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 All the non-MPSAFE filesystems have been disconnected from
29 the build. The full list includes: codafs, hpfs, ntfs, nwfs,
33 The interface cloning API and ABI has changed. The following
34 modules need to be recompiled together with kernel:
35 ipfw(4), pfsync(4), pflog(4), usb(4), wlan(4), stf(4),
36 vlan(4), disc(4), edsc(4), if_bridge(4), gif(4), tap(4),
37 faith(4), epair(4), enc(4), tun(4), if_lagg(4), gre(4).
40 The sdhci driver was split in two parts: sdhci (generic SD Host
41 Controller logic) and sdhci_pci (actual hardware driver).
42 No kernel config modifications are required, but if you
43 load sdhc as a module you must switch to sdhci_pci instead.
46 Import the FUSE kernel and userland support into base system.
49 The GNU sort(1) program has been removed since the BSD-licensed
50 sort(1) has been the default for quite some time and no serious
51 problems have been reported. The corresponding WITH_GNU_SORT
55 The pfil(9) API/ABI for AF_INET family has been changed. Packet
56 filtering modules: pf(4), ipfw(4), ipfilter(4) need to be recompiled
60 The net80211(4) ABI has been changed to allow for improved driver
61 PS-POLL and power-save support. All wireless drivers need to be
62 recompiled to work with the new kernel.
65 The random(4) support for the VIA hardware random number
66 generator (`PADLOCK') is no longer enabled unconditionally.
67 Add the PADLOCK_RNG option in the custom kernel config if
68 needed. The GENERIC kernels on i386 and amd64 do include the
69 option, so the change only affects the custom kernel
73 The pf(4) packet filter ABI has been changed. pfctl(8) and
74 snmp_pf module need to be recompiled to work with new kernel.
77 A new ZFS feature flag "com.delphix:empty_bpobj" has been merged
78 to -HEAD. Pools that have empty_bpobj in active state can not be
79 imported read-write with ZFS implementations that do not support
80 this feature. For more information read the zpool-features(5)
84 The sparc64 ZFS loader has been changed to no longer try to auto-
85 detect ZFS providers based on diskN aliases but now requires these
86 to be explicitly listed in the OFW boot-device environment variable.
89 The OpenSSL has been upgraded to 1.0.1c. Any binaries requiring
90 libcrypto.so.6 or libssl.so.6 must be recompiled. Also, there are
91 configuration changes. Make sure to merge /etc/ssl/openssl.cnf.
94 The following sysctls and tunables have been renamed for consistency
96 kern.cam.da.da_send_ordered -> kern.cam.da.send_ordered
97 kern.cam.ada.ada_send_ordered -> kern.cam.ada.send_ordered
100 The sort utility has been replaced with BSD sort. For now, GNU sort
101 is also available as "gnusort" or the default can be set back to
102 GNU sort by setting WITH_GNU_SORT. In this case, BSD sort will be
103 installed as "bsdsort".
106 A new version of ZFS (pool version 5000) has been merged to -HEAD.
107 Starting with this version the old system of ZFS pool versioning
108 is superseded by "feature flags". This concept enables forward
109 compatibility against certain future changes in functionality of ZFS
110 pools. The first read-only compatible "feature flag" for ZFS pools
111 is named "com.delphix:async_destroy". For more information
112 read the new zpool-features(5) manual page.
113 Please refer to the "ZFS notes" section of this file for information
114 on upgrading boot ZFS pools.
117 The malloc(3) implementation embedded in libc now uses sources imported
118 as contrib/jemalloc. The most disruptive API change is to
119 /etc/malloc.conf. If your system has an old-style /etc/malloc.conf,
120 delete it prior to installworld, and optionally re-create it using the
121 new format after rebooting. See malloc.conf(5) for details
122 (specifically the TUNING section and the "opt.*" entries in the MALLCTL
126 Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb
127 is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is
128 now spelled mips. This is to aid compatibility with third-party
129 software that expects this naming scheme in uname(3). Little-endian
130 settings are unchanged.
133 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
137 Now unix domain sockets behave "as expected" on nullfs(5). Previously
138 nullfs(5) did not pass through all behaviours to the underlying layer,
139 as a result if we bound to a socket on the lower layer we could connect
140 only to the lower path; if we bound to the upper layer we could connect
141 only to the upper path. The new behavior is one can connect to both the
142 lower and the upper paths regardless what layer path one binds to.
145 The getifaddrs upgrade path broken with 20111215 has been restored.
146 If you have upgraded in between 20111215 and 20120209 you need to
147 recompile libc again with your kernel. You still need to recompile
148 world to be able to configure CARP but this restriction already
152 The set_rcvar() function has been removed from /etc/rc.subr. All
153 base and ports rc.d scripts have been updated, so if you have a
154 port installed with a script in /usr/local/etc/rc.d you can either
155 hand-edit the rcvar= line, or reinstall the port.
157 An easy way to handle the mass-update of /etc/rc.d:
158 rm /etc/rc.d/* && mergemaster -i
161 panic(9) now stops other CPUs in the SMP systems, disables interrupts
162 on the current CPU and prevents other threads from running.
163 This behavior can be reverted using the kern.stop_scheduler_on_panic
165 The new behavior can be incompatible with kern.sync_on_panic.
168 The carp(4) facility has been changed significantly. Configuration
169 of the CARP protocol via ifconfig(8) has changed, as well as format
170 of CARP events submitted to devd(8) has changed. See manual pages
171 for more information. The arpbalance feature of carp(4) is currently
172 not supported anymore.
174 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
175 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
176 need to be recompiled.
179 The acpi_wmi(4) status device /dev/wmistat has been renamed to
183 The option VFS_ALLOW_NONMPSAFE option has been added in order to
184 explicitely support non-MPSAFE filesystems.
185 It is on by default for all supported platform at this present
189 The broken amd(4) driver has been replaced with esp(4) in the amd64,
190 i386 and pc98 GENERIC kernel configuration files.
193 sysinstall has been removed
196 The stable/9 branch created in subversion. This corresponds to the
197 RELENG_9 branch in CVS.
200 This commit modifies vfs_register() so that it uses a hash
201 calculation to set vfc_typenum, which is enabled by default.
202 The first time a system is booted after this change, the
203 vfc_typenum values will change for all file systems. The
204 main effect of this is a change to the NFS server file handles
205 for file systems that use vfc_typenum in their fsid, such as ZFS.
206 It will, however, prevent vfc_typenum from changing when file
207 systems are loaded in a different order for subsequent reboots.
208 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
209 until you are ready to remount all NFS clients after a reboot.
212 Bump the shared library version numbers for libraries that
213 do not use symbol versioning, have changed the ABI compared
214 to stable/8 and which shared library version was not bumped.
215 Done as part of 9.0-RELEASE cycle.
218 During the merge of Capsicum features, the fget(9) KPI was modified.
219 This may require the rebuilding of out-of-tree device drivers --
220 issues have been reported specifically with the nVidia device driver.
221 __FreeBSD_version is bumped to 900041.
223 Also, there is a period between 20110811 and 20110814 where the
224 special devices /dev/{stdin,stdout,stderr} did not work correctly.
225 Building world from a kernel during that window may not work.
228 The packet filter (pf) code has been updated to OpenBSD 4.5.
229 You need to update userland tools to be in sync with kernel.
230 This update breaks backward compatibility with earlier pfsync(4)
231 versions. Care must be taken when updating redundant firewall setups.
234 The following sysctls and tunables are retired on x86 platforms:
236 machdep.hlt_logical_cpus
237 The following sysctl is retired:
238 machdep.hyperthreading_allowed
239 The sysctls were supposed to provide a way to dynamically offline and
240 online selected CPUs on x86 platforms, but the implementation has not
241 been reliable especially with SCHED_ULE scheduler.
242 machdep.hyperthreading_allowed tunable is still available to ignore
243 hyperthreading CPUs at OS level.
244 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
245 where X is an APIC ID of a CPU. Be advised, though, that disabling
246 CPUs in non-uniform fashion will result in non-uniform topology and
247 may lead to sub-optimal system performance with SCHED_ULE, which is
251 cpumask_t type is retired and cpuset_t is used in order to describe
255 Changes to ifconfig(8) for dynamic address family detection mandate
256 that you are running a kernel of 20110525 or later. Make sure to
257 follow the update procedure to boot a new kernel before installing
261 Support for sun4v architecture is officially dropped
264 Several KPI breaking changes have been committed to the mii(4) layer,
265 the PHY drivers and consequently some Ethernet drivers using mii(4).
266 This means that miibus.ko and the modules of the affected Ethernet
267 drivers need to be recompiled.
269 Note to kernel developers: Given that the OUI bit reversion problem
270 was fixed as part of these changes all mii(4) commits related to OUIs,
271 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
272 handling, no longer can be merged verbatim to stable/8 and previous
276 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
277 into their kernel configurations along with 'device pci'.
280 The default NFS client is now the new NFS client, so fstype "newnfs"
281 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
282 Although mounts via fstype "nfs" will usually work without userland
283 changes, it is recommended that the mount(8) and mount_nfs(8)
284 commands be rebuilt from sources and that a link to mount_nfs called
285 mount_oldnfs be created. The new client is compiled into the
286 kernel with "options NFSCL" and this is needed for diskless root
287 file systems. The GENERIC kernel configs have been changed to use
288 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
289 To use the regular/old client, you can "mount -t oldnfs ...". For
290 a diskless root file system, you must also include a line like:
292 vfs.root.mountfrom="oldnfs:"
294 in the boot/loader.conf on the root fs on the NFS server to make
295 a diskless root fs use the old client.
298 The GENERIC kernels for all architectures now default to the new
299 CAM-based ATA stack. It means that all legacy ATA drivers were
300 removed and replaced by respective CAM drivers. If you are using
301 ATA device names in /etc/fstab or other places, make sure to update
302 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
303 where 'Y's are the sequential numbers starting from zero for each type
304 in order of detection, unless configured otherwise with tunables,
305 see cam(4)). There will be symbolic links created in /dev/ to map
306 old adX devices to the respective adaY. They should provide basic
307 compatibility for file systems mounting in most cases, but they do
308 not support old user-level APIs and do not have respective providers
309 in GEOM. Consider using updated management tools with new device names.
311 It is possible to load devices ahci, ata, siis and mvs as modules,
312 but option ATA_CAM should remain in kernel configuration to make ata
313 module work as CAM driver supporting legacy ATA controllers. Device ata
314 still can be used in modular fashion (atacore + ...). Modules atadisk
315 and atapi* are not used and won't affect operation in ATA_CAM mode.
316 Note that to use CAM-based ATA kernel should include CAM devices
317 scbus, pass, da (or explicitly ada), cd and optionally others. All of
318 them are parts of the cam module.
320 ataraid(4) functionality is now supported by the RAID GEOM class.
321 To use it you can load geom_raid kernel module and use graid(8) tool
322 for management. Instead of /dev/arX device names, use /dev/raid/rX.
324 No kernel config options or code have been removed, so if a problem
325 arises, please report it and optionally revert to the old ATA stack.
326 In order to do it you can remove from the kernel config:
331 , and instead add back:
332 device atadisk # ATA disk drives
333 device ataraid # ATA RAID drives
334 device atapicd # ATAPI CDROM drives
335 device atapifd # ATAPI floppy drives
336 device atapist # ATAPI tape drives
339 The default NFS server has been changed to the new server, which
340 was referred to as the experimental server. If you need to switch
341 back to the old NFS server, you must now put the "-o" option on
342 both the mountd and nfsd commands. This can be done using the
343 mountd_flags and nfs_server_flags rc.conf variables until an
344 update to the rc scripts is committed, which is coming soon.
347 The GNU Objective-C runtime library (libobjc), and other Objective-C
348 related components have been removed from the base system. If you
349 require an Objective-C library, please use one of the available ports.
352 ath(4) has been split into bus- and device- modules. if_ath contains
353 the HAL, the TX rate control and the network device code. if_ath_pci
354 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
355 contains the AHB glue. Users need to load both if_ath_pci and if_ath
356 in order to use ath on everything else.
358 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
359 need to load if_ath and if_ath_pci for ath(4) operation.
362 As part of the replacement of sysinstall, the process of building
363 release media has changed significantly. For details, please re-read
364 release(7), which has been updated to reflect the new build process.
367 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
368 is the last available version under GPLv2. It brings a number of new
369 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
370 4.1 and SSE 4.2), better support for powerpc64, a number of new
371 directives, and lots of other small improvements. See the ChangeLog
372 file in contrib/binutils for the full details.
375 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
376 compliant, and will now use half of hash for authentication.
377 This will break interoperability with all stacks (including all
378 actual FreeBSD versions) who implement
379 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
381 The only workaround with such peers is to use another HMAC
382 algorithm for IPsec ("phase 2") authentication.
385 Remove the uio_yield prototype and symbol. This function has
386 been misnamed since it was introduced and should not be
387 globally exposed with this name. The equivalent functionality
388 is now available using kern_yield(curthread->td_user_pri).
389 The function remains undocumented.
392 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
393 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
394 for scalar sysctls is defined but disabled. Code that needs
395 UQUAD to pass the type checking that must compile on older
396 systems where the define is not present can check against
397 __FreeBSD_version >= 900030.
399 The system dialog(1) has been replaced with a new version previously
400 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
401 with the previous version, but the libdialog associated with it has
402 a largely incompatible API. As such, the original version of libdialog
403 will be kept temporarily as libodialog, until its base system consumers
404 are replaced or updated. Bump __FreeBSD_version to 900030.
407 If you are trying to run make universe on a -stable system, and you get
408 the following warning:
409 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
410 unknown. config(8) likely too old."
411 or something similar to it, then you must upgrade your -stable system
412 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
413 stable/8) or build the config from the latest stable/8 branch and
414 install it on your system.
416 Prior to this date, building a current universe on 8-stable system from
417 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
418 error in the first kernel build phase. A new config on those old
419 systems will fix that problem for older versions of -current.
422 The TCP stack has been modified to allow Khelp modules to interact with
423 it via helper hook points and store per-connection data in the TCP
424 control block. Bump __FreeBSD_version to 900029. User space tools that
425 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
429 Generic IEEE 802.3 annex 31B full duplex flow control support has been
430 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
431 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
432 to take advantage of it instead of using custom implementations. This
433 means that these drivers now no longer unconditionally advertise
434 support for flow control but only do so if flow control is a selected
435 media option. This was implemented in the generic support that way in
436 order to allow flow control to be switched on and off via ifconfig(8)
437 with the PHY specific default to typically off in order to protect
438 from unwanted effects. Consequently, if you used flow control with
439 one of the above mentioned drivers you now need to explicitly enable
441 ifconfig bge0 media auto mediaopt flowcontrol
443 Along with the above mentioned changes generic support for setting
444 1000baseT master mode also has been added and brgphy(4), ciphy(4),
445 e1000phy(4) as well as ip1000phy(4) have been converted to take
446 advantage of it. This means that these drivers now no longer take the
447 link0 parameter for selecting master mode but the master media option
448 has to be used instead, for example like in the following:
449 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
451 Selection of master mode now is also available with all other PHY
452 drivers supporting 1000baseT.
455 The TCP stack has received a significant update to add support for
456 modularised congestion control and generally improve the clarity of
457 congestion control decisions. Bump __FreeBSD_version to 900025. User
458 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
459 sockstat) need to be recompiled.
462 The man(1) utility has been replaced by a new version that no longer
463 uses /etc/manpath.config. Please consult man.conf(5) for how to
464 migrate local entries to the new format.
467 The copyright strings printed by login(1) and sshd(8) at the time of a
468 new connection have been removed to follow other operating systems and
472 A workaround for a fixed ld bug has been removed in kernel code,
473 so make sure that your system ld is built from sources after
474 revision 210245 from 2010-07-19 (r211583 if building head kernel
475 on stable/8, r211584 for stable/7; both from 2010-08-21).
476 A symptom of incorrect ld version is different addresses for
477 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
480 The $ipv6_prefer variable in rc.conf(5) has been split into
481 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
483 The $ip6addrctl_policy is a variable to choose a pre-defined
484 address selection policy set by ip6addrctl(8). A value
485 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
488 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
489 flag (see an entry of 20090926) is set on an interface with no
490 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
491 security reason. If you want IPv6 link-local address on all
492 interfaces by default, set this to "YES".
494 The old ipv6_prefer="YES" is equivalent to
495 ipv6_activate_all_interfaces="YES" and
496 ip6addrctl_policy="ipv6_prefer".
499 DTrace has grown support for userland tracing. Due to this, DTrace is
500 now i386 and amd64 only.
501 dtruss(1) is now installed by default on those systems and a new
502 kernel module is needed for userland tracing: fasttrap.
503 No changes to your kernel config file are necessary to enable
504 userland tracing, but you might consider adding 'STRIP=' and
505 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
506 to have informative userland stack traces in DTrace (ustack).
509 The acpi_aiboost(4) driver has been removed in favor of the new
510 aibs(4) driver. You should update your kernel configuration file.
513 BSD grep has been imported to the base system and it is built by
514 default. It is completely BSD licensed, highly GNU-compatible, uses
515 less memory than its GNU counterpart and has a small codebase.
516 However, it is slower than its GNU counterpart, which is mostly
517 noticeable for larger searches, for smaller ones it is measurable
518 but not significant. The reason is complex, the most important factor
519 is that we lack a modern and efficient regex library and GNU
520 overcomes this by optimizing the searches internally. Future work
521 on improving the regex performance is planned, for the meantime,
522 users that need better performance, can build GNU grep instead by
523 setting the WITH_GNU_GREP knob.
526 Due to the import of powerpc64 support, all existing powerpc kernel
527 configuration files must be updated with a machine directive like this:
528 machine powerpc powerpc
530 In addition, an updated config(8) is required to build powerpc kernels
534 A new version of ZFS (version 15) has been merged to -HEAD.
535 This version uses a python library for the following subcommands:
536 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
537 For full functionality of these commands the following port must
538 be installed: sysutils/py-zfs
541 'vm_page's are now hashed by physical address to an array of mutexes.
542 Currently this is only used to serialize access to hold_count. Over
543 time the page queue mutex will be peeled away. This changes the size
544 of pmap on every architecture. And requires all callers of vm_page_hold
545 and vm_page_unhold to be updated.
548 WITH_CTF can now be specified in src.conf (not recommended, there
549 are some problems with static executables), make.conf (would also
550 affect ports which do not use GNU make and do not override the
551 compile targets) or in the kernel config (via "makeoptions
553 When WITH_CTF was specified there before this was silently ignored,
554 so make sure that WITH_CTF is not used in places which could lead
555 to unwanted behavior.
558 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
559 to allow 32-bit compatibility on non-x86 platforms. All kernel
560 configurations on amd64 and ia64 platforms using these options must
561 be modified accordingly.
564 The utmp user accounting database has been replaced with utmpx,
565 the user accounting interface standardized by POSIX.
566 Unfortunately the semantics of utmp and utmpx don't match,
567 making it practically impossible to support both interfaces.
568 The user accounting database is used by tools like finger(1),
569 last(1), talk(1), w(1) and ac(8).
571 All applications in the base system use utmpx. This means only
572 local binaries (e.g. from the ports tree) may still use these
573 utmp database files. These applications must be rebuilt to make
576 After the system has been upgraded, it is safe to remove the old
577 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
578 assuming their contents is of no importance anymore. Old wtmp
579 databases can only be used by last(1) and ac(8) after they have
580 been converted to the new format using wtmpcvt(1).
583 Introduce the kernel thread "deadlock resolver" (which can be enabled
584 via the DEADLKRES option, see NOTES for more details) and the
585 sleepq_type() function for sleepqueues.
588 The rc.firewall and rc.firewall6 were unified, and
589 rc.firewall6 and rc.d/ip6fw were removed.
590 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
591 variables are obsoleted. Instead, the following new rc
592 variables are added to rc.d/ipfw:
594 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
595 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
596 firewall_simple_onet_ipv6, firewall_trusted_ipv6
598 The meanings correspond to the relevant IPv4 variables.
604 The default terminal emulation for syscons(4) has been changed
605 from cons25 to xterm on all platforms except pc98. This means
606 that the /etc/ttys file needs to be updated to ensure correct
607 operation of applications on the console.
609 The terminal emulation style can be toggled per window by using
610 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
611 options can be used to change the compile-time default back to
614 To prevent graphical artifacts, make sure the TERM environment
615 variable is set to match the terminal emulation that is being
616 performed by syscons(4).
619 The layout of the structure ieee80211req_scan_result has changed.
620 Applications that require wireless scan results (e.g. ifconfig(8))
621 from net80211 need to be recompiled.
623 Applications such as wpa_supplicant(8) may require a full world
624 build without using NO_CLEAN in order to get synchronized with the
628 The iwn(4) driver has been updated to support the 5000 and 5150 series.
629 There's one kernel module for each firmware. Adding "device iwnfw"
630 to the kernel configuration file means including all three firmware
631 images inside the kernel. If you want to include just the one for
632 your wireless card, use the devices iwn4965fw, iwn5000fw or
636 The rc.d/network_ipv6, IPv6 configuration script has been integrated
637 into rc.d/netif. The changes are the following:
639 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
640 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
641 Note that both variables need the "inet6" keyword at the head.
643 Do not set $ipv6_network_interfaces manually if you do not
644 understand what you are doing. It is not needed in most cases.
646 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
649 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
650 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
652 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
653 all configured interfaces have "inet6 accept_rtadv" in the
654 $ifconfig_IF_ipv6. These are for backward compatibility.
656 3. A new variable $ipv6_prefer has been added. If NO, IPv6
657 functionality of interfaces with no corresponding
658 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
659 and the default address selection policy of ip6addrctl(8)
660 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
661 Note that if you want to configure IPv6 functionality on the
662 disabled interfaces after boot, first you need to clear the flag by
663 using ifconfig(8) like:
665 ifconfig em0 inet6 -ifdisabled
667 If YES, the default address selection policy is set as
670 The default value of $ipv6_prefer is NO.
672 4. If your system need to receive Router Advertisement messages,
673 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
674 scripts automatically invoke rtsol(8) when the interface becomes
675 UP. The Router Advertisement messages are used for SLAAC
676 (State-Less Address AutoConfiguration).
679 802.11s D3.03 support was committed. This is incompatible with the
680 previous code, which was based on D3.0.
683 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
684 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
685 control whether accepting Router Advertisement messages or not.
686 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
687 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
688 The ifconfig(8) utility now supports these flags.
691 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
692 mount(8) and -a option for df(1) to see them.
695 The old tunable hw.bus.devctl_disable has been superseded by
696 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
697 replaced by hw.bus.devctl_queue=0. The default for this new tunable
701 Remove the option STOP_NMI. The default action is now to use NMI only
702 for KDB via the newly introduced function stop_cpus_hard() and
703 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
706 The stable/8 branch created in subversion. This corresponds to the
707 RELENG_8 branch in CVS.
710 Bump the shared library version numbers for all libraries that do not
711 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
712 __FreeBSD_version to 800105.
715 Due to changes in the implementation of virtual network stack support,
716 all network-related kernel modules must be recompiled. As this change
717 breaks the ABI, bump __FreeBSD_version to 800104.
720 The TOE interface to the TCP syncache has been modified to remove
721 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
722 The cxgb driver is the only TOE consumer affected by this change, and
723 needs to be recompiled along with the kernel. As this change breaks
724 the ABI, bump __FreeBSD_version to 800103.
727 Padding has been added to struct tcpcb, sackhint and tcpstat in
728 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
729 maintaining the ABI. However, this change breaks the ABI, so bump
730 __FreeBSD_version to 800102. User space tools that rely on the size of
731 any of these structs (e.g. sockstat) need to be recompiled.
734 The NFS_LEGACYRPC option has been removed along with the old kernel
735 RPC implementation that this option selected. Kernel configurations
736 may need to be adjusted.
739 The network interface device nodes at /dev/net/<interface> have been
740 removed. All ioctl operations can be performed the normal way using
741 routing sockets. The kqueue functionality can generally be replaced
742 with routing sockets.
745 The documentation from the FreeBSD Documentation Project (Handbook,
746 FAQ, etc.) is now installed via packages by sysinstall(8) and under
747 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
750 The ABI of various structures related to the SYSV IPC API have been
751 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
752 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
756 Layout of struct vnet has changed as routing related variables were
757 moved to their own Vimage module. Modules need to be recompiled. Bump
758 __FreeBSD_version to 800099.
761 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
762 respectively. As long as no more than 16 groups per process are used,
763 no changes should be visible. When more than 16 groups are used, old
764 binaries may fail if they call getgroups() or getgrouplist() with
765 statically sized storage. Recompiling will work around this, but
766 applications should be modified to use dynamically allocated storage
767 for group arrays as POSIX.1-2008 does not cap an implementation's
768 number of supported groups at NGROUPS_MAX+1 as previous versions did.
770 NFS and portalfs mounts may also be affected as the list of groups is
771 truncated to 16. Users of NFS who use more than 16 groups, should
772 take care that negative group permissions are not used on the exported
773 file systems as they will not be reliable unless a GSSAPI based
774 authentication method is used.
777 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
778 option compiles in the support for adaptive spinning for lockmgrs
779 which want to enable it. The lockinit() function now accepts the flag
780 LK_ADAPTIVE in order to make the lock object subject to adaptive
781 spinning when both held in write and read mode.
784 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
785 changed. User applications that use this ioctl need to be rebuilt.
788 The layout of struct thread has changed. Kernel and modules need to
792 The layout of structs ifnet, domain, protosw and vnet_net has changed.
793 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
796 window(1) has been removed from the base system. It can now be
797 installed from ports. The port is called misc/window.
800 The way we are storing and accessing `routing table' entries has
801 changed. Programs reading the FIB, like netstat, need to be
805 A new netisr implementation has been added for FreeBSD 8. Network
806 file system modules, such as igmp, ipdivert, and others, should be
808 Bump __FreeBSD_version to 800096.
811 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
815 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
816 Bump __FreeBSD_version to 800094.
819 Add mnt_xflag field to 'struct mount'. File system modules need to be
821 Bump __FreeBSD_version to 800093.
824 The compiling option ADAPTIVE_SX has been retired while it has been
825 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
826 The KPI for sx_init_flags() changes as accepting flags:
827 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
828 been introduced in order to handle the reversed logic.
829 Bump __FreeBSD_version to 800092.
832 Add support for hierarchical jails. Remove global securelevel.
833 Bump __FreeBSD_version to 800091.
836 The layout of struct vnet_net has changed, therefore modules
838 Bump __FreeBSD_version to 800090.
841 The newly imported zic(8) produces a new format in the output. Please
842 run tzsetup(8) to install the newly created data to /etc/localtime.
845 The sysctl tree for the usb stack has renamed from hw.usb2.* to
846 hw.usb.* and is now consistent again with previous releases.
849 802.11 monitor mode support was revised and driver api's were changed.
850 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
851 of DLT_IEEE802_11. No user-visible data structures were changed but
852 applications that use DLT_IEEE802_11 may require changes.
853 Bump __FreeBSD_version to 800088.
856 The layout of the following structs has changed: sysctl_oid,
857 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
858 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
859 panics may be experienced. World rebuild is required for
860 correctly checking networking state from userland.
861 Bump __FreeBSD_version to 800085.
864 MLDv2 and Source-Specific Multicast (SSM) have been merged
865 to the IPv6 stack. VIMAGE hooks are in but not yet used.
866 The implementation of SSM within FreeBSD's IPv6 stack closely
867 follows the IPv4 implementation.
869 For kernel developers:
871 * The most important changes are that the ip6_output() and
872 ip6_input() paths no longer take the IN6_MULTI_LOCK,
873 and this lock has been downgraded to a non-recursive mutex.
875 * As with the changes to the IPv4 stack to support SSM, filtering
876 of inbound multicast traffic must now be performed by transport
877 protocols within the IPv6 stack. This does not apply to TCP and
878 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
880 * The KPIs used by IPv6 multicast are similar to those used by
881 the IPv4 stack, with the following differences:
882 * im6o_mc_filter() is analogous to imo_multicast_filter().
883 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
884 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
885 * IN6_LOOKUP_MULTI() has been deprecated and removed.
886 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
887 for MLDv1 have an additional 'timer' argument which is used to
888 jitter the initial membership report for the solicited-node
889 multicast membership on-link.
890 * This is not strictly needed for MLDv2, which already jitters
891 its report transmissions. However, the 'timer' argument is
892 preserved in case MLDv1 is active on the interface.
894 * The KAME linked-list based IPv6 membership implementation has
895 been refactored to use a vector similar to that used by the IPv4
897 Code which maintains a list of its own multicast memberships
898 internally, e.g. carp, has been updated to reflect the new
901 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
902 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
903 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
904 implementation constraint which needs to be addressed in HEAD.
906 For application developers:
908 * The changes are broadly similar to those made for the IPv4
911 * The use of IPv4 and IPv6 multicast socket options on the same
912 socket, using mapped addresses, HAS NOT been tested or supported.
914 * There are a number of issues with the implementation of various
915 IPv6 multicast APIs which need to be resolved in the API surface
916 before the implementation is fully compatible with KAME userland
917 use, and these are mostly to do with interface index treatment.
919 * The literature available discusses the use of either the delta / ASM
920 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
921 using setsourcefilter(3)/getsourcefilter(3). For more information
922 please refer to RFC 3768, 'Socket Interface Extensions for
923 Multicast Source Filters'.
925 * Applications which use the published RFC 3678 APIs should be fine.
927 For systems administrators:
929 * The mtest(8) utility has been refactored to support IPv6, in
930 addition to IPv4. Interface addresses are no longer accepted
931 as arguments, their names must be used instead. The utility
932 will map the interface name to its first IPv4 address as
933 returned by getifaddrs(3).
935 * The ifmcstat(8) utility has also been updated to print the MLDv2
936 endpoint state and source filter lists via sysctl(3).
938 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
939 loopback of IPv6 multicast datagrams by default; it defaults to 1
940 to preserve the existing behaviour. Disabling multicast loopback is
941 recommended for optimal system performance.
943 * The IPv6 MROUTING code has been changed to examine this sysctl
944 instead of attempting to perform a group lookup before looping
945 back forwarded datagrams.
947 Bump __FreeBSD_version to 800084.
950 Implement low-level Bluetooth HCI API.
951 Bump __FreeBSD_version to 800083.
954 The layout of struct malloc_type, used by modules to register new
955 memory allocation types, has changed. Most modules will need to
956 be rebuilt or panics may be experienced.
957 Bump __FreeBSD_version to 800081.
960 Anticipate overflowing inp_flags - add inp_flags2.
961 This changes most offsets in inpcb, so checking v4 connection
962 state will require a world rebuild.
963 Bump __FreeBSD_version to 800080.
966 Add an llentry to struct route and struct route_in6. Modules
967 embedding a struct route will need to be recompiled.
968 Bump __FreeBSD_version to 800079.
971 The size of rt_metrics_lite and by extension rtentry has changed.
972 Networking administration apps will need to be recompiled.
973 The route command now supports show as an alias for get, weighting
974 of routes, sticky and nostick flags to alter the behavior of stateful
976 Bump __FreeBSD_version to 800078.
979 Do not use Giant for kbdmux(4) locking. This is wrong and
980 apparently causing more problems than it solves. This will
981 re-open the issue where interrupt handlers may race with
982 kbdmux(4) in polling mode. Typical symptoms include (but
983 not limited to) duplicated and/or missing characters when
984 low level console functions (such as gets) are used while
985 interrupts are enabled (for example geli password prompt,
986 mountroot prompt etc.). Disabling kbdmux(4) may help.
989 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
990 kernel modules referencing any of the above need to be recompiled.
991 Bump __FreeBSD_version to 800075.
994 GEOM_PART has become the default partition slicer for storage devices,
995 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
996 introduces some changes:
998 MSDOS/EBR: the devices created from MSDOS extended partition entries
999 (EBR) can be named differently than with GEOM_MBR and are now symlinks
1000 to devices with offset-based names. fstabs may need to be modified.
1002 BSD: the "geometry does not match label" warning is harmless in most
1003 cases but it points to problems in file system misalignment with
1004 disk geometry. The "c" partition is now implicit, covers the whole
1005 top-level drive and cannot be (mis)used by users.
1007 General: Kernel dumps are now not allowed to be written to devices
1008 whose partition types indicate they are meant to be used for file
1009 systems (or, in case of MSDOS partitions, as something else than
1012 Most of these changes date approximately from 200812.
1015 The uscanner(4) driver has been removed from the kernel. This follows
1016 Linux removing theirs in 2.6 and making libusb the default interface
1017 (supported by sane).
1020 The multicast forwarding code has been cleaned up. netstat(1)
1021 only relies on KVM now for printing bandwidth upcall meters.
1022 The IPv4 and IPv6 modules are split into ip_mroute_mod and
1023 ip6_mroute_mod respectively. The config(5) options for statically
1024 compiling this code remain the same, i.e. 'options MROUTING'.
1027 Support for the IFF_NEEDSGIANT network interface flag has been
1028 removed, which means that non-MPSAFE network device drivers are no
1029 longer supported. In particular, if_ar, if_sr, and network device
1030 drivers from the old (legacy) USB stack can no longer be built or
1034 POSIX.1 Native Language Support (NLS) has been enabled in libc and
1035 a bunch of new language catalog files have also been added.
1036 This means that some common libc messages are now localized and
1037 they depend on the LC_MESSAGES environmental variable.
1040 The k8temp(4) driver has been renamed to amdtemp(4) since
1041 support for Family 10 and Family 11 CPU families was added.
1044 IGMPv3 and Source-Specific Multicast (SSM) have been merged
1045 to the IPv4 stack. VIMAGE hooks are in but not yet used.
1047 For kernel developers, the most important changes are that the
1048 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
1049 and this lock has been downgraded to a non-recursive mutex.
1051 Transport protocols (UDP, Raw IP) are now responsible for filtering
1052 inbound multicast traffic according to group membership and source
1053 filters. The imo_multicast_filter() KPI exists for this purpose.
1054 Transports which do not use multicast (SCTP, TCP) already reject
1055 multicast by default. Forwarding and receive performance may improve
1056 as a mutex acquisition is no longer needed in the ip_input()
1057 low-level input path. in_addmulti() and in_delmulti() are shimmed
1058 to new KPIs which exist to support SSM in-kernel.
1060 For application developers, it is recommended that loopback of
1061 multicast datagrams be disabled for best performance, as this
1062 will still cause the lock to be taken for each looped-back
1063 datagram transmission. The net.inet.ip.mcast.loop sysctl may
1064 be tuned to 0 to disable loopback by default; it defaults to 1
1065 to preserve the existing behaviour.
1067 For systems administrators, to obtain best performance with
1068 multicast reception and multiple groups, it is always recommended
1069 that a card with a suitably precise hash filter is used. Hash
1070 collisions will still result in the lock being taken within the
1071 transport protocol input path to check group membership.
1073 If deploying FreeBSD in an environment with IGMP snooping switches,
1074 it is recommended that the net.inet.igmp.sendlocal sysctl remain
1075 enabled; this forces 224.0.0.0/24 group membership to be announced
1078 The size of 'struct igmpstat' has changed; netstat needs to be
1079 recompiled to reflect this.
1080 Bump __FreeBSD_version to 800070.
1083 libusb20.so.1 is now installed as libusb.so.1 and the ports system
1084 updated to use it. This requires a buildworld/installworld in order to
1085 update the library and dependencies (usbconfig, etc). Its advisable to
1086 rebuild all ports which uses libusb. More specific directions are given
1087 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
1088 libusb are no longer required and can be removed.
1091 A workaround is committed to allow the creation of System V shared
1092 memory segment of size > 2 GB on the 64-bit architectures.
1093 Due to a limitation of the existing ABI, the shm_segsz member
1094 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
1095 wrong for large segments. Note that limits must be explicitly
1096 raised to allow such segments to be created.
1099 The layout of struct ifnet has changed, requiring a rebuild of all
1100 network device driver modules.
1103 The /dev handling for the new USB stack has changed, a
1104 buildworld/installworld is required for libusb20.
1107 The new USB2 stack has now been permanently moved in and all kernel and
1108 module names reverted to their previous values (eg, usb, ehci, ohci,
1109 ums, ...). The old usb stack can be compiled in by prefixing the name
1110 with the letter 'o', the old usb modules have been removed.
1111 Updating entry 20090216 for xorg and 20090215 for libmap may still
1115 The rc.conf(5) option if_up_delay has been renamed to
1116 defaultroute_delay to better reflect its purpose. If you have
1117 customized this setting in /etc/rc.conf you need to update it to
1121 xorg 7.4 wants to configure its input devices via hald which does not
1122 yet work with USB2. If the keyboard/mouse does not work in xorg then
1124 Option "AllowEmptyInput" "off"
1125 to your ServerLayout section. This will cause X to use the configured
1126 kbd and mouse sections from your xorg.conf.
1129 The GENERIC kernels for all architectures now default to the new USB2
1130 stack. No kernel config options or code have been removed so if a
1131 problem arises please report it and optionally revert to the old USB
1132 stack. If you are loading USB kernel modules or have a custom kernel
1133 that includes GENERIC then ensure that usb names are also changed over,
1134 eg uftdi -> usb2_serial_ftdi.
1136 Older programs linked against the ports libusb 0.1 need to be
1137 redirected to the new stack's libusb20. /etc/libmap.conf can
1139 # Map old usb library to new one for usb2 stack
1140 libusb-0.1.so.8 libusb20.so.1
1143 All USB ethernet devices now attach as interfaces under the name ueN
1144 (eg. ue0). This is to provide a predictable name as vendors often
1145 change usb chipsets in a product without notice.
1148 The ichsmb(4) driver has been changed to require SMBus slave
1149 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1150 All of the other SMBus controller drivers require left-justified
1151 slave addresses, so this change makes all the drivers provide the
1155 INET6 statistics (struct ip6stat) was updated.
1156 netstat(1) needs to be recompiled.
1159 NTFS has been removed from GENERIC kernel on amd64 to match
1160 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1161 will load ntfs.ko module automatically when NTFS support is
1162 actually needed, unless ntfs.ko is not installed or security
1163 level prohibits loading kernel modules. If either is the case,
1164 "options NTFS" has to be added into kernel config.
1167 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1168 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1169 800061. User space tools that rely on the size of struct tcpcb in
1170 tcp_var.h (e.g. sockstat) need to be recompiled.
1173 ng_tty(4) module updated to match the new TTY subsystem.
1174 Due to API change, user-level applications must be updated.
1175 New API support added to mpd5 CVS and expected to be present
1176 in next mpd5.3 release.
1179 With __FreeBSD_version 800060 the makefs tool is part of
1180 the base system (it was a port).
1183 The afdata and ifnet locks have been changed from mutexes to
1184 rwlocks, network modules will need to be re-compiled.
1187 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1188 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1189 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1190 architecture and 40 bytes on 64-bit architecture. The userland
1191 applications "arp" and "ndp" have been updated accordingly.
1192 The output from "netstat -r" shows only routing entries and
1193 none of the L2 information.
1196 __FreeBSD_version 800057 marks the switchover from the
1197 binary ath hal to source code. Users must add the line:
1199 options AH_SUPPORT_AR5416
1201 to their kernel config files when specifying:
1205 The ath_hal module no longer exists; the code is now compiled
1206 together with the driver in the ath module. It is now
1207 possible to tailor chip support (i.e. reduce the set of chips
1208 and thereby the code size); consult ath_hal(4) for details.
1211 __FreeBSD_version 800054 adds memory barriers to
1212 <machine/atomic.h>, new interfaces to ifnet to facilitate
1213 multiple hardware transmit queues for cards that support
1214 them, and a lock-less ring-buffer implementation to
1215 enable drivers to more efficiently manage queueing of
1219 A new version of ZFS (version 13) has been merged to -HEAD.
1220 This version has zpool attribute "listsnapshots" off by
1221 default, which means "zfs list" does not show snapshots,
1222 and is the same as Solaris behavior.
1225 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1228 The uhci, ohci, ehci and slhci USB Host controller drivers have
1229 been put into separate modules. If you load the usb module
1230 separately through loader.conf you will need to load the
1231 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1232 controller add the following to loader.conf:
1238 The ABI used by the PMC toolset has changed. Please keep
1239 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1243 atapci kernel module now includes only generic PCI ATA
1244 driver. AHCI driver moved to ataahci kernel module.
1245 All vendor-specific code moved into separate kernel modules:
1246 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1247 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1248 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1249 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1252 The TTY subsystem of the kernel has been replaced by a new
1253 implementation, which provides better scalability and an
1254 improved driver model. Most common drivers have been migrated to
1255 the new TTY subsystem, while others have not. The following
1256 drivers have not yet been ported to the new TTY layer:
1259 cy, digi, rc, rp, sio
1265 ng_h4, ng_tty, ppp, sl, snp
1267 Adding these drivers to your kernel configuration file shall
1268 cause compilation to fail.
1271 ntpd has been upgraded to 4.2.4p5.
1274 OpenSSH has been upgraded to 5.1p1.
1276 For many years, FreeBSD's version of OpenSSH preferred DSA
1277 over RSA for host and user authentication keys. With this
1278 upgrade, we've switched to the vendor's default of RSA over
1279 DSA. This may cause upgraded clients to warn about unknown
1280 host keys even for previously known hosts. Users should
1281 follow the usual procedure for verifying host keys before
1282 accepting the RSA key.
1284 This can be circumvented by setting the "HostKeyAlgorithms"
1285 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1288 Please note that the sequence of keys offered for
1289 authentication has been changed as well. You may want to
1290 specify IdentityFile in a different order to revert this
1294 The sio(4) driver has been removed from the i386 and amd64
1295 kernel configuration files. This means uart(4) is now the
1296 default serial port driver on those platforms as well.
1298 To prevent collisions with the sio(4) driver, the uart(4) driver
1299 uses different names for its device nodes. This means the
1300 onboard serial port will now most likely be called "ttyu0"
1301 instead of "ttyd0". You may need to reconfigure applications to
1302 use the new device names.
1304 When using the serial port as a boot console, be sure to update
1305 /boot/device.hints and /etc/ttys before booting the new kernel.
1306 If you forget to do so, you can still manually specify the hints
1307 at the loader prompt:
1309 set hint.uart.0.at="isa"
1310 set hint.uart.0.port="0x3F8"
1311 set hint.uart.0.flags="0x10"
1312 set hint.uart.0.irq="4"
1316 The gpt(8) utility has been removed. Use gpart(8) to partition
1320 The version that Linuxulator emulates was changed from 2.4.2
1321 to 2.6.16. If you experience any problems with Linux binaries
1322 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1323 if it fixes the problem contact emulation mailing list.
1326 ISDN4BSD (I4B) was removed from the src tree. You may need to
1327 update a your kernel configuration and remove relevant entries.
1330 I have checked in code to support multiple routing tables.
1331 See the man pages setfib(1) and setfib(2).
1332 This is a hopefully backwards compatible version,
1333 but to make use of it you need to compile your kernel
1334 with options ROUTETABLES=2 (or more up to 16).
1337 The 802.11 wireless support was redone to enable multi-bss
1338 operation on devices that are capable. The underlying device
1339 is no longer used directly but instead wlanX devices are
1340 cloned with ifconfig. This requires changes to rc.conf files.
1341 For example, change:
1342 ifconfig_ath0="WPA DHCP"
1345 ifconfig_wlan0="WPA DHCP"
1346 see rc.conf(5) for more details. In addition, mergemaster of
1347 /etc/rc.d is highly recommended. Simultaneous update of userland
1348 and kernel wouldn't hurt either.
1350 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1351 modules were merged into the base wlan module. All references
1352 to these modules (e.g. in kernel config files) must be removed.
1355 psm(4) has gained write(2) support in native operation level.
1356 Arbitrary commands can be written to /dev/psm%d and status can
1357 be read back from it. Therefore, an application is responsible
1358 for status validation and error recovery. It is a no-op in
1359 other operation levels.
1362 Support for KSE threading has been removed from the kernel. To
1363 run legacy applications linked against KSE libmap.conf may
1364 be used. The following libmap.conf may be used to ensure
1365 compatibility with any prior release:
1367 libpthread.so.1 libthr.so.1
1368 libpthread.so.2 libthr.so.2
1369 libkse.so.3 libthr.so.3
1372 The layout of struct vmspace has changed. This affects libkvm
1373 and any executables that link against libkvm and use the
1374 kvm_getprocs() function. In particular, but not exclusively,
1375 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1376 The effects are minimal, but it's advisable to upgrade world
1380 The latest em driver no longer has support in it for the
1381 82575 adapter, this is now moved to the igb driver. The
1382 split was done to make new features that are incompatible
1383 with older hardware easier to do.
1386 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1387 likewise the kernel option is now GEOM_LINUX_LVM.
1390 The default NFS mount mode has changed from UDP to TCP for
1391 increased reliability. If you rely on (insecurely) NFS
1392 mounting across a firewall you may need to update your
1396 Belatedly note the addition of m_collapse for compacting
1400 The fts(3) structures have been changed to use adequate
1401 integer types for their members and so to be able to cope
1402 with huge file trees. The old fts(3) ABI is preserved
1403 through symbol versioning in libc, so third-party binaries
1404 using fts(3) should still work, although they will not take
1405 advantage of the extended types. At the same time, some
1406 third-party software might fail to build after this change
1407 due to unportable assumptions made in its source code about
1408 fts(3) structure members. Such software should be fixed
1409 by its vendor or, in the worst case, in the ports tree.
1410 FreeBSD_version 800015 marks this change for the unlikely
1411 case that a portable fix is impossible.
1414 To upgrade to -current after this date, you must be running
1415 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1416 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1419 The ADAPTIVE_GIANT kernel option has been retired because its
1420 functionality is the default now.
1423 The AT keyboard emulation of sunkbd(4) has been turned on
1424 by default. In order to make the special symbols of the Sun
1425 keyboards driven by sunkbd(4) work under X these now have
1426 to be configured the same way as Sun USB keyboards driven
1427 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1429 Option "XkbLayout" "us"
1430 Option "XkbRules" "xorg"
1431 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1434 It has been decided that it is desirable to provide ABI
1435 backwards compatibility to the FreeBSD 4/5/6 versions of the
1436 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1437 broken with the introduction of PCI domain support (see the
1438 20070930 entry). Unfortunately, this required the ABI of
1439 PCIOCGETCONF to be broken again in order to be able to
1440 provide backwards compatibility to the old version of that
1441 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1442 again. As for prominent ports this affects neither pciutils
1443 nor xorg-server this time, the hal port needs to be rebuilt
1447 The misnamed kthread_create() and friends have been renamed
1448 to kproc_create() etc. Many of the callers already
1449 used kproc_start()..
1450 I will return kthread_create() and friends in a while
1451 with implementations that actually create threads, not procs.
1452 Renaming corresponds with version 800002.
1461 Avoid using make -j when upgrading. While generally safe, there are
1462 sometimes problems using -j to upgrade. If your upgrade fails with
1463 -j, please try again without -j. From time to time in the past there
1464 have been problems using -j with buildworld and/or installworld. This
1465 is especially true when upgrading between "distant" versions (eg one
1466 that cross a major release boundary or several minor releases, or when
1467 several months have passed on the -current branch).
1469 Sometimes, obscure build problems are the result of environment
1470 poisoning. This can happen because the make utility reads its
1471 environment when searching for values for global variables. To run
1472 your build attempts in an "environmental clean room", prefix all make
1473 commands with 'env -i '. See the env(1) manual page for more details.
1475 When upgrading from one major version to another it is generally best
1476 to upgrade to the latest code in the currently installed branch first,
1477 then do an upgrade to the new branch. This is the best-tested upgrade
1478 path, and has the highest probability of being successful. Please try
1479 this approach before reporting problems with a major version upgrade.
1483 When upgrading the boot ZFS pool to a new version, always follow
1486 1.) recompile and reinstall the ZFS boot loader and boot block
1487 (this is part of "make buildworld" and "make installworld")
1489 2.) update the ZFS boot block on your boot drive
1491 The following example updates the ZFS boot block on the first
1492 partition (freebsd-boot) of a GPT partitioned drive ad0:
1493 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1495 Non-boot pools do not need these updates.
1499 If you are updating from a prior version of FreeBSD (even one just
1500 a few days old), you should follow this procedure. It is the most
1501 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1503 make kernel-toolchain
1504 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1505 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1507 To test a kernel once
1508 ---------------------
1509 If you just want to boot a kernel once (because you are not sure
1510 if it works, or if you want to boot a known bad kernel to provide
1511 debugging information) run
1512 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1513 nextboot -k testkernel
1515 To just build a kernel when you know that it won't mess you up
1516 --------------------------------------------------------------
1517 This assumes you are already running a CURRENT system. Replace
1518 ${arch} with the architecture of your machine (e.g. "i386",
1519 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1521 cd src/sys/${arch}/conf
1522 config KERNEL_NAME_HERE
1523 cd ../compile/KERNEL_NAME_HERE
1528 If this fails, go to the "To build a kernel" section.
1530 To rebuild everything and install it on the current system.
1531 -----------------------------------------------------------
1532 # Note: sometimes if you are running current you gotta do more than
1533 # is listed here if you are upgrading from a really old current.
1535 <make sure you have good level 0 dumps>
1537 make kernel KERNCONF=YOUR_KERNEL_HERE
1539 <reboot in single user> [3]
1547 To cross-install current onto a separate partition
1548 --------------------------------------------------
1549 # In this approach we use a separate partition to hold
1550 # current's root, 'usr', and 'var' directories. A partition
1551 # holding "/", "/usr" and "/var" should be about 2GB in
1554 <make sure you have good level 0 dumps>
1557 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1558 <maybe newfs current's root partition>
1559 <mount current's root partition on directory ${CURRENT_ROOT}>
1560 make installworld DESTDIR=${CURRENT_ROOT}
1561 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1562 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1563 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1564 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1565 <reboot into current>
1566 <do a "native" rebuild/install as described in the previous section>
1567 <maybe install compatibility libraries from ports/misc/compat*>
1571 To upgrade in-place from 8.x-stable to current
1572 ----------------------------------------------
1573 <make sure you have good level 0 dumps>
1575 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1577 <reboot in single user> [3]
1584 Make sure that you've read the UPDATING file to understand the
1585 tweaks to various things you need. At this point in the life
1586 cycle of current, things change often and you are on your own
1587 to cope. The defaults can also change, so please read ALL of
1588 the UPDATING entries.
1590 Also, if you are tracking -current, you must be subscribed to
1591 freebsd-current@freebsd.org. Make sure that before you update
1592 your sources that you have read and understood all the recent
1593 messages there. If in doubt, please track -stable which has
1594 much fewer pitfalls.
1596 [1] If you have third party modules, such as vmware, you
1597 should disable them at this point so they don't crash your
1600 [3] From the bootblocks, boot -s, and then do
1605 adjkerntz -i # if CMOS is wall time
1606 Also, when doing a major release upgrade, it is required that
1607 you boot into single user mode to do the installworld.
1609 [4] Note: This step is non-optional. Failure to do this step
1610 can result in a significant reduction in the functionality of the
1611 system. Attempting to do it by hand is not recommended and those
1612 that pursue this avenue should read this file carefully, as well
1613 as the archives of freebsd-current and freebsd-hackers mailing lists
1614 for potential gotchas. The -U option is also useful to consider.
1615 See mergemaster(8) for more information.
1617 [5] Usually this step is a noop. However, from time to time
1618 you may need to do this if you get unknown user in the following
1619 step. It never hurts to do it all the time. You may need to
1620 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1621 install) after the buildworld before this step if you last updated
1622 from current before 20020224 or from -stable before 20020408.
1624 [6] This only deletes old files and directories. Old libraries
1625 can be deleted by "make delete-old-libs", but you have to make
1626 sure that no program is using those libraries anymore.
1628 [8] In order to have a kernel that can run the 4.x binaries needed to
1629 do an installworld, you must include the COMPAT_FREEBSD4 option in
1630 your kernel. Failure to do so may leave you with a system that is
1631 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1632 required to run the 5.x binaries on more recent kernels. And so on
1633 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1635 Make sure that you merge any new devices from GENERIC since the
1636 last time you updated your kernel config file.
1638 [9] When checking out sources, you must include the -P flag to have
1639 cvs prune empty directories.
1641 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1642 "?=" instead of the "=" assignment operator, so that buildworld can
1643 override the CPUTYPE if it needs to.
1645 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1646 not on the command line, or in /etc/make.conf. buildworld will
1647 warn if it is improperly defined.
1650 This file contains a list, in reverse chronological order, of major
1651 breakages in tracking -current. It is not guaranteed to be a complete
1652 list of such breakages, and only contains entries since October 10, 2007.
1653 If you need to see UPDATING entries from before that date, you will need
1654 to fetch an UPDATING file from an older FreeBSD release.
1656 Copyright information:
1658 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1660 Redistribution, publication, translation and use, with or without
1661 modification, in full or in part, in any form or format of this
1662 document are permitted without further permission from the author.
1664 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1665 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1666 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1667 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1668 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1669 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1670 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1671 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1672 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1673 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1674 POSSIBILITY OF SUCH DAMAGE.
1676 Contact Warner Losh if you have any questions about your use of