
1418 this approach before reporting problems with a major version upgrade.
1422 When upgrading the boot ZFS pool to a new version, always follow
1425 1.) recompile and reinstall the ZFS boot loader and boot block
1426 (this is part of "make buildworld" and "make installworld")
1428 2.) update the ZFS boot block on your boot drive
1430 The following example updates the ZFS boot block on the first
1431 partition (freebsd-boot) of a GPT partitioned drive ad0:
1432 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1434 Non-boot pools do not need these updates.
1438 If you are updating from a prior version of FreeBSD (even one just
1439 a few days old), you should follow this procedure. It is the most
1440 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1442 make kernel-toolchain
1443 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1444 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1446 To test a kernel once
1447 ---------------------
1448 If you just want to boot a kernel once (because you are not sure
1449 if it works, or if you want to boot a known bad kernel to provide
1450 debugging information) run
1451 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1452 nextboot -k testkernel
1454 To just build a kernel when you know that it won't mess you up
1455 --------------------------------------------------------------
1456 This assumes you are already running a CURRENT system. Replace
1457 ${arch} with the architecture of your machine (e.g. "i386",
1458 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1460 cd src/sys/${arch}/conf
1461 config KERNEL_NAME_HERE
1462 cd ../compile/KERNEL_NAME_HERE
1467 If this fails, go to the "To build a kernel" section.
1469 To rebuild everything and install it on the current system.
1470 -----------------------------------------------------------
1471 # Note: sometimes if you are running current you gotta do more than
1472 # is listed here if you are upgrading from a really old current.
1474 <make sure you have good level 0 dumps>
1476 make kernel KERNCONF=YOUR_KERNEL_HERE
1478 <reboot in single user> [3]
1486 To cross-install current onto a separate partition
1487 --------------------------------------------------
1488 # In this approach we use a separate partition to hold
1489 # current's root, 'usr', and 'var' directories. A partition
1490 # holding "/", "/usr" and "/var" should be about 2GB in
1493 <make sure you have good level 0 dumps>
1496 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1497 <maybe newfs current's root partition>
1498 <mount current's root partition on directory ${CURRENT_ROOT}>
1499 make installworld DESTDIR=${CURRENT_ROOT}
1500 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1501 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1502 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1503 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1504 <reboot into current>
1505 <do a "native" rebuild/install as described in the previous section>
1506 <maybe install compatibility libraries from ports/misc/compat*>
1510 To upgrade in-place from 8.x-stable to current
1511 ----------------------------------------------
1512 <make sure you have good level 0 dumps>
1514 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1516 <reboot in single user> [3]
1523 Make sure that you've read the UPDATING file to understand the
1524 tweaks to various things you need. At this point in the life
1525 cycle of current, things change often and you are on your own
1526 to cope. The defaults can also change, so please read ALL of
1527 the UPDATING entries.
1529 Also, if you are tracking -current, you must be subscribed to
1530 freebsd-current@freebsd.org. Make sure that before you update
1531 your sources that you have read and understood all the recent
1532 messages there. If in doubt, please track -stable which has
1533 much fewer pitfalls.
1535 [1] If you have third party modules, such as vmware, you
1536 should disable them at this point so they don't crash your
1539 [3] From the bootblocks, boot -s, and then do
1544 adjkerntz -i # if CMOS is wall time
1545 Also, when doing a major release upgrade, it is required that
1546 you boot into single user mode to do the installworld.
1548 [4] Note: This step is non-optional. Failure to do this step
1549 can result in a significant reduction in the functionality of the
1550 system. Attempting to do it by hand is not recommended and those
1551 that pursue this avenue should read this file carefully, as well
1552 as the archives of freebsd-current and freebsd-hackers mailing lists
1553 for potential gotchas. The -U option is also useful to consider.
1554 See mergemaster(8) for more information.
1556 [5] Usually this step is a noop. However, from time to time
1557 you may need to do this if you get unknown user in the following
1558 step. It never hurts to do it all the time. You may need to
1559 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1560 install) after the buildworld before this step if you last updated
1561 from current before 20020224 or from -stable before 20020408.
1563 [6] This only deletes old files and directories. Old libraries
1564 can be deleted by "make delete-old-libs", but you have to make
1565 sure that no program is using those libraries anymore.
1567 [8] In order to have a kernel that can run the 4.x binaries needed to
1568 do an installworld, you must include the COMPAT_FREEBSD4 option in
1569 your kernel. Failure to do so may leave you with a system that is
1570 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1571 required to run the 5.x binaries on more recent kernels. And so on
1572 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1574 Make sure that you merge any new devices from GENERIC since the
1575 last time you updated your kernel config file.
1577 [9] When checking out sources, you must include the -P flag to have
1578 cvs prune empty directories.
1580 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1581 "?=" instead of the "=" assignment operator, so that buildworld can
1582 override the CPUTYPE if it needs to.
1584 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1585 not on the command line, or in /etc/make.conf. buildworld will
1586 warn if it is improperly defined.
1589 This file contains a list, in reverse chronological order, of major
1590 breakages in tracking -current. It is not guaranteed to be a complete
1591 list of such breakages, and only contains entries since October 10, 2007.
1592 If you need to see UPDATING entries from before that date, you will need
1593 to fetch an UPDATING file from an older FreeBSD release.
1595 Copyright information:
1597 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1599 Redistribution, publication, translation and use, with or without
1600 modification, in full or in part, in any form or format of this
1601 document are permitted without further permission from the author.
1603 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1604 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1605 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1606 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1607 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1608 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1609 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1610 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1611 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1612 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1613 POSSIBILITY OF SUCH DAMAGE.
1615 Contact Warner Losh if you have any questions about your use of