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 Support for sun4v architecture is officially dropped
29 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
30 into their kernel configurations along with 'device pci'.
33 The default NFS client is now the new NFS client, so fstype "newnfs"
34 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
35 Although mounts via fstype "nfs" will usually work without userland
36 changes, it is recommended that the mount(8) and mount_nfs(8)
37 commands be rebuilt from sources and that a link to mount_nfs called
38 mount_oldnfs be created. The new client is compiled into the
39 kernel with "options NFSCL" and this is needed for diskless root
40 file systems. The GENERIC kernel configs have been changed to use
41 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
42 To use the regular/old client, you can "mount -t oldnfs ...". For
43 a diskless root file system, you must also include a line like:
45 vfs.root.mountfrom="oldnfs:"
47 in the boot/loader.conf on the root fs on the NFS server to make
48 a diskless root fs use the old client.
51 The GENERIC kernels for all architectures now default to the new
52 CAM-based ATA stack. It means that all legacy ATA drivers were
53 removed and replaced by respective CAM drivers. If you are using
54 ATA device names in /etc/fstab or other places, make sure to update
55 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
56 where 'Y's are the sequential numbers starting from zero for each type
57 in order of detection, unless configured otherwise with tunables,
58 see cam(4)). There will be symbolic links created in /dev/ to map
59 old adX devices to the respective adaY. They should provide basic
60 compatibility for file systems mounting in most cases, but they do
61 not support old user-level APIs and do not have respective providers
62 in GEOM. Consider using updated management tools with new device names.
64 It is possible to load devices ahci, ata, siis and mvs as modules,
65 but option ATA_CAM should remain in kernel configuration to make ata
66 module work as CAM driver supporting legacy ATA controllers. Device ata
67 still can be used in modular fashion (atacore + ...). Modules atadisk
68 and atapi* are not used and won't affect operation in ATA_CAM mode.
69 Note that to use CAM-based ATA kernel should include CAM devices
70 scbus, pass, da (or explicitly ada), cd and optionally others. All of
71 them are parts of the cam module.
73 ataraid(4) functionality is now supported by the RAID GEOM class.
74 To use it you can load geom_raid kernel module and use graid(8) tool
75 for management. Instead of /dev/arX device names, use /dev/raid/rX.
77 No kernel config options or code have been removed, so if a problem
78 arises, please report it and optionally revert to the old ATA stack.
79 In order to do it you can remove from the kernel config:
84 , and instead add back:
85 device atadisk # ATA disk drives
86 device ataraid # ATA RAID drives
87 device atapicd # ATAPI CDROM drives
88 device atapifd # ATAPI floppy drives
89 device atapist # ATAPI tape drives
92 The default NFS server has been changed to the new server, which
93 was referred to as the experimental server. If you need to switch
94 back to the old NFS server, you must now put the "-o" option on
95 both the mountd and nfsd commands. This can be done using the
96 mountd_flags and nfs_server_flags rc.conf variables until an
97 update to the rc scripts is committed, which is coming soon.
100 The GNU Objective-C runtime library (libobjc), and other Objective-C
101 related components have been removed from the base system. If you
102 require an Objective-C library, please use one of the available ports.
105 ath(4) has been split into bus- and device- modules. if_ath contains
106 the HAL, the TX rate control and the network device code. if_ath_pci
107 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
108 contains the AHB glue. Users need to load both if_ath_pci and if_ath
109 in order to use ath on everything else.
111 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
112 need to load if_ath and if_ath_pci for ath(4) operation.
115 As part of the replacement of sysinstall, the process of building
116 release media has changed significantly. For details, please re-read
117 release(7), which has been updated to reflect the new build process.
120 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
121 is the last available version under GPLv2. It brings a number of new
122 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
123 4.1 and SSE 4.2), better support for powerpc64, a number of new
124 directives, and lots of other small improvements. See the ChangeLog
125 file in contrib/binutils for the full details.
128 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
129 compliant, and will now use half of hash for authentication.
130 This will break interoperability with all stacks (including all
131 actual FreeBSD versions) who implement
132 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
134 The only workaround with such peers is to use another HMAC
135 algorithm for IPsec ("phase 2") authentication.
138 Remove the uio_yield prototype and symbol. This function has
139 been misnamed since it was introduced and should not be
140 globally exposed with this name. The equivalent functionality
141 is now available using kern_yield(curthread->td_user_pri).
142 The function remains undocumented.
145 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
146 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
147 for scalar sysctls is defined but disabled. Code that needs
148 UQUAD to pass the type checking that must compile on older
149 systems where the define is not present can check against
150 __FreeBSD_version >= 900030.
152 The system dialog(1) has been replaced with a new version previously
153 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
154 with the previous version, but the libdialog associated with it has
155 a largely incompatible API. As such, the original version of libdialog
156 will be kept temporarily as libodialog, until its base system consumers
157 are replaced or updated. Bump __FreeBSD_version to 900030.
160 If you are trying to run make universe on a -stable system, and you get
161 the following warning:
162 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
163 unknown. config(8) likely too old."
164 or something similar to it, then you must upgrade your -stable system
165 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
166 stable/8) or build the config from the latest stable/8 branch and
167 install it on your system.
169 Prior to this date, building a current universe on 8-stable system from
170 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
171 error in the first kernel build phase. A new config on those old
172 systems will fix that problem for older versions of -current.
175 The TCP stack has been modified to allow Khelp modules to interact with
176 it via helper hook points and store per-connection data in the TCP
177 control block. Bump __FreeBSD_version to 900029. User space tools that
178 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
182 Generic IEEE 802.3 annex 31B full duplex flow control support has been
183 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
184 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
185 to take advantage of it instead of using custom implementations. This
186 means that these drivers now no longer unconditionally advertise
187 support for flow control but only do so if flow control is a selected
188 media option. This was implemented in the generic support that way in
189 order to allow flow control to be switched on and off via ifconfig(8)
190 with the PHY specific default to typically off in order to protect
191 from unwanted effects. Consequently, if you used flow control with
192 one of the above mentioned drivers you now need to explicitly enable
194 ifconfig bge0 media auto mediaopt flowcontrol
196 Along with the above mentioned changes generic support for setting
197 1000baseT master mode also has been added and brgphy(4), ciphy(4),
198 e1000phy(4) as well as ip1000phy(4) have been converted to take
199 advantage of it. This means that these drivers now no longer take the
200 link0 parameter for selecting master mode but the master media option
201 has to be used instead, for example like in the following:
202 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
204 Selection of master mode now is also available with all other PHY
205 drivers supporting 1000baseT.
208 The TCP stack has received a significant update to add support for
209 modularised congestion control and generally improve the clarity of
210 congestion control decisions. Bump __FreeBSD_version to 900025. User
211 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
212 sockstat) need to be recompiled.
215 The man(1) utility has been replaced by a new version that no longer
216 uses /etc/manpath.config. Please consult man.conf(5) for how to
217 migrate local entries to the new format.
220 The copyright strings printed by login(1) and sshd(8) at the time of a
221 new connection have been removed to follow other operating systems and
225 A workaround for a fixed ld bug has been removed in kernel code,
226 so make sure that your system ld is built from sources after
227 revision 210245 from 2010-07-19 (r211583 if building head kernel
228 on stable/8, r211584 for stable/7; both from 2010-08-21).
229 A symptom of incorrect ld version is different addresses for
230 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
233 The $ipv6_prefer variable in rc.conf(5) has been split into
234 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
236 The $ip6addrctl_policy is a variable to choose a pre-defined
237 address selection policy set by ip6addrctl(8). A value
238 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
241 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
242 flag (see an entry of 20090926) is set on an interface with no
243 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
244 security reason. If you want IPv6 link-local address on all
245 interfaces by default, set this to "YES".
247 The old ipv6_prefer="YES" is equivalent to
248 ipv6_activate_all_interfaces="YES" and
249 ip6addrctl_policy="ipv6_prefer".
252 DTrace has grown support for userland tracing. Due to this, DTrace is
253 now i386 and amd64 only.
254 dtruss(1) is now installed by default on those systems and a new
255 kernel module is needed for userland tracing: fasttrap.
256 No changes to your kernel config file are necessary to enable
257 userland tracing, but you might consider adding 'STRIP=' and
258 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
259 to have informative userland stack traces in DTrace (ustack).
262 The acpi_aiboost(4) driver has been removed in favor of the new
263 aibs(4) driver. You should update your kernel configuration file.
266 BSD grep has been imported to the base system and it is built by
267 default. It is completely BSD licensed, highly GNU-compatible, uses
268 less memory than its GNU counterpart and has a small codebase.
269 However, it is slower than its GNU counterpart, which is mostly
270 noticeable for larger searches, for smaller ones it is measurable
271 but not significant. The reason is complex, the most important factor
272 is that we lack a modern and efficient regex library and GNU
273 overcomes this by optimizing the searches internally. Future work
274 on improving the regex performance is planned, for the meantime,
275 users that need better performance, can build GNU grep instead by
276 setting the WITH_GNU_GREP knob.
279 Due to the import of powerpc64 support, all existing powerpc kernel
280 configuration files must be updated with a machine directive like this:
281 machine powerpc powerpc
283 In addition, an updated config(8) is required to build powerpc kernels
287 A new version of ZFS (version 15) has been merged to -HEAD.
288 This version uses a python library for the following subcommands:
289 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
290 For full functionality of these commands the following port must
291 be installed: sysutils/py-zfs
294 'vm_page's are now hashed by physical address to an array of mutexes.
295 Currently this is only used to serialize access to hold_count. Over
296 time the page queue mutex will be peeled away. This changes the size
297 of pmap on every architecture. And requires all callers of vm_page_hold
298 and vm_page_unhold to be updated.
301 WITH_CTF can now be specified in src.conf (not recommended, there
302 are some problems with static executables), make.conf (would also
303 affect ports which do not use GNU make and do not override the
304 compile targets) or in the kernel config (via "makeoptions
306 When WITH_CTF was specified there before this was silently ignored,
307 so make sure that WITH_CTF is not used in places which could lead
308 to unwanted behavior.
311 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
312 to allow 32-bit compatibility on non-x86 platforms. All kernel
313 configurations on amd64 and ia64 platforms using these options must
314 be modified accordingly.
317 The utmp user accounting database has been replaced with utmpx,
318 the user accounting interface standardized by POSIX.
319 Unfortunately the semantics of utmp and utmpx don't match,
320 making it practically impossible to support both interfaces.
321 The user accounting database is used by tools like finger(1),
322 last(1), talk(1), w(1) and ac(8).
324 All applications in the base system use utmpx. This means only
325 local binaries (e.g. from the ports tree) may still use these
326 utmp database files. These applications must be rebuilt to make
329 After the system has been upgraded, it is safe to remove the old
330 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
331 assuming their contents is of no importance anymore. Old wtmp
332 databases can only be used by last(1) and ac(8) after they have
333 been converted to the new format using wtmpcvt(1).
336 Introduce the kernel thread "deadlock resolver" (which can be enabled
337 via the DEADLKRES option, see NOTES for more details) and the
338 sleepq_type() function for sleepqueues.
341 The rc.firewall and rc.firewall6 were unified, and
342 rc.firewall6 and rc.d/ip6fw were removed.
343 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
344 variables are obsoleted. Instead, the following new rc
345 variables are added to rc.d/ipfw:
347 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
348 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
349 firewall_simple_onet_ipv6, firewall_trusted_ipv6
351 The meanings correspond to the relevant IPv4 variables.
357 The default terminal emulation for syscons(4) has been changed
358 from cons25 to xterm on all platforms except pc98. This means
359 that the /etc/ttys file needs to be updated to ensure correct
360 operation of applications on the console.
362 The terminal emulation style can be toggled per window by using
363 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
364 options can be used to change the compile-time default back to
367 To prevent graphical artifacts, make sure the TERM environment
368 variable is set to match the terminal emulation that is being
369 performed by syscons(4).
372 The layout of the structure ieee80211req_scan_result has changed.
373 Applications that require wireless scan results (e.g. ifconfig(8))
374 from net80211 need to be recompiled.
376 Applications such as wpa_supplicant(8) may require a full world
377 build without using NO_CLEAN in order to get synchronized with the
381 The iwn(4) driver has been updated to support the 5000 and 5150 series.
382 There's one kernel module for each firmware. Adding "device iwnfw"
383 to the kernel configuration file means including all three firmware
384 images inside the kernel. If you want to include just the one for
385 your wireless card, use the the devices iwn4965fw, iwn5000fw or
389 The rc.d/network_ipv6, IPv6 configuration script has been integrated
390 into rc.d/netif. The changes are the following:
392 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
393 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
394 Note that both variables need the "inet6" keyword at the head.
396 Do not set $ipv6_network_interfaces manually if you do not
397 understand what you are doing. It is not needed in most cases.
399 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
402 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
403 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
405 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
406 all configured interfaces have "inet6 accept_rtadv" in the
407 $ifconfig_IF_ipv6. These are for backward compatibility.
409 3. A new variable $ipv6_prefer has been added. If NO, IPv6
410 functionality of interfaces with no corresponding
411 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
412 and the default address selection policy of ip6addrctl(8)
413 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
414 Note that if you want to configure IPv6 functionality on the
415 disabled interfaces after boot, first you need to clear the flag by
416 using ifconfig(8) like:
418 ifconfig em0 inet6 -ifdisabled
420 If YES, the default address selection policy is set as
423 The default value of $ipv6_prefer is NO.
425 4. If your system need to receive Router Advertisement messages,
426 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
427 scripts automatically invoke rtsol(8) when the interface becomes
428 UP. The Router Advertisement messages are used for SLAAC
429 (State-Less Address AutoConfiguration).
432 802.11s D3.03 support was committed. This is incompatible with the
433 previous code, which was based on D3.0.
436 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
437 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
438 control whether accepting Router Advertisement messages or not.
439 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
440 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
441 The ifconfig(8) utility now supports these flags.
444 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
445 mount(8) and -a option for df(1) to see them.
448 The old tunable hw.bus.devctl_disable has been superseded by
449 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
450 replaced by hw.bus.devctl_queue=0. The default for this new tunable
454 Remove the option STOP_NMI. The default action is now to use NMI only
455 for KDB via the newly introduced function stop_cpus_hard() and
456 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
459 The stable/8 branch created in subversion. This corresponds to the
460 RELENG_8 branch in CVS.
463 Bump the shared library version numbers for all libraries that do not
464 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
465 __FreeBSD_version to 800105.
468 Due to changes in the implementation of virtual network stack support,
469 all network-related kernel modules must be recompiled. As this change
470 breaks the ABI, bump __FreeBSD_version to 800104.
473 The TOE interface to the TCP syncache has been modified to remove
474 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
475 The cxgb driver is the only TOE consumer affected by this change, and
476 needs to be recompiled along with the kernel. As this change breaks
477 the ABI, bump __FreeBSD_version to 800103.
480 Padding has been added to struct tcpcb, sackhint and tcpstat in
481 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
482 maintaining the ABI. However, this change breaks the ABI, so bump
483 __FreeBSD_version to 800102. User space tools that rely on the size of
484 any of these structs (e.g. sockstat) need to be recompiled.
487 The NFS_LEGACYRPC option has been removed along with the old kernel
488 RPC implementation that this option selected. Kernel configurations
489 may need to be adjusted.
492 The network interface device nodes at /dev/net/<interface> have been
493 removed. All ioctl operations can be performed the normal way using
494 routing sockets. The kqueue functionality can generally be replaced
495 with routing sockets.
498 The documentation from the FreeBSD Documentation Project (Handbook,
499 FAQ, etc.) is now installed via packages by sysinstall(8) and under
500 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
503 The ABI of various structures related to the SYSV IPC API have been
504 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
505 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
509 Layout of struct vnet has changed as routing related variables were
510 moved to their own Vimage module. Modules need to be recompiled. Bump
511 __FreeBSD_version to 800099.
514 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
515 respectively. As long as no more than 16 groups per process are used,
516 no changes should be visible. When more than 16 groups are used, old
517 binaries may fail if they call getgroups() or getgrouplist() with
518 statically sized storage. Recompiling will work around this, but
519 applications should be modified to use dynamically allocated storage
520 for group arrays as POSIX.1-2008 does not cap an implementation's
521 number of supported groups at NGROUPS_MAX+1 as previous versions did.
523 NFS and portalfs mounts may also be affected as the list of groups is
524 truncated to 16. Users of NFS who use more than 16 groups, should
525 take care that negative group permissions are not used on the exported
526 file systems as they will not be reliable unless a GSSAPI based
527 authentication method is used.
530 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
531 option compiles in the support for adaptive spinning for lockmgrs
532 which want to enable it. The lockinit() function now accepts the flag
533 LK_ADAPTIVE in order to make the lock object subject to adaptive
534 spinning when both held in write and read mode.
537 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
538 changed. User applications that use this ioctl need to be rebuilt.
541 The layout of struct thread has changed. Kernel and modules need to
545 The layout of structs ifnet, domain, protosw and vnet_net has changed.
546 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
549 window(1) has been removed from the base system. It can now be
550 installed from ports. The port is called misc/window.
553 The way we are storing and accessing `routing table' entries has
554 changed. Programs reading the FIB, like netstat, need to be
558 A new netisr implementation has been added for FreeBSD 8. Network
559 file system modules, such as igmp, ipdivert, and others, should be
561 Bump __FreeBSD_version to 800096.
564 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
568 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
569 Bump __FreeBSD_version to 800094.
572 Add mnt_xflag field to 'struct mount'. File system modules need to be
574 Bump __FreeBSD_version to 800093.
577 The compiling option ADAPTIVE_SX has been retired while it has been
578 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
579 The KPI for sx_init_flags() changes as accepting flags:
580 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
581 been introduced in order to handle the reversed logic.
582 Bump __FreeBSD_version to 800092.
585 Add support for hierarchical jails. Remove global securelevel.
586 Bump __FreeBSD_version to 800091.
589 The layout of struct vnet_net has changed, therefore modules
591 Bump __FreeBSD_version to 800090.
594 The newly imported zic(8) produces a new format in the output. Please
595 run tzsetup(8) to install the newly created data to /etc/localtime.
598 The sysctl tree for the usb stack has renamed from hw.usb2.* to
599 hw.usb.* and is now consistent again with previous releases.
602 802.11 monitor mode support was revised and driver api's were changed.
603 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
604 of DLT_IEEE802_11. No user-visible data structures were changed but
605 applications that use DLT_IEEE802_11 may require changes.
606 Bump __FreeBSD_version to 800088.
609 The layout of the following structs has changed: sysctl_oid,
610 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
611 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
612 panics may be experienced. World rebuild is required for
613 correctly checking networking state from userland.
614 Bump __FreeBSD_version to 800085.
617 MLDv2 and Source-Specific Multicast (SSM) have been merged
618 to the IPv6 stack. VIMAGE hooks are in but not yet used.
619 The implementation of SSM within FreeBSD's IPv6 stack closely
620 follows the IPv4 implementation.
622 For kernel developers:
624 * The most important changes are that the ip6_output() and
625 ip6_input() paths no longer take the IN6_MULTI_LOCK,
626 and this lock has been downgraded to a non-recursive mutex.
628 * As with the changes to the IPv4 stack to support SSM, filtering
629 of inbound multicast traffic must now be performed by transport
630 protocols within the IPv6 stack. This does not apply to TCP and
631 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
633 * The KPIs used by IPv6 multicast are similar to those used by
634 the IPv4 stack, with the following differences:
635 * im6o_mc_filter() is analogous to imo_multicast_filter().
636 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
637 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
638 * IN6_LOOKUP_MULTI() has been deprecated and removed.
639 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
640 for MLDv1 have an additional 'timer' argument which is used to
641 jitter the initial membership report for the solicited-node
642 multicast membership on-link.
643 * This is not strictly needed for MLDv2, which already jitters
644 its report transmissions. However, the 'timer' argument is
645 preserved in case MLDv1 is active on the interface.
647 * The KAME linked-list based IPv6 membership implementation has
648 been refactored to use a vector similar to that used by the IPv4
650 Code which maintains a list of its own multicast memberships
651 internally, e.g. carp, has been updated to reflect the new
654 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
655 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
656 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
657 implementation constraint which needs to be addressed in HEAD.
659 For application developers:
661 * The changes are broadly similar to those made for the IPv4
664 * The use of IPv4 and IPv6 multicast socket options on the same
665 socket, using mapped addresses, HAS NOT been tested or supported.
667 * There are a number of issues with the implementation of various
668 IPv6 multicast APIs which need to be resolved in the API surface
669 before the implementation is fully compatible with KAME userland
670 use, and these are mostly to do with interface index treatment.
672 * The literature available discusses the use of either the delta / ASM
673 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
674 using setsourcefilter(3)/getsourcefilter(3). For more information
675 please refer to RFC 3768, 'Socket Interface Extensions for
676 Multicast Source Filters'.
678 * Applications which use the published RFC 3678 APIs should be fine.
680 For systems administrators:
682 * The mtest(8) utility has been refactored to support IPv6, in
683 addition to IPv4. Interface addresses are no longer accepted
684 as arguments, their names must be used instead. The utility
685 will map the interface name to its first IPv4 address as
686 returned by getifaddrs(3).
688 * The ifmcstat(8) utility has also been updated to print the MLDv2
689 endpoint state and source filter lists via sysctl(3).
691 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
692 loopback of IPv6 multicast datagrams by default; it defaults to 1
693 to preserve the existing behaviour. Disabling multicast loopback is
694 recommended for optimal system performance.
696 * The IPv6 MROUTING code has been changed to examine this sysctl
697 instead of attempting to perform a group lookup before looping
698 back forwarded datagrams.
700 Bump __FreeBSD_version to 800084.
703 Implement low-level Bluetooth HCI API.
704 Bump __FreeBSD_version to 800083.
707 The layout of struct malloc_type, used by modules to register new
708 memory allocation types, has changed. Most modules will need to
709 be rebuilt or panics may be experienced.
710 Bump __FreeBSD_version to 800081.
713 Anticipate overflowing inp_flags - add inp_flags2.
714 This changes most offsets in inpcb, so checking v4 connection
715 state will require a world rebuild.
716 Bump __FreeBSD_version to 800080.
719 Add an llentry to struct route and struct route_in6. Modules
720 embedding a struct route will need to be recompiled.
721 Bump __FreeBSD_version to 800079.
724 The size of rt_metrics_lite and by extension rtentry has changed.
725 Networking administration apps will need to be recompiled.
726 The route command now supports show as an alias for get, weighting
727 of routes, sticky and nostick flags to alter the behavior of stateful
729 Bump __FreeBSD_version to 800078.
732 Do not use Giant for kbdmux(4) locking. This is wrong and
733 apparently causing more problems than it solves. This will
734 re-open the issue where interrupt handlers may race with
735 kbdmux(4) in polling mode. Typical symptoms include (but
736 not limited to) duplicated and/or missing characters when
737 low level console functions (such as gets) are used while
738 interrupts are enabled (for example geli password prompt,
739 mountroot prompt etc.). Disabling kbdmux(4) may help.
742 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
743 kernel modules referencing any of the above need to be recompiled.
744 Bump __FreeBSD_version to 800075.
747 GEOM_PART has become the default partition slicer for storage devices,
748 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
749 introduces some changes:
751 MSDOS/EBR: the devices created from MSDOS extended partition entries
752 (EBR) can be named differently than with GEOM_MBR and are now symlinks
753 to devices with offset-based names. fstabs may need to be modified.
755 BSD: the "geometry does not match label" warning is harmless in most
756 cases but it points to problems in file system misalignment with
757 disk geometry. The "c" partition is now implicit, covers the whole
758 top-level drive and cannot be (mis)used by users.
760 General: Kernel dumps are now not allowed to be written to devices
761 whose partition types indicate they are meant to be used for file
762 systems (or, in case of MSDOS partitions, as something else than
765 Most of these changes date approximately from 200812.
768 The uscanner(4) driver has been removed from the kernel. This follows
769 Linux removing theirs in 2.6 and making libusb the default interface
773 The multicast forwarding code has been cleaned up. netstat(1)
774 only relies on KVM now for printing bandwidth upcall meters.
775 The IPv4 and IPv6 modules are split into ip_mroute_mod and
776 ip6_mroute_mod respectively. The config(5) options for statically
777 compiling this code remain the same, i.e. 'options MROUTING'.
780 Support for the IFF_NEEDSGIANT network interface flag has been
781 removed, which means that non-MPSAFE network device drivers are no
782 longer supported. In particular, if_ar, if_sr, and network device
783 drivers from the old (legacy) USB stack can no longer be built or
787 POSIX.1 Native Language Support (NLS) has been enabled in libc and
788 a bunch of new language catalog files have also been added.
789 This means that some common libc messages are now localized and
790 they depend on the LC_MESSAGES environmental variable.
793 The k8temp(4) driver has been renamed to amdtemp(4) since
794 support for Family 10 and Family 11 CPU families was added.
797 IGMPv3 and Source-Specific Multicast (SSM) have been merged
798 to the IPv4 stack. VIMAGE hooks are in but not yet used.
800 For kernel developers, the most important changes are that the
801 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
802 and this lock has been downgraded to a non-recursive mutex.
804 Transport protocols (UDP, Raw IP) are now responsible for filtering
805 inbound multicast traffic according to group membership and source
806 filters. The imo_multicast_filter() KPI exists for this purpose.
807 Transports which do not use multicast (SCTP, TCP) already reject
808 multicast by default. Forwarding and receive performance may improve
809 as a mutex acquisition is no longer needed in the ip_input()
810 low-level input path. in_addmulti() and in_delmulti() are shimmed
811 to new KPIs which exist to support SSM in-kernel.
813 For application developers, it is recommended that loopback of
814 multicast datagrams be disabled for best performance, as this
815 will still cause the lock to be taken for each looped-back
816 datagram transmission. The net.inet.ip.mcast.loop sysctl may
817 be tuned to 0 to disable loopback by default; it defaults to 1
818 to preserve the existing behaviour.
820 For systems administrators, to obtain best performance with
821 multicast reception and multiple groups, it is always recommended
822 that a card with a suitably precise hash filter is used. Hash
823 collisions will still result in the lock being taken within the
824 transport protocol input path to check group membership.
826 If deploying FreeBSD in an environment with IGMP snooping switches,
827 it is recommended that the net.inet.igmp.sendlocal sysctl remain
828 enabled; this forces 224.0.0.0/24 group membership to be announced
831 The size of 'struct igmpstat' has changed; netstat needs to be
832 recompiled to reflect this.
833 Bump __FreeBSD_version to 800070.
836 libusb20.so.1 is now installed as libusb.so.1 and the ports system
837 updated to use it. This requires a buildworld/installworld in order to
838 update the library and dependencies (usbconfig, etc). Its advisable to
839 rebuild all ports which uses libusb. More specific directions are given
840 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
841 libusb are no longer required and can be removed.
844 A workaround is committed to allow the creation of System V shared
845 memory segment of size > 2 GB on the 64-bit architectures.
846 Due to a limitation of the existing ABI, the shm_segsz member
847 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
848 wrong for large segments. Note that limits must be explicitly
849 raised to allow such segments to be created.
852 The layout of struct ifnet has changed, requiring a rebuild of all
853 network device driver modules.
856 The /dev handling for the new USB stack has changed, a
857 buildworld/installworld is required for libusb20.
860 The new USB2 stack has now been permanently moved in and all kernel and
861 module names reverted to their previous values (eg, usb, ehci, ohci,
862 ums, ...). The old usb stack can be compiled in by prefixing the name
863 with the letter 'o', the old usb modules have been removed.
864 Updating entry 20090216 for xorg and 20090215 for libmap may still
868 The rc.conf(5) option if_up_delay has been renamed to
869 defaultroute_delay to better reflect its purpose. If you have
870 customized this setting in /etc/rc.conf you need to update it to
874 xorg 7.4 wants to configure its input devices via hald which does not
875 yet work with USB2. If the keyboard/mouse does not work in xorg then
877 Option "AllowEmptyInput" "off"
878 to your ServerLayout section. This will cause X to use the configured
879 kbd and mouse sections from your xorg.conf.
882 The GENERIC kernels for all architectures now default to the new USB2
883 stack. No kernel config options or code have been removed so if a
884 problem arises please report it and optionally revert to the old USB
885 stack. If you are loading USB kernel modules or have a custom kernel
886 that includes GENERIC then ensure that usb names are also changed over,
887 eg uftdi -> usb2_serial_ftdi.
889 Older programs linked against the ports libusb 0.1 need to be
890 redirected to the new stack's libusb20. /etc/libmap.conf can
892 # Map old usb library to new one for usb2 stack
893 libusb-0.1.so.8 libusb20.so.1
896 All USB ethernet devices now attach as interfaces under the name ueN
897 (eg. ue0). This is to provide a predictable name as vendors often
898 change usb chipsets in a product without notice.
901 The ichsmb(4) driver has been changed to require SMBus slave
902 addresses be left-justified (xxxxxxx0b) rather than right-justified.
903 All of the other SMBus controller drivers require left-justified
904 slave addresses, so this change makes all the drivers provide the
908 INET6 statistics (struct ip6stat) was updated.
909 netstat(1) needs to be recompiled.
912 NTFS has been removed from GENERIC kernel on amd64 to match
913 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
914 will load ntfs.ko module automatically when NTFS support is
915 actually needed, unless ntfs.ko is not installed or security
916 level prohibits loading kernel modules. If either is the case,
917 "options NTFS" has to be added into kernel config.
920 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
921 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
922 800061. User space tools that rely on the size of struct tcpcb in
923 tcp_var.h (e.g. sockstat) need to be recompiled.
926 ng_tty(4) module updated to match the new TTY subsystem.
927 Due to API change, user-level applications must be updated.
928 New API support added to mpd5 CVS and expected to be present
929 in next mpd5.3 release.
932 With __FreeBSD_version 800060 the makefs tool is part of
933 the base system (it was a port).
936 The afdata and ifnet locks have been changed from mutexes to
937 rwlocks, network modules will need to be re-compiled.
940 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
941 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
942 The new code reduced struct rtentry{} by 16 bytes on 32-bit
943 architecture and 40 bytes on 64-bit architecture. The userland
944 applications "arp" and "ndp" have been updated accordingly.
945 The output from "netstat -r" shows only routing entries and
946 none of the L2 information.
949 __FreeBSD_version 800057 marks the switchover from the
950 binary ath hal to source code. Users must add the line:
952 options AH_SUPPORT_AR5416
954 to their kernel config files when specifying:
958 The ath_hal module no longer exists; the code is now compiled
959 together with the driver in the ath module. It is now
960 possible to tailor chip support (i.e. reduce the set of chips
961 and thereby the code size); consult ath_hal(4) for details.
964 __FreeBSD_version 800054 adds memory barriers to
965 <machine/atomic.h>, new interfaces to ifnet to facilitate
966 multiple hardware transmit queues for cards that support
967 them, and a lock-less ring-buffer implementation to
968 enable drivers to more efficiently manage queueing of
972 A new version of ZFS (version 13) has been merged to -HEAD.
973 This version has zpool attribute "listsnapshots" off by
974 default, which means "zfs list" does not show snapshots,
975 and is the same as Solaris behavior.
978 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
981 The uhci, ohci, ehci and slhci USB Host controller drivers have
982 been put into separate modules. If you load the usb module
983 separately through loader.conf you will need to load the
984 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
985 controller add the following to loader.conf:
991 The ABI used by the PMC toolset has changed. Please keep
992 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
996 atapci kernel module now includes only generic PCI ATA
997 driver. AHCI driver moved to ataahci kernel module.
998 All vendor-specific code moved into separate kernel modules:
999 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1000 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1001 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1002 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1005 The TTY subsystem of the kernel has been replaced by a new
1006 implementation, which provides better scalability and an
1007 improved driver model. Most common drivers have been migrated to
1008 the new TTY subsystem, while others have not. The following
1009 drivers have not yet been ported to the new TTY layer:
1012 cy, digi, rc, rp, sio
1018 ng_h4, ng_tty, ppp, sl, snp
1020 Adding these drivers to your kernel configuration file shall
1021 cause compilation to fail.
1024 ntpd has been upgraded to 4.2.4p5.
1027 OpenSSH has been upgraded to 5.1p1.
1029 For many years, FreeBSD's version of OpenSSH preferred DSA
1030 over RSA for host and user authentication keys. With this
1031 upgrade, we've switched to the vendor's default of RSA over
1032 DSA. This may cause upgraded clients to warn about unknown
1033 host keys even for previously known hosts. Users should
1034 follow the usual procedure for verifying host keys before
1035 accepting the RSA key.
1037 This can be circumvented by setting the "HostKeyAlgorithms"
1038 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1041 Please note that the sequence of keys offered for
1042 authentication has been changed as well. You may want to
1043 specify IdentityFile in a different order to revert this
1047 The sio(4) driver has been removed from the i386 and amd64
1048 kernel configuration files. This means uart(4) is now the
1049 default serial port driver on those platforms as well.
1051 To prevent collisions with the sio(4) driver, the uart(4) driver
1052 uses different names for its device nodes. This means the
1053 onboard serial port will now most likely be called "ttyu0"
1054 instead of "ttyd0". You may need to reconfigure applications to
1055 use the new device names.
1057 When using the serial port as a boot console, be sure to update
1058 /boot/device.hints and /etc/ttys before booting the new kernel.
1059 If you forget to do so, you can still manually specify the hints
1060 at the loader prompt:
1062 set hint.uart.0.at="isa"
1063 set hint.uart.0.port="0x3F8"
1064 set hint.uart.0.flags="0x10"
1065 set hint.uart.0.irq="4"
1069 The gpt(8) utility has been removed. Use gpart(8) to partition
1073 The version that Linuxulator emulates was changed from 2.4.2
1074 to 2.6.16. If you experience any problems with Linux binaries
1075 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1076 if it fixes the problem contact emulation mailing list.
1079 ISDN4BSD (I4B) was removed from the src tree. You may need to
1080 update a your kernel configuration and remove relevant entries.
1083 I have checked in code to support multiple routing tables.
1084 See the man pages setfib(1) and setfib(2).
1085 This is a hopefully backwards compatible version,
1086 but to make use of it you need to compile your kernel
1087 with options ROUTETABLES=2 (or more up to 16).
1090 The 802.11 wireless support was redone to enable multi-bss
1091 operation on devices that are capable. The underlying device
1092 is no longer used directly but instead wlanX devices are
1093 cloned with ifconfig. This requires changes to rc.conf files.
1094 For example, change:
1095 ifconfig_ath0="WPA DHCP"
1098 ifconfig_wlan0="WPA DHCP"
1099 see rc.conf(5) for more details. In addition, mergemaster of
1100 /etc/rc.d is highly recommended. Simultaneous update of userland
1101 and kernel wouldn't hurt either.
1103 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1104 modules were merged into the base wlan module. All references
1105 to these modules (e.g. in kernel config files) must be removed.
1108 psm(4) has gained write(2) support in native operation level.
1109 Arbitrary commands can be written to /dev/psm%d and status can
1110 be read back from it. Therefore, an application is responsible
1111 for status validation and error recovery. It is a no-op in
1112 other operation levels.
1115 Support for KSE threading has been removed from the kernel. To
1116 run legacy applications linked against KSE libmap.conf may
1117 be used. The following libmap.conf may be used to ensure
1118 compatibility with any prior release:
1120 libpthread.so.1 libthr.so.1
1121 libpthread.so.2 libthr.so.2
1122 libkse.so.3 libthr.so.3
1125 The layout of struct vmspace has changed. This affects libkvm
1126 and any executables that link against libkvm and use the
1127 kvm_getprocs() function. In particular, but not exclusively,
1128 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1129 The effects are minimal, but it's advisable to upgrade world
1133 The latest em driver no longer has support in it for the
1134 82575 adapter, this is now moved to the igb driver. The
1135 split was done to make new features that are incompatible
1136 with older hardware easier to do.
1139 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1140 likewise the kernel option is now GEOM_LINUX_LVM.
1143 The default NFS mount mode has changed from UDP to TCP for
1144 increased reliability. If you rely on (insecurely) NFS
1145 mounting across a firewall you may need to update your
1149 Belatedly note the addition of m_collapse for compacting
1153 The fts(3) structures have been changed to use adequate
1154 integer types for their members and so to be able to cope
1155 with huge file trees. The old fts(3) ABI is preserved
1156 through symbol versioning in libc, so third-party binaries
1157 using fts(3) should still work, although they will not take
1158 advantage of the extended types. At the same time, some
1159 third-party software might fail to build after this change
1160 due to unportable assumptions made in its source code about
1161 fts(3) structure members. Such software should be fixed
1162 by its vendor or, in the worst case, in the ports tree.
1163 FreeBSD_version 800015 marks this change for the unlikely
1164 case that a portable fix is impossible.
1167 To upgrade to -current after this date, you must be running
1168 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1169 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1172 The ADAPTIVE_GIANT kernel option has been retired because its
1173 functionality is the default now.
1176 The AT keyboard emulation of sunkbd(4) has been turned on
1177 by default. In order to make the special symbols of the Sun
1178 keyboards driven by sunkbd(4) work under X these now have
1179 to be configured the same way as Sun USB keyboards driven
1180 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1182 Option "XkbLayout" "us"
1183 Option "XkbRules" "xorg"
1184 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1187 It has been decided that it is desirable to provide ABI
1188 backwards compatibility to the FreeBSD 4/5/6 versions of the
1189 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1190 broken with the introduction of PCI domain support (see the
1191 20070930 entry). Unfortunately, this required the ABI of
1192 PCIOCGETCONF to be broken again in order to be able to
1193 provide backwards compatibility to the old version of that
1194 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1195 again. As for prominent ports this affects neither pciutils
1196 nor xorg-server this time, the hal port needs to be rebuilt
1200 The misnamed kthread_create() and friends have been renamed
1201 to kproc_create() etc. Many of the callers already
1202 used kproc_start()..
1203 I will return kthread_create() and friends in a while
1204 with implementations that actually create threads, not procs.
1205 Renaming corresponds with version 800002.
1214 Avoid using make -j when upgrading. While generally safe, there are
1215 sometimes problems using -j to upgrade. If your upgrade fails with
1216 -j, please try again without -j. From time to time in the past there
1217 have been problems using -j with buildworld and/or installworld. This
1218 is especially true when upgrading between "distant" versions (eg one
1219 that cross a major release boundary or several minor releases, or when
1220 several months have passed on the -current branch).
1222 Sometimes, obscure build problems are the result of environment
1223 poisoning. This can happen because the make utility reads its
1224 environment when searching for values for global variables. To run
1225 your build attempts in an "environmental clean room", prefix all make
1226 commands with 'env -i '. See the env(1) manual page for more details.
1228 When upgrading from one major version to another it is generally best
1229 to upgrade to the latest code in the currently installed branch first,
1230 then do an upgrade to the new branch. This is the best-tested upgrade
1231 path, and has the highest probability of being successful. Please try
1232 this approach before reporting problems with a major version upgrade.
1236 When upgrading the boot ZFS pool to a new version, always follow
1239 1.) recompile and reinstall the ZFS boot loader and boot block
1240 (this is part of "make buildworld" and "make installworld")
1242 2.) update the ZFS boot block on your boot drive
1244 The following example updates the ZFS boot block on the first
1245 partition (freebsd-boot) of a GPT partitioned drive ad0:
1246 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1248 Non-boot pools do not need these updates.
1252 If you are updating from a prior version of FreeBSD (even one just
1253 a few days old), you should follow this procedure. It is the most
1254 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1256 make kernel-toolchain
1257 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1258 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1260 To test a kernel once
1261 ---------------------
1262 If you just want to boot a kernel once (because you are not sure
1263 if it works, or if you want to boot a known bad kernel to provide
1264 debugging information) run
1265 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1266 nextboot -k testkernel
1268 To just build a kernel when you know that it won't mess you up
1269 --------------------------------------------------------------
1270 This assumes you are already running a CURRENT system. Replace
1271 ${arch} with the architecture of your machine (e.g. "i386",
1272 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1274 cd src/sys/${arch}/conf
1275 config KERNEL_NAME_HERE
1276 cd ../compile/KERNEL_NAME_HERE
1281 If this fails, go to the "To build a kernel" section.
1283 To rebuild everything and install it on the current system.
1284 -----------------------------------------------------------
1285 # Note: sometimes if you are running current you gotta do more than
1286 # is listed here if you are upgrading from a really old current.
1288 <make sure you have good level 0 dumps>
1290 make kernel KERNCONF=YOUR_KERNEL_HERE
1292 <reboot in single user> [3]
1300 To cross-install current onto a separate partition
1301 --------------------------------------------------
1302 # In this approach we use a separate partition to hold
1303 # current's root, 'usr', and 'var' directories. A partition
1304 # holding "/", "/usr" and "/var" should be about 2GB in
1307 <make sure you have good level 0 dumps>
1310 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1311 <maybe newfs current's root partition>
1312 <mount current's root partition on directory ${CURRENT_ROOT}>
1313 make installworld DESTDIR=${CURRENT_ROOT}
1314 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1315 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1316 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1317 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1318 <reboot into current>
1319 <do a "native" rebuild/install as described in the previous section>
1320 <maybe install compatibility libraries from ports/misc/compat*>
1324 To upgrade in-place from 8.x-stable to current
1325 ----------------------------------------------
1326 <make sure you have good level 0 dumps>
1328 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1330 <reboot in single user> [3]
1337 Make sure that you've read the UPDATING file to understand the
1338 tweaks to various things you need. At this point in the life
1339 cycle of current, things change often and you are on your own
1340 to cope. The defaults can also change, so please read ALL of
1341 the UPDATING entries.
1343 Also, if you are tracking -current, you must be subscribed to
1344 freebsd-current@freebsd.org. Make sure that before you update
1345 your sources that you have read and understood all the recent
1346 messages there. If in doubt, please track -stable which has
1347 much fewer pitfalls.
1349 [1] If you have third party modules, such as vmware, you
1350 should disable them at this point so they don't crash your
1353 [3] From the bootblocks, boot -s, and then do
1358 adjkerntz -i # if CMOS is wall time
1359 Also, when doing a major release upgrade, it is required that
1360 you boot into single user mode to do the installworld.
1362 [4] Note: This step is non-optional. Failure to do this step
1363 can result in a significant reduction in the functionality of the
1364 system. Attempting to do it by hand is not recommended and those
1365 that pursue this avenue should read this file carefully, as well
1366 as the archives of freebsd-current and freebsd-hackers mailing lists
1367 for potential gotchas. The -U option is also useful to consider.
1368 See mergemaster(8) for more information.
1370 [5] Usually this step is a noop. However, from time to time
1371 you may need to do this if you get unknown user in the following
1372 step. It never hurts to do it all the time. You may need to
1373 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1374 install) after the buildworld before this step if you last updated
1375 from current before 20020224 or from -stable before 20020408.
1377 [6] This only deletes old files and directories. Old libraries
1378 can be deleted by "make delete-old-libs", but you have to make
1379 sure that no program is using those libraries anymore.
1381 [8] In order to have a kernel that can run the 4.x binaries needed to
1382 do an installworld, you must include the COMPAT_FREEBSD4 option in
1383 your kernel. Failure to do so may leave you with a system that is
1384 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1385 required to run the 5.x binaries on more recent kernels. And so on
1386 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1388 Make sure that you merge any new devices from GENERIC since the
1389 last time you updated your kernel config file.
1391 [9] When checking out sources, you must include the -P flag to have
1392 cvs prune empty directories.
1394 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1395 "?=" instead of the "=" assignment operator, so that buildworld can
1396 override the CPUTYPE if it needs to.
1398 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1399 not on the command line, or in /etc/make.conf. buildworld will
1400 warn if it is improperly defined.
1403 This file contains a list, in reverse chronological order, of major
1404 breakages in tracking -current. Not all things will be listed here,
1405 and it only starts on October 16, 2004. Updating files can found in
1406 previous releases if your system is older than this.
1408 Copyright information:
1410 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1412 Redistribution, publication, translation and use, with or without
1413 modification, in full or in part, in any form or format of this
1414 document are permitted without further permission from the author.
1416 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1417 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1418 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1419 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1420 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1421 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1422 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1423 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1424 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1425 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1426 POSSIBILITY OF SUCH DAMAGE.
1428 Contact Warner Losh if you have any questions about your use of