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.
14 NOTE: FreeBSD has switched from gcc to clang. If you have trouble bootstrapping
15 from older versions of FreeBSD, try WITHOUT_CLANG to bootstrap to the tip of
16 head, and then rebuild without this option. The bootstrap process from
17 older version of current is a bit fragile.
19 NOTE TO PEOPLE WHO THINK THAT FreeBSD 10.x IS SLOW:
20 FreeBSD 10.x has many debugging features turned on, in both the kernel
21 and userland. These features attempt to detect incorrect use of
22 system primitives, and encourage loud failure through extra sanity
23 checking and fail stop semantics. They also substantially impact
24 system performance. If you want to do performance measurement,
25 benchmarking, and optimization, you'll want to turn them off. This
26 includes various WITNESS- related kernel options, INVARIANTS, malloc
27 debugging flags in userland, and various verbose features in the
28 kernel. Many developers choose to disable these features on build
29 machines to maximize performance. (To completely disable malloc
30 debugging, define MALLOC_PRODUCTION in /etc/make.conf, or to merely
31 disable the most expensive debugging functionality run
32 "ln -s 'abort:false,junk:false' /etc/malloc.conf".)
35 The GNU Compiler Collection and C++ standard library (libstdc++)
36 are no longer built by default on platforms where clang is the system
37 compiler. You can enable them with the WITH_GCC and WITH_GNUCXX
41 The PROCDESC kernel option is now part of the GENERIC kernel
42 configuration and is required for the rwhod(8) to work.
43 If you are using custom kernel configuration, you should include
47 The API and ABI related to the Capsicum framework was modified
48 in backward incompatible way. The userland libraries and programs
49 have to be recompiled to work with the new kernel. This includes the
50 following libraries and programs, but the whole buildworld is
51 advised: libc, libprocstat, dhclient, tcpdump, hastd, hastctl,
52 kdump, procstat, rwho, rwhod, uniq.
55 AES-NI intrinsic support has been added to gcc. The AES-NI module
56 has been updated to use this support. A new gcc is required to build
57 the aesni module on both i386 and amd64.
60 Thomas Dickey (vendor author thereof) reports that dialog(1) since
61 2011/10/18 has a bug in handling --hline. Testers and I noticed the
62 --hline is not ignored but displayed as a NULL string, regardless of
63 value. This will cause confusion in some bsdconfig dialogs where the
64 --hline is used to inform users which keybindings to use. This will
65 likewise affect any other persons relying on --hline. It also looks
66 rather strange seeing "[]" at the bottom of dialog(1) widgets when
67 passing --hline "anything". Thomas said he will have a look in a few
68 weeks. NOTE: The "[]" brackets appear with the left-edge where it
69 would normally appear given the width of text to display, but the
70 displayed text is not there (part of the bug).
73 The PADLOCK_RNG and RDRAND_RNG kernel options are now devices.
74 Thus "device padlock_rng" and "device rdrand_rng" should be
75 used instead of "options PADLOCK_RNG" & "options RDRAND_RNG".
78 WITH_ICONV has been split into two feature sets. WITH_ICONV now
79 enables just the iconv* functionality and is now on by default.
80 WITH_LIBICONV_COMPAT enables the libiconv api and link time
81 compatability. Set WITHOUT_ICONV to build the old way.
82 If you have been using WITH_ICONV before, you will very likely
83 need to turn on WITH_LIBICONV_COMPAT.
86 INVARIANTS option now enables DEBUG for code with OpenSolaris and
87 Illumos origin, including ZFS. If you have INVARIANTS in your
88 kernel configuration, then there is no need to set DEBUG or ZFS_DEBUG
90 DEBUG used to enable witness(9) tracking of OpenSolaris (mostly ZFS)
91 locks if WITNESS option was set. Because that generated a lot of
92 witness(9) reports and all of them were believed to be false
93 positives, this is no longer done. New option OPENSOLARIS_WITNESS
94 can be used to achieve the previous behavior.
97 Timer values in IPv6 data structures now use time_uptime instead
98 of time_second. Although this is not a user-visible functional
99 change, userland utilities which directly use them---ndp(8),
100 rtadvd(8), and rtsold(8) in the base system---need to be updated
104 find -delete can now delete the pathnames given as arguments,
105 instead of only files found below them or if the pathname did
106 not contain any slashes. Formerly, the following error message
109 find: -delete: <path>: relative path potentially not safe
111 Deleting the pathnames given as arguments can be prevented
112 without error messages using -mindepth 1 or by changing
113 directory and passing "." as argument to find. This works in the
114 old as well as the new version of find.
117 Behavior of devfs rules path matching has been changed.
118 Pattern is now always matched against fully qualified devfs
119 path and slash characters must be explicitly matched by
120 slashes in pattern (FNM_PATHNAME). Rulesets involving devfs
121 subdirectories must be reviewed.
124 The default ARM ABI has changed to the ARM EABI. The old ABI is
125 incompatible with the ARM EABI and all programs and modules will
126 need to be rebuilt to work with a new kernel.
128 To keep using the old ABI ensure the WITHOUT_ARM_EABI knob is set.
130 NOTE: Support for the old ABI will be removed in the future and
131 users are advised to upgrade.
134 pkg_install has been disconnected from the build if you really need it
135 you should add WITH_PKGTOOLS in your src.conf(5).
138 Most of network statistics structures were changed to be able
139 keep 64-bits counters. Thus all tools, that work with networking
140 statistics, must be rebuilt (netstat(1), bsnmpd(1), etc.)
143 Fix targets that run multiple make's to use && rather than ;
144 so that subsequent steps depend on success of previous.
146 NOTE: if building 'universe' with -j* on stable/8 or stable/9
147 it would be better to start the build using bmake, to avoid
148 overloading the machine.
151 Fix a bug that allowed a tracing process (e.g. gdb) to write
152 to a memory-mapped file in the traced process's address space
153 even if neither the traced process nor the tracing process had
154 write access to that file.
157 CVS has been removed from the base system. An exact copy
158 of the code is available from the devel/cvs port.
161 Some people report the following error after the switch to bmake:
163 make: illegal option -- J
164 usage: make [-BPSXeiknpqrstv] [-C directory] [-D variable]
166 *** [buildworld] Error code 2
168 this likely due to an old instance of make in
169 ${MAKEPATH} (${MAKEOBJDIRPREFIX}${.CURDIR}/make.${MACHINE})
170 which src/Makefile will use that blindly, if it exists, so if
171 you see the above error:
173 rm -rf `make -V MAKEPATH`
178 Use bmake by default.
179 Whereas before one could choose to build with bmake via
180 -DWITH_BMAKE one must now use -DWITHOUT_BMAKE to use the old
181 make. The goal is to remove these knobs for 10-RELEASE.
183 It is worth noting that bmake (like gmake) treats the command
184 line as the unit of failure, rather than statements within the
185 command line. Thus '(cd some/where && dosomething)' is safer
186 than 'cd some/where; dosomething'. The '()' allows consistent
187 behavior in parallel build.
190 Fix a bug that allows NFS clients to issue READDIR on files.
193 The WITHOUT_IDEA option has been removed because
194 the IDEA patent expired.
197 The sysctl which controls TRIM support under ZFS has been renamed
198 from vfs.zfs.trim_disable -> vfs.zfs.trim.enabled and has been
202 The mergemaster command now uses the default MAKEOBJDIRPREFIX
203 rather than creating it's own in the temporary directory in
204 order allow access to bootstrapped versions of tools such as
205 install and mtree. When upgrading from version of FreeBSD where
206 the install command does not support -l, you will need to
207 install a new mergemaster command if mergemaster -p is required.
208 This can be accomplished with the command (cd src/usr.sbin/mergemaster
212 Legacy ATA stack, disabled and replaced by new CAM-based one since
213 FreeBSD 9.0, completely removed from the sources. Kernel modules
214 atadisk and atapi*, user-level tools atacontrol and burncd are
215 removed. Kernel option `options ATA_CAM` is now permanently enabled
219 SOCK_CLOEXEC and SOCK_NONBLOCK flags have been added to socket(2)
220 and socketpair(2). Software, in particular Kerberos, may
221 automatically detect and use these during building. The resulting
222 binaries will not work on older kernels.
225 CTL_DISABLE has also been added to the sparc64 GENERIC (for further
226 information, see the respective 20130304 entry).
229 Recent commits to callout(9) changed the size of struct callout,
230 so the KBI is probably heavily disturbed. Also, some functions
231 in callout(9)/sleep(9)/sleepqueue(9)/condvar(9) KPIs were replaced
232 by macros. Every kernel module using it won't load, so rebuild
235 The ctl device has been re-enabled in GENERIC for i386 and amd64,
236 but does not initialize by default (because of the new CTL_DISABLE
237 option) to save memory. To re-enable it, remove the CTL_DISABLE
238 option from the kernel config file or set kern.cam.ctl.disable=0
239 in /boot/loader.conf.
242 The ctl device has been disabled in GENERIC for i386 and amd64.
243 This was done due to the extra memory being allocated at system
244 initialisation time by the ctl driver which was only used if
245 a CAM target device was created. This makes a FreeBSD system
246 unusable on 128MB or less of RAM.
249 A new compression method (lz4) has been merged to -HEAD. Please
250 refer to zpool-features(7) for more information.
252 Please refer to the "ZFS notes" section of this file for information
253 on upgrading boot ZFS pools.
256 A BSD-licensed patch(1) variant has been added and is installed
257 as bsdpatch, being the GNU version the default patch.
258 To inverse the logic and use the BSD-licensed one as default,
259 while having the GNU version installed as gnupatch, rebuild
260 and install world with the WITH_BSD_PATCH knob set.
263 Due to the use of the new -l option to install(1) during build
264 and install, you must take care not to directly set the INSTALL
265 make variable in your /etc/make.conf, /etc/src.conf, or on the
266 command line. If you wish to use the -C flag for all installs
267 you may be able to add INSTALL+=-C to /etc/make.conf or
271 The install(1) option -M has changed meaning and now takes an
272 argument that is a file or path to append logs to. In the
273 unlikely event that -M was the last option on the command line
274 and the command line contained at least two files and a target
275 directory the first file will have logs appended to it. The -M
276 option served little practical purpose in the last decade so its
277 use is expected to be extremely rare.
280 After switching to Clang as the default compiler some users of ZFS
281 on i386 systems started to experience stack overflow kernel panics.
282 Please consider using 'options KSTACK_PAGES=4' in such configurations.
285 GEOM_LABEL now mangles label names read from file system metadata.
286 Mangling affect labels containing spaces, non-printable characters,
287 '%' or '"'. Device names in /etc/fstab and other places may need to
291 By default, only the 10 most recent kernel dumps will be saved. To
292 restore the previous behaviour (no limit on the number of kernel dumps
293 stored in the dump directory) add the following line to /etc/rc.conf:
298 With the addition of auditdistd(8), a new auditdistd user is now
299 depended on during installworld. "mergemaster -p" can be used to add
300 the user prior to installworld, as documented in the handbook.
303 The sin6_scope_id member variable in struct sockaddr_in6 is now
304 filled by the kernel before passing the structure to the userland via
305 sysctl or routing socket. This means the KAME-specific embedded scope
306 id in sin6_addr.s6_addr[2] is always cleared in userland application.
307 This behavior can be controlled by net.inet6.ip6.deembed_scopeid.
308 __FreeBSD_version is bumped to 1000025.
311 On i386 and amd64 systems WITH_CLANG_IS_CC is now the default.
312 This means that the world and kernel will be compiled with clang
313 and that clang will be installed as /usr/bin/cc, /usr/bin/c++,
314 and /usr/bin/cpp. To disable this behavior and revert to building
315 with gcc, compile with WITHOUT_CLANG_IS_CC. Really old versions
316 of current may need to bootstrap WITHOUT_CLANG first if the clang
317 build fails (its compatibility window doesn't extend to the 9 stable
321 The IPFIREWALL_FORWARD kernel option has been removed. Its
322 functionality now turned on by default.
325 The ZERO_COPY_SOCKET kernel option has been removed and
326 split into SOCKET_SEND_COW and SOCKET_RECV_PFLIP.
327 NB: SOCKET_SEND_COW uses the VM page based copy-on-write
328 mechanism which is not safe and may result in kernel crashes.
329 NB: The SOCKET_RECV_PFLIP mechanism is useless as no current
330 driver supports disposeable external page sized mbuf storage.
331 Proper replacements for both zero-copy mechanisms are under
332 consideration and will eventually lead to complete removal
333 of the two kernel options.
336 The IPv4 network stack has been converted to network byte
337 order. The following modules need to be recompiled together
338 with kernel: carp(4), divert(4), gif(4), siftr(4), gre(4),
339 pf(4), ipfw(4), ng_ipfw(4), stf(4).
342 Support for non-MPSAFE filesystems was removed from VFS. The
343 VFS_VERSION was bumped, all filesystem modules shall be
347 All the non-MPSAFE filesystems have been disconnected from
348 the build. The full list includes: codafs, hpfs, ntfs, nwfs,
349 portalfs, smbfs, xfs.
352 The interface cloning API and ABI has changed. The following
353 modules need to be recompiled together with kernel:
354 ipfw(4), pfsync(4), pflog(4), usb(4), wlan(4), stf(4),
355 vlan(4), disc(4), edsc(4), if_bridge(4), gif(4), tap(4),
356 faith(4), epair(4), enc(4), tun(4), if_lagg(4), gre(4).
359 The sdhci driver was split in two parts: sdhci (generic SD Host
360 Controller logic) and sdhci_pci (actual hardware driver).
361 No kernel config modifications are required, but if you
362 load sdhc as a module you must switch to sdhci_pci instead.
365 Import the FUSE kernel and userland support into base system.
368 The GNU sort(1) program has been removed since the BSD-licensed
369 sort(1) has been the default for quite some time and no serious
370 problems have been reported. The corresponding WITH_GNU_SORT
374 The pfil(9) API/ABI for AF_INET family has been changed. Packet
375 filtering modules: pf(4), ipfw(4), ipfilter(4) need to be recompiled
379 The net80211(4) ABI has been changed to allow for improved driver
380 PS-POLL and power-save support. All wireless drivers need to be
381 recompiled to work with the new kernel.
384 The random(4) support for the VIA hardware random number
385 generator (`PADLOCK') is no longer enabled unconditionally.
386 Add the padlock_rng device in the custom kernel config if
387 needed. The GENERIC kernels on i386 and amd64 do include the
388 device, so the change only affects the custom kernel
392 The pf(4) packet filter ABI has been changed. pfctl(8) and
393 snmp_pf module need to be recompiled to work with new kernel.
396 A new ZFS feature flag "com.delphix:empty_bpobj" has been merged
397 to -HEAD. Pools that have empty_bpobj in active state can not be
398 imported read-write with ZFS implementations that do not support
399 this feature. For more information read the zpool-features(5)
403 The sparc64 ZFS loader has been changed to no longer try to auto-
404 detect ZFS providers based on diskN aliases but now requires these
405 to be explicitly listed in the OFW boot-device environment variable.
408 The OpenSSL has been upgraded to 1.0.1c. Any binaries requiring
409 libcrypto.so.6 or libssl.so.6 must be recompiled. Also, there are
410 configuration changes. Make sure to merge /etc/ssl/openssl.cnf.
413 The following sysctls and tunables have been renamed for consistency
414 with other variables:
415 kern.cam.da.da_send_ordered -> kern.cam.da.send_ordered
416 kern.cam.ada.ada_send_ordered -> kern.cam.ada.send_ordered
419 The sort utility has been replaced with BSD sort. For now, GNU sort
420 is also available as "gnusort" or the default can be set back to
421 GNU sort by setting WITH_GNU_SORT. In this case, BSD sort will be
422 installed as "bsdsort".
425 A new version of ZFS (pool version 5000) has been merged to -HEAD.
426 Starting with this version the old system of ZFS pool versioning
427 is superseded by "feature flags". This concept enables forward
428 compatibility against certain future changes in functionality of ZFS
429 pools. The first read-only compatible "feature flag" for ZFS pools
430 is named "com.delphix:async_destroy". For more information
431 read the new zpool-features(5) manual page.
432 Please refer to the "ZFS notes" section of this file for information
433 on upgrading boot ZFS pools.
436 The malloc(3) implementation embedded in libc now uses sources imported
437 as contrib/jemalloc. The most disruptive API change is to
438 /etc/malloc.conf. If your system has an old-style /etc/malloc.conf,
439 delete it prior to installworld, and optionally re-create it using the
440 new format after rebooting. See malloc.conf(5) for details
441 (specifically the TUNING section and the "opt.*" entries in the MALLCTL
445 Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb
446 is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is
447 now spelled mips. This is to aid compatibility with third-party
448 software that expects this naming scheme in uname(3). Little-endian
449 settings are unchanged. If you are updating a big-endian mips64 machine
450 from before this change, you may need to set MACHINE_ARCH=mips64 in
451 your environment before the new build system will recognize your machine.
454 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
458 Now unix domain sockets behave "as expected" on nullfs(5). Previously
459 nullfs(5) did not pass through all behaviours to the underlying layer,
460 as a result if we bound to a socket on the lower layer we could connect
461 only to the lower path; if we bound to the upper layer we could connect
462 only to the upper path. The new behavior is one can connect to both the
463 lower and the upper paths regardless what layer path one binds to.
466 The getifaddrs upgrade path broken with 20111215 has been restored.
467 If you have upgraded in between 20111215 and 20120209 you need to
468 recompile libc again with your kernel. You still need to recompile
469 world to be able to configure CARP but this restriction already
473 The set_rcvar() function has been removed from /etc/rc.subr. All
474 base and ports rc.d scripts have been updated, so if you have a
475 port installed with a script in /usr/local/etc/rc.d you can either
476 hand-edit the rcvar= line, or reinstall the port.
478 An easy way to handle the mass-update of /etc/rc.d:
479 rm /etc/rc.d/* && mergemaster -i
482 panic(9) now stops other CPUs in the SMP systems, disables interrupts
483 on the current CPU and prevents other threads from running.
484 This behavior can be reverted using the kern.stop_scheduler_on_panic
486 The new behavior can be incompatible with kern.sync_on_panic.
489 The carp(4) facility has been changed significantly. Configuration
490 of the CARP protocol via ifconfig(8) has changed, as well as format
491 of CARP events submitted to devd(8) has changed. See manual pages
492 for more information. The arpbalance feature of carp(4) is currently
493 not supported anymore.
495 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
496 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
497 need to be recompiled.
500 The acpi_wmi(4) status device /dev/wmistat has been renamed to
504 The option VFS_ALLOW_NONMPSAFE option has been added in order to
505 explicitely support non-MPSAFE filesystems.
506 It is on by default for all supported platform at this present
510 The broken amd(4) driver has been replaced with esp(4) in the amd64,
511 i386 and pc98 GENERIC kernel configuration files.
514 sysinstall has been removed
517 The stable/9 branch created in subversion. This corresponds to the
518 RELENG_9 branch in CVS.
521 This commit modifies vfs_register() so that it uses a hash
522 calculation to set vfc_typenum, which is enabled by default.
523 The first time a system is booted after this change, the
524 vfc_typenum values will change for all file systems. The
525 main effect of this is a change to the NFS server file handles
526 for file systems that use vfc_typenum in their fsid, such as ZFS.
527 It will, however, prevent vfc_typenum from changing when file
528 systems are loaded in a different order for subsequent reboots.
529 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
530 until you are ready to remount all NFS clients after a reboot.
533 Bump the shared library version numbers for libraries that
534 do not use symbol versioning, have changed the ABI compared
535 to stable/8 and which shared library version was not bumped.
536 Done as part of 9.0-RELEASE cycle.
539 During the merge of Capsicum features, the fget(9) KPI was modified.
540 This may require the rebuilding of out-of-tree device drivers --
541 issues have been reported specifically with the nVidia device driver.
542 __FreeBSD_version is bumped to 900041.
544 Also, there is a period between 20110811 and 20110814 where the
545 special devices /dev/{stdin,stdout,stderr} did not work correctly.
546 Building world from a kernel during that window may not work.
549 The packet filter (pf) code has been updated to OpenBSD 4.5.
550 You need to update userland tools to be in sync with kernel.
551 This update breaks backward compatibility with earlier pfsync(4)
552 versions. Care must be taken when updating redundant firewall setups.
555 The following sysctls and tunables are retired on x86 platforms:
557 machdep.hlt_logical_cpus
558 The following sysctl is retired:
559 machdep.hyperthreading_allowed
560 The sysctls were supposed to provide a way to dynamically offline and
561 online selected CPUs on x86 platforms, but the implementation has not
562 been reliable especially with SCHED_ULE scheduler.
563 machdep.hyperthreading_allowed tunable is still available to ignore
564 hyperthreading CPUs at OS level.
565 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
566 where X is an APIC ID of a CPU. Be advised, though, that disabling
567 CPUs in non-uniform fashion will result in non-uniform topology and
568 may lead to sub-optimal system performance with SCHED_ULE, which is
572 cpumask_t type is retired and cpuset_t is used in order to describe
576 Changes to ifconfig(8) for dynamic address family detection mandate
577 that you are running a kernel of 20110525 or later. Make sure to
578 follow the update procedure to boot a new kernel before installing
582 Support for sun4v architecture is officially dropped
585 Several KPI breaking changes have been committed to the mii(4) layer,
586 the PHY drivers and consequently some Ethernet drivers using mii(4).
587 This means that miibus.ko and the modules of the affected Ethernet
588 drivers need to be recompiled.
590 Note to kernel developers: Given that the OUI bit reversion problem
591 was fixed as part of these changes all mii(4) commits related to OUIs,
592 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
593 handling, no longer can be merged verbatim to stable/8 and previous
597 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
598 into their kernel configurations along with 'device pci'.
601 The default NFS client is now the new NFS client, so fstype "newnfs"
602 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
603 Although mounts via fstype "nfs" will usually work without userland
604 changes, it is recommended that the mount(8) and mount_nfs(8)
605 commands be rebuilt from sources and that a link to mount_nfs called
606 mount_oldnfs be created. The new client is compiled into the
607 kernel with "options NFSCL" and this is needed for diskless root
608 file systems. The GENERIC kernel configs have been changed to use
609 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
610 To use the regular/old client, you can "mount -t oldnfs ...". For
611 a diskless root file system, you must also include a line like:
613 vfs.root.mountfrom="oldnfs:"
615 in the boot/loader.conf on the root fs on the NFS server to make
616 a diskless root fs use the old client.
619 The GENERIC kernels for all architectures now default to the new
620 CAM-based ATA stack. It means that all legacy ATA drivers were
621 removed and replaced by respective CAM drivers. If you are using
622 ATA device names in /etc/fstab or other places, make sure to update
623 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
624 where 'Y's are the sequential numbers starting from zero for each type
625 in order of detection, unless configured otherwise with tunables,
626 see cam(4)). There will be symbolic links created in /dev/ to map
627 old adX devices to the respective adaY. They should provide basic
628 compatibility for file systems mounting in most cases, but they do
629 not support old user-level APIs and do not have respective providers
630 in GEOM. Consider using updated management tools with new device names.
632 It is possible to load devices ahci, ata, siis and mvs as modules,
633 but option ATA_CAM should remain in kernel configuration to make ata
634 module work as CAM driver supporting legacy ATA controllers. Device ata
635 still can be used in modular fashion (atacore + ...). Modules atadisk
636 and atapi* are not used and won't affect operation in ATA_CAM mode.
637 Note that to use CAM-based ATA kernel should include CAM devices
638 scbus, pass, da (or explicitly ada), cd and optionally others. All of
639 them are parts of the cam module.
641 ataraid(4) functionality is now supported by the RAID GEOM class.
642 To use it you can load geom_raid kernel module and use graid(8) tool
643 for management. Instead of /dev/arX device names, use /dev/raid/rX.
645 No kernel config options or code have been removed, so if a problem
646 arises, please report it and optionally revert to the old ATA stack.
647 In order to do it you can remove from the kernel config:
652 , and instead add back:
653 device atadisk # ATA disk drives
654 device ataraid # ATA RAID drives
655 device atapicd # ATAPI CDROM drives
656 device atapifd # ATAPI floppy drives
657 device atapist # ATAPI tape drives
660 The default NFS server has been changed to the new server, which
661 was referred to as the experimental server. If you need to switch
662 back to the old NFS server, you must now put the "-o" option on
663 both the mountd and nfsd commands. This can be done using the
664 mountd_flags and nfs_server_flags rc.conf variables until an
665 update to the rc scripts is committed, which is coming soon.
668 The GNU Objective-C runtime library (libobjc), and other Objective-C
669 related components have been removed from the base system. If you
670 require an Objective-C library, please use one of the available ports.
673 ath(4) has been split into bus- and device- modules. if_ath contains
674 the HAL, the TX rate control and the network device code. if_ath_pci
675 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
676 contains the AHB glue. Users need to load both if_ath_pci and if_ath
677 in order to use ath on everything else.
679 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
680 need to load if_ath and if_ath_pci for ath(4) operation.
683 As part of the replacement of sysinstall, the process of building
684 release media has changed significantly. For details, please re-read
685 release(7), which has been updated to reflect the new build process.
688 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
689 is the last available version under GPLv2. It brings a number of new
690 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
691 4.1 and SSE 4.2), better support for powerpc64, a number of new
692 directives, and lots of other small improvements. See the ChangeLog
693 file in contrib/binutils for the full details.
696 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
697 compliant, and will now use half of hash for authentication.
698 This will break interoperability with all stacks (including all
699 actual FreeBSD versions) who implement
700 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
702 The only workaround with such peers is to use another HMAC
703 algorithm for IPsec ("phase 2") authentication.
706 Remove the uio_yield prototype and symbol. This function has
707 been misnamed since it was introduced and should not be
708 globally exposed with this name. The equivalent functionality
709 is now available using kern_yield(curthread->td_user_pri).
710 The function remains undocumented.
713 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
714 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
715 for scalar sysctls is defined but disabled. Code that needs
716 UQUAD to pass the type checking that must compile on older
717 systems where the define is not present can check against
718 __FreeBSD_version >= 900030.
720 The system dialog(1) has been replaced with a new version previously
721 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
722 with the previous version, but the libdialog associated with it has
723 a largely incompatible API. As such, the original version of libdialog
724 will be kept temporarily as libodialog, until its base system consumers
725 are replaced or updated. Bump __FreeBSD_version to 900030.
728 If you are trying to run make universe on a -stable system, and you get
729 the following warning:
730 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
731 unknown. config(8) likely too old."
732 or something similar to it, then you must upgrade your -stable system
733 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
734 stable/8) or build the config from the latest stable/8 branch and
735 install it on your system.
737 Prior to this date, building a current universe on 8-stable system from
738 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
739 error in the first kernel build phase. A new config on those old
740 systems will fix that problem for older versions of -current.
743 The TCP stack has been modified to allow Khelp modules to interact with
744 it via helper hook points and store per-connection data in the TCP
745 control block. Bump __FreeBSD_version to 900029. User space tools that
746 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
750 Generic IEEE 802.3 annex 31B full duplex flow control support has been
751 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
752 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
753 to take advantage of it instead of using custom implementations. This
754 means that these drivers now no longer unconditionally advertise
755 support for flow control but only do so if flow control is a selected
756 media option. This was implemented in the generic support that way in
757 order to allow flow control to be switched on and off via ifconfig(8)
758 with the PHY specific default to typically off in order to protect
759 from unwanted effects. Consequently, if you used flow control with
760 one of the above mentioned drivers you now need to explicitly enable
762 ifconfig bge0 media auto mediaopt flowcontrol
764 Along with the above mentioned changes generic support for setting
765 1000baseT master mode also has been added and brgphy(4), ciphy(4),
766 e1000phy(4) as well as ip1000phy(4) have been converted to take
767 advantage of it. This means that these drivers now no longer take the
768 link0 parameter for selecting master mode but the master media option
769 has to be used instead, for example like in the following:
770 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
772 Selection of master mode now is also available with all other PHY
773 drivers supporting 1000baseT.
776 The TCP stack has received a significant update to add support for
777 modularised congestion control and generally improve the clarity of
778 congestion control decisions. Bump __FreeBSD_version to 900025. User
779 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
780 sockstat) need to be recompiled.
783 The man(1) utility has been replaced by a new version that no longer
784 uses /etc/manpath.config. Please consult man.conf(5) for how to
785 migrate local entries to the new format.
788 The copyright strings printed by login(1) and sshd(8) at the time of a
789 new connection have been removed to follow other operating systems and
793 A workaround for a fixed ld bug has been removed in kernel code,
794 so make sure that your system ld is built from sources after
795 revision 210245 from 2010-07-19 (r211583 if building head kernel
796 on stable/8, r211584 for stable/7; both from 2010-08-21).
797 A symptom of incorrect ld version is different addresses for
798 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
801 The $ipv6_prefer variable in rc.conf(5) has been split into
802 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
804 The $ip6addrctl_policy is a variable to choose a pre-defined
805 address selection policy set by ip6addrctl(8). A value
806 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
809 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
810 flag (see an entry of 20090926) is set on an interface with no
811 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
812 security reason. If you want IPv6 link-local address on all
813 interfaces by default, set this to "YES".
815 The old ipv6_prefer="YES" is equivalent to
816 ipv6_activate_all_interfaces="YES" and
817 ip6addrctl_policy="ipv6_prefer".
820 DTrace has grown support for userland tracing. Due to this, DTrace is
821 now i386 and amd64 only.
822 dtruss(1) is now installed by default on those systems and a new
823 kernel module is needed for userland tracing: fasttrap.
824 No changes to your kernel config file are necessary to enable
825 userland tracing, but you might consider adding 'STRIP=' and
826 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
827 to have informative userland stack traces in DTrace (ustack).
830 The acpi_aiboost(4) driver has been removed in favor of the new
831 aibs(4) driver. You should update your kernel configuration file.
834 BSD grep has been imported to the base system and it is built by
835 default. It is completely BSD licensed, highly GNU-compatible, uses
836 less memory than its GNU counterpart and has a small codebase.
837 However, it is slower than its GNU counterpart, which is mostly
838 noticeable for larger searches, for smaller ones it is measurable
839 but not significant. The reason is complex, the most important factor
840 is that we lack a modern and efficient regex library and GNU
841 overcomes this by optimizing the searches internally. Future work
842 on improving the regex performance is planned, for the meantime,
843 users that need better performance, can build GNU grep instead by
844 setting the WITH_GNU_GREP knob.
847 Due to the import of powerpc64 support, all existing powerpc kernel
848 configuration files must be updated with a machine directive like this:
849 machine powerpc powerpc
851 In addition, an updated config(8) is required to build powerpc kernels
855 A new version of ZFS (version 15) has been merged to -HEAD.
856 This version uses a python library for the following subcommands:
857 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
858 For full functionality of these commands the following port must
859 be installed: sysutils/py-zfs
862 'vm_page's are now hashed by physical address to an array of mutexes.
863 Currently this is only used to serialize access to hold_count. Over
864 time the page queue mutex will be peeled away. This changes the size
865 of pmap on every architecture. And requires all callers of vm_page_hold
866 and vm_page_unhold to be updated.
869 WITH_CTF can now be specified in src.conf (not recommended, there
870 are some problems with static executables), make.conf (would also
871 affect ports which do not use GNU make and do not override the
872 compile targets) or in the kernel config (via "makeoptions
874 When WITH_CTF was specified there before this was silently ignored,
875 so make sure that WITH_CTF is not used in places which could lead
876 to unwanted behavior.
879 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
880 to allow 32-bit compatibility on non-x86 platforms. All kernel
881 configurations on amd64 and ia64 platforms using these options must
882 be modified accordingly.
885 The utmp user accounting database has been replaced with utmpx,
886 the user accounting interface standardized by POSIX.
887 Unfortunately the semantics of utmp and utmpx don't match,
888 making it practically impossible to support both interfaces.
889 The user accounting database is used by tools like finger(1),
890 last(1), talk(1), w(1) and ac(8).
892 All applications in the base system use utmpx. This means only
893 local binaries (e.g. from the ports tree) may still use these
894 utmp database files. These applications must be rebuilt to make
897 After the system has been upgraded, it is safe to remove the old
898 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
899 assuming their contents is of no importance anymore. Old wtmp
900 databases can only be used by last(1) and ac(8) after they have
901 been converted to the new format using wtmpcvt(1).
904 Introduce the kernel thread "deadlock resolver" (which can be enabled
905 via the DEADLKRES option, see NOTES for more details) and the
906 sleepq_type() function for sleepqueues.
909 The rc.firewall and rc.firewall6 were unified, and
910 rc.firewall6 and rc.d/ip6fw were removed.
911 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
912 variables are obsoleted. Instead, the following new rc
913 variables are added to rc.d/ipfw:
915 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
916 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
917 firewall_simple_onet_ipv6, firewall_trusted_ipv6
919 The meanings correspond to the relevant IPv4 variables.
925 The default terminal emulation for syscons(4) has been changed
926 from cons25 to xterm on all platforms except pc98. This means
927 that the /etc/ttys file needs to be updated to ensure correct
928 operation of applications on the console.
930 The terminal emulation style can be toggled per window by using
931 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
932 options can be used to change the compile-time default back to
935 To prevent graphical artifacts, make sure the TERM environment
936 variable is set to match the terminal emulation that is being
937 performed by syscons(4).
940 The layout of the structure ieee80211req_scan_result has changed.
941 Applications that require wireless scan results (e.g. ifconfig(8))
942 from net80211 need to be recompiled.
944 Applications such as wpa_supplicant(8) may require a full world
945 build without using NO_CLEAN in order to get synchronized with the
949 The iwn(4) driver has been updated to support the 5000 and 5150 series.
950 There's one kernel module for each firmware. Adding "device iwnfw"
951 to the kernel configuration file means including all three firmware
952 images inside the kernel. If you want to include just the one for
953 your wireless card, use the devices iwn4965fw, iwn5000fw or
957 The rc.d/network_ipv6, IPv6 configuration script has been integrated
958 into rc.d/netif. The changes are the following:
960 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
961 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
962 Note that both variables need the "inet6" keyword at the head.
964 Do not set $ipv6_network_interfaces manually if you do not
965 understand what you are doing. It is not needed in most cases.
967 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
970 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
971 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
973 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
974 all configured interfaces have "inet6 accept_rtadv" in the
975 $ifconfig_IF_ipv6. These are for backward compatibility.
977 3. A new variable $ipv6_prefer has been added. If NO, IPv6
978 functionality of interfaces with no corresponding
979 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
980 and the default address selection policy of ip6addrctl(8)
981 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
982 Note that if you want to configure IPv6 functionality on the
983 disabled interfaces after boot, first you need to clear the flag by
984 using ifconfig(8) like:
986 ifconfig em0 inet6 -ifdisabled
988 If YES, the default address selection policy is set as
991 The default value of $ipv6_prefer is NO.
993 4. If your system need to receive Router Advertisement messages,
994 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
995 scripts automatically invoke rtsol(8) when the interface becomes
996 UP. The Router Advertisement messages are used for SLAAC
997 (State-Less Address AutoConfiguration).
1000 802.11s D3.03 support was committed. This is incompatible with the
1001 previous code, which was based on D3.0.
1004 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
1005 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
1006 control whether accepting Router Advertisement messages or not.
1007 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
1008 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
1009 The ifconfig(8) utility now supports these flags.
1012 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
1013 mount(8) and -a option for df(1) to see them.
1016 The old tunable hw.bus.devctl_disable has been superseded by
1017 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
1018 replaced by hw.bus.devctl_queue=0. The default for this new tunable
1022 Remove the option STOP_NMI. The default action is now to use NMI only
1023 for KDB via the newly introduced function stop_cpus_hard() and
1024 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
1027 The stable/8 branch created in subversion. This corresponds to the
1028 RELENG_8 branch in CVS.
1031 Bump the shared library version numbers for all libraries that do not
1032 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
1033 __FreeBSD_version to 800105.
1036 Due to changes in the implementation of virtual network stack support,
1037 all network-related kernel modules must be recompiled. As this change
1038 breaks the ABI, bump __FreeBSD_version to 800104.
1041 The TOE interface to the TCP syncache has been modified to remove
1042 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
1043 The cxgb driver is the only TOE consumer affected by this change, and
1044 needs to be recompiled along with the kernel. As this change breaks
1045 the ABI, bump __FreeBSD_version to 800103.
1048 Padding has been added to struct tcpcb, sackhint and tcpstat in
1049 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
1050 maintaining the ABI. However, this change breaks the ABI, so bump
1051 __FreeBSD_version to 800102. User space tools that rely on the size of
1052 any of these structs (e.g. sockstat) need to be recompiled.
1055 The NFS_LEGACYRPC option has been removed along with the old kernel
1056 RPC implementation that this option selected. Kernel configurations
1057 may need to be adjusted.
1060 The network interface device nodes at /dev/net/<interface> have been
1061 removed. All ioctl operations can be performed the normal way using
1062 routing sockets. The kqueue functionality can generally be replaced
1063 with routing sockets.
1066 The documentation from the FreeBSD Documentation Project (Handbook,
1067 FAQ, etc.) is now installed via packages by sysinstall(8) and under
1068 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
1071 The ABI of various structures related to the SYSV IPC API have been
1072 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
1073 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
1077 Layout of struct vnet has changed as routing related variables were
1078 moved to their own Vimage module. Modules need to be recompiled. Bump
1079 __FreeBSD_version to 800099.
1082 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
1083 respectively. As long as no more than 16 groups per process are used,
1084 no changes should be visible. When more than 16 groups are used, old
1085 binaries may fail if they call getgroups() or getgrouplist() with
1086 statically sized storage. Recompiling will work around this, but
1087 applications should be modified to use dynamically allocated storage
1088 for group arrays as POSIX.1-2008 does not cap an implementation's
1089 number of supported groups at NGROUPS_MAX+1 as previous versions did.
1091 NFS and portalfs mounts may also be affected as the list of groups is
1092 truncated to 16. Users of NFS who use more than 16 groups, should
1093 take care that negative group permissions are not used on the exported
1094 file systems as they will not be reliable unless a GSSAPI based
1095 authentication method is used.
1098 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
1099 option compiles in the support for adaptive spinning for lockmgrs
1100 which want to enable it. The lockinit() function now accepts the flag
1101 LK_ADAPTIVE in order to make the lock object subject to adaptive
1102 spinning when both held in write and read mode.
1105 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
1106 changed. User applications that use this ioctl need to be rebuilt.
1109 The layout of struct thread has changed. Kernel and modules need to
1113 The layout of structs ifnet, domain, protosw and vnet_net has changed.
1114 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
1117 window(1) has been removed from the base system. It can now be
1118 installed from ports. The port is called misc/window.
1121 The way we are storing and accessing `routing table' entries has
1122 changed. Programs reading the FIB, like netstat, need to be
1126 A new netisr implementation has been added for FreeBSD 8. Network
1127 file system modules, such as igmp, ipdivert, and others, should be
1129 Bump __FreeBSD_version to 800096.
1132 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
1136 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
1137 Bump __FreeBSD_version to 800094.
1140 Add mnt_xflag field to 'struct mount'. File system modules need to be
1142 Bump __FreeBSD_version to 800093.
1145 The compiling option ADAPTIVE_SX has been retired while it has been
1146 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
1147 The KPI for sx_init_flags() changes as accepting flags:
1148 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
1149 been introduced in order to handle the reversed logic.
1150 Bump __FreeBSD_version to 800092.
1153 Add support for hierarchical jails. Remove global securelevel.
1154 Bump __FreeBSD_version to 800091.
1157 The layout of struct vnet_net has changed, therefore modules
1159 Bump __FreeBSD_version to 800090.
1162 The newly imported zic(8) produces a new format in the output. Please
1163 run tzsetup(8) to install the newly created data to /etc/localtime.
1166 The sysctl tree for the usb stack has renamed from hw.usb2.* to
1167 hw.usb.* and is now consistent again with previous releases.
1170 802.11 monitor mode support was revised and driver api's were changed.
1171 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
1172 of DLT_IEEE802_11. No user-visible data structures were changed but
1173 applications that use DLT_IEEE802_11 may require changes.
1174 Bump __FreeBSD_version to 800088.
1177 The layout of the following structs has changed: sysctl_oid,
1178 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
1179 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
1180 panics may be experienced. World rebuild is required for
1181 correctly checking networking state from userland.
1182 Bump __FreeBSD_version to 800085.
1185 MLDv2 and Source-Specific Multicast (SSM) have been merged
1186 to the IPv6 stack. VIMAGE hooks are in but not yet used.
1187 The implementation of SSM within FreeBSD's IPv6 stack closely
1188 follows the IPv4 implementation.
1190 For kernel developers:
1192 * The most important changes are that the ip6_output() and
1193 ip6_input() paths no longer take the IN6_MULTI_LOCK,
1194 and this lock has been downgraded to a non-recursive mutex.
1196 * As with the changes to the IPv4 stack to support SSM, filtering
1197 of inbound multicast traffic must now be performed by transport
1198 protocols within the IPv6 stack. This does not apply to TCP and
1199 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
1201 * The KPIs used by IPv6 multicast are similar to those used by
1202 the IPv4 stack, with the following differences:
1203 * im6o_mc_filter() is analogous to imo_multicast_filter().
1204 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
1205 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
1206 * IN6_LOOKUP_MULTI() has been deprecated and removed.
1207 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
1208 for MLDv1 have an additional 'timer' argument which is used to
1209 jitter the initial membership report for the solicited-node
1210 multicast membership on-link.
1211 * This is not strictly needed for MLDv2, which already jitters
1212 its report transmissions. However, the 'timer' argument is
1213 preserved in case MLDv1 is active on the interface.
1215 * The KAME linked-list based IPv6 membership implementation has
1216 been refactored to use a vector similar to that used by the IPv4
1218 Code which maintains a list of its own multicast memberships
1219 internally, e.g. carp, has been updated to reflect the new
1222 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
1223 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
1224 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
1225 implementation constraint which needs to be addressed in HEAD.
1227 For application developers:
1229 * The changes are broadly similar to those made for the IPv4
1232 * The use of IPv4 and IPv6 multicast socket options on the same
1233 socket, using mapped addresses, HAS NOT been tested or supported.
1235 * There are a number of issues with the implementation of various
1236 IPv6 multicast APIs which need to be resolved in the API surface
1237 before the implementation is fully compatible with KAME userland
1238 use, and these are mostly to do with interface index treatment.
1240 * The literature available discusses the use of either the delta / ASM
1241 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
1242 using setsourcefilter(3)/getsourcefilter(3). For more information
1243 please refer to RFC 3768, 'Socket Interface Extensions for
1244 Multicast Source Filters'.
1246 * Applications which use the published RFC 3678 APIs should be fine.
1248 For systems administrators:
1250 * The mtest(8) utility has been refactored to support IPv6, in
1251 addition to IPv4. Interface addresses are no longer accepted
1252 as arguments, their names must be used instead. The utility
1253 will map the interface name to its first IPv4 address as
1254 returned by getifaddrs(3).
1256 * The ifmcstat(8) utility has also been updated to print the MLDv2
1257 endpoint state and source filter lists via sysctl(3).
1259 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
1260 loopback of IPv6 multicast datagrams by default; it defaults to 1
1261 to preserve the existing behaviour. Disabling multicast loopback is
1262 recommended for optimal system performance.
1264 * The IPv6 MROUTING code has been changed to examine this sysctl
1265 instead of attempting to perform a group lookup before looping
1266 back forwarded datagrams.
1268 Bump __FreeBSD_version to 800084.
1271 Implement low-level Bluetooth HCI API.
1272 Bump __FreeBSD_version to 800083.
1275 The layout of struct malloc_type, used by modules to register new
1276 memory allocation types, has changed. Most modules will need to
1277 be rebuilt or panics may be experienced.
1278 Bump __FreeBSD_version to 800081.
1281 Anticipate overflowing inp_flags - add inp_flags2.
1282 This changes most offsets in inpcb, so checking v4 connection
1283 state will require a world rebuild.
1284 Bump __FreeBSD_version to 800080.
1287 Add an llentry to struct route and struct route_in6. Modules
1288 embedding a struct route will need to be recompiled.
1289 Bump __FreeBSD_version to 800079.
1292 The size of rt_metrics_lite and by extension rtentry has changed.
1293 Networking administration apps will need to be recompiled.
1294 The route command now supports show as an alias for get, weighting
1295 of routes, sticky and nostick flags to alter the behavior of stateful
1297 Bump __FreeBSD_version to 800078.
1300 Do not use Giant for kbdmux(4) locking. This is wrong and
1301 apparently causing more problems than it solves. This will
1302 re-open the issue where interrupt handlers may race with
1303 kbdmux(4) in polling mode. Typical symptoms include (but
1304 not limited to) duplicated and/or missing characters when
1305 low level console functions (such as gets) are used while
1306 interrupts are enabled (for example geli password prompt,
1307 mountroot prompt etc.). Disabling kbdmux(4) may help.
1310 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
1311 kernel modules referencing any of the above need to be recompiled.
1312 Bump __FreeBSD_version to 800075.
1315 GEOM_PART has become the default partition slicer for storage devices,
1316 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
1317 introduces some changes:
1319 MSDOS/EBR: the devices created from MSDOS extended partition entries
1320 (EBR) can be named differently than with GEOM_MBR and are now symlinks
1321 to devices with offset-based names. fstabs may need to be modified.
1323 BSD: the "geometry does not match label" warning is harmless in most
1324 cases but it points to problems in file system misalignment with
1325 disk geometry. The "c" partition is now implicit, covers the whole
1326 top-level drive and cannot be (mis)used by users.
1328 General: Kernel dumps are now not allowed to be written to devices
1329 whose partition types indicate they are meant to be used for file
1330 systems (or, in case of MSDOS partitions, as something else than
1333 Most of these changes date approximately from 200812.
1336 The uscanner(4) driver has been removed from the kernel. This follows
1337 Linux removing theirs in 2.6 and making libusb the default interface
1338 (supported by sane).
1341 The multicast forwarding code has been cleaned up. netstat(1)
1342 only relies on KVM now for printing bandwidth upcall meters.
1343 The IPv4 and IPv6 modules are split into ip_mroute_mod and
1344 ip6_mroute_mod respectively. The config(5) options for statically
1345 compiling this code remain the same, i.e. 'options MROUTING'.
1348 Support for the IFF_NEEDSGIANT network interface flag has been
1349 removed, which means that non-MPSAFE network device drivers are no
1350 longer supported. In particular, if_ar, if_sr, and network device
1351 drivers from the old (legacy) USB stack can no longer be built or
1355 POSIX.1 Native Language Support (NLS) has been enabled in libc and
1356 a bunch of new language catalog files have also been added.
1357 This means that some common libc messages are now localized and
1358 they depend on the LC_MESSAGES environmental variable.
1361 The k8temp(4) driver has been renamed to amdtemp(4) since
1362 support for Family 10 and Family 11 CPU families was added.
1365 IGMPv3 and Source-Specific Multicast (SSM) have been merged
1366 to the IPv4 stack. VIMAGE hooks are in but not yet used.
1368 For kernel developers, the most important changes are that the
1369 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
1370 and this lock has been downgraded to a non-recursive mutex.
1372 Transport protocols (UDP, Raw IP) are now responsible for filtering
1373 inbound multicast traffic according to group membership and source
1374 filters. The imo_multicast_filter() KPI exists for this purpose.
1375 Transports which do not use multicast (SCTP, TCP) already reject
1376 multicast by default. Forwarding and receive performance may improve
1377 as a mutex acquisition is no longer needed in the ip_input()
1378 low-level input path. in_addmulti() and in_delmulti() are shimmed
1379 to new KPIs which exist to support SSM in-kernel.
1381 For application developers, it is recommended that loopback of
1382 multicast datagrams be disabled for best performance, as this
1383 will still cause the lock to be taken for each looped-back
1384 datagram transmission. The net.inet.ip.mcast.loop sysctl may
1385 be tuned to 0 to disable loopback by default; it defaults to 1
1386 to preserve the existing behaviour.
1388 For systems administrators, to obtain best performance with
1389 multicast reception and multiple groups, it is always recommended
1390 that a card with a suitably precise hash filter is used. Hash
1391 collisions will still result in the lock being taken within the
1392 transport protocol input path to check group membership.
1394 If deploying FreeBSD in an environment with IGMP snooping switches,
1395 it is recommended that the net.inet.igmp.sendlocal sysctl remain
1396 enabled; this forces 224.0.0.0/24 group membership to be announced
1399 The size of 'struct igmpstat' has changed; netstat needs to be
1400 recompiled to reflect this.
1401 Bump __FreeBSD_version to 800070.
1404 libusb20.so.1 is now installed as libusb.so.1 and the ports system
1405 updated to use it. This requires a buildworld/installworld in order to
1406 update the library and dependencies (usbconfig, etc). Its advisable to
1407 rebuild all ports which uses libusb. More specific directions are given
1408 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
1409 libusb are no longer required and can be removed.
1412 A workaround is committed to allow the creation of System V shared
1413 memory segment of size > 2 GB on the 64-bit architectures.
1414 Due to a limitation of the existing ABI, the shm_segsz member
1415 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
1416 wrong for large segments. Note that limits must be explicitly
1417 raised to allow such segments to be created.
1420 The layout of struct ifnet has changed, requiring a rebuild of all
1421 network device driver modules.
1424 The /dev handling for the new USB stack has changed, a
1425 buildworld/installworld is required for libusb20.
1428 The new USB2 stack has now been permanently moved in and all kernel and
1429 module names reverted to their previous values (eg, usb, ehci, ohci,
1430 ums, ...). The old usb stack can be compiled in by prefixing the name
1431 with the letter 'o', the old usb modules have been removed.
1432 Updating entry 20090216 for xorg and 20090215 for libmap may still
1436 The rc.conf(5) option if_up_delay has been renamed to
1437 defaultroute_delay to better reflect its purpose. If you have
1438 customized this setting in /etc/rc.conf you need to update it to
1442 xorg 7.4 wants to configure its input devices via hald which does not
1443 yet work with USB2. If the keyboard/mouse does not work in xorg then
1445 Option "AllowEmptyInput" "off"
1446 to your ServerLayout section. This will cause X to use the configured
1447 kbd and mouse sections from your xorg.conf.
1450 The GENERIC kernels for all architectures now default to the new USB2
1451 stack. No kernel config options or code have been removed so if a
1452 problem arises please report it and optionally revert to the old USB
1453 stack. If you are loading USB kernel modules or have a custom kernel
1454 that includes GENERIC then ensure that usb names are also changed over,
1455 eg uftdi -> usb2_serial_ftdi.
1457 Older programs linked against the ports libusb 0.1 need to be
1458 redirected to the new stack's libusb20. /etc/libmap.conf can
1460 # Map old usb library to new one for usb2 stack
1461 libusb-0.1.so.8 libusb20.so.1
1464 All USB ethernet devices now attach as interfaces under the name ueN
1465 (eg. ue0). This is to provide a predictable name as vendors often
1466 change usb chipsets in a product without notice.
1469 The ichsmb(4) driver has been changed to require SMBus slave
1470 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1471 All of the other SMBus controller drivers require left-justified
1472 slave addresses, so this change makes all the drivers provide the
1476 INET6 statistics (struct ip6stat) was updated.
1477 netstat(1) needs to be recompiled.
1480 NTFS has been removed from GENERIC kernel on amd64 to match
1481 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1482 will load ntfs.ko module automatically when NTFS support is
1483 actually needed, unless ntfs.ko is not installed or security
1484 level prohibits loading kernel modules. If either is the case,
1485 "options NTFS" has to be added into kernel config.
1488 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1489 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1490 800061. User space tools that rely on the size of struct tcpcb in
1491 tcp_var.h (e.g. sockstat) need to be recompiled.
1494 ng_tty(4) module updated to match the new TTY subsystem.
1495 Due to API change, user-level applications must be updated.
1496 New API support added to mpd5 CVS and expected to be present
1497 in next mpd5.3 release.
1500 With __FreeBSD_version 800060 the makefs tool is part of
1501 the base system (it was a port).
1504 The afdata and ifnet locks have been changed from mutexes to
1505 rwlocks, network modules will need to be re-compiled.
1508 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1509 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1510 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1511 architecture and 40 bytes on 64-bit architecture. The userland
1512 applications "arp" and "ndp" have been updated accordingly.
1513 The output from "netstat -r" shows only routing entries and
1514 none of the L2 information.
1517 __FreeBSD_version 800057 marks the switchover from the
1518 binary ath hal to source code. Users must add the line:
1520 options AH_SUPPORT_AR5416
1522 to their kernel config files when specifying:
1526 The ath_hal module no longer exists; the code is now compiled
1527 together with the driver in the ath module. It is now
1528 possible to tailor chip support (i.e. reduce the set of chips
1529 and thereby the code size); consult ath_hal(4) for details.
1532 __FreeBSD_version 800054 adds memory barriers to
1533 <machine/atomic.h>, new interfaces to ifnet to facilitate
1534 multiple hardware transmit queues for cards that support
1535 them, and a lock-less ring-buffer implementation to
1536 enable drivers to more efficiently manage queueing of
1540 A new version of ZFS (version 13) has been merged to -HEAD.
1541 This version has zpool attribute "listsnapshots" off by
1542 default, which means "zfs list" does not show snapshots,
1543 and is the same as Solaris behavior.
1546 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1549 The uhci, ohci, ehci and slhci USB Host controller drivers have
1550 been put into separate modules. If you load the usb module
1551 separately through loader.conf you will need to load the
1552 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1553 controller add the following to loader.conf:
1559 The ABI used by the PMC toolset has changed. Please keep
1560 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1564 atapci kernel module now includes only generic PCI ATA
1565 driver. AHCI driver moved to ataahci kernel module.
1566 All vendor-specific code moved into separate kernel modules:
1567 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1568 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1569 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1570 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1573 The TTY subsystem of the kernel has been replaced by a new
1574 implementation, which provides better scalability and an
1575 improved driver model. Most common drivers have been migrated to
1576 the new TTY subsystem, while others have not. The following
1577 drivers have not yet been ported to the new TTY layer:
1580 cy, digi, rc, rp, sio
1586 ng_h4, ng_tty, ppp, sl, snp
1588 Adding these drivers to your kernel configuration file shall
1589 cause compilation to fail.
1592 ntpd has been upgraded to 4.2.4p5.
1595 OpenSSH has been upgraded to 5.1p1.
1597 For many years, FreeBSD's version of OpenSSH preferred DSA
1598 over RSA for host and user authentication keys. With this
1599 upgrade, we've switched to the vendor's default of RSA over
1600 DSA. This may cause upgraded clients to warn about unknown
1601 host keys even for previously known hosts. Users should
1602 follow the usual procedure for verifying host keys before
1603 accepting the RSA key.
1605 This can be circumvented by setting the "HostKeyAlgorithms"
1606 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1609 Please note that the sequence of keys offered for
1610 authentication has been changed as well. You may want to
1611 specify IdentityFile in a different order to revert this
1615 The sio(4) driver has been removed from the i386 and amd64
1616 kernel configuration files. This means uart(4) is now the
1617 default serial port driver on those platforms as well.
1619 To prevent collisions with the sio(4) driver, the uart(4) driver
1620 uses different names for its device nodes. This means the
1621 onboard serial port will now most likely be called "ttyu0"
1622 instead of "ttyd0". You may need to reconfigure applications to
1623 use the new device names.
1625 When using the serial port as a boot console, be sure to update
1626 /boot/device.hints and /etc/ttys before booting the new kernel.
1627 If you forget to do so, you can still manually specify the hints
1628 at the loader prompt:
1630 set hint.uart.0.at="isa"
1631 set hint.uart.0.port="0x3F8"
1632 set hint.uart.0.flags="0x10"
1633 set hint.uart.0.irq="4"
1637 The gpt(8) utility has been removed. Use gpart(8) to partition
1641 The version that Linuxulator emulates was changed from 2.4.2
1642 to 2.6.16. If you experience any problems with Linux binaries
1643 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1644 if it fixes the problem contact emulation mailing list.
1647 ISDN4BSD (I4B) was removed from the src tree. You may need to
1648 update a your kernel configuration and remove relevant entries.
1651 I have checked in code to support multiple routing tables.
1652 See the man pages setfib(1) and setfib(2).
1653 This is a hopefully backwards compatible version,
1654 but to make use of it you need to compile your kernel
1655 with options ROUTETABLES=2 (or more up to 16).
1658 The 802.11 wireless support was redone to enable multi-bss
1659 operation on devices that are capable. The underlying device
1660 is no longer used directly but instead wlanX devices are
1661 cloned with ifconfig. This requires changes to rc.conf files.
1662 For example, change:
1663 ifconfig_ath0="WPA DHCP"
1666 ifconfig_wlan0="WPA DHCP"
1667 see rc.conf(5) for more details. In addition, mergemaster of
1668 /etc/rc.d is highly recommended. Simultaneous update of userland
1669 and kernel wouldn't hurt either.
1671 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1672 modules were merged into the base wlan module. All references
1673 to these modules (e.g. in kernel config files) must be removed.
1676 psm(4) has gained write(2) support in native operation level.
1677 Arbitrary commands can be written to /dev/psm%d and status can
1678 be read back from it. Therefore, an application is responsible
1679 for status validation and error recovery. It is a no-op in
1680 other operation levels.
1683 Support for KSE threading has been removed from the kernel. To
1684 run legacy applications linked against KSE libmap.conf may
1685 be used. The following libmap.conf may be used to ensure
1686 compatibility with any prior release:
1688 libpthread.so.1 libthr.so.1
1689 libpthread.so.2 libthr.so.2
1690 libkse.so.3 libthr.so.3
1693 The layout of struct vmspace has changed. This affects libkvm
1694 and any executables that link against libkvm and use the
1695 kvm_getprocs() function. In particular, but not exclusively,
1696 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1697 The effects are minimal, but it's advisable to upgrade world
1701 The latest em driver no longer has support in it for the
1702 82575 adapter, this is now moved to the igb driver. The
1703 split was done to make new features that are incompatible
1704 with older hardware easier to do.
1707 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1708 likewise the kernel option is now GEOM_LINUX_LVM.
1711 The default NFS mount mode has changed from UDP to TCP for
1712 increased reliability. If you rely on (insecurely) NFS
1713 mounting across a firewall you may need to update your
1717 Belatedly note the addition of m_collapse for compacting
1721 The fts(3) structures have been changed to use adequate
1722 integer types for their members and so to be able to cope
1723 with huge file trees. The old fts(3) ABI is preserved
1724 through symbol versioning in libc, so third-party binaries
1725 using fts(3) should still work, although they will not take
1726 advantage of the extended types. At the same time, some
1727 third-party software might fail to build after this change
1728 due to unportable assumptions made in its source code about
1729 fts(3) structure members. Such software should be fixed
1730 by its vendor or, in the worst case, in the ports tree.
1731 FreeBSD_version 800015 marks this change for the unlikely
1732 case that a portable fix is impossible.
1735 To upgrade to -current after this date, you must be running
1736 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1737 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1740 The ADAPTIVE_GIANT kernel option has been retired because its
1741 functionality is the default now.
1744 The AT keyboard emulation of sunkbd(4) has been turned on
1745 by default. In order to make the special symbols of the Sun
1746 keyboards driven by sunkbd(4) work under X these now have
1747 to be configured the same way as Sun USB keyboards driven
1748 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1750 Option "XkbLayout" "us"
1751 Option "XkbRules" "xorg"
1752 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1755 It has been decided that it is desirable to provide ABI
1756 backwards compatibility to the FreeBSD 4/5/6 versions of the
1757 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1758 broken with the introduction of PCI domain support (see the
1759 20070930 entry). Unfortunately, this required the ABI of
1760 PCIOCGETCONF to be broken again in order to be able to
1761 provide backwards compatibility to the old version of that
1762 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1763 again. As for prominent ports this affects neither pciutils
1764 nor xorg-server this time, the hal port needs to be rebuilt
1768 The misnamed kthread_create() and friends have been renamed
1769 to kproc_create() etc. Many of the callers already
1770 used kproc_start()..
1771 I will return kthread_create() and friends in a while
1772 with implementations that actually create threads, not procs.
1773 Renaming corresponds with version 800002.
1782 Avoid using make -j when upgrading. While generally safe, there are
1783 sometimes problems using -j to upgrade. If your upgrade fails with
1784 -j, please try again without -j. From time to time in the past there
1785 have been problems using -j with buildworld and/or installworld. This
1786 is especially true when upgrading between "distant" versions (eg one
1787 that cross a major release boundary or several minor releases, or when
1788 several months have passed on the -current branch).
1790 Sometimes, obscure build problems are the result of environment
1791 poisoning. This can happen because the make utility reads its
1792 environment when searching for values for global variables. To run
1793 your build attempts in an "environmental clean room", prefix all make
1794 commands with 'env -i '. See the env(1) manual page for more details.
1796 When upgrading from one major version to another it is generally best
1797 to upgrade to the latest code in the currently installed branch first,
1798 then do an upgrade to the new branch. This is the best-tested upgrade
1799 path, and has the highest probability of being successful. Please try
1800 this approach before reporting problems with a major version upgrade.
1802 When upgrading a live system, having a root shell around before
1803 installing anything can help undo problems. Not having a root shell
1804 around can lead to problems if pam has changed too much from your
1805 starting point to allow continued authentication after the upgrade.
1809 When upgrading the boot ZFS pool to a new version, always follow
1812 1.) recompile and reinstall the ZFS boot loader and boot block
1813 (this is part of "make buildworld" and "make installworld")
1815 2.) update the ZFS boot block on your boot drive
1817 The following example updates the ZFS boot block on the first
1818 partition (freebsd-boot) of a GPT partitioned drive ad0:
1819 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1821 Non-boot pools do not need these updates.
1825 If you are updating from a prior version of FreeBSD (even one just
1826 a few days old), you should follow this procedure. It is the most
1827 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1829 make kernel-toolchain
1830 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1831 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1833 To test a kernel once
1834 ---------------------
1835 If you just want to boot a kernel once (because you are not sure
1836 if it works, or if you want to boot a known bad kernel to provide
1837 debugging information) run
1838 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1839 nextboot -k testkernel
1841 To just build a kernel when you know that it won't mess you up
1842 --------------------------------------------------------------
1843 This assumes you are already running a CURRENT system. Replace
1844 ${arch} with the architecture of your machine (e.g. "i386",
1845 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1847 cd src/sys/${arch}/conf
1848 config KERNEL_NAME_HERE
1849 cd ../compile/KERNEL_NAME_HERE
1854 If this fails, go to the "To build a kernel" section.
1856 To rebuild everything and install it on the current system.
1857 -----------------------------------------------------------
1858 # Note: sometimes if you are running current you gotta do more than
1859 # is listed here if you are upgrading from a really old current.
1861 <make sure you have good level 0 dumps>
1863 make kernel KERNCONF=YOUR_KERNEL_HERE
1865 <reboot in single user> [3]
1872 To cross-install current onto a separate partition
1873 --------------------------------------------------
1874 # In this approach we use a separate partition to hold
1875 # current's root, 'usr', and 'var' directories. A partition
1876 # holding "/", "/usr" and "/var" should be about 2GB in
1879 <make sure you have good level 0 dumps>
1882 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1883 <maybe newfs current's root partition>
1884 <mount current's root partition on directory ${CURRENT_ROOT}>
1885 make installworld DESTDIR=${CURRENT_ROOT}
1886 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1887 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1888 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1889 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1890 <reboot into current>
1891 <do a "native" rebuild/install as described in the previous section>
1892 <maybe install compatibility libraries from ports/misc/compat*>
1896 To upgrade in-place from stable to current
1897 ----------------------------------------------
1898 <make sure you have good level 0 dumps>
1900 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1902 <reboot in single user> [3]
1909 Make sure that you've read the UPDATING file to understand the
1910 tweaks to various things you need. At this point in the life
1911 cycle of current, things change often and you are on your own
1912 to cope. The defaults can also change, so please read ALL of
1913 the UPDATING entries.
1915 Also, if you are tracking -current, you must be subscribed to
1916 freebsd-current@freebsd.org. Make sure that before you update
1917 your sources that you have read and understood all the recent
1918 messages there. If in doubt, please track -stable which has
1919 much fewer pitfalls.
1921 [1] If you have third party modules, such as vmware, you
1922 should disable them at this point so they don't crash your
1925 [3] From the bootblocks, boot -s, and then do
1930 adjkerntz -i # if CMOS is wall time
1931 Also, when doing a major release upgrade, it is required that
1932 you boot into single user mode to do the installworld.
1934 [4] Note: This step is non-optional. Failure to do this step
1935 can result in a significant reduction in the functionality of the
1936 system. Attempting to do it by hand is not recommended and those
1937 that pursue this avenue should read this file carefully, as well
1938 as the archives of freebsd-current and freebsd-hackers mailing lists
1939 for potential gotchas. The -U option is also useful to consider.
1940 See mergemaster(8) for more information.
1942 [5] Usually this step is a noop. However, from time to time
1943 you may need to do this if you get unknown user in the following
1944 step. It never hurts to do it all the time. You may need to
1945 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1946 install) after the buildworld before this step if you last updated
1947 from current before 20130425 or from -stable before 20130430.
1949 [6] This only deletes old files and directories. Old libraries
1950 can be deleted by "make delete-old-libs", but you have to make
1951 sure that no program is using those libraries anymore.
1953 [8] In order to have a kernel that can run the 4.x binaries needed to
1954 do an installworld, you must include the COMPAT_FREEBSD4 option in
1955 your kernel. Failure to do so may leave you with a system that is
1956 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1957 required to run the 5.x binaries on more recent kernels. And so on
1958 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1960 Make sure that you merge any new devices from GENERIC since the
1961 last time you updated your kernel config file.
1963 [9] When checking out sources, you must include the -P flag to have
1964 cvs prune empty directories.
1966 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1967 "?=" instead of the "=" assignment operator, so that buildworld can
1968 override the CPUTYPE if it needs to.
1970 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1971 not on the command line, or in /etc/make.conf. buildworld will
1972 warn if it is improperly defined.
1975 This file contains a list, in reverse chronological order, of major
1976 breakages in tracking -current. It is not guaranteed to be a complete
1977 list of such breakages, and only contains entries since October 10, 2007.
1978 If you need to see UPDATING entries from before that date, you will need
1979 to fetch an UPDATING file from an older FreeBSD release.
1981 Copyright information:
1983 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1985 Redistribution, publication, translation and use, with or without
1986 modification, in full or in part, in any form or format of this
1987 document are permitted without further permission from the author.
1989 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1990 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1991 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1992 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1993 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1994 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1995 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1996 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1997 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1998 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1999 POSSIBILITY OF SUCH DAMAGE.
2001 Contact Warner Losh if you have any questions about your use of