1 Updating Information for FreeBSD current users
3 This file is maintained and copyrighted by M. Warner Losh <imp@freebsd.org>.
4 See end of file for further details. For commonly done items, please see the
5 COMMON ITEMS: section later in the file. These instructions assume that you
6 basically know what you are doing. If not, then please consult the FreeBSD
9 Items affecting the ports and packages system can be found in
10 /usr/ports/UPDATING. Please read that file before running portupgrade.
12 NOTE TO PEOPLE WHO THINK THAT FreeBSD 9.x IS SLOW:
13 FreeBSD 9.x has many debugging features turned on, in both the kernel
14 and userland. These features attempt to detect incorrect use of
15 system primitives, and encourage loud failure through extra sanity
16 checking and fail stop semantics. They also substantially impact
17 system performance. If you want to do performance measurement,
18 benchmarking, and optimization, you'll want to turn them off. This
19 includes various WITNESS- related kernel options, INVARIANTS, malloc
20 debugging flags in userland, and various verbose features in the
21 kernel. Many developers choose to disable these features on build
22 machines to maximize performance. (To disable malloc debugging, run
23 ln -s aj /etc/malloc.conf.)
26 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
27 into their kernel configurations along with 'device pci'.
30 The default NFS client is now the new NFS client, so fstype "newnfs"
31 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
32 Although mounts via fstype "nfs" will usually work without userland
33 changes, it is recommended that the mount(8) and mount_nfs(8)
34 commands be rebuilt from sources and that a link to mount_nfs called
35 mount_oldnfs be created. The new client is compiled into the
36 kernel with "options NFSCL" and this is needed for diskless root
37 file systems. The GENERIC kernel configs have been changed to use
38 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
39 To use the regular/old client, you can "mount -t oldnfs ...". For
40 a diskless root file system, you must also include a line like:
42 vfs.root.mountfrom="oldnfs:"
44 in the boot/loader.conf on the root fs on the NFS server to make
45 a diskless root fs use the old client.
48 The GENERIC kernels for all architectures now default to the new
49 CAM-based ATA stack. It means that all legacy ATA drivers were
50 removed and replaced by respective CAM drivers. If you are using
51 ATA device names in /etc/fstab or other places, make sure to update
52 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
53 where 'Y's are the sequential numbers starting from zero for each type
54 in order of detection, unless configured otherwise with tunables,
55 see cam(4)). There will be symbolic links created in /dev/ to map
56 old adX devices to the respective adaY. They should provide basic
57 compatibility for file systems mounting in most cases, but they do
58 not support old user-level APIs and do not have respective providers
59 in GEOM. Consider using updated management tools with new device names.
61 It is possible to load devices ahci, ata, siis and mvs as modules,
62 but option ATA_CAM should remain in kernel configuration to make ata
63 module work as CAM driver supporting legacy ATA controllers. Device ata
64 still can be used in modular fashion (atacore + ...). Modules atadisk
65 and atapi* are not used and won't affect operation in ATA_CAM mode.
66 Note that to use CAM-based ATA kernel should include CAM devices
67 scbus, pass, da (or explicitly ada), cd and optionally others. All of
68 them are parts of the cam module.
70 ataraid(4) functionality is now supported by the RAID GEOM class.
71 To use it you can load geom_raid kernel module and use graid(8) tool
72 for management. Instead of /dev/arX device names, use /dev/raid/rX.
74 No kernel config options or code have been removed, so if a problem
75 arises, please report it and optionally revert to the old ATA stack.
76 In order to do it you can remove from the kernel config:
81 , and instead add back:
82 device atadisk # ATA disk drives
83 device ataraid # ATA RAID drives
84 device atapicd # ATAPI CDROM drives
85 device atapifd # ATAPI floppy drives
86 device atapist # ATAPI tape drives
89 The default NFS server has been changed to the new server, which
90 was referred to as the experimental server. If you need to switch
91 back to the old NFS server, you must now put the "-o" option on
92 both the mountd and nfsd commands. This can be done using the
93 mountd_flags and nfs_server_flags rc.conf variables until an
94 update to the rc scripts is committed, which is coming soon.
97 The GNU Objective-C runtime library (libobjc), and other Objective-C
98 related components have been removed from the base system. If you
99 require an Objective-C library, please use one of the available ports.
102 ath(4) has been split into bus- and device- modules. if_ath contains
103 the HAL, the TX rate control and the network device code. if_ath_pci
104 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
105 contains the AHB glue. Users need to load both if_ath_pci and if_ath
106 in order to use ath on everything else.
108 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
109 need to load if_ath and if_ath_pci for ath(4) operation.
112 As part of the replacement of sysinstall, the process of building
113 release media has changed significantly. For details, please re-read
114 release(7), which has been updated to reflect the new build process.
117 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
118 is the last available version under GPLv2. It brings a number of new
119 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
120 4.1 and SSE 4.2), better support for powerpc64, a number of new
121 directives, and lots of other small improvements. See the ChangeLog
122 file in contrib/binutils for the full details.
125 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
126 compliant, and will now use half of hash for authentication.
127 This will break interoperability with all stacks (including all
128 actual FreeBSD versions) who implement
129 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
131 The only workaround with such peers is to use another HMAC
132 algorithm for IPsec ("phase 2") authentication.
135 Remove the uio_yield prototype and symbol. This function has
136 been misnamed since it was introduced and should not be
137 globally exposed with this name. The equivalent functionality
138 is now available using kern_yield(curthread->td_user_pri).
139 The function remains undocumented.
142 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
143 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
144 for scalar sysctls is defined but disabled. Code that needs
145 UQUAD to pass the type checking that must compile on older
146 systems where the define is not present can check against
147 __FreeBSD_version >= 900030.
149 The system dialog(1) has been replaced with a new version previously
150 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
151 with the previous version, but the libdialog associated with it has
152 a largely incompatible API. As such, the original version of libdialog
153 will be kept temporarily as libodialog, until its base system consumers
154 are replaced or updated. Bump __FreeBSD_version to 900030.
157 If you are trying to run make universe on a -stable system, and you get
158 the following warning:
159 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
160 unknown. config(8) likely too old."
161 or something similar to it, then you must upgrade your -stable system
162 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
163 stable/8) or build the config from the latest stable/8 branch and
164 install it on your system.
166 Prior to this date, building a current universe on 8-stable system from
167 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
168 error in the first kernel build phase. A new config on those old
169 systems will fix that problem for older versions of -current.
172 The TCP stack has been modified to allow Khelp modules to interact with
173 it via helper hook points and store per-connection data in the TCP
174 control block. Bump __FreeBSD_version to 900029. User space tools that
175 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
179 Generic IEEE 802.3 annex 31B full duplex flow control support has been
180 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
181 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
182 to take advantage of it instead of using custom implementations. This
183 means that these drivers now no longer unconditionally advertise
184 support for flow control but only do so if flow control is a selected
185 media option. This was implemented in the generic support that way in
186 order to allow flow control to be switched on and off via ifconfig(8)
187 with the PHY specific default to typically off in order to protect
188 from unwanted effects. Consequently, if you used flow control with
189 one of the above mentioned drivers you now need to explicitly enable
191 ifconfig bge0 media auto mediaopt flowcontrol
193 Along with the above mentioned changes generic support for setting
194 1000baseT master mode also has been added and brgphy(4), ciphy(4),
195 e1000phy(4) as well as ip1000phy(4) have been converted to take
196 advantage of it. This means that these drivers now no longer take the
197 link0 parameter for selecting master mode but the master media option
198 has to be used instead, for example like in the following:
199 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
201 Selection of master mode now is also available with all other PHY
202 drivers supporting 1000baseT.
205 The TCP stack has received a significant update to add support for
206 modularised congestion control and generally improve the clarity of
207 congestion control decisions. Bump __FreeBSD_version to 900025. User
208 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
209 sockstat) need to be recompiled.
212 The man(1) utility has been replaced by a new version that no longer
213 uses /etc/manpath.config. Please consult man.conf(5) for how to
214 migrate local entries to the new format.
217 The copyright strings printed by login(1) and sshd(8) at the time of a
218 new connection have been removed to follow other operating systems and
222 A workaround for a fixed ld bug has been removed in kernel code,
223 so make sure that your system ld is built from sources after
224 revision 210245 from 2010-07-19 (r211583 if building head kernel
225 on stable/8, r211584 for stable/7; both from 2010-08-21).
226 A symptom of incorrect ld version is different addresses for
227 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
230 The $ipv6_prefer variable in rc.conf(5) has been split into
231 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
233 The $ip6addrctl_policy is a variable to choose a pre-defined
234 address selection policy set by ip6addrctl(8). A value
235 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
238 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
239 flag (see an entry of 20090926) is set on an interface with no
240 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
241 security reason. If you want IPv6 link-local address on all
242 interfaces by default, set this to "YES".
244 The old ipv6_prefer="YES" is equivalent to
245 ipv6_activate_all_interfaces="YES" and
246 ip6addrctl_policy="ipv6_prefer".
249 DTrace has grown support for userland tracing. Due to this, DTrace is
250 now i386 and amd64 only.
251 dtruss(1) is now installed by default on those systems and a new
252 kernel module is needed for userland tracing: fasttrap.
253 No changes to your kernel config file are necessary to enable
254 userland tracing, but you might consider adding 'STRIP=' and
255 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
256 to have informative userland stack traces in DTrace (ustack).
259 The acpi_aiboost(4) driver has been removed in favor of the new
260 aibs(4) driver. You should update your kernel configuration file.
263 BSD grep has been imported to the base system and it is built by
264 default. It is completely BSD licensed, highly GNU-compatible, uses
265 less memory than its GNU counterpart and has a small codebase.
266 However, it is slower than its GNU counterpart, which is mostly
267 noticeable for larger searches, for smaller ones it is measurable
268 but not significant. The reason is complex, the most important factor
269 is that we lack a modern and efficient regex library and GNU
270 overcomes this by optimizing the searches internally. Future work
271 on improving the regex performance is planned, for the meantime,
272 users that need better performance, can build GNU grep instead by
273 setting the WITH_GNU_GREP knob.
276 Due to the import of powerpc64 support, all existing powerpc kernel
277 configuration files must be updated with a machine directive like this:
278 machine powerpc powerpc
280 In addition, an updated config(8) is required to build powerpc kernels
284 A new version of ZFS (version 15) has been merged to -HEAD.
285 This version uses a python library for the following subcommands:
286 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
287 For full functionality of these commands the following port must
288 be installed: sysutils/py-zfs
291 'vm_page's are now hashed by physical address to an array of mutexes.
292 Currently this is only used to serialize access to hold_count. Over
293 time the page queue mutex will be peeled away. This changes the size
294 of pmap on every architecture. And requires all callers of vm_page_hold
295 and vm_page_unhold to be updated.
298 WITH_CTF can now be specified in src.conf (not recommended, there
299 are some problems with static executables), make.conf (would also
300 affect ports which do not use GNU make and do not override the
301 compile targets) or in the kernel config (via "makeoptions
303 When WITH_CTF was specified there before this was silently ignored,
304 so make sure that WITH_CTF is not used in places which could lead
305 to unwanted behavior.
308 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
309 to allow 32-bit compatibility on non-x86 platforms. All kernel
310 configurations on amd64 and ia64 platforms using these options must
311 be modified accordingly.
314 The utmp user accounting database has been replaced with utmpx,
315 the user accounting interface standardized by POSIX.
316 Unfortunately the semantics of utmp and utmpx don't match,
317 making it practically impossible to support both interfaces.
318 The user accounting database is used by tools like finger(1),
319 last(1), talk(1), w(1) and ac(8).
321 All applications in the base system use utmpx. This means only
322 local binaries (e.g. from the ports tree) may still use these
323 utmp database files. These applications must be rebuilt to make
326 After the system has been upgraded, it is safe to remove the old
327 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
328 assuming their contents is of no importance anymore. Old wtmp
329 databases can only be used by last(1) and ac(8) after they have
330 been converted to the new format using wtmpcvt(1).
333 Introduce the kernel thread "deadlock resolver" (which can be enabled
334 via the DEADLKRES option, see NOTES for more details) and the
335 sleepq_type() function for sleepqueues.
338 The rc.firewall and rc.firewall6 were unified, and
339 rc.firewall6 and rc.d/ip6fw were removed.
340 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
341 variables are obsoleted. Instead, the following new rc
342 variables are added to rc.d/ipfw:
344 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
345 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
346 firewall_simple_onet_ipv6, firewall_trusted_ipv6
348 The meanings correspond to the relevant IPv4 variables.
354 The default terminal emulation for syscons(4) has been changed
355 from cons25 to xterm on all platforms except pc98. This means
356 that the /etc/ttys file needs to be updated to ensure correct
357 operation of applications on the console.
359 The terminal emulation style can be toggled per window by using
360 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
361 options can be used to change the compile-time default back to
364 To prevent graphical artifacts, make sure the TERM environment
365 variable is set to match the terminal emulation that is being
366 performed by syscons(4).
369 The layout of the structure ieee80211req_scan_result has changed.
370 Applications that require wireless scan results (e.g. ifconfig(8))
371 from net80211 need to be recompiled.
373 Applications such as wpa_supplicant(8) may require a full world
374 build without using NO_CLEAN in order to get synchronized with the
378 The iwn(4) driver has been updated to support the 5000 and 5150 series.
379 There's one kernel module for each firmware. Adding "device iwnfw"
380 to the kernel configuration file means including all three firmware
381 images inside the kernel. If you want to include just the one for
382 your wireless card, use the the devices iwn4965fw, iwn5000fw or
386 The rc.d/network_ipv6, IPv6 configuration script has been integrated
387 into rc.d/netif. The changes are the following:
389 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
390 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
391 Note that both variables need the "inet6" keyword at the head.
393 Do not set $ipv6_network_interfaces manually if you do not
394 understand what you are doing. It is not needed in most cases.
396 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
399 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
400 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
402 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
403 all configured interfaces have "inet6 accept_rtadv" in the
404 $ifconfig_IF_ipv6. These are for backward compatibility.
406 3. A new variable $ipv6_prefer has been added. If NO, IPv6
407 functionality of interfaces with no corresponding
408 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
409 and the default address selection policy of ip6addrctl(8)
410 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
411 Note that if you want to configure IPv6 functionality on the
412 disabled interfaces after boot, first you need to clear the flag by
413 using ifconfig(8) like:
415 ifconfig em0 inet6 -ifdisabled
417 If YES, the default address selection policy is set as
420 The default value of $ipv6_prefer is NO.
422 4. If your system need to receive Router Advertisement messages,
423 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
424 scripts automatically invoke rtsol(8) when the interface becomes
425 UP. The Router Advertisement messages are used for SLAAC
426 (State-Less Address AutoConfiguration).
429 802.11s D3.03 support was committed. This is incompatible with the
430 previous code, which was based on D3.0.
433 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
434 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
435 control whether accepting Router Advertisement messages or not.
436 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
437 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
438 The ifconfig(8) utility now supports these flags.
441 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
442 mount(8) and -a option for df(1) to see them.
445 The old tunable hw.bus.devctl_disable has been superseded by
446 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
447 replaced by hw.bus.devctl_queue=0. The default for this new tunable
451 Remove the option STOP_NMI. The default action is now to use NMI only
452 for KDB via the newly introduced function stop_cpus_hard() and
453 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
456 The stable/8 branch created in subversion. This corresponds to the
457 RELENG_8 branch in CVS.
460 Bump the shared library version numbers for all libraries that do not
461 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
462 __FreeBSD_version to 800105.
465 Due to changes in the implementation of virtual network stack support,
466 all network-related kernel modules must be recompiled. As this change
467 breaks the ABI, bump __FreeBSD_version to 800104.
470 The TOE interface to the TCP syncache has been modified to remove
471 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
472 The cxgb driver is the only TOE consumer affected by this change, and
473 needs to be recompiled along with the kernel. As this change breaks
474 the ABI, bump __FreeBSD_version to 800103.
477 Padding has been added to struct tcpcb, sackhint and tcpstat in
478 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
479 maintaining the ABI. However, this change breaks the ABI, so bump
480 __FreeBSD_version to 800102. User space tools that rely on the size of
481 any of these structs (e.g. sockstat) need to be recompiled.
484 The NFS_LEGACYRPC option has been removed along with the old kernel
485 RPC implementation that this option selected. Kernel configurations
486 may need to be adjusted.
489 The network interface device nodes at /dev/net/<interface> have been
490 removed. All ioctl operations can be performed the normal way using
491 routing sockets. The kqueue functionality can generally be replaced
492 with routing sockets.
495 The documentation from the FreeBSD Documentation Project (Handbook,
496 FAQ, etc.) is now installed via packages by sysinstall(8) and under
497 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
500 The ABI of various structures related to the SYSV IPC API have been
501 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
502 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
506 Layout of struct vnet has changed as routing related variables were
507 moved to their own Vimage module. Modules need to be recompiled. Bump
508 __FreeBSD_version to 800099.
511 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
512 respectively. As long as no more than 16 groups per process are used,
513 no changes should be visible. When more than 16 groups are used, old
514 binaries may fail if they call getgroups() or getgrouplist() with
515 statically sized storage. Recompiling will work around this, but
516 applications should be modified to use dynamically allocated storage
517 for group arrays as POSIX.1-2008 does not cap an implementation's
518 number of supported groups at NGROUPS_MAX+1 as previous versions did.
520 NFS and portalfs mounts may also be affected as the list of groups is
521 truncated to 16. Users of NFS who use more than 16 groups, should
522 take care that negative group permissions are not used on the exported
523 file systems as they will not be reliable unless a GSSAPI based
524 authentication method is used.
527 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
528 option compiles in the support for adaptive spinning for lockmgrs
529 which want to enable it. The lockinit() function now accepts the flag
530 LK_ADAPTIVE in order to make the lock object subject to adaptive
531 spinning when both held in write and read mode.
534 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
535 changed. User applications that use this ioctl need to be rebuilt.
538 The layout of struct thread has changed. Kernel and modules need to
542 The layout of structs ifnet, domain, protosw and vnet_net has changed.
543 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
546 window(1) has been removed from the base system. It can now be
547 installed from ports. The port is called misc/window.
550 The way we are storing and accessing `routing table' entries has
551 changed. Programs reading the FIB, like netstat, need to be
555 A new netisr implementation has been added for FreeBSD 8. Network
556 file system modules, such as igmp, ipdivert, and others, should be
558 Bump __FreeBSD_version to 800096.
561 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
565 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
566 Bump __FreeBSD_version to 800094.
569 Add mnt_xflag field to 'struct mount'. File system modules need to be
571 Bump __FreeBSD_version to 800093.
574 The compiling option ADAPTIVE_SX has been retired while it has been
575 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
576 The KPI for sx_init_flags() changes as accepting flags:
577 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
578 been introduced in order to handle the reversed logic.
579 Bump __FreeBSD_version to 800092.
582 Add support for hierarchical jails. Remove global securelevel.
583 Bump __FreeBSD_version to 800091.
586 The layout of struct vnet_net has changed, therefore modules
588 Bump __FreeBSD_version to 800090.
591 The newly imported zic(8) produces a new format in the output. Please
592 run tzsetup(8) to install the newly created data to /etc/localtime.
595 The sysctl tree for the usb stack has renamed from hw.usb2.* to
596 hw.usb.* and is now consistent again with previous releases.
599 802.11 monitor mode support was revised and driver api's were changed.
600 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
601 of DLT_IEEE802_11. No user-visible data structures were changed but
602 applications that use DLT_IEEE802_11 may require changes.
603 Bump __FreeBSD_version to 800088.
606 The layout of the following structs has changed: sysctl_oid,
607 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
608 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
609 panics may be experienced. World rebuild is required for
610 correctly checking networking state from userland.
611 Bump __FreeBSD_version to 800085.
614 MLDv2 and Source-Specific Multicast (SSM) have been merged
615 to the IPv6 stack. VIMAGE hooks are in but not yet used.
616 The implementation of SSM within FreeBSD's IPv6 stack closely
617 follows the IPv4 implementation.
619 For kernel developers:
621 * The most important changes are that the ip6_output() and
622 ip6_input() paths no longer take the IN6_MULTI_LOCK,
623 and this lock has been downgraded to a non-recursive mutex.
625 * As with the changes to the IPv4 stack to support SSM, filtering
626 of inbound multicast traffic must now be performed by transport
627 protocols within the IPv6 stack. This does not apply to TCP and
628 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
630 * The KPIs used by IPv6 multicast are similar to those used by
631 the IPv4 stack, with the following differences:
632 * im6o_mc_filter() is analogous to imo_multicast_filter().
633 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
634 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
635 * IN6_LOOKUP_MULTI() has been deprecated and removed.
636 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
637 for MLDv1 have an additional 'timer' argument which is used to
638 jitter the initial membership report for the solicited-node
639 multicast membership on-link.
640 * This is not strictly needed for MLDv2, which already jitters
641 its report transmissions. However, the 'timer' argument is
642 preserved in case MLDv1 is active on the interface.
644 * The KAME linked-list based IPv6 membership implementation has
645 been refactored to use a vector similar to that used by the IPv4
647 Code which maintains a list of its own multicast memberships
648 internally, e.g. carp, has been updated to reflect the new
651 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
652 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
653 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
654 implementation constraint which needs to be addressed in HEAD.
656 For application developers:
658 * The changes are broadly similar to those made for the IPv4
661 * The use of IPv4 and IPv6 multicast socket options on the same
662 socket, using mapped addresses, HAS NOT been tested or supported.
664 * There are a number of issues with the implementation of various
665 IPv6 multicast APIs which need to be resolved in the API surface
666 before the implementation is fully compatible with KAME userland
667 use, and these are mostly to do with interface index treatment.
669 * The literature available discusses the use of either the delta / ASM
670 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
671 using setsourcefilter(3)/getsourcefilter(3). For more information
672 please refer to RFC 3768, 'Socket Interface Extensions for
673 Multicast Source Filters'.
675 * Applications which use the published RFC 3678 APIs should be fine.
677 For systems administrators:
679 * The mtest(8) utility has been refactored to support IPv6, in
680 addition to IPv4. Interface addresses are no longer accepted
681 as arguments, their names must be used instead. The utility
682 will map the interface name to its first IPv4 address as
683 returned by getifaddrs(3).
685 * The ifmcstat(8) utility has also been updated to print the MLDv2
686 endpoint state and source filter lists via sysctl(3).
688 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
689 loopback of IPv6 multicast datagrams by default; it defaults to 1
690 to preserve the existing behaviour. Disabling multicast loopback is
691 recommended for optimal system performance.
693 * The IPv6 MROUTING code has been changed to examine this sysctl
694 instead of attempting to perform a group lookup before looping
695 back forwarded datagrams.
697 Bump __FreeBSD_version to 800084.
700 Implement low-level Bluetooth HCI API.
701 Bump __FreeBSD_version to 800083.
704 The layout of struct malloc_type, used by modules to register new
705 memory allocation types, has changed. Most modules will need to
706 be rebuilt or panics may be experienced.
707 Bump __FreeBSD_version to 800081.
710 Anticipate overflowing inp_flags - add inp_flags2.
711 This changes most offsets in inpcb, so checking v4 connection
712 state will require a world rebuild.
713 Bump __FreeBSD_version to 800080.
716 Add an llentry to struct route and struct route_in6. Modules
717 embedding a struct route will need to be recompiled.
718 Bump __FreeBSD_version to 800079.
721 The size of rt_metrics_lite and by extension rtentry has changed.
722 Networking administration apps will need to be recompiled.
723 The route command now supports show as an alias for get, weighting
724 of routes, sticky and nostick flags to alter the behavior of stateful
726 Bump __FreeBSD_version to 800078.
729 Do not use Giant for kbdmux(4) locking. This is wrong and
730 apparently causing more problems than it solves. This will
731 re-open the issue where interrupt handlers may race with
732 kbdmux(4) in polling mode. Typical symptoms include (but
733 not limited to) duplicated and/or missing characters when
734 low level console functions (such as gets) are used while
735 interrupts are enabled (for example geli password prompt,
736 mountroot prompt etc.). Disabling kbdmux(4) may help.
739 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
740 kernel modules referencing any of the above need to be recompiled.
741 Bump __FreeBSD_version to 800075.
744 GEOM_PART has become the default partition slicer for storage devices,
745 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
746 introduces some changes:
748 MSDOS/EBR: the devices created from MSDOS extended partition entries
749 (EBR) can be named differently than with GEOM_MBR and are now symlinks
750 to devices with offset-based names. fstabs may need to be modified.
752 BSD: the "geometry does not match label" warning is harmless in most
753 cases but it points to problems in file system misalignment with
754 disk geometry. The "c" partition is now implicit, covers the whole
755 top-level drive and cannot be (mis)used by users.
757 General: Kernel dumps are now not allowed to be written to devices
758 whose partition types indicate they are meant to be used for file
759 systems (or, in case of MSDOS partitions, as something else than
762 Most of these changes date approximately from 200812.
765 The uscanner(4) driver has been removed from the kernel. This follows
766 Linux removing theirs in 2.6 and making libusb the default interface
770 The multicast forwarding code has been cleaned up. netstat(1)
771 only relies on KVM now for printing bandwidth upcall meters.
772 The IPv4 and IPv6 modules are split into ip_mroute_mod and
773 ip6_mroute_mod respectively. The config(5) options for statically
774 compiling this code remain the same, i.e. 'options MROUTING'.
777 Support for the IFF_NEEDSGIANT network interface flag has been
778 removed, which means that non-MPSAFE network device drivers are no
779 longer supported. In particular, if_ar, if_sr, and network device
780 drivers from the old (legacy) USB stack can no longer be built or
784 POSIX.1 Native Language Support (NLS) has been enabled in libc and
785 a bunch of new language catalog files have also been added.
786 This means that some common libc messages are now localized and
787 they depend on the LC_MESSAGES environmental variable.
790 The k8temp(4) driver has been renamed to amdtemp(4) since
791 support for Family 10 and Family 11 CPU families was added.
794 IGMPv3 and Source-Specific Multicast (SSM) have been merged
795 to the IPv4 stack. VIMAGE hooks are in but not yet used.
797 For kernel developers, the most important changes are that the
798 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
799 and this lock has been downgraded to a non-recursive mutex.
801 Transport protocols (UDP, Raw IP) are now responsible for filtering
802 inbound multicast traffic according to group membership and source
803 filters. The imo_multicast_filter() KPI exists for this purpose.
804 Transports which do not use multicast (SCTP, TCP) already reject
805 multicast by default. Forwarding and receive performance may improve
806 as a mutex acquisition is no longer needed in the ip_input()
807 low-level input path. in_addmulti() and in_delmulti() are shimmed
808 to new KPIs which exist to support SSM in-kernel.
810 For application developers, it is recommended that loopback of
811 multicast datagrams be disabled for best performance, as this
812 will still cause the lock to be taken for each looped-back
813 datagram transmission. The net.inet.ip.mcast.loop sysctl may
814 be tuned to 0 to disable loopback by default; it defaults to 1
815 to preserve the existing behaviour.
817 For systems administrators, to obtain best performance with
818 multicast reception and multiple groups, it is always recommended
819 that a card with a suitably precise hash filter is used. Hash
820 collisions will still result in the lock being taken within the
821 transport protocol input path to check group membership.
823 If deploying FreeBSD in an environment with IGMP snooping switches,
824 it is recommended that the net.inet.igmp.sendlocal sysctl remain
825 enabled; this forces 224.0.0.0/24 group membership to be announced
828 The size of 'struct igmpstat' has changed; netstat needs to be
829 recompiled to reflect this.
830 Bump __FreeBSD_version to 800070.
833 libusb20.so.1 is now installed as libusb.so.1 and the ports system
834 updated to use it. This requires a buildworld/installworld in order to
835 update the library and dependencies (usbconfig, etc). Its advisable to
836 rebuild all ports which uses libusb. More specific directions are given
837 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
838 libusb are no longer required and can be removed.
841 A workaround is committed to allow the creation of System V shared
842 memory segment of size > 2 GB on the 64-bit architectures.
843 Due to a limitation of the existing ABI, the shm_segsz member
844 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
845 wrong for large segments. Note that limits must be explicitly
846 raised to allow such segments to be created.
849 The layout of struct ifnet has changed, requiring a rebuild of all
850 network device driver modules.
853 The /dev handling for the new USB stack has changed, a
854 buildworld/installworld is required for libusb20.
857 The new USB2 stack has now been permanently moved in and all kernel and
858 module names reverted to their previous values (eg, usb, ehci, ohci,
859 ums, ...). The old usb stack can be compiled in by prefixing the name
860 with the letter 'o', the old usb modules have been removed.
861 Updating entry 20090216 for xorg and 20090215 for libmap may still
865 The rc.conf(5) option if_up_delay has been renamed to
866 defaultroute_delay to better reflect its purpose. If you have
867 customized this setting in /etc/rc.conf you need to update it to
871 xorg 7.4 wants to configure its input devices via hald which does not
872 yet work with USB2. If the keyboard/mouse does not work in xorg then
874 Option "AllowEmptyInput" "off"
875 to your ServerLayout section. This will cause X to use the configured
876 kbd and mouse sections from your xorg.conf.
879 The GENERIC kernels for all architectures now default to the new USB2
880 stack. No kernel config options or code have been removed so if a
881 problem arises please report it and optionally revert to the old USB
882 stack. If you are loading USB kernel modules or have a custom kernel
883 that includes GENERIC then ensure that usb names are also changed over,
884 eg uftdi -> usb2_serial_ftdi.
886 Older programs linked against the ports libusb 0.1 need to be
887 redirected to the new stack's libusb20. /etc/libmap.conf can
889 # Map old usb library to new one for usb2 stack
890 libusb-0.1.so.8 libusb20.so.1
893 All USB ethernet devices now attach as interfaces under the name ueN
894 (eg. ue0). This is to provide a predictable name as vendors often
895 change usb chipsets in a product without notice.
898 The ichsmb(4) driver has been changed to require SMBus slave
899 addresses be left-justified (xxxxxxx0b) rather than right-justified.
900 All of the other SMBus controller drivers require left-justified
901 slave addresses, so this change makes all the drivers provide the
905 INET6 statistics (struct ip6stat) was updated.
906 netstat(1) needs to be recompiled.
909 NTFS has been removed from GENERIC kernel on amd64 to match
910 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
911 will load ntfs.ko module automatically when NTFS support is
912 actually needed, unless ntfs.ko is not installed or security
913 level prohibits loading kernel modules. If either is the case,
914 "options NTFS" has to be added into kernel config.
917 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
918 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
919 800061. User space tools that rely on the size of struct tcpcb in
920 tcp_var.h (e.g. sockstat) need to be recompiled.
923 ng_tty(4) module updated to match the new TTY subsystem.
924 Due to API change, user-level applications must be updated.
925 New API support added to mpd5 CVS and expected to be present
926 in next mpd5.3 release.
929 With __FreeBSD_version 800060 the makefs tool is part of
930 the base system (it was a port).
933 The afdata and ifnet locks have been changed from mutexes to
934 rwlocks, network modules will need to be re-compiled.
937 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
938 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
939 The new code reduced struct rtentry{} by 16 bytes on 32-bit
940 architecture and 40 bytes on 64-bit architecture. The userland
941 applications "arp" and "ndp" have been updated accordingly.
942 The output from "netstat -r" shows only routing entries and
943 none of the L2 information.
946 __FreeBSD_version 800057 marks the switchover from the
947 binary ath hal to source code. Users must add the line:
949 options AH_SUPPORT_AR5416
951 to their kernel config files when specifying:
955 The ath_hal module no longer exists; the code is now compiled
956 together with the driver in the ath module. It is now
957 possible to tailor chip support (i.e. reduce the set of chips
958 and thereby the code size); consult ath_hal(4) for details.
961 __FreeBSD_version 800054 adds memory barriers to
962 <machine/atomic.h>, new interfaces to ifnet to facilitate
963 multiple hardware transmit queues for cards that support
964 them, and a lock-less ring-buffer implementation to
965 enable drivers to more efficiently manage queueing of
969 A new version of ZFS (version 13) has been merged to -HEAD.
970 This version has zpool attribute "listsnapshots" off by
971 default, which means "zfs list" does not show snapshots,
972 and is the same as Solaris behavior.
975 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
978 The uhci, ohci, ehci and slhci USB Host controller drivers have
979 been put into separate modules. If you load the usb module
980 separately through loader.conf you will need to load the
981 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
982 controller add the following to loader.conf:
988 The ABI used by the PMC toolset has changed. Please keep
989 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
993 atapci kernel module now includes only generic PCI ATA
994 driver. AHCI driver moved to ataahci kernel module.
995 All vendor-specific code moved into separate kernel modules:
996 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
997 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
998 atamarvell, atamicron, atanational, atanetcell, atanvidia,
999 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1002 The TTY subsystem of the kernel has been replaced by a new
1003 implementation, which provides better scalability and an
1004 improved driver model. Most common drivers have been migrated to
1005 the new TTY subsystem, while others have not. The following
1006 drivers have not yet been ported to the new TTY layer:
1009 cy, digi, rc, rp, sio
1015 ng_h4, ng_tty, ppp, sl, snp
1017 Adding these drivers to your kernel configuration file shall
1018 cause compilation to fail.
1021 ntpd has been upgraded to 4.2.4p5.
1024 OpenSSH has been upgraded to 5.1p1.
1026 For many years, FreeBSD's version of OpenSSH preferred DSA
1027 over RSA for host and user authentication keys. With this
1028 upgrade, we've switched to the vendor's default of RSA over
1029 DSA. This may cause upgraded clients to warn about unknown
1030 host keys even for previously known hosts. Users should
1031 follow the usual procedure for verifying host keys before
1032 accepting the RSA key.
1034 This can be circumvented by setting the "HostKeyAlgorithms"
1035 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1038 Please note that the sequence of keys offered for
1039 authentication has been changed as well. You may want to
1040 specify IdentityFile in a different order to revert this
1044 The sio(4) driver has been removed from the i386 and amd64
1045 kernel configuration files. This means uart(4) is now the
1046 default serial port driver on those platforms as well.
1048 To prevent collisions with the sio(4) driver, the uart(4) driver
1049 uses different names for its device nodes. This means the
1050 onboard serial port will now most likely be called "ttyu0"
1051 instead of "ttyd0". You may need to reconfigure applications to
1052 use the new device names.
1054 When using the serial port as a boot console, be sure to update
1055 /boot/device.hints and /etc/ttys before booting the new kernel.
1056 If you forget to do so, you can still manually specify the hints
1057 at the loader prompt:
1059 set hint.uart.0.at="isa"
1060 set hint.uart.0.port="0x3F8"
1061 set hint.uart.0.flags="0x10"
1062 set hint.uart.0.irq="4"
1066 The gpt(8) utility has been removed. Use gpart(8) to partition
1070 The version that Linuxulator emulates was changed from 2.4.2
1071 to 2.6.16. If you experience any problems with Linux binaries
1072 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1073 if it fixes the problem contact emulation mailing list.
1076 ISDN4BSD (I4B) was removed from the src tree. You may need to
1077 update a your kernel configuration and remove relevant entries.
1080 I have checked in code to support multiple routing tables.
1081 See the man pages setfib(1) and setfib(2).
1082 This is a hopefully backwards compatible version,
1083 but to make use of it you need to compile your kernel
1084 with options ROUTETABLES=2 (or more up to 16).
1087 The 802.11 wireless support was redone to enable multi-bss
1088 operation on devices that are capable. The underlying device
1089 is no longer used directly but instead wlanX devices are
1090 cloned with ifconfig. This requires changes to rc.conf files.
1091 For example, change:
1092 ifconfig_ath0="WPA DHCP"
1095 ifconfig_wlan0="WPA DHCP"
1096 see rc.conf(5) for more details. In addition, mergemaster of
1097 /etc/rc.d is highly recommended. Simultaneous update of userland
1098 and kernel wouldn't hurt either.
1100 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1101 modules were merged into the base wlan module. All references
1102 to these modules (e.g. in kernel config files) must be removed.
1105 psm(4) has gained write(2) support in native operation level.
1106 Arbitrary commands can be written to /dev/psm%d and status can
1107 be read back from it. Therefore, an application is responsible
1108 for status validation and error recovery. It is a no-op in
1109 other operation levels.
1112 Support for KSE threading has been removed from the kernel. To
1113 run legacy applications linked against KSE libmap.conf may
1114 be used. The following libmap.conf may be used to ensure
1115 compatibility with any prior release:
1117 libpthread.so.1 libthr.so.1
1118 libpthread.so.2 libthr.so.2
1119 libkse.so.3 libthr.so.3
1122 The layout of struct vmspace has changed. This affects libkvm
1123 and any executables that link against libkvm and use the
1124 kvm_getprocs() function. In particular, but not exclusively,
1125 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1126 The effects are minimal, but it's advisable to upgrade world
1130 The latest em driver no longer has support in it for the
1131 82575 adapter, this is now moved to the igb driver. The
1132 split was done to make new features that are incompatible
1133 with older hardware easier to do.
1136 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1137 likewise the kernel option is now GEOM_LINUX_LVM.
1140 The default NFS mount mode has changed from UDP to TCP for
1141 increased reliability. If you rely on (insecurely) NFS
1142 mounting across a firewall you may need to update your
1146 Belatedly note the addition of m_collapse for compacting
1150 The fts(3) structures have been changed to use adequate
1151 integer types for their members and so to be able to cope
1152 with huge file trees. The old fts(3) ABI is preserved
1153 through symbol versioning in libc, so third-party binaries
1154 using fts(3) should still work, although they will not take
1155 advantage of the extended types. At the same time, some
1156 third-party software might fail to build after this change
1157 due to unportable assumptions made in its source code about
1158 fts(3) structure members. Such software should be fixed
1159 by its vendor or, in the worst case, in the ports tree.
1160 FreeBSD_version 800015 marks this change for the unlikely
1161 case that a portable fix is impossible.
1164 To upgrade to -current after this date, you must be running
1165 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1166 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1169 The ADAPTIVE_GIANT kernel option has been retired because its
1170 functionality is the default now.
1173 The AT keyboard emulation of sunkbd(4) has been turned on
1174 by default. In order to make the special symbols of the Sun
1175 keyboards driven by sunkbd(4) work under X these now have
1176 to be configured the same way as Sun USB keyboards driven
1177 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1179 Option "XkbLayout" "us"
1180 Option "XkbRules" "xorg"
1181 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1184 It has been decided that it is desirable to provide ABI
1185 backwards compatibility to the FreeBSD 4/5/6 versions of the
1186 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1187 broken with the introduction of PCI domain support (see the
1188 20070930 entry). Unfortunately, this required the ABI of
1189 PCIOCGETCONF to be broken again in order to be able to
1190 provide backwards compatibility to the old version of that
1191 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1192 again. As for prominent ports this affects neither pciutils
1193 nor xorg-server this time, the hal port needs to be rebuilt
1197 The misnamed kthread_create() and friends have been renamed
1198 to kproc_create() etc. Many of the callers already
1199 used kproc_start()..
1200 I will return kthread_create() and friends in a while
1201 with implementations that actually create threads, not procs.
1202 Renaming corresponds with version 800002.
1211 Avoid using make -j when upgrading. While generally safe, there are
1212 sometimes problems using -j to upgrade. If your upgrade fails with
1213 -j, please try again without -j. From time to time in the past there
1214 have been problems using -j with buildworld and/or installworld. This
1215 is especially true when upgrading between "distant" versions (eg one
1216 that cross a major release boundary or several minor releases, or when
1217 several months have passed on the -current branch).
1219 Sometimes, obscure build problems are the result of environment
1220 poisoning. This can happen because the make utility reads its
1221 environment when searching for values for global variables. To run
1222 your build attempts in an "environmental clean room", prefix all make
1223 commands with 'env -i '. See the env(1) manual page for more details.
1225 When upgrading from one major version to another it is generally best
1226 to upgrade to the latest code in the currently installed branch first,
1227 then do an upgrade to the new branch. This is the best-tested upgrade
1228 path, and has the highest probability of being successful. Please try
1229 this approach before reporting problems with a major version upgrade.
1233 When upgrading the boot ZFS pool to a new version, always follow
1236 1.) recompile and reinstall the ZFS boot loader and boot block
1237 (this is part of "make buildworld" and "make installworld")
1239 2.) update the ZFS boot block on your boot drive
1241 The following example updates the ZFS boot block on the first
1242 partition (freebsd-boot) of a GPT partitioned drive ad0:
1243 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1245 Non-boot pools do not need these updates.
1249 If you are updating from a prior version of FreeBSD (even one just
1250 a few days old), you should follow this procedure. It is the most
1251 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1253 make kernel-toolchain
1254 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1255 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1257 To test a kernel once
1258 ---------------------
1259 If you just want to boot a kernel once (because you are not sure
1260 if it works, or if you want to boot a known bad kernel to provide
1261 debugging information) run
1262 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1263 nextboot -k testkernel
1265 To just build a kernel when you know that it won't mess you up
1266 --------------------------------------------------------------
1267 This assumes you are already running a CURRENT system. Replace
1268 ${arch} with the architecture of your machine (e.g. "i386",
1269 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1271 cd src/sys/${arch}/conf
1272 config KERNEL_NAME_HERE
1273 cd ../compile/KERNEL_NAME_HERE
1278 If this fails, go to the "To build a kernel" section.
1280 To rebuild everything and install it on the current system.
1281 -----------------------------------------------------------
1282 # Note: sometimes if you are running current you gotta do more than
1283 # is listed here if you are upgrading from a really old current.
1285 <make sure you have good level 0 dumps>
1287 make kernel KERNCONF=YOUR_KERNEL_HERE
1289 <reboot in single user> [3]
1297 To cross-install current onto a separate partition
1298 --------------------------------------------------
1299 # In this approach we use a separate partition to hold
1300 # current's root, 'usr', and 'var' directories. A partition
1301 # holding "/", "/usr" and "/var" should be about 2GB in
1304 <make sure you have good level 0 dumps>
1307 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1308 <maybe newfs current's root partition>
1309 <mount current's root partition on directory ${CURRENT_ROOT}>
1310 make installworld DESTDIR=${CURRENT_ROOT}
1311 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1312 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1313 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1314 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1315 <reboot into current>
1316 <do a "native" rebuild/install as described in the previous section>
1317 <maybe install compatibility libraries from ports/misc/compat*>
1321 To upgrade in-place from 8.x-stable to current
1322 ----------------------------------------------
1323 <make sure you have good level 0 dumps>
1325 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1327 <reboot in single user> [3]
1334 Make sure that you've read the UPDATING file to understand the
1335 tweaks to various things you need. At this point in the life
1336 cycle of current, things change often and you are on your own
1337 to cope. The defaults can also change, so please read ALL of
1338 the UPDATING entries.
1340 Also, if you are tracking -current, you must be subscribed to
1341 freebsd-current@freebsd.org. Make sure that before you update
1342 your sources that you have read and understood all the recent
1343 messages there. If in doubt, please track -stable which has
1344 much fewer pitfalls.
1346 [1] If you have third party modules, such as vmware, you
1347 should disable them at this point so they don't crash your
1350 [3] From the bootblocks, boot -s, and then do
1355 adjkerntz -i # if CMOS is wall time
1356 Also, when doing a major release upgrade, it is required that
1357 you boot into single user mode to do the installworld.
1359 [4] Note: This step is non-optional. Failure to do this step
1360 can result in a significant reduction in the functionality of the
1361 system. Attempting to do it by hand is not recommended and those
1362 that pursue this avenue should read this file carefully, as well
1363 as the archives of freebsd-current and freebsd-hackers mailing lists
1364 for potential gotchas. The -U option is also useful to consider.
1365 See mergemaster(8) for more information.
1367 [5] Usually this step is a noop. However, from time to time
1368 you may need to do this if you get unknown user in the following
1369 step. It never hurts to do it all the time. You may need to
1370 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1371 install) after the buildworld before this step if you last updated
1372 from current before 20020224 or from -stable before 20020408.
1374 [6] This only deletes old files and directories. Old libraries
1375 can be deleted by "make delete-old-libs", but you have to make
1376 sure that no program is using those libraries anymore.
1378 [8] In order to have a kernel that can run the 4.x binaries needed to
1379 do an installworld, you must include the COMPAT_FREEBSD4 option in
1380 your kernel. Failure to do so may leave you with a system that is
1381 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1382 required to run the 5.x binaries on more recent kernels. And so on
1383 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1385 Make sure that you merge any new devices from GENERIC since the
1386 last time you updated your kernel config file.
1388 [9] When checking out sources, you must include the -P flag to have
1389 cvs prune empty directories.
1391 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1392 "?=" instead of the "=" assignment operator, so that buildworld can
1393 override the CPUTYPE if it needs to.
1395 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1396 not on the command line, or in /etc/make.conf. buildworld will
1397 warn if it is improperly defined.
1400 This file contains a list, in reverse chronological order, of major
1401 breakages in tracking -current. Not all things will be listed here,
1402 and it only starts on October 16, 2004. Updating files can found in
1403 previous releases if your system is older than this.
1405 Copyright information:
1407 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1409 Redistribution, publication, translation and use, with or without
1410 modification, in full or in part, in any form or format of this
1411 document are permitted without further permission from the author.
1413 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1414 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1415 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1416 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1417 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1418 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1419 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1420 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1421 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1422 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1423 POSSIBILITY OF SUCH DAMAGE.
1425 Contact Warner Losh if you have any questions about your use of