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 http://www.freebsd.org/doc/en_US.ISO8859-1/books/handbook/makeworld.html
11 Items affecting the ports and packages system can be found in
12 /usr/ports/UPDATING. Please read that file before running portupgrade.
15 Clang and llvm have been upgraded to 3.4 release. Please note that
16 clang 3.4 now defaults to DWARF4 debug information format when you
17 specify -g. Since kgdb(1) only supports DWARF2, you should update any
18 customized kernel configurations which include debug information to
19 explicitly use -gdwarf-2, e.g:
21 makeoptions DEBUG=-gdwarf-2
23 This has already been applied to the appropriate GENERIC configuration
24 files, so if you inherit from those, no changes are required.
27 The nve(4) driver for NVIDIA nForce MCP Ethernet adapters has
28 been deprecated and will not be part of FreeBSD 11.0 and later
29 releases. If you use this driver, please consider switching to
30 the nfe(4) driver instead.
33 The behavior of gss_pseudo_random() for the krb5 mechanism
34 has changed, for applications requesting a longer random string
35 than produced by the underlying enctype's pseudo-random() function.
36 In particular, the random string produced from a session key of
37 enctype aes256-cts-hmac-sha1-96 or aes256-cts-hmac-sha1-96 will
38 be different at the 17th octet and later, after this change.
39 The counter used in the PRF+ construction is now encoded as a
40 big-endian integer in accordance with RFC 4402.
41 __FreeBSD_version is bumped to 902505.
47 Behavior of devfs rules path matching has been changed.
48 Pattern is now always matched against fully qualified devfs
49 path and slash characters must be explicitly matched by
50 slashes in pattern (FNM_PATHNAME). Rulesets involving devfs
51 subdirectories must be reviewed.
54 hastctl(8)'s `status' command output changed to terse one-liner format.
55 Scripts using this should switch to `list' command or be rewritten.
58 Fix a bug that allowed a tracing process (e.g. gdb) to write
59 to a memory-mapped file in the traced process's address space
60 even if neither the traced process nor the tracing process had
61 write access to that file.
64 Added ZFS TRIM support which is enabled by default. To disable
65 ZFS TRIM support set vfs.zfs.trim.enabled=0 in loader.conf.
67 Creating new ZFS pools and adding new devices to existing pools
68 first performs a full device level TRIM which can take a significant
69 amount of time. The sysctl vfs.zfs.vdev.trim_on_init can be set to 0
70 to disable this behaviour.
72 ZFS TRIM requires the underlying device support BIO_DELETE which
73 is currently provided by methods such as ATA TRIM and SCSI UNMAP
74 via CAM, which are typically supported by SSD's.
76 Stats for ZFS TRIM can be monitored by looking at the sysctl's
77 under kstat.zfs.misc.zio_trim.
80 `list' command has been added to hastctl(8). For now, it is full
81 equivalent of `status' command.
82 WARNING: in the near future the output of hastctl's status command
83 will change to more terse format. If you use `hastctl status'
84 for parsing in your scripts, switch to `hastctl list'.
87 The mergemaster command now uses the default MAKEOBJDIRPREFIX
88 rather than creating it's own in the temporary directory in
89 order allow access to bootstrapped versions of tools such as
90 install and mtree. When upgrading from version of FreeBSD where
91 the install command does not support -l, you will need to
92 install a new mergemaster command if mergemaster -p is required.
93 This can be accomplished with the command (cd src/usr.sbin/mergemaster
96 Due to the use of the new -l option to install(1) during build
97 and install, you must take care not to directly set the INSTALL
98 make variable in your /etc/make.conf, /etc/src.conf, or on the
99 command line. If you wish to use the -C flag for all installs
100 you may be able to add INSTALL+=-C to /etc/make.conf or
104 Fix a bug that allows NFS clients to issue READDIR on files.
107 The install(1) option -M has changed meaning and now takes an
108 argument that is a file or path to append logs to. In the
109 unlikely event that -M was the last option on the command line
110 and the command line contained at least two files and a target
111 directory the first file will have logs appended to it. The -M
112 option served little practical purpose in the last decade so it's
113 used expected to be extremely rare.
116 A new compression method (lz4) has been merged to. Please refer to
117 zpool-features(7) for more information.
119 Please refer to the "ZFS notes" section of this file for information
120 on upgrading boot ZFS pools.
123 The VFS KBI was changed with the merge of several nullfs
124 optimizations and fixes. All filesystem modules must be
128 With the addition of auditdistd(8), a new auditdistd user is now
129 depended on during installworld. "mergemaster -p" can be used to add
130 the user prior to installworld, as documented in the handbook.
136 A new version of ZFS (pool version 5000) has been merged to 9-STABLE.
137 Starting with this version the old system of ZFS pool versioning
138 is superseded by "feature flags". This concept enables forward
139 compatibility against certain future changes in functionality of ZFS
140 pools. The first two read-only compatible "feature flags" for ZFS
141 pools are "com.delphix:async_destroy" and "com.delphix:empty_bpobj".
142 For more information read the new zpool-features(7) manual page.
143 Please refer to the "ZFS notes" section of this file for information
144 on upgrading boot ZFS pools.
147 The commit introducing bsd.compiler.mk breaks the traditional
148 building of kernels before this point. Add -m ${SRC}/share/mk
149 (for the right value of SRC) to your command lines to work
150 around; update your useland to a point after this; or use the
151 buildkernel/installkernel top-level targets. See also 20120829.
154 The IPFIREWALL_FORWARD kernel option has been removed. Its
155 functionality now turned on by default.
158 The random(4) support for the VIA hardware random number
159 generator (`PADLOCK') is no longer enabled unconditionally.
160 Add the PADLOCK_RNG option in the custom kernel config if
161 needed. The GENERIC kernels on i386 and amd64 do include the
162 option, so the change only affects the custom kernel
166 The amd64 kernel now uses xsetbv, xrstor instructions. To compile with
167 the traditional method, you must update your system with an installworld
168 before the kernel will build. The documented make buildkernel/installkernel
169 interfaces (coupled with fresh make kernel-toolchain) continue to work.
172 The sparc64 ZFS loader has been changed to no longer try to auto-
173 detect ZFS providers based on diskN aliases but now requires these
174 to be explicitly listed in the OFW boot-device environment variable.
177 Now unix domain sockets behave "as expected" on nullfs(5). Previously
178 nullfs(5) did not pass through all behaviours to the underlying layer,
179 as a result if we bound to a socket on the lower layer we could connect
180 only to the lower path; if we bound to the upper layer we could connect
181 only to the upper path. The new behavior is one can connect to both the
182 lower and the upper paths regardless what layer path one binds to.
185 The acpi_wmi(4) status device /dev/wmistat has been renamed to
189 A new VOP_ADVISE() was added to support posix_fadvise(2). All
190 filesystem modules must be recompiled.
193 The interface of the VOP_VPTOCNP(9) changed, now the returned
194 vnode shall be referenced, previously it was required to be
195 only held. All in-tree filesystems are converted.
201 The broken amd(4) driver has been replaced with esp(4) in the amd64,
202 i386 and pc98 GENERIC kernel configuration files.
205 This commit modifies vfs_register() so that it uses a hash
206 calculation to set vfc_typenum, which is enabled by default.
207 The first time a system is booted after this change, the
208 vfc_typenum values will change for all file systems. The
209 main effect of this is a change to the NFS server file handles
210 for file systems that use vfc_typenum in their fsid, such as ZFS.
211 It will, however, prevent vfc_typenum from changing when file
212 systems are loaded in a different order for subsequent reboots.
213 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
214 until you are ready to remount all NFS clients after a reboot.
217 Bump the shared library version numbers for libraries that
218 do not use symbol versioning, have changed the ABI compared
219 to stable/8 and which shared library version was not bumped.
220 Done as part of 9.0-RELEASE cycle.
223 During the merge of Capsicum features, the fget(9) KPI was modified.
224 This may require the rebuilding of out-of-tree device drivers --
225 issues have been reported specifically with the nVidia device driver.
226 __FreeBSD_version is bumped to 900041.
228 Also, there is a period between 20110811 and 20110814 where the
229 special devices /dev/{stdin,stdout,stderr} did not work correctly.
230 Building world from a kernel during that window may not work.
233 The packet filter (pf) code has been updated to OpenBSD 4.5.
234 You need to update userland tools to be in sync with kernel.
235 This update breaks backward compatibility with earlier pfsync(4)
236 versions. Care must be taken when updating redundant firewall setups.
239 The following sysctls and tunables are retired on x86 platforms:
241 machdep.hlt_logical_cpus
242 The following sysctl is retired:
243 machdep.hyperthreading_allowed
244 The sysctls were supposed to provide a way to dynamically offline and
245 online selected CPUs on x86 platforms, but the implementation has not
246 been reliable especially with SCHED_ULE scheduler.
247 machdep.hyperthreading_allowed tunable is still available to ignore
248 hyperthreading CPUs at OS level.
249 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
250 where X is an APIC ID of a CPU. Be advised, though, that disabling
251 CPUs in non-uniform fashion will result in non-uniform topology and
252 may lead to sub-optimal system performance with SCHED_ULE, which is
256 cpumask_t type is retired and cpuset_t is used in order to describe
260 Changes to ifconfig(8) for dynamic address family detection mandate
261 that you are running a kernel of 20110525 or later. Make sure to
262 follow the update procedure to boot a new kernel before installing
266 Support for sun4v architecture is officially dropped
269 Several KPI breaking changes have been committed to the mii(4) layer,
270 the PHY drivers and consequently some Ethernet drivers using mii(4).
271 This means that miibus.ko and the modules of the affected Ethernet
272 drivers need to be recompiled.
274 Note to kernel developers: Given that the OUI bit reversion problem
275 was fixed as part of these changes all mii(4) commits related to OUIs,
276 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
277 handling, no longer can be merged verbatim to stable/8 and previous
281 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
282 into their kernel configurations along with 'device pci'.
285 The default NFS client is now the new NFS client, so fstype "newnfs"
286 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
287 Although mounts via fstype "nfs" will usually work without userland
288 changes, it is recommended that the mount(8) and mount_nfs(8)
289 commands be rebuilt from sources and that a link to mount_nfs called
290 mount_oldnfs be created. The new client is compiled into the
291 kernel with "options NFSCL" and this is needed for diskless root
292 file systems. The GENERIC kernel configs have been changed to use
293 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
294 To use the regular/old client, you can "mount -t oldnfs ...". For
295 a diskless root file system, you must also include a line like:
297 vfs.root.mountfrom="oldnfs:"
299 in the boot/loader.conf on the root fs on the NFS server to make
300 a diskless root fs use the old client.
303 The GENERIC kernels for all architectures now default to the new
304 CAM-based ATA stack. It means that all legacy ATA drivers were
305 removed and replaced by respective CAM drivers. If you are using
306 ATA device names in /etc/fstab or other places, make sure to update
307 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
308 where 'Y's are the sequential numbers starting from zero for each type
309 in order of detection, unless configured otherwise with tunables,
310 see cam(4)). There will be symbolic links created in /dev/ to map
311 old adX devices to the respective adaY. They should provide basic
312 compatibility for file systems mounting in most cases, but they do
313 not support old user-level APIs and do not have respective providers
314 in GEOM. Consider using updated management tools with new device names.
316 It is possible to load devices ahci, ata, siis and mvs as modules,
317 but option ATA_CAM should remain in kernel configuration to make ata
318 module work as CAM driver supporting legacy ATA controllers. Device ata
319 still can be used in modular fashion (atacore + ...). Modules atadisk
320 and atapi* are not used and won't affect operation in ATA_CAM mode.
321 Note that to use CAM-based ATA kernel should include CAM devices
322 scbus, pass, da (or explicitly ada), cd and optionally others. All of
323 them are parts of the cam module.
325 ataraid(4) functionality is now supported by the RAID GEOM class.
326 To use it you can load geom_raid kernel module and use graid(8) tool
327 for management. Instead of /dev/arX device names, use /dev/raid/rX.
329 No kernel config options or code have been removed, so if a problem
330 arises, please report it and optionally revert to the old ATA stack.
331 In order to do it you can remove from the kernel config:
336 , and instead add back:
337 device atadisk # ATA disk drives
338 device ataraid # ATA RAID drives
339 device atapicd # ATAPI CDROM drives
340 device atapifd # ATAPI floppy drives
341 device atapist # ATAPI tape drives
344 The default NFS server has been changed to the new server, which
345 was referred to as the experimental server. If you need to switch
346 back to the old NFS server, you must now put the "-o" option on
347 both the mountd and nfsd commands. This can be done using the
348 mountd_flags and nfs_server_flags rc.conf variables until an
349 update to the rc scripts is committed, which is coming soon.
352 The GNU Objective-C runtime library (libobjc), and other Objective-C
353 related components have been removed from the base system. If you
354 require an Objective-C library, please use one of the available ports.
357 ath(4) has been split into bus- and device- modules. if_ath contains
358 the HAL, the TX rate control and the network device code. if_ath_pci
359 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
360 contains the AHB glue. Users need to load both if_ath_pci and if_ath
361 in order to use ath on everything else.
363 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
364 need to load if_ath and if_ath_pci for ath(4) operation.
367 As part of the replacement of sysinstall, the process of building
368 release media has changed significantly. For details, please re-read
369 release(7), which has been updated to reflect the new build process.
372 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
373 is the last available version under GPLv2. It brings a number of new
374 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
375 4.1 and SSE 4.2), better support for powerpc64, a number of new
376 directives, and lots of other small improvements. See the ChangeLog
377 file in contrib/binutils for the full details.
380 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
381 compliant, and will now use half of hash for authentication.
382 This will break interoperability with all stacks (including all
383 actual FreeBSD versions) who implement
384 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
386 The only workaround with such peers is to use another HMAC
387 algorithm for IPsec ("phase 2") authentication.
390 Remove the uio_yield prototype and symbol. This function has
391 been misnamed since it was introduced and should not be
392 globally exposed with this name. The equivalent functionality
393 is now available using kern_yield(curthread->td_user_pri).
394 The function remains undocumented.
397 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
398 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
399 for scalar sysctls is defined but disabled. Code that needs
400 UQUAD to pass the type checking that must compile on older
401 systems where the define is not present can check against
402 __FreeBSD_version >= 900030.
404 The system dialog(1) has been replaced with a new version previously
405 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
406 with the previous version, but the libdialog associated with it has
407 a largely incompatible API. As such, the original version of libdialog
408 will be kept temporarily as libodialog, until its base system consumers
409 are replaced or updated. Bump __FreeBSD_version to 900030.
412 If you are trying to run make universe on a -stable system, and you get
413 the following warning:
414 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
415 unknown. config(8) likely too old."
416 or something similar to it, then you must upgrade your -stable system
417 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
418 stable/8) or build the config from the latest stable/8 branch and
419 install it on your system.
421 Prior to this date, building a current universe on 8-stable system from
422 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
423 error in the first kernel build phase. A new config on those old
424 systems will fix that problem for older versions of -current.
427 The TCP stack has been modified to allow Khelp modules to interact with
428 it via helper hook points and store per-connection data in the TCP
429 control block. Bump __FreeBSD_version to 900029. User space tools that
430 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
434 Generic IEEE 802.3 annex 31B full duplex flow control support has been
435 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
436 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
437 to take advantage of it instead of using custom implementations. This
438 means that these drivers now no longer unconditionally advertise
439 support for flow control but only do so if flow control is a selected
440 media option. This was implemented in the generic support that way in
441 order to allow flow control to be switched on and off via ifconfig(8)
442 with the PHY specific default to typically off in order to protect
443 from unwanted effects. Consequently, if you used flow control with
444 one of the above mentioned drivers you now need to explicitly enable
446 ifconfig bge0 media auto mediaopt flowcontrol
448 Along with the above mentioned changes generic support for setting
449 1000baseT master mode also has been added and brgphy(4), ciphy(4),
450 e1000phy(4) as well as ip1000phy(4) have been converted to take
451 advantage of it. This means that these drivers now no longer take the
452 link0 parameter for selecting master mode but the master media option
453 has to be used instead, for example like in the following:
454 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
456 Selection of master mode now is also available with all other PHY
457 drivers supporting 1000baseT.
460 The TCP stack has received a significant update to add support for
461 modularised congestion control and generally improve the clarity of
462 congestion control decisions. Bump __FreeBSD_version to 900025. User
463 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
464 sockstat) need to be recompiled.
467 The man(1) utility has been replaced by a new version that no longer
468 uses /etc/manpath.config. Please consult man.conf(5) for how to
469 migrate local entries to the new format.
472 The copyright strings printed by login(1) and sshd(8) at the time of a
473 new connection have been removed to follow other operating systems and
477 A workaround for a fixed ld bug has been removed in kernel code,
478 so make sure that your system ld is built from sources after
479 revision 210245 from 2010-07-19 (r211583 if building head kernel
480 on stable/8, r211584 for stable/7; both from 2010-08-21).
481 A symptom of incorrect ld version is different addresses for
482 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
485 The $ipv6_prefer variable in rc.conf(5) has been split into
486 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
488 The $ip6addrctl_policy is a variable to choose a pre-defined
489 address selection policy set by ip6addrctl(8). A value
490 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
493 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
494 flag (see an entry of 20090926) is set on an interface with no
495 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
496 security reason. If you want IPv6 link-local address on all
497 interfaces by default, set this to "YES".
499 The old ipv6_prefer="YES" is equivalent to
500 ipv6_activate_all_interfaces="YES" and
501 ip6addrctl_policy="ipv6_prefer".
504 DTrace has grown support for userland tracing. Due to this, DTrace is
505 now i386 and amd64 only.
506 dtruss(1) is now installed by default on those systems and a new
507 kernel module is needed for userland tracing: fasttrap.
508 No changes to your kernel config file are necessary to enable
509 userland tracing, but you might consider adding 'STRIP=' and
510 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
511 to have informative userland stack traces in DTrace (ustack).
514 The acpi_aiboost(4) driver has been removed in favor of the new
515 aibs(4) driver. You should update your kernel configuration file.
518 BSD grep has been imported to the base system and it is built by
519 default. It is completely BSD licensed, highly GNU-compatible, uses
520 less memory than its GNU counterpart and has a small codebase.
521 However, it is slower than its GNU counterpart, which is mostly
522 noticeable for larger searches, for smaller ones it is measurable
523 but not significant. The reason is complex, the most important factor
524 is that we lack a modern and efficient regex library and GNU
525 overcomes this by optimizing the searches internally. Future work
526 on improving the regex performance is planned, for the meantime,
527 users that need better performance, can build GNU grep instead by
528 setting the WITH_GNU_GREP knob.
531 Due to the import of powerpc64 support, all existing powerpc kernel
532 configuration files must be updated with a machine directive like this:
533 machine powerpc powerpc
535 In addition, an updated config(8) is required to build powerpc kernels
539 A new version of ZFS (version 15) has been merged to -HEAD.
540 This version uses a python library for the following subcommands:
541 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
542 For full functionality of these commands the following port must
543 be installed: sysutils/py-zfs
546 'vm_page's are now hashed by physical address to an array of mutexes.
547 Currently this is only used to serialize access to hold_count. Over
548 time the page queue mutex will be peeled away. This changes the size
549 of pmap on every architecture. And requires all callers of vm_page_hold
550 and vm_page_unhold to be updated.
553 WITH_CTF can now be specified in src.conf (not recommended, there
554 are some problems with static executables), make.conf (would also
555 affect ports which do not use GNU make and do not override the
556 compile targets) or in the kernel config (via "makeoptions
558 When WITH_CTF was specified there before this was silently ignored,
559 so make sure that WITH_CTF is not used in places which could lead
560 to unwanted behavior.
563 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
564 to allow 32-bit compatibility on non-x86 platforms. All kernel
565 configurations on amd64 and ia64 platforms using these options must
566 be modified accordingly.
569 The utmp user accounting database has been replaced with utmpx,
570 the user accounting interface standardized by POSIX.
571 Unfortunately the semantics of utmp and utmpx don't match,
572 making it practically impossible to support both interfaces.
573 The user accounting database is used by tools like finger(1),
574 last(1), talk(1), w(1) and ac(8).
576 All applications in the base system use utmpx. This means only
577 local binaries (e.g. from the ports tree) may still use these
578 utmp database files. These applications must be rebuilt to make
581 After the system has been upgraded, it is safe to remove the old
582 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
583 assuming their contents is of no importance anymore. Old wtmp
584 databases can only be used by last(1) and ac(8) after they have
585 been converted to the new format using wtmpcvt(1).
588 Introduce the kernel thread "deadlock resolver" (which can be enabled
589 via the DEADLKRES option, see NOTES for more details) and the
590 sleepq_type() function for sleepqueues.
593 The rc.firewall and rc.firewall6 were unified, and
594 rc.firewall6 and rc.d/ip6fw were removed.
595 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
596 variables are obsoleted. Instead, the following new rc
597 variables are added to rc.d/ipfw:
599 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
600 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
601 firewall_simple_onet_ipv6, firewall_trusted_ipv6
603 The meanings correspond to the relevant IPv4 variables.
609 The default terminal emulation for syscons(4) has been changed
610 from cons25 to xterm on all platforms except pc98. This means
611 that the /etc/ttys file needs to be updated to ensure correct
612 operation of applications on the console.
614 The terminal emulation style can be toggled per window by using
615 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
616 options can be used to change the compile-time default back to
619 To prevent graphical artifacts, make sure the TERM environment
620 variable is set to match the terminal emulation that is being
621 performed by syscons(4).
624 The layout of the structure ieee80211req_scan_result has changed.
625 Applications that require wireless scan results (e.g. ifconfig(8))
626 from net80211 need to be recompiled.
628 Applications such as wpa_supplicant(8) may require a full world
629 build without using NO_CLEAN in order to get synchronized with the
633 The iwn(4) driver has been updated to support the 5000 and 5150 series.
634 There's one kernel module for each firmware. Adding "device iwnfw"
635 to the kernel configuration file means including all three firmware
636 images inside the kernel. If you want to include just the one for
637 your wireless card, use the devices iwn4965fw, iwn5000fw or
641 The rc.d/network_ipv6, IPv6 configuration script has been integrated
642 into rc.d/netif. The changes are the following:
644 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
645 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
646 Note that both variables need the "inet6" keyword at the head.
648 Do not set $ipv6_network_interfaces manually if you do not
649 understand what you are doing. It is not needed in most cases.
651 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
654 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
655 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
657 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
658 all configured interfaces have "inet6 accept_rtadv" in the
659 $ifconfig_IF_ipv6. These are for backward compatibility.
661 3. A new variable $ipv6_prefer has been added. If NO, IPv6
662 functionality of interfaces with no corresponding
663 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
664 and the default address selection policy of ip6addrctl(8)
665 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
666 Note that if you want to configure IPv6 functionality on the
667 disabled interfaces after boot, first you need to clear the flag by
668 using ifconfig(8) like:
670 ifconfig em0 inet6 -ifdisabled
672 If YES, the default address selection policy is set as
675 The default value of $ipv6_prefer is NO.
677 4. If your system need to receive Router Advertisement messages,
678 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
679 scripts automatically invoke rtsol(8) when the interface becomes
680 UP. The Router Advertisement messages are used for SLAAC
681 (State-Less Address AutoConfiguration).
684 802.11s D3.03 support was committed. This is incompatible with the
685 previous code, which was based on D3.0.
688 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
689 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
690 control whether accepting Router Advertisement messages or not.
691 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
692 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
693 The ifconfig(8) utility now supports these flags.
696 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
697 mount(8) and -a option for df(1) to see them.
700 The old tunable hw.bus.devctl_disable has been superseded by
701 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
702 replaced by hw.bus.devctl_queue=0. The default for this new tunable
706 Remove the option STOP_NMI. The default action is now to use NMI only
707 for KDB via the newly introduced function stop_cpus_hard() and
708 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
711 The stable/8 branch created in subversion. This corresponds to the
712 RELENG_8 branch in CVS.
715 Bump the shared library version numbers for all libraries that do not
716 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
717 __FreeBSD_version to 800105.
720 Due to changes in the implementation of virtual network stack support,
721 all network-related kernel modules must be recompiled. As this change
722 breaks the ABI, bump __FreeBSD_version to 800104.
725 The TOE interface to the TCP syncache has been modified to remove
726 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
727 The cxgb driver is the only TOE consumer affected by this change, and
728 needs to be recompiled along with the kernel. As this change breaks
729 the ABI, bump __FreeBSD_version to 800103.
732 Padding has been added to struct tcpcb, sackhint and tcpstat in
733 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
734 maintaining the ABI. However, this change breaks the ABI, so bump
735 __FreeBSD_version to 800102. User space tools that rely on the size of
736 any of these structs (e.g. sockstat) need to be recompiled.
739 The NFS_LEGACYRPC option has been removed along with the old kernel
740 RPC implementation that this option selected. Kernel configurations
741 may need to be adjusted.
744 The network interface device nodes at /dev/net/<interface> have been
745 removed. All ioctl operations can be performed the normal way using
746 routing sockets. The kqueue functionality can generally be replaced
747 with routing sockets.
750 The documentation from the FreeBSD Documentation Project (Handbook,
751 FAQ, etc.) is now installed via packages by sysinstall(8) and under
752 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
755 The ABI of various structures related to the SYSV IPC API have been
756 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
757 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
761 Layout of struct vnet has changed as routing related variables were
762 moved to their own Vimage module. Modules need to be recompiled. Bump
763 __FreeBSD_version to 800099.
766 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
767 respectively. As long as no more than 16 groups per process are used,
768 no changes should be visible. When more than 16 groups are used, old
769 binaries may fail if they call getgroups() or getgrouplist() with
770 statically sized storage. Recompiling will work around this, but
771 applications should be modified to use dynamically allocated storage
772 for group arrays as POSIX.1-2008 does not cap an implementation's
773 number of supported groups at NGROUPS_MAX+1 as previous versions did.
775 NFS and portalfs mounts may also be affected as the list of groups is
776 truncated to 16. Users of NFS who use more than 16 groups, should
777 take care that negative group permissions are not used on the exported
778 file systems as they will not be reliable unless a GSSAPI based
779 authentication method is used.
782 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
783 option compiles in the support for adaptive spinning for lockmgrs
784 which want to enable it. The lockinit() function now accepts the flag
785 LK_ADAPTIVE in order to make the lock object subject to adaptive
786 spinning when both held in write and read mode.
789 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
790 changed. User applications that use this ioctl need to be rebuilt.
793 The layout of struct thread has changed. Kernel and modules need to
797 The layout of structs ifnet, domain, protosw and vnet_net has changed.
798 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
801 window(1) has been removed from the base system. It can now be
802 installed from ports. The port is called misc/window.
805 The way we are storing and accessing `routing table' entries has
806 changed. Programs reading the FIB, like netstat, need to be
810 A new netisr implementation has been added for FreeBSD 8. Network
811 file system modules, such as igmp, ipdivert, and others, should be
813 Bump __FreeBSD_version to 800096.
816 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
820 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
821 Bump __FreeBSD_version to 800094.
824 Add mnt_xflag field to 'struct mount'. File system modules need to be
826 Bump __FreeBSD_version to 800093.
829 The compiling option ADAPTIVE_SX has been retired while it has been
830 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
831 The KPI for sx_init_flags() changes as accepting flags:
832 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
833 been introduced in order to handle the reversed logic.
834 Bump __FreeBSD_version to 800092.
837 Add support for hierarchical jails. Remove global securelevel.
838 Bump __FreeBSD_version to 800091.
841 The layout of struct vnet_net has changed, therefore modules
843 Bump __FreeBSD_version to 800090.
846 The newly imported zic(8) produces a new format in the output. Please
847 run tzsetup(8) to install the newly created data to /etc/localtime.
850 The sysctl tree for the usb stack has renamed from hw.usb2.* to
851 hw.usb.* and is now consistent again with previous releases.
854 802.11 monitor mode support was revised and driver api's were changed.
855 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
856 of DLT_IEEE802_11. No user-visible data structures were changed but
857 applications that use DLT_IEEE802_11 may require changes.
858 Bump __FreeBSD_version to 800088.
861 The layout of the following structs has changed: sysctl_oid,
862 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
863 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
864 panics may be experienced. World rebuild is required for
865 correctly checking networking state from userland.
866 Bump __FreeBSD_version to 800085.
869 MLDv2 and Source-Specific Multicast (SSM) have been merged
870 to the IPv6 stack. VIMAGE hooks are in but not yet used.
871 The implementation of SSM within FreeBSD's IPv6 stack closely
872 follows the IPv4 implementation.
874 For kernel developers:
876 * The most important changes are that the ip6_output() and
877 ip6_input() paths no longer take the IN6_MULTI_LOCK,
878 and this lock has been downgraded to a non-recursive mutex.
880 * As with the changes to the IPv4 stack to support SSM, filtering
881 of inbound multicast traffic must now be performed by transport
882 protocols within the IPv6 stack. This does not apply to TCP and
883 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
885 * The KPIs used by IPv6 multicast are similar to those used by
886 the IPv4 stack, with the following differences:
887 * im6o_mc_filter() is analogous to imo_multicast_filter().
888 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
889 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
890 * IN6_LOOKUP_MULTI() has been deprecated and removed.
891 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
892 for MLDv1 have an additional 'timer' argument which is used to
893 jitter the initial membership report for the solicited-node
894 multicast membership on-link.
895 * This is not strictly needed for MLDv2, which already jitters
896 its report transmissions. However, the 'timer' argument is
897 preserved in case MLDv1 is active on the interface.
899 * The KAME linked-list based IPv6 membership implementation has
900 been refactored to use a vector similar to that used by the IPv4
902 Code which maintains a list of its own multicast memberships
903 internally, e.g. carp, has been updated to reflect the new
906 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
907 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
908 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
909 implementation constraint which needs to be addressed in HEAD.
911 For application developers:
913 * The changes are broadly similar to those made for the IPv4
916 * The use of IPv4 and IPv6 multicast socket options on the same
917 socket, using mapped addresses, HAS NOT been tested or supported.
919 * There are a number of issues with the implementation of various
920 IPv6 multicast APIs which need to be resolved in the API surface
921 before the implementation is fully compatible with KAME userland
922 use, and these are mostly to do with interface index treatment.
924 * The literature available discusses the use of either the delta / ASM
925 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
926 using setsourcefilter(3)/getsourcefilter(3). For more information
927 please refer to RFC 3768, 'Socket Interface Extensions for
928 Multicast Source Filters'.
930 * Applications which use the published RFC 3678 APIs should be fine.
932 For systems administrators:
934 * The mtest(8) utility has been refactored to support IPv6, in
935 addition to IPv4. Interface addresses are no longer accepted
936 as arguments, their names must be used instead. The utility
937 will map the interface name to its first IPv4 address as
938 returned by getifaddrs(3).
940 * The ifmcstat(8) utility has also been updated to print the MLDv2
941 endpoint state and source filter lists via sysctl(3).
943 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
944 loopback of IPv6 multicast datagrams by default; it defaults to 1
945 to preserve the existing behaviour. Disabling multicast loopback is
946 recommended for optimal system performance.
948 * The IPv6 MROUTING code has been changed to examine this sysctl
949 instead of attempting to perform a group lookup before looping
950 back forwarded datagrams.
952 Bump __FreeBSD_version to 800084.
955 Implement low-level Bluetooth HCI API.
956 Bump __FreeBSD_version to 800083.
959 The layout of struct malloc_type, used by modules to register new
960 memory allocation types, has changed. Most modules will need to
961 be rebuilt or panics may be experienced.
962 Bump __FreeBSD_version to 800081.
965 Anticipate overflowing inp_flags - add inp_flags2.
966 This changes most offsets in inpcb, so checking v4 connection
967 state will require a world rebuild.
968 Bump __FreeBSD_version to 800080.
971 Add an llentry to struct route and struct route_in6. Modules
972 embedding a struct route will need to be recompiled.
973 Bump __FreeBSD_version to 800079.
976 The size of rt_metrics_lite and by extension rtentry has changed.
977 Networking administration apps will need to be recompiled.
978 The route command now supports show as an alias for get, weighting
979 of routes, sticky and nostick flags to alter the behavior of stateful
981 Bump __FreeBSD_version to 800078.
984 Do not use Giant for kbdmux(4) locking. This is wrong and
985 apparently causing more problems than it solves. This will
986 re-open the issue where interrupt handlers may race with
987 kbdmux(4) in polling mode. Typical symptoms include (but
988 not limited to) duplicated and/or missing characters when
989 low level console functions (such as gets) are used while
990 interrupts are enabled (for example geli password prompt,
991 mountroot prompt etc.). Disabling kbdmux(4) may help.
994 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
995 kernel modules referencing any of the above need to be recompiled.
996 Bump __FreeBSD_version to 800075.
999 GEOM_PART has become the default partition slicer for storage devices,
1000 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
1001 introduces some changes:
1003 MSDOS/EBR: the devices created from MSDOS extended partition entries
1004 (EBR) can be named differently than with GEOM_MBR and are now symlinks
1005 to devices with offset-based names. fstabs may need to be modified.
1007 BSD: the "geometry does not match label" warning is harmless in most
1008 cases but it points to problems in file system misalignment with
1009 disk geometry. The "c" partition is now implicit, covers the whole
1010 top-level drive and cannot be (mis)used by users.
1012 General: Kernel dumps are now not allowed to be written to devices
1013 whose partition types indicate they are meant to be used for file
1014 systems (or, in case of MSDOS partitions, as something else than
1017 Most of these changes date approximately from 200812.
1020 The uscanner(4) driver has been removed from the kernel. This follows
1021 Linux removing theirs in 2.6 and making libusb the default interface
1022 (supported by sane).
1025 The multicast forwarding code has been cleaned up. netstat(1)
1026 only relies on KVM now for printing bandwidth upcall meters.
1027 The IPv4 and IPv6 modules are split into ip_mroute_mod and
1028 ip6_mroute_mod respectively. The config(5) options for statically
1029 compiling this code remain the same, i.e. 'options MROUTING'.
1032 Support for the IFF_NEEDSGIANT network interface flag has been
1033 removed, which means that non-MPSAFE network device drivers are no
1034 longer supported. In particular, if_ar, if_sr, and network device
1035 drivers from the old (legacy) USB stack can no longer be built or
1039 POSIX.1 Native Language Support (NLS) has been enabled in libc and
1040 a bunch of new language catalog files have also been added.
1041 This means that some common libc messages are now localized and
1042 they depend on the LC_MESSAGES environmental variable.
1045 The k8temp(4) driver has been renamed to amdtemp(4) since
1046 support for Family 10 and Family 11 CPU families was added.
1049 IGMPv3 and Source-Specific Multicast (SSM) have been merged
1050 to the IPv4 stack. VIMAGE hooks are in but not yet used.
1052 For kernel developers, the most important changes are that the
1053 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
1054 and this lock has been downgraded to a non-recursive mutex.
1056 Transport protocols (UDP, Raw IP) are now responsible for filtering
1057 inbound multicast traffic according to group membership and source
1058 filters. The imo_multicast_filter() KPI exists for this purpose.
1059 Transports which do not use multicast (SCTP, TCP) already reject
1060 multicast by default. Forwarding and receive performance may improve
1061 as a mutex acquisition is no longer needed in the ip_input()
1062 low-level input path. in_addmulti() and in_delmulti() are shimmed
1063 to new KPIs which exist to support SSM in-kernel.
1065 For application developers, it is recommended that loopback of
1066 multicast datagrams be disabled for best performance, as this
1067 will still cause the lock to be taken for each looped-back
1068 datagram transmission. The net.inet.ip.mcast.loop sysctl may
1069 be tuned to 0 to disable loopback by default; it defaults to 1
1070 to preserve the existing behaviour.
1072 For systems administrators, to obtain best performance with
1073 multicast reception and multiple groups, it is always recommended
1074 that a card with a suitably precise hash filter is used. Hash
1075 collisions will still result in the lock being taken within the
1076 transport protocol input path to check group membership.
1078 If deploying FreeBSD in an environment with IGMP snooping switches,
1079 it is recommended that the net.inet.igmp.sendlocal sysctl remain
1080 enabled; this forces 224.0.0.0/24 group membership to be announced
1083 The size of 'struct igmpstat' has changed; netstat needs to be
1084 recompiled to reflect this.
1085 Bump __FreeBSD_version to 800070.
1088 libusb20.so.1 is now installed as libusb.so.1 and the ports system
1089 updated to use it. This requires a buildworld/installworld in order to
1090 update the library and dependencies (usbconfig, etc). Its advisable to
1091 rebuild all ports which uses libusb. More specific directions are given
1092 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
1093 libusb are no longer required and can be removed.
1096 A workaround is committed to allow the creation of System V shared
1097 memory segment of size > 2 GB on the 64-bit architectures.
1098 Due to a limitation of the existing ABI, the shm_segsz member
1099 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
1100 wrong for large segments. Note that limits must be explicitly
1101 raised to allow such segments to be created.
1104 The layout of struct ifnet has changed, requiring a rebuild of all
1105 network device driver modules.
1108 The /dev handling for the new USB stack has changed, a
1109 buildworld/installworld is required for libusb20.
1112 The new USB2 stack has now been permanently moved in and all kernel and
1113 module names reverted to their previous values (eg, usb, ehci, ohci,
1114 ums, ...). The old usb stack can be compiled in by prefixing the name
1115 with the letter 'o', the old usb modules have been removed.
1116 Updating entry 20090216 for xorg and 20090215 for libmap may still
1120 The rc.conf(5) option if_up_delay has been renamed to
1121 defaultroute_delay to better reflect its purpose. If you have
1122 customized this setting in /etc/rc.conf you need to update it to
1126 xorg 7.4 wants to configure its input devices via hald which does not
1127 yet work with USB2. If the keyboard/mouse does not work in xorg then
1129 Option "AllowEmptyInput" "off"
1130 to your ServerLayout section. This will cause X to use the configured
1131 kbd and mouse sections from your xorg.conf.
1134 The GENERIC kernels for all architectures now default to the new USB2
1135 stack. No kernel config options or code have been removed so if a
1136 problem arises please report it and optionally revert to the old USB
1137 stack. If you are loading USB kernel modules or have a custom kernel
1138 that includes GENERIC then ensure that usb names are also changed over,
1139 eg uftdi -> usb2_serial_ftdi.
1141 Older programs linked against the ports libusb 0.1 need to be
1142 redirected to the new stack's libusb20. /etc/libmap.conf can
1144 # Map old usb library to new one for usb2 stack
1145 libusb-0.1.so.8 libusb20.so.1
1148 All USB ethernet devices now attach as interfaces under the name ueN
1149 (eg. ue0). This is to provide a predictable name as vendors often
1150 change usb chipsets in a product without notice.
1153 The ichsmb(4) driver has been changed to require SMBus slave
1154 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1155 All of the other SMBus controller drivers require left-justified
1156 slave addresses, so this change makes all the drivers provide the
1160 INET6 statistics (struct ip6stat) was updated.
1161 netstat(1) needs to be recompiled.
1164 NTFS has been removed from GENERIC kernel on amd64 to match
1165 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1166 will load ntfs.ko module automatically when NTFS support is
1167 actually needed, unless ntfs.ko is not installed or security
1168 level prohibits loading kernel modules. If either is the case,
1169 "options NTFS" has to be added into kernel config.
1172 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1173 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1174 800061. User space tools that rely on the size of struct tcpcb in
1175 tcp_var.h (e.g. sockstat) need to be recompiled.
1178 ng_tty(4) module updated to match the new TTY subsystem.
1179 Due to API change, user-level applications must be updated.
1180 New API support added to mpd5 CVS and expected to be present
1181 in next mpd5.3 release.
1184 With __FreeBSD_version 800060 the makefs tool is part of
1185 the base system (it was a port).
1188 The afdata and ifnet locks have been changed from mutexes to
1189 rwlocks, network modules will need to be re-compiled.
1192 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1193 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1194 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1195 architecture and 40 bytes on 64-bit architecture. The userland
1196 applications "arp" and "ndp" have been updated accordingly.
1197 The output from "netstat -r" shows only routing entries and
1198 none of the L2 information.
1201 __FreeBSD_version 800057 marks the switchover from the
1202 binary ath hal to source code. Users must add the line:
1204 options AH_SUPPORT_AR5416
1206 to their kernel config files when specifying:
1210 The ath_hal module no longer exists; the code is now compiled
1211 together with the driver in the ath module. It is now
1212 possible to tailor chip support (i.e. reduce the set of chips
1213 and thereby the code size); consult ath_hal(4) for details.
1216 __FreeBSD_version 800054 adds memory barriers to
1217 <machine/atomic.h>, new interfaces to ifnet to facilitate
1218 multiple hardware transmit queues for cards that support
1219 them, and a lock-less ring-buffer implementation to
1220 enable drivers to more efficiently manage queueing of
1224 A new version of ZFS (version 13) has been merged to -HEAD.
1225 This version has zpool attribute "listsnapshots" off by
1226 default, which means "zfs list" does not show snapshots,
1227 and is the same as Solaris behavior.
1230 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1233 The uhci, ohci, ehci and slhci USB Host controller drivers have
1234 been put into separate modules. If you load the usb module
1235 separately through loader.conf you will need to load the
1236 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1237 controller add the following to loader.conf:
1243 The ABI used by the PMC toolset has changed. Please keep
1244 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1248 atapci kernel module now includes only generic PCI ATA
1249 driver. AHCI driver moved to ataahci kernel module.
1250 All vendor-specific code moved into separate kernel modules:
1251 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1252 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1253 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1254 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1257 The TTY subsystem of the kernel has been replaced by a new
1258 implementation, which provides better scalability and an
1259 improved driver model. Most common drivers have been migrated to
1260 the new TTY subsystem, while others have not. The following
1261 drivers have not yet been ported to the new TTY layer:
1264 cy, digi, rc, rp, sio
1270 ng_h4, ng_tty, ppp, sl, snp
1272 Adding these drivers to your kernel configuration file shall
1273 cause compilation to fail.
1276 ntpd has been upgraded to 4.2.4p5.
1279 OpenSSH has been upgraded to 5.1p1.
1281 For many years, FreeBSD's version of OpenSSH preferred DSA
1282 over RSA for host and user authentication keys. With this
1283 upgrade, we've switched to the vendor's default of RSA over
1284 DSA. This may cause upgraded clients to warn about unknown
1285 host keys even for previously known hosts. Users should
1286 follow the usual procedure for verifying host keys before
1287 accepting the RSA key.
1289 This can be circumvented by setting the "HostKeyAlgorithms"
1290 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1293 Please note that the sequence of keys offered for
1294 authentication has been changed as well. You may want to
1295 specify IdentityFile in a different order to revert this
1299 The sio(4) driver has been removed from the i386 and amd64
1300 kernel configuration files. This means uart(4) is now the
1301 default serial port driver on those platforms as well.
1303 To prevent collisions with the sio(4) driver, the uart(4) driver
1304 uses different names for its device nodes. This means the
1305 onboard serial port will now most likely be called "ttyu0"
1306 instead of "ttyd0". You may need to reconfigure applications to
1307 use the new device names.
1309 When using the serial port as a boot console, be sure to update
1310 /boot/device.hints and /etc/ttys before booting the new kernel.
1311 If you forget to do so, you can still manually specify the hints
1312 at the loader prompt:
1314 set hint.uart.0.at="isa"
1315 set hint.uart.0.port="0x3F8"
1316 set hint.uart.0.flags="0x10"
1317 set hint.uart.0.irq="4"
1321 The gpt(8) utility has been removed. Use gpart(8) to partition
1325 The version that Linuxulator emulates was changed from 2.4.2
1326 to 2.6.16. If you experience any problems with Linux binaries
1327 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1328 if it fixes the problem contact emulation mailing list.
1331 ISDN4BSD (I4B) was removed from the src tree. You may need to
1332 update a your kernel configuration and remove relevant entries.
1335 I have checked in code to support multiple routing tables.
1336 See the man pages setfib(1) and setfib(2).
1337 This is a hopefully backwards compatible version,
1338 but to make use of it you need to compile your kernel
1339 with options ROUTETABLES=2 (or more up to 16).
1342 The 802.11 wireless support was redone to enable multi-bss
1343 operation on devices that are capable. The underlying device
1344 is no longer used directly but instead wlanX devices are
1345 cloned with ifconfig. This requires changes to rc.conf files.
1346 For example, change:
1347 ifconfig_ath0="WPA DHCP"
1350 ifconfig_wlan0="WPA DHCP"
1351 see rc.conf(5) for more details. In addition, mergemaster of
1352 /etc/rc.d is highly recommended. Simultaneous update of userland
1353 and kernel wouldn't hurt either.
1355 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1356 modules were merged into the base wlan module. All references
1357 to these modules (e.g. in kernel config files) must be removed.
1360 psm(4) has gained write(2) support in native operation level.
1361 Arbitrary commands can be written to /dev/psm%d and status can
1362 be read back from it. Therefore, an application is responsible
1363 for status validation and error recovery. It is a no-op in
1364 other operation levels.
1367 Support for KSE threading has been removed from the kernel. To
1368 run legacy applications linked against KSE libmap.conf may
1369 be used. The following libmap.conf may be used to ensure
1370 compatibility with any prior release:
1372 libpthread.so.1 libthr.so.1
1373 libpthread.so.2 libthr.so.2
1374 libkse.so.3 libthr.so.3
1377 The layout of struct vmspace has changed. This affects libkvm
1378 and any executables that link against libkvm and use the
1379 kvm_getprocs() function. In particular, but not exclusively,
1380 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1381 The effects are minimal, but it's advisable to upgrade world
1385 The latest em driver no longer has support in it for the
1386 82575 adapter, this is now moved to the igb driver. The
1387 split was done to make new features that are incompatible
1388 with older hardware easier to do.
1391 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1392 likewise the kernel option is now GEOM_LINUX_LVM.
1395 The default NFS mount mode has changed from UDP to TCP for
1396 increased reliability. If you rely on (insecurely) NFS
1397 mounting across a firewall you may need to update your
1401 Belatedly note the addition of m_collapse for compacting
1405 The fts(3) structures have been changed to use adequate
1406 integer types for their members and so to be able to cope
1407 with huge file trees. The old fts(3) ABI is preserved
1408 through symbol versioning in libc, so third-party binaries
1409 using fts(3) should still work, although they will not take
1410 advantage of the extended types. At the same time, some
1411 third-party software might fail to build after this change
1412 due to unportable assumptions made in its source code about
1413 fts(3) structure members. Such software should be fixed
1414 by its vendor or, in the worst case, in the ports tree.
1415 FreeBSD_version 800015 marks this change for the unlikely
1416 case that a portable fix is impossible.
1419 To upgrade to -current after this date, you must be running
1420 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1421 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1424 The ADAPTIVE_GIANT kernel option has been retired because its
1425 functionality is the default now.
1428 The AT keyboard emulation of sunkbd(4) has been turned on
1429 by default. In order to make the special symbols of the Sun
1430 keyboards driven by sunkbd(4) work under X these now have
1431 to be configured the same way as Sun USB keyboards driven
1432 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1434 Option "XkbLayout" "us"
1435 Option "XkbRules" "xorg"
1436 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1439 It has been decided that it is desirable to provide ABI
1440 backwards compatibility to the FreeBSD 4/5/6 versions of the
1441 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1442 broken with the introduction of PCI domain support (see the
1443 20070930 entry). Unfortunately, this required the ABI of
1444 PCIOCGETCONF to be broken again in order to be able to
1445 provide backwards compatibility to the old version of that
1446 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1447 again. As for prominent ports this affects neither pciutils
1448 nor xorg-server this time, the hal port needs to be rebuilt
1452 The misnamed kthread_create() and friends have been renamed
1453 to kproc_create() etc. Many of the callers already
1454 used kproc_start()..
1455 I will return kthread_create() and friends in a while
1456 with implementations that actually create threads, not procs.
1457 Renaming corresponds with version 800002.
1466 Avoid using make -j when upgrading. While generally safe, there are
1467 sometimes problems using -j to upgrade. If your upgrade fails with
1468 -j, please try again without -j. From time to time in the past there
1469 have been problems using -j with buildworld and/or installworld. This
1470 is especially true when upgrading between "distant" versions (eg one
1471 that cross a major release boundary or several minor releases, or when
1472 several months have passed on the -current branch).
1474 Sometimes, obscure build problems are the result of environment
1475 poisoning. This can happen because the make utility reads its
1476 environment when searching for values for global variables. To run
1477 your build attempts in an "environmental clean room", prefix all make
1478 commands with 'env -i '. See the env(1) manual page for more details.
1480 When upgrading from one major version to another it is generally best
1481 to upgrade to the latest code in the currently installed branch first,
1482 then do an upgrade to the new branch. This is the best-tested upgrade
1483 path, and has the highest probability of being successful. Please try
1484 this approach before reporting problems with a major version upgrade.
1488 When upgrading the boot ZFS pool to a new version, always follow
1491 1.) recompile and reinstall the ZFS boot loader and boot block
1492 (this is part of "make buildworld" and "make installworld")
1494 2.) update the ZFS boot block on your boot drive
1496 The following example updates the ZFS boot block on the first
1497 partition (freebsd-boot) of a GPT partitioned drive ad0:
1498 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1500 Non-boot pools do not need these updates.
1504 If you are updating from a prior version of FreeBSD (even one just
1505 a few days old), you should follow this procedure. It is the most
1506 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1508 make kernel-toolchain
1509 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1510 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1512 To test a kernel once
1513 ---------------------
1514 If you just want to boot a kernel once (because you are not sure
1515 if it works, or if you want to boot a known bad kernel to provide
1516 debugging information) run
1517 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1518 nextboot -k testkernel
1520 To just build a kernel when you know that it won't mess you up
1521 --------------------------------------------------------------
1522 This assumes you are already running a CURRENT system. Replace
1523 ${arch} with the architecture of your machine (e.g. "i386",
1524 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1526 cd src/sys/${arch}/conf
1527 config KERNEL_NAME_HERE
1528 cd ../compile/KERNEL_NAME_HERE
1533 If this fails, go to the "To build a kernel" section.
1535 To rebuild everything and install it on the current system.
1536 -----------------------------------------------------------
1537 # Note: sometimes if you are running current you gotta do more than
1538 # is listed here if you are upgrading from a really old current.
1540 <make sure you have good level 0 dumps>
1542 make kernel KERNCONF=YOUR_KERNEL_HERE
1544 <reboot in single user> [3]
1552 To cross-install current onto a separate partition
1553 --------------------------------------------------
1554 # In this approach we use a separate partition to hold
1555 # current's root, 'usr', and 'var' directories. A partition
1556 # holding "/", "/usr" and "/var" should be about 2GB in
1559 <make sure you have good level 0 dumps>
1562 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1563 <maybe newfs current's root partition>
1564 <mount current's root partition on directory ${CURRENT_ROOT}>
1565 make installworld DESTDIR=${CURRENT_ROOT}
1566 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1567 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1568 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1569 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1570 <reboot into current>
1571 <do a "native" rebuild/install as described in the previous section>
1572 <maybe install compatibility libraries from ports/misc/compat*>
1576 To upgrade in-place from stable to current
1577 ----------------------------------------------
1578 <make sure you have good level 0 dumps>
1580 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1582 <reboot in single user> [3]
1589 Make sure that you've read the UPDATING file to understand the
1590 tweaks to various things you need. At this point in the life
1591 cycle of current, things change often and you are on your own
1592 to cope. The defaults can also change, so please read ALL of
1593 the UPDATING entries.
1595 Also, if you are tracking -current, you must be subscribed to
1596 freebsd-current@freebsd.org. Make sure that before you update
1597 your sources that you have read and understood all the recent
1598 messages there. If in doubt, please track -stable which has
1599 much fewer pitfalls.
1601 [1] If you have third party modules, such as vmware, you
1602 should disable them at this point so they don't crash your
1605 [3] From the bootblocks, boot -s, and then do
1610 adjkerntz -i # if CMOS is wall time
1611 Also, when doing a major release upgrade, it is required that
1612 you boot into single user mode to do the installworld.
1614 [4] Note: This step is non-optional. Failure to do this step
1615 can result in a significant reduction in the functionality of the
1616 system. Attempting to do it by hand is not recommended and those
1617 that pursue this avenue should read this file carefully, as well
1618 as the archives of freebsd-current and freebsd-hackers mailing lists
1619 for potential gotchas. The -U option is also useful to consider.
1620 See mergemaster(8) for more information.
1622 [5] Usually this step is a noop. However, from time to time
1623 you may need to do this if you get unknown user in the following
1624 step. It never hurts to do it all the time. You may need to
1625 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1626 install) after the buildworld before this step if you last updated
1627 from [78]-stable or 9-stable before 20130430.
1629 [6] This only deletes old files and directories. Old libraries
1630 can be deleted by "make delete-old-libs", but you have to make
1631 sure that no program is using those libraries anymore.
1633 [8] In order to have a kernel that can run the 4.x binaries needed to
1634 do an installworld, you must include the COMPAT_FREEBSD4 option in
1635 your kernel. Failure to do so may leave you with a system that is
1636 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1637 required to run the 5.x binaries on more recent kernels. And so on
1638 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1640 Make sure that you merge any new devices from GENERIC since the
1641 last time you updated your kernel config file.
1643 [9] When checking out sources, you must include the -P flag to have
1644 cvs prune empty directories.
1646 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1647 "?=" instead of the "=" assignment operator, so that buildworld can
1648 override the CPUTYPE if it needs to.
1650 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1651 not on the command line, or in /etc/make.conf. buildworld will
1652 warn if it is improperly defined.
1655 This file contains a list, in reverse chronological order, of major
1656 breakages in tracking -current. It is not guaranteed to be a complete
1657 list of such breakages, and only contains entries since October 10, 2007.
1658 If you need to see UPDATING entries from before that date, you will need
1659 to fetch an UPDATING file from an older FreeBSD release.
1661 Copyright information:
1663 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1665 Redistribution, publication, translation and use, with or without
1666 modification, in full or in part, in any form or format of this
1667 document are permitted without further permission from the author.
1669 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1670 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1671 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1672 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1673 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1674 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1675 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1676 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1677 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1678 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1679 POSSIBILITY OF SUCH DAMAGE.
1681 Contact Warner Losh if you have any questions about your use of