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.
14 The random(4) support for the VIA hardware random number
15 generator (`PADLOCK') is no longer enabled unconditionally.
16 Add the PADLOCK_RNG option in the custom kernel config if
17 needed. The GENERIC kernels on i386 and amd64 do include the
18 option, so the change only affects the custom kernel
22 The sparc64 ZFS loader has been changed to no longer try to auto-
23 detect ZFS providers based on diskN aliases but now requires these
24 to be explicitly listed in the OFW boot-device environment variable.
27 Now unix domain sockets behave "as expected" on nullfs(5). Previously
28 nullfs(5) did not pass through all behaviours to the underlying layer,
29 as a result if we bound to a socket on the lower layer we could connect
30 only to the lower path; if we bound to the upper layer we could connect
31 only to the upper path. The new behavior is one can connect to both the
32 lower and the upper paths regardless what layer path one binds to.
35 The acpi_wmi(4) status device /dev/wmistat has been renamed to
39 A new VOP_ADVISE() was added to support posix_fadvise(2). All
40 filesystem modules must be recompiled.
43 The interface of the VOP_VPTOCNP(9) changed, now the returned
44 vnode shall be referenced, previously it was required to be
45 only held. All in-tree filesystems are converted.
51 The broken amd(4) driver has been replaced with esp(4) in the amd64,
52 i386 and pc98 GENERIC kernel configuration files.
55 This commit modifies vfs_register() so that it uses a hash
56 calculation to set vfc_typenum, which is enabled by default.
57 The first time a system is booted after this change, the
58 vfc_typenum values will change for all file systems. The
59 main effect of this is a change to the NFS server file handles
60 for file systems that use vfc_typenum in their fsid, such as ZFS.
61 It will, however, prevent vfc_typenum from changing when file
62 systems are loaded in a different order for subsequent reboots.
63 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
64 until you are ready to remount all NFS clients after a reboot.
67 Bump the shared library version numbers for libraries that
68 do not use symbol versioning, have changed the ABI compared
69 to stable/8 and which shared library version was not bumped.
70 Done as part of 9.0-RELEASE cycle.
73 During the merge of Capsicum features, the fget(9) KPI was modified.
74 This may require the rebuilding of out-of-tree device drivers --
75 issues have been reported specifically with the nVidia device driver.
76 __FreeBSD_version is bumped to 900041.
78 Also, there is a period between 20110811 and 20110814 where the
79 special devices /dev/{stdin,stdout,stderr} did not work correctly.
80 Building world from a kernel during that window may not work.
83 The packet filter (pf) code has been updated to OpenBSD 4.5.
84 You need to update userland tools to be in sync with kernel.
85 This update breaks backward compatibility with earlier pfsync(4)
86 versions. Care must be taken when updating redundant firewall setups.
89 The following sysctls and tunables are retired on x86 platforms:
91 machdep.hlt_logical_cpus
92 The following sysctl is retired:
93 machdep.hyperthreading_allowed
94 The sysctls were supposed to provide a way to dynamically offline and
95 online selected CPUs on x86 platforms, but the implementation has not
96 been reliable especially with SCHED_ULE scheduler.
97 machdep.hyperthreading_allowed tunable is still available to ignore
98 hyperthreading CPUs at OS level.
99 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
100 where X is an APIC ID of a CPU. Be advised, though, that disabling
101 CPUs in non-uniform fashion will result in non-uniform topology and
102 may lead to sub-optimal system performance with SCHED_ULE, which is
106 cpumask_t type is retired and cpuset_t is used in order to describe
110 Changes to ifconfig(8) for dynamic address family detection mandate
111 that you are running a kernel of 20110525 or later. Make sure to
112 follow the update procedure to boot a new kernel before installing
116 Support for sun4v architecture is officially dropped
119 Several KPI breaking changes have been committed to the mii(4) layer,
120 the PHY drivers and consequently some Ethernet drivers using mii(4).
121 This means that miibus.ko and the modules of the affected Ethernet
122 drivers need to be recompiled.
124 Note to kernel developers: Given that the OUI bit reversion problem
125 was fixed as part of these changes all mii(4) commits related to OUIs,
126 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
127 handling, no longer can be merged verbatim to stable/8 and previous
131 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
132 into their kernel configurations along with 'device pci'.
135 The default NFS client is now the new NFS client, so fstype "newnfs"
136 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
137 Although mounts via fstype "nfs" will usually work without userland
138 changes, it is recommended that the mount(8) and mount_nfs(8)
139 commands be rebuilt from sources and that a link to mount_nfs called
140 mount_oldnfs be created. The new client is compiled into the
141 kernel with "options NFSCL" and this is needed for diskless root
142 file systems. The GENERIC kernel configs have been changed to use
143 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
144 To use the regular/old client, you can "mount -t oldnfs ...". For
145 a diskless root file system, you must also include a line like:
147 vfs.root.mountfrom="oldnfs:"
149 in the boot/loader.conf on the root fs on the NFS server to make
150 a diskless root fs use the old client.
153 The GENERIC kernels for all architectures now default to the new
154 CAM-based ATA stack. It means that all legacy ATA drivers were
155 removed and replaced by respective CAM drivers. If you are using
156 ATA device names in /etc/fstab or other places, make sure to update
157 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
158 where 'Y's are the sequential numbers starting from zero for each type
159 in order of detection, unless configured otherwise with tunables,
160 see cam(4)). There will be symbolic links created in /dev/ to map
161 old adX devices to the respective adaY. They should provide basic
162 compatibility for file systems mounting in most cases, but they do
163 not support old user-level APIs and do not have respective providers
164 in GEOM. Consider using updated management tools with new device names.
166 It is possible to load devices ahci, ata, siis and mvs as modules,
167 but option ATA_CAM should remain in kernel configuration to make ata
168 module work as CAM driver supporting legacy ATA controllers. Device ata
169 still can be used in modular fashion (atacore + ...). Modules atadisk
170 and atapi* are not used and won't affect operation in ATA_CAM mode.
171 Note that to use CAM-based ATA kernel should include CAM devices
172 scbus, pass, da (or explicitly ada), cd and optionally others. All of
173 them are parts of the cam module.
175 ataraid(4) functionality is now supported by the RAID GEOM class.
176 To use it you can load geom_raid kernel module and use graid(8) tool
177 for management. Instead of /dev/arX device names, use /dev/raid/rX.
179 No kernel config options or code have been removed, so if a problem
180 arises, please report it and optionally revert to the old ATA stack.
181 In order to do it you can remove from the kernel config:
186 , and instead add back:
187 device atadisk # ATA disk drives
188 device ataraid # ATA RAID drives
189 device atapicd # ATAPI CDROM drives
190 device atapifd # ATAPI floppy drives
191 device atapist # ATAPI tape drives
194 The default NFS server has been changed to the new server, which
195 was referred to as the experimental server. If you need to switch
196 back to the old NFS server, you must now put the "-o" option on
197 both the mountd and nfsd commands. This can be done using the
198 mountd_flags and nfs_server_flags rc.conf variables until an
199 update to the rc scripts is committed, which is coming soon.
202 The GNU Objective-C runtime library (libobjc), and other Objective-C
203 related components have been removed from the base system. If you
204 require an Objective-C library, please use one of the available ports.
207 ath(4) has been split into bus- and device- modules. if_ath contains
208 the HAL, the TX rate control and the network device code. if_ath_pci
209 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
210 contains the AHB glue. Users need to load both if_ath_pci and if_ath
211 in order to use ath on everything else.
213 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
214 need to load if_ath and if_ath_pci for ath(4) operation.
217 As part of the replacement of sysinstall, the process of building
218 release media has changed significantly. For details, please re-read
219 release(7), which has been updated to reflect the new build process.
222 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
223 is the last available version under GPLv2. It brings a number of new
224 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
225 4.1 and SSE 4.2), better support for powerpc64, a number of new
226 directives, and lots of other small improvements. See the ChangeLog
227 file in contrib/binutils for the full details.
230 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
231 compliant, and will now use half of hash for authentication.
232 This will break interoperability with all stacks (including all
233 actual FreeBSD versions) who implement
234 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
236 The only workaround with such peers is to use another HMAC
237 algorithm for IPsec ("phase 2") authentication.
240 Remove the uio_yield prototype and symbol. This function has
241 been misnamed since it was introduced and should not be
242 globally exposed with this name. The equivalent functionality
243 is now available using kern_yield(curthread->td_user_pri).
244 The function remains undocumented.
247 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
248 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
249 for scalar sysctls is defined but disabled. Code that needs
250 UQUAD to pass the type checking that must compile on older
251 systems where the define is not present can check against
252 __FreeBSD_version >= 900030.
254 The system dialog(1) has been replaced with a new version previously
255 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
256 with the previous version, but the libdialog associated with it has
257 a largely incompatible API. As such, the original version of libdialog
258 will be kept temporarily as libodialog, until its base system consumers
259 are replaced or updated. Bump __FreeBSD_version to 900030.
262 If you are trying to run make universe on a -stable system, and you get
263 the following warning:
264 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
265 unknown. config(8) likely too old."
266 or something similar to it, then you must upgrade your -stable system
267 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
268 stable/8) or build the config from the latest stable/8 branch and
269 install it on your system.
271 Prior to this date, building a current universe on 8-stable system from
272 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
273 error in the first kernel build phase. A new config on those old
274 systems will fix that problem for older versions of -current.
277 The TCP stack has been modified to allow Khelp modules to interact with
278 it via helper hook points and store per-connection data in the TCP
279 control block. Bump __FreeBSD_version to 900029. User space tools that
280 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
284 Generic IEEE 802.3 annex 31B full duplex flow control support has been
285 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
286 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
287 to take advantage of it instead of using custom implementations. This
288 means that these drivers now no longer unconditionally advertise
289 support for flow control but only do so if flow control is a selected
290 media option. This was implemented in the generic support that way in
291 order to allow flow control to be switched on and off via ifconfig(8)
292 with the PHY specific default to typically off in order to protect
293 from unwanted effects. Consequently, if you used flow control with
294 one of the above mentioned drivers you now need to explicitly enable
296 ifconfig bge0 media auto mediaopt flowcontrol
298 Along with the above mentioned changes generic support for setting
299 1000baseT master mode also has been added and brgphy(4), ciphy(4),
300 e1000phy(4) as well as ip1000phy(4) have been converted to take
301 advantage of it. This means that these drivers now no longer take the
302 link0 parameter for selecting master mode but the master media option
303 has to be used instead, for example like in the following:
304 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
306 Selection of master mode now is also available with all other PHY
307 drivers supporting 1000baseT.
310 The TCP stack has received a significant update to add support for
311 modularised congestion control and generally improve the clarity of
312 congestion control decisions. Bump __FreeBSD_version to 900025. User
313 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
314 sockstat) need to be recompiled.
317 The man(1) utility has been replaced by a new version that no longer
318 uses /etc/manpath.config. Please consult man.conf(5) for how to
319 migrate local entries to the new format.
322 The copyright strings printed by login(1) and sshd(8) at the time of a
323 new connection have been removed to follow other operating systems and
327 A workaround for a fixed ld bug has been removed in kernel code,
328 so make sure that your system ld is built from sources after
329 revision 210245 from 2010-07-19 (r211583 if building head kernel
330 on stable/8, r211584 for stable/7; both from 2010-08-21).
331 A symptom of incorrect ld version is different addresses for
332 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
335 The $ipv6_prefer variable in rc.conf(5) has been split into
336 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
338 The $ip6addrctl_policy is a variable to choose a pre-defined
339 address selection policy set by ip6addrctl(8). A value
340 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
343 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
344 flag (see an entry of 20090926) is set on an interface with no
345 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
346 security reason. If you want IPv6 link-local address on all
347 interfaces by default, set this to "YES".
349 The old ipv6_prefer="YES" is equivalent to
350 ipv6_activate_all_interfaces="YES" and
351 ip6addrctl_policy="ipv6_prefer".
354 DTrace has grown support for userland tracing. Due to this, DTrace is
355 now i386 and amd64 only.
356 dtruss(1) is now installed by default on those systems and a new
357 kernel module is needed for userland tracing: fasttrap.
358 No changes to your kernel config file are necessary to enable
359 userland tracing, but you might consider adding 'STRIP=' and
360 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
361 to have informative userland stack traces in DTrace (ustack).
364 The acpi_aiboost(4) driver has been removed in favor of the new
365 aibs(4) driver. You should update your kernel configuration file.
368 BSD grep has been imported to the base system and it is built by
369 default. It is completely BSD licensed, highly GNU-compatible, uses
370 less memory than its GNU counterpart and has a small codebase.
371 However, it is slower than its GNU counterpart, which is mostly
372 noticeable for larger searches, for smaller ones it is measurable
373 but not significant. The reason is complex, the most important factor
374 is that we lack a modern and efficient regex library and GNU
375 overcomes this by optimizing the searches internally. Future work
376 on improving the regex performance is planned, for the meantime,
377 users that need better performance, can build GNU grep instead by
378 setting the WITH_GNU_GREP knob.
381 Due to the import of powerpc64 support, all existing powerpc kernel
382 configuration files must be updated with a machine directive like this:
383 machine powerpc powerpc
385 In addition, an updated config(8) is required to build powerpc kernels
389 A new version of ZFS (version 15) has been merged to -HEAD.
390 This version uses a python library for the following subcommands:
391 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
392 For full functionality of these commands the following port must
393 be installed: sysutils/py-zfs
396 'vm_page's are now hashed by physical address to an array of mutexes.
397 Currently this is only used to serialize access to hold_count. Over
398 time the page queue mutex will be peeled away. This changes the size
399 of pmap on every architecture. And requires all callers of vm_page_hold
400 and vm_page_unhold to be updated.
403 WITH_CTF can now be specified in src.conf (not recommended, there
404 are some problems with static executables), make.conf (would also
405 affect ports which do not use GNU make and do not override the
406 compile targets) or in the kernel config (via "makeoptions
408 When WITH_CTF was specified there before this was silently ignored,
409 so make sure that WITH_CTF is not used in places which could lead
410 to unwanted behavior.
413 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
414 to allow 32-bit compatibility on non-x86 platforms. All kernel
415 configurations on amd64 and ia64 platforms using these options must
416 be modified accordingly.
419 The utmp user accounting database has been replaced with utmpx,
420 the user accounting interface standardized by POSIX.
421 Unfortunately the semantics of utmp and utmpx don't match,
422 making it practically impossible to support both interfaces.
423 The user accounting database is used by tools like finger(1),
424 last(1), talk(1), w(1) and ac(8).
426 All applications in the base system use utmpx. This means only
427 local binaries (e.g. from the ports tree) may still use these
428 utmp database files. These applications must be rebuilt to make
431 After the system has been upgraded, it is safe to remove the old
432 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
433 assuming their contents is of no importance anymore. Old wtmp
434 databases can only be used by last(1) and ac(8) after they have
435 been converted to the new format using wtmpcvt(1).
438 Introduce the kernel thread "deadlock resolver" (which can be enabled
439 via the DEADLKRES option, see NOTES for more details) and the
440 sleepq_type() function for sleepqueues.
443 The rc.firewall and rc.firewall6 were unified, and
444 rc.firewall6 and rc.d/ip6fw were removed.
445 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
446 variables are obsoleted. Instead, the following new rc
447 variables are added to rc.d/ipfw:
449 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
450 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
451 firewall_simple_onet_ipv6, firewall_trusted_ipv6
453 The meanings correspond to the relevant IPv4 variables.
459 The default terminal emulation for syscons(4) has been changed
460 from cons25 to xterm on all platforms except pc98. This means
461 that the /etc/ttys file needs to be updated to ensure correct
462 operation of applications on the console.
464 The terminal emulation style can be toggled per window by using
465 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
466 options can be used to change the compile-time default back to
469 To prevent graphical artifacts, make sure the TERM environment
470 variable is set to match the terminal emulation that is being
471 performed by syscons(4).
474 The layout of the structure ieee80211req_scan_result has changed.
475 Applications that require wireless scan results (e.g. ifconfig(8))
476 from net80211 need to be recompiled.
478 Applications such as wpa_supplicant(8) may require a full world
479 build without using NO_CLEAN in order to get synchronized with the
483 The iwn(4) driver has been updated to support the 5000 and 5150 series.
484 There's one kernel module for each firmware. Adding "device iwnfw"
485 to the kernel configuration file means including all three firmware
486 images inside the kernel. If you want to include just the one for
487 your wireless card, use the devices iwn4965fw, iwn5000fw or
491 The rc.d/network_ipv6, IPv6 configuration script has been integrated
492 into rc.d/netif. The changes are the following:
494 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
495 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
496 Note that both variables need the "inet6" keyword at the head.
498 Do not set $ipv6_network_interfaces manually if you do not
499 understand what you are doing. It is not needed in most cases.
501 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
504 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
505 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
507 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
508 all configured interfaces have "inet6 accept_rtadv" in the
509 $ifconfig_IF_ipv6. These are for backward compatibility.
511 3. A new variable $ipv6_prefer has been added. If NO, IPv6
512 functionality of interfaces with no corresponding
513 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
514 and the default address selection policy of ip6addrctl(8)
515 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
516 Note that if you want to configure IPv6 functionality on the
517 disabled interfaces after boot, first you need to clear the flag by
518 using ifconfig(8) like:
520 ifconfig em0 inet6 -ifdisabled
522 If YES, the default address selection policy is set as
525 The default value of $ipv6_prefer is NO.
527 4. If your system need to receive Router Advertisement messages,
528 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
529 scripts automatically invoke rtsol(8) when the interface becomes
530 UP. The Router Advertisement messages are used for SLAAC
531 (State-Less Address AutoConfiguration).
534 802.11s D3.03 support was committed. This is incompatible with the
535 previous code, which was based on D3.0.
538 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
539 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
540 control whether accepting Router Advertisement messages or not.
541 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
542 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
543 The ifconfig(8) utility now supports these flags.
546 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
547 mount(8) and -a option for df(1) to see them.
550 The old tunable hw.bus.devctl_disable has been superseded by
551 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
552 replaced by hw.bus.devctl_queue=0. The default for this new tunable
556 Remove the option STOP_NMI. The default action is now to use NMI only
557 for KDB via the newly introduced function stop_cpus_hard() and
558 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
561 The stable/8 branch created in subversion. This corresponds to the
562 RELENG_8 branch in CVS.
565 Bump the shared library version numbers for all libraries that do not
566 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
567 __FreeBSD_version to 800105.
570 Due to changes in the implementation of virtual network stack support,
571 all network-related kernel modules must be recompiled. As this change
572 breaks the ABI, bump __FreeBSD_version to 800104.
575 The TOE interface to the TCP syncache has been modified to remove
576 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
577 The cxgb driver is the only TOE consumer affected by this change, and
578 needs to be recompiled along with the kernel. As this change breaks
579 the ABI, bump __FreeBSD_version to 800103.
582 Padding has been added to struct tcpcb, sackhint and tcpstat in
583 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
584 maintaining the ABI. However, this change breaks the ABI, so bump
585 __FreeBSD_version to 800102. User space tools that rely on the size of
586 any of these structs (e.g. sockstat) need to be recompiled.
589 The NFS_LEGACYRPC option has been removed along with the old kernel
590 RPC implementation that this option selected. Kernel configurations
591 may need to be adjusted.
594 The network interface device nodes at /dev/net/<interface> have been
595 removed. All ioctl operations can be performed the normal way using
596 routing sockets. The kqueue functionality can generally be replaced
597 with routing sockets.
600 The documentation from the FreeBSD Documentation Project (Handbook,
601 FAQ, etc.) is now installed via packages by sysinstall(8) and under
602 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
605 The ABI of various structures related to the SYSV IPC API have been
606 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
607 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
611 Layout of struct vnet has changed as routing related variables were
612 moved to their own Vimage module. Modules need to be recompiled. Bump
613 __FreeBSD_version to 800099.
616 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
617 respectively. As long as no more than 16 groups per process are used,
618 no changes should be visible. When more than 16 groups are used, old
619 binaries may fail if they call getgroups() or getgrouplist() with
620 statically sized storage. Recompiling will work around this, but
621 applications should be modified to use dynamically allocated storage
622 for group arrays as POSIX.1-2008 does not cap an implementation's
623 number of supported groups at NGROUPS_MAX+1 as previous versions did.
625 NFS and portalfs mounts may also be affected as the list of groups is
626 truncated to 16. Users of NFS who use more than 16 groups, should
627 take care that negative group permissions are not used on the exported
628 file systems as they will not be reliable unless a GSSAPI based
629 authentication method is used.
632 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
633 option compiles in the support for adaptive spinning for lockmgrs
634 which want to enable it. The lockinit() function now accepts the flag
635 LK_ADAPTIVE in order to make the lock object subject to adaptive
636 spinning when both held in write and read mode.
639 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
640 changed. User applications that use this ioctl need to be rebuilt.
643 The layout of struct thread has changed. Kernel and modules need to
647 The layout of structs ifnet, domain, protosw and vnet_net has changed.
648 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
651 window(1) has been removed from the base system. It can now be
652 installed from ports. The port is called misc/window.
655 The way we are storing and accessing `routing table' entries has
656 changed. Programs reading the FIB, like netstat, need to be
660 A new netisr implementation has been added for FreeBSD 8. Network
661 file system modules, such as igmp, ipdivert, and others, should be
663 Bump __FreeBSD_version to 800096.
666 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
670 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
671 Bump __FreeBSD_version to 800094.
674 Add mnt_xflag field to 'struct mount'. File system modules need to be
676 Bump __FreeBSD_version to 800093.
679 The compiling option ADAPTIVE_SX has been retired while it has been
680 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
681 The KPI for sx_init_flags() changes as accepting flags:
682 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
683 been introduced in order to handle the reversed logic.
684 Bump __FreeBSD_version to 800092.
687 Add support for hierarchical jails. Remove global securelevel.
688 Bump __FreeBSD_version to 800091.
691 The layout of struct vnet_net has changed, therefore modules
693 Bump __FreeBSD_version to 800090.
696 The newly imported zic(8) produces a new format in the output. Please
697 run tzsetup(8) to install the newly created data to /etc/localtime.
700 The sysctl tree for the usb stack has renamed from hw.usb2.* to
701 hw.usb.* and is now consistent again with previous releases.
704 802.11 monitor mode support was revised and driver api's were changed.
705 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
706 of DLT_IEEE802_11. No user-visible data structures were changed but
707 applications that use DLT_IEEE802_11 may require changes.
708 Bump __FreeBSD_version to 800088.
711 The layout of the following structs has changed: sysctl_oid,
712 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
713 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
714 panics may be experienced. World rebuild is required for
715 correctly checking networking state from userland.
716 Bump __FreeBSD_version to 800085.
719 MLDv2 and Source-Specific Multicast (SSM) have been merged
720 to the IPv6 stack. VIMAGE hooks are in but not yet used.
721 The implementation of SSM within FreeBSD's IPv6 stack closely
722 follows the IPv4 implementation.
724 For kernel developers:
726 * The most important changes are that the ip6_output() and
727 ip6_input() paths no longer take the IN6_MULTI_LOCK,
728 and this lock has been downgraded to a non-recursive mutex.
730 * As with the changes to the IPv4 stack to support SSM, filtering
731 of inbound multicast traffic must now be performed by transport
732 protocols within the IPv6 stack. This does not apply to TCP and
733 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
735 * The KPIs used by IPv6 multicast are similar to those used by
736 the IPv4 stack, with the following differences:
737 * im6o_mc_filter() is analogous to imo_multicast_filter().
738 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
739 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
740 * IN6_LOOKUP_MULTI() has been deprecated and removed.
741 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
742 for MLDv1 have an additional 'timer' argument which is used to
743 jitter the initial membership report for the solicited-node
744 multicast membership on-link.
745 * This is not strictly needed for MLDv2, which already jitters
746 its report transmissions. However, the 'timer' argument is
747 preserved in case MLDv1 is active on the interface.
749 * The KAME linked-list based IPv6 membership implementation has
750 been refactored to use a vector similar to that used by the IPv4
752 Code which maintains a list of its own multicast memberships
753 internally, e.g. carp, has been updated to reflect the new
756 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
757 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
758 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
759 implementation constraint which needs to be addressed in HEAD.
761 For application developers:
763 * The changes are broadly similar to those made for the IPv4
766 * The use of IPv4 and IPv6 multicast socket options on the same
767 socket, using mapped addresses, HAS NOT been tested or supported.
769 * There are a number of issues with the implementation of various
770 IPv6 multicast APIs which need to be resolved in the API surface
771 before the implementation is fully compatible with KAME userland
772 use, and these are mostly to do with interface index treatment.
774 * The literature available discusses the use of either the delta / ASM
775 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
776 using setsourcefilter(3)/getsourcefilter(3). For more information
777 please refer to RFC 3768, 'Socket Interface Extensions for
778 Multicast Source Filters'.
780 * Applications which use the published RFC 3678 APIs should be fine.
782 For systems administrators:
784 * The mtest(8) utility has been refactored to support IPv6, in
785 addition to IPv4. Interface addresses are no longer accepted
786 as arguments, their names must be used instead. The utility
787 will map the interface name to its first IPv4 address as
788 returned by getifaddrs(3).
790 * The ifmcstat(8) utility has also been updated to print the MLDv2
791 endpoint state and source filter lists via sysctl(3).
793 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
794 loopback of IPv6 multicast datagrams by default; it defaults to 1
795 to preserve the existing behaviour. Disabling multicast loopback is
796 recommended for optimal system performance.
798 * The IPv6 MROUTING code has been changed to examine this sysctl
799 instead of attempting to perform a group lookup before looping
800 back forwarded datagrams.
802 Bump __FreeBSD_version to 800084.
805 Implement low-level Bluetooth HCI API.
806 Bump __FreeBSD_version to 800083.
809 The layout of struct malloc_type, used by modules to register new
810 memory allocation types, has changed. Most modules will need to
811 be rebuilt or panics may be experienced.
812 Bump __FreeBSD_version to 800081.
815 Anticipate overflowing inp_flags - add inp_flags2.
816 This changes most offsets in inpcb, so checking v4 connection
817 state will require a world rebuild.
818 Bump __FreeBSD_version to 800080.
821 Add an llentry to struct route and struct route_in6. Modules
822 embedding a struct route will need to be recompiled.
823 Bump __FreeBSD_version to 800079.
826 The size of rt_metrics_lite and by extension rtentry has changed.
827 Networking administration apps will need to be recompiled.
828 The route command now supports show as an alias for get, weighting
829 of routes, sticky and nostick flags to alter the behavior of stateful
831 Bump __FreeBSD_version to 800078.
834 Do not use Giant for kbdmux(4) locking. This is wrong and
835 apparently causing more problems than it solves. This will
836 re-open the issue where interrupt handlers may race with
837 kbdmux(4) in polling mode. Typical symptoms include (but
838 not limited to) duplicated and/or missing characters when
839 low level console functions (such as gets) are used while
840 interrupts are enabled (for example geli password prompt,
841 mountroot prompt etc.). Disabling kbdmux(4) may help.
844 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
845 kernel modules referencing any of the above need to be recompiled.
846 Bump __FreeBSD_version to 800075.
849 GEOM_PART has become the default partition slicer for storage devices,
850 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
851 introduces some changes:
853 MSDOS/EBR: the devices created from MSDOS extended partition entries
854 (EBR) can be named differently than with GEOM_MBR and are now symlinks
855 to devices with offset-based names. fstabs may need to be modified.
857 BSD: the "geometry does not match label" warning is harmless in most
858 cases but it points to problems in file system misalignment with
859 disk geometry. The "c" partition is now implicit, covers the whole
860 top-level drive and cannot be (mis)used by users.
862 General: Kernel dumps are now not allowed to be written to devices
863 whose partition types indicate they are meant to be used for file
864 systems (or, in case of MSDOS partitions, as something else than
867 Most of these changes date approximately from 200812.
870 The uscanner(4) driver has been removed from the kernel. This follows
871 Linux removing theirs in 2.6 and making libusb the default interface
875 The multicast forwarding code has been cleaned up. netstat(1)
876 only relies on KVM now for printing bandwidth upcall meters.
877 The IPv4 and IPv6 modules are split into ip_mroute_mod and
878 ip6_mroute_mod respectively. The config(5) options for statically
879 compiling this code remain the same, i.e. 'options MROUTING'.
882 Support for the IFF_NEEDSGIANT network interface flag has been
883 removed, which means that non-MPSAFE network device drivers are no
884 longer supported. In particular, if_ar, if_sr, and network device
885 drivers from the old (legacy) USB stack can no longer be built or
889 POSIX.1 Native Language Support (NLS) has been enabled in libc and
890 a bunch of new language catalog files have also been added.
891 This means that some common libc messages are now localized and
892 they depend on the LC_MESSAGES environmental variable.
895 The k8temp(4) driver has been renamed to amdtemp(4) since
896 support for Family 10 and Family 11 CPU families was added.
899 IGMPv3 and Source-Specific Multicast (SSM) have been merged
900 to the IPv4 stack. VIMAGE hooks are in but not yet used.
902 For kernel developers, the most important changes are that the
903 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
904 and this lock has been downgraded to a non-recursive mutex.
906 Transport protocols (UDP, Raw IP) are now responsible for filtering
907 inbound multicast traffic according to group membership and source
908 filters. The imo_multicast_filter() KPI exists for this purpose.
909 Transports which do not use multicast (SCTP, TCP) already reject
910 multicast by default. Forwarding and receive performance may improve
911 as a mutex acquisition is no longer needed in the ip_input()
912 low-level input path. in_addmulti() and in_delmulti() are shimmed
913 to new KPIs which exist to support SSM in-kernel.
915 For application developers, it is recommended that loopback of
916 multicast datagrams be disabled for best performance, as this
917 will still cause the lock to be taken for each looped-back
918 datagram transmission. The net.inet.ip.mcast.loop sysctl may
919 be tuned to 0 to disable loopback by default; it defaults to 1
920 to preserve the existing behaviour.
922 For systems administrators, to obtain best performance with
923 multicast reception and multiple groups, it is always recommended
924 that a card with a suitably precise hash filter is used. Hash
925 collisions will still result in the lock being taken within the
926 transport protocol input path to check group membership.
928 If deploying FreeBSD in an environment with IGMP snooping switches,
929 it is recommended that the net.inet.igmp.sendlocal sysctl remain
930 enabled; this forces 224.0.0.0/24 group membership to be announced
933 The size of 'struct igmpstat' has changed; netstat needs to be
934 recompiled to reflect this.
935 Bump __FreeBSD_version to 800070.
938 libusb20.so.1 is now installed as libusb.so.1 and the ports system
939 updated to use it. This requires a buildworld/installworld in order to
940 update the library and dependencies (usbconfig, etc). Its advisable to
941 rebuild all ports which uses libusb. More specific directions are given
942 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
943 libusb are no longer required and can be removed.
946 A workaround is committed to allow the creation of System V shared
947 memory segment of size > 2 GB on the 64-bit architectures.
948 Due to a limitation of the existing ABI, the shm_segsz member
949 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
950 wrong for large segments. Note that limits must be explicitly
951 raised to allow such segments to be created.
954 The layout of struct ifnet has changed, requiring a rebuild of all
955 network device driver modules.
958 The /dev handling for the new USB stack has changed, a
959 buildworld/installworld is required for libusb20.
962 The new USB2 stack has now been permanently moved in and all kernel and
963 module names reverted to their previous values (eg, usb, ehci, ohci,
964 ums, ...). The old usb stack can be compiled in by prefixing the name
965 with the letter 'o', the old usb modules have been removed.
966 Updating entry 20090216 for xorg and 20090215 for libmap may still
970 The rc.conf(5) option if_up_delay has been renamed to
971 defaultroute_delay to better reflect its purpose. If you have
972 customized this setting in /etc/rc.conf you need to update it to
976 xorg 7.4 wants to configure its input devices via hald which does not
977 yet work with USB2. If the keyboard/mouse does not work in xorg then
979 Option "AllowEmptyInput" "off"
980 to your ServerLayout section. This will cause X to use the configured
981 kbd and mouse sections from your xorg.conf.
984 The GENERIC kernels for all architectures now default to the new USB2
985 stack. No kernel config options or code have been removed so if a
986 problem arises please report it and optionally revert to the old USB
987 stack. If you are loading USB kernel modules or have a custom kernel
988 that includes GENERIC then ensure that usb names are also changed over,
989 eg uftdi -> usb2_serial_ftdi.
991 Older programs linked against the ports libusb 0.1 need to be
992 redirected to the new stack's libusb20. /etc/libmap.conf can
994 # Map old usb library to new one for usb2 stack
995 libusb-0.1.so.8 libusb20.so.1
998 All USB ethernet devices now attach as interfaces under the name ueN
999 (eg. ue0). This is to provide a predictable name as vendors often
1000 change usb chipsets in a product without notice.
1003 The ichsmb(4) driver has been changed to require SMBus slave
1004 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1005 All of the other SMBus controller drivers require left-justified
1006 slave addresses, so this change makes all the drivers provide the
1010 INET6 statistics (struct ip6stat) was updated.
1011 netstat(1) needs to be recompiled.
1014 NTFS has been removed from GENERIC kernel on amd64 to match
1015 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1016 will load ntfs.ko module automatically when NTFS support is
1017 actually needed, unless ntfs.ko is not installed or security
1018 level prohibits loading kernel modules. If either is the case,
1019 "options NTFS" has to be added into kernel config.
1022 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1023 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1024 800061. User space tools that rely on the size of struct tcpcb in
1025 tcp_var.h (e.g. sockstat) need to be recompiled.
1028 ng_tty(4) module updated to match the new TTY subsystem.
1029 Due to API change, user-level applications must be updated.
1030 New API support added to mpd5 CVS and expected to be present
1031 in next mpd5.3 release.
1034 With __FreeBSD_version 800060 the makefs tool is part of
1035 the base system (it was a port).
1038 The afdata and ifnet locks have been changed from mutexes to
1039 rwlocks, network modules will need to be re-compiled.
1042 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1043 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1044 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1045 architecture and 40 bytes on 64-bit architecture. The userland
1046 applications "arp" and "ndp" have been updated accordingly.
1047 The output from "netstat -r" shows only routing entries and
1048 none of the L2 information.
1051 __FreeBSD_version 800057 marks the switchover from the
1052 binary ath hal to source code. Users must add the line:
1054 options AH_SUPPORT_AR5416
1056 to their kernel config files when specifying:
1060 The ath_hal module no longer exists; the code is now compiled
1061 together with the driver in the ath module. It is now
1062 possible to tailor chip support (i.e. reduce the set of chips
1063 and thereby the code size); consult ath_hal(4) for details.
1066 __FreeBSD_version 800054 adds memory barriers to
1067 <machine/atomic.h>, new interfaces to ifnet to facilitate
1068 multiple hardware transmit queues for cards that support
1069 them, and a lock-less ring-buffer implementation to
1070 enable drivers to more efficiently manage queueing of
1074 A new version of ZFS (version 13) has been merged to -HEAD.
1075 This version has zpool attribute "listsnapshots" off by
1076 default, which means "zfs list" does not show snapshots,
1077 and is the same as Solaris behavior.
1080 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1083 The uhci, ohci, ehci and slhci USB Host controller drivers have
1084 been put into separate modules. If you load the usb module
1085 separately through loader.conf you will need to load the
1086 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1087 controller add the following to loader.conf:
1093 The ABI used by the PMC toolset has changed. Please keep
1094 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1098 atapci kernel module now includes only generic PCI ATA
1099 driver. AHCI driver moved to ataahci kernel module.
1100 All vendor-specific code moved into separate kernel modules:
1101 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1102 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1103 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1104 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1107 The TTY subsystem of the kernel has been replaced by a new
1108 implementation, which provides better scalability and an
1109 improved driver model. Most common drivers have been migrated to
1110 the new TTY subsystem, while others have not. The following
1111 drivers have not yet been ported to the new TTY layer:
1114 cy, digi, rc, rp, sio
1120 ng_h4, ng_tty, ppp, sl, snp
1122 Adding these drivers to your kernel configuration file shall
1123 cause compilation to fail.
1126 ntpd has been upgraded to 4.2.4p5.
1129 OpenSSH has been upgraded to 5.1p1.
1131 For many years, FreeBSD's version of OpenSSH preferred DSA
1132 over RSA for host and user authentication keys. With this
1133 upgrade, we've switched to the vendor's default of RSA over
1134 DSA. This may cause upgraded clients to warn about unknown
1135 host keys even for previously known hosts. Users should
1136 follow the usual procedure for verifying host keys before
1137 accepting the RSA key.
1139 This can be circumvented by setting the "HostKeyAlgorithms"
1140 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1143 Please note that the sequence of keys offered for
1144 authentication has been changed as well. You may want to
1145 specify IdentityFile in a different order to revert this
1149 The sio(4) driver has been removed from the i386 and amd64
1150 kernel configuration files. This means uart(4) is now the
1151 default serial port driver on those platforms as well.
1153 To prevent collisions with the sio(4) driver, the uart(4) driver
1154 uses different names for its device nodes. This means the
1155 onboard serial port will now most likely be called "ttyu0"
1156 instead of "ttyd0". You may need to reconfigure applications to
1157 use the new device names.
1159 When using the serial port as a boot console, be sure to update
1160 /boot/device.hints and /etc/ttys before booting the new kernel.
1161 If you forget to do so, you can still manually specify the hints
1162 at the loader prompt:
1164 set hint.uart.0.at="isa"
1165 set hint.uart.0.port="0x3F8"
1166 set hint.uart.0.flags="0x10"
1167 set hint.uart.0.irq="4"
1171 The gpt(8) utility has been removed. Use gpart(8) to partition
1175 The version that Linuxulator emulates was changed from 2.4.2
1176 to 2.6.16. If you experience any problems with Linux binaries
1177 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1178 if it fixes the problem contact emulation mailing list.
1181 ISDN4BSD (I4B) was removed from the src tree. You may need to
1182 update a your kernel configuration and remove relevant entries.
1185 I have checked in code to support multiple routing tables.
1186 See the man pages setfib(1) and setfib(2).
1187 This is a hopefully backwards compatible version,
1188 but to make use of it you need to compile your kernel
1189 with options ROUTETABLES=2 (or more up to 16).
1192 The 802.11 wireless support was redone to enable multi-bss
1193 operation on devices that are capable. The underlying device
1194 is no longer used directly but instead wlanX devices are
1195 cloned with ifconfig. This requires changes to rc.conf files.
1196 For example, change:
1197 ifconfig_ath0="WPA DHCP"
1200 ifconfig_wlan0="WPA DHCP"
1201 see rc.conf(5) for more details. In addition, mergemaster of
1202 /etc/rc.d is highly recommended. Simultaneous update of userland
1203 and kernel wouldn't hurt either.
1205 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1206 modules were merged into the base wlan module. All references
1207 to these modules (e.g. in kernel config files) must be removed.
1210 psm(4) has gained write(2) support in native operation level.
1211 Arbitrary commands can be written to /dev/psm%d and status can
1212 be read back from it. Therefore, an application is responsible
1213 for status validation and error recovery. It is a no-op in
1214 other operation levels.
1217 Support for KSE threading has been removed from the kernel. To
1218 run legacy applications linked against KSE libmap.conf may
1219 be used. The following libmap.conf may be used to ensure
1220 compatibility with any prior release:
1222 libpthread.so.1 libthr.so.1
1223 libpthread.so.2 libthr.so.2
1224 libkse.so.3 libthr.so.3
1227 The layout of struct vmspace has changed. This affects libkvm
1228 and any executables that link against libkvm and use the
1229 kvm_getprocs() function. In particular, but not exclusively,
1230 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1231 The effects are minimal, but it's advisable to upgrade world
1235 The latest em driver no longer has support in it for the
1236 82575 adapter, this is now moved to the igb driver. The
1237 split was done to make new features that are incompatible
1238 with older hardware easier to do.
1241 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1242 likewise the kernel option is now GEOM_LINUX_LVM.
1245 The default NFS mount mode has changed from UDP to TCP for
1246 increased reliability. If you rely on (insecurely) NFS
1247 mounting across a firewall you may need to update your
1251 Belatedly note the addition of m_collapse for compacting
1255 The fts(3) structures have been changed to use adequate
1256 integer types for their members and so to be able to cope
1257 with huge file trees. The old fts(3) ABI is preserved
1258 through symbol versioning in libc, so third-party binaries
1259 using fts(3) should still work, although they will not take
1260 advantage of the extended types. At the same time, some
1261 third-party software might fail to build after this change
1262 due to unportable assumptions made in its source code about
1263 fts(3) structure members. Such software should be fixed
1264 by its vendor or, in the worst case, in the ports tree.
1265 FreeBSD_version 800015 marks this change for the unlikely
1266 case that a portable fix is impossible.
1269 To upgrade to -current after this date, you must be running
1270 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1271 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1274 The ADAPTIVE_GIANT kernel option has been retired because its
1275 functionality is the default now.
1278 The AT keyboard emulation of sunkbd(4) has been turned on
1279 by default. In order to make the special symbols of the Sun
1280 keyboards driven by sunkbd(4) work under X these now have
1281 to be configured the same way as Sun USB keyboards driven
1282 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1284 Option "XkbLayout" "us"
1285 Option "XkbRules" "xorg"
1286 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1289 It has been decided that it is desirable to provide ABI
1290 backwards compatibility to the FreeBSD 4/5/6 versions of the
1291 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1292 broken with the introduction of PCI domain support (see the
1293 20070930 entry). Unfortunately, this required the ABI of
1294 PCIOCGETCONF to be broken again in order to be able to
1295 provide backwards compatibility to the old version of that
1296 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1297 again. As for prominent ports this affects neither pciutils
1298 nor xorg-server this time, the hal port needs to be rebuilt
1302 The misnamed kthread_create() and friends have been renamed
1303 to kproc_create() etc. Many of the callers already
1304 used kproc_start()..
1305 I will return kthread_create() and friends in a while
1306 with implementations that actually create threads, not procs.
1307 Renaming corresponds with version 800002.
1316 Avoid using make -j when upgrading. While generally safe, there are
1317 sometimes problems using -j to upgrade. If your upgrade fails with
1318 -j, please try again without -j. From time to time in the past there
1319 have been problems using -j with buildworld and/or installworld. This
1320 is especially true when upgrading between "distant" versions (eg one
1321 that cross a major release boundary or several minor releases, or when
1322 several months have passed on the -current branch).
1324 Sometimes, obscure build problems are the result of environment
1325 poisoning. This can happen because the make utility reads its
1326 environment when searching for values for global variables. To run
1327 your build attempts in an "environmental clean room", prefix all make
1328 commands with 'env -i '. See the env(1) manual page for more details.
1330 When upgrading from one major version to another it is generally best
1331 to upgrade to the latest code in the currently installed branch first,
1332 then do an upgrade to the new branch. This is the best-tested upgrade
1333 path, and has the highest probability of being successful. Please try
1334 this approach before reporting problems with a major version upgrade.
1338 When upgrading the boot ZFS pool to a new version, always follow
1341 1.) recompile and reinstall the ZFS boot loader and boot block
1342 (this is part of "make buildworld" and "make installworld")
1344 2.) update the ZFS boot block on your boot drive
1346 The following example updates the ZFS boot block on the first
1347 partition (freebsd-boot) of a GPT partitioned drive ad0:
1348 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1350 Non-boot pools do not need these updates.
1354 If you are updating from a prior version of FreeBSD (even one just
1355 a few days old), you should follow this procedure. It is the most
1356 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1358 make kernel-toolchain
1359 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1360 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1362 To test a kernel once
1363 ---------------------
1364 If you just want to boot a kernel once (because you are not sure
1365 if it works, or if you want to boot a known bad kernel to provide
1366 debugging information) run
1367 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1368 nextboot -k testkernel
1370 To just build a kernel when you know that it won't mess you up
1371 --------------------------------------------------------------
1372 This assumes you are already running a CURRENT system. Replace
1373 ${arch} with the architecture of your machine (e.g. "i386",
1374 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1376 cd src/sys/${arch}/conf
1377 config KERNEL_NAME_HERE
1378 cd ../compile/KERNEL_NAME_HERE
1383 If this fails, go to the "To build a kernel" section.
1385 To rebuild everything and install it on the current system.
1386 -----------------------------------------------------------
1387 # Note: sometimes if you are running current you gotta do more than
1388 # is listed here if you are upgrading from a really old current.
1390 <make sure you have good level 0 dumps>
1392 make kernel KERNCONF=YOUR_KERNEL_HERE
1394 <reboot in single user> [3]
1402 To cross-install current onto a separate partition
1403 --------------------------------------------------
1404 # In this approach we use a separate partition to hold
1405 # current's root, 'usr', and 'var' directories. A partition
1406 # holding "/", "/usr" and "/var" should be about 2GB in
1409 <make sure you have good level 0 dumps>
1412 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1413 <maybe newfs current's root partition>
1414 <mount current's root partition on directory ${CURRENT_ROOT}>
1415 make installworld DESTDIR=${CURRENT_ROOT}
1416 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1417 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1418 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1419 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1420 <reboot into current>
1421 <do a "native" rebuild/install as described in the previous section>
1422 <maybe install compatibility libraries from ports/misc/compat*>
1426 To upgrade in-place from 8.x-stable to current
1427 ----------------------------------------------
1428 <make sure you have good level 0 dumps>
1430 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1432 <reboot in single user> [3]
1439 Make sure that you've read the UPDATING file to understand the
1440 tweaks to various things you need. At this point in the life
1441 cycle of current, things change often and you are on your own
1442 to cope. The defaults can also change, so please read ALL of
1443 the UPDATING entries.
1445 Also, if you are tracking -current, you must be subscribed to
1446 freebsd-current@freebsd.org. Make sure that before you update
1447 your sources that you have read and understood all the recent
1448 messages there. If in doubt, please track -stable which has
1449 much fewer pitfalls.
1451 [1] If you have third party modules, such as vmware, you
1452 should disable them at this point so they don't crash your
1455 [3] From the bootblocks, boot -s, and then do
1460 adjkerntz -i # if CMOS is wall time
1461 Also, when doing a major release upgrade, it is required that
1462 you boot into single user mode to do the installworld.
1464 [4] Note: This step is non-optional. Failure to do this step
1465 can result in a significant reduction in the functionality of the
1466 system. Attempting to do it by hand is not recommended and those
1467 that pursue this avenue should read this file carefully, as well
1468 as the archives of freebsd-current and freebsd-hackers mailing lists
1469 for potential gotchas. The -U option is also useful to consider.
1470 See mergemaster(8) for more information.
1472 [5] Usually this step is a noop. However, from time to time
1473 you may need to do this if you get unknown user in the following
1474 step. It never hurts to do it all the time. You may need to
1475 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1476 install) after the buildworld before this step if you last updated
1477 from current before 20020224 or from -stable before 20020408.
1479 [6] This only deletes old files and directories. Old libraries
1480 can be deleted by "make delete-old-libs", but you have to make
1481 sure that no program is using those libraries anymore.
1483 [8] In order to have a kernel that can run the 4.x binaries needed to
1484 do an installworld, you must include the COMPAT_FREEBSD4 option in
1485 your kernel. Failure to do so may leave you with a system that is
1486 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1487 required to run the 5.x binaries on more recent kernels. And so on
1488 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1490 Make sure that you merge any new devices from GENERIC since the
1491 last time you updated your kernel config file.
1493 [9] When checking out sources, you must include the -P flag to have
1494 cvs prune empty directories.
1496 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1497 "?=" instead of the "=" assignment operator, so that buildworld can
1498 override the CPUTYPE if it needs to.
1500 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1501 not on the command line, or in /etc/make.conf. buildworld will
1502 warn if it is improperly defined.
1505 This file contains a list, in reverse chronological order, of major
1506 breakages in tracking -current. It is not guaranteed to be a complete
1507 list of such breakages, and only contains entries since October 10, 2007.
1508 If you need to see UPDATING entries from before that date, you will need
1509 to fetch an UPDATING file from an older FreeBSD release.
1511 Copyright information:
1513 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1515 Redistribution, publication, translation and use, with or without
1516 modification, in full or in part, in any form or format of this
1517 document are permitted without further permission from the author.
1519 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1520 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1521 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1522 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1523 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1524 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1525 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1526 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1527 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1528 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1529 POSSIBILITY OF SUCH DAMAGE.
1531 Contact Warner Losh if you have any questions about your use of