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 10.x IS SLOW:
13 FreeBSD 10.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 Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb
27 is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is
28 now spelled mips. This is to aid compatibility with third-party
29 software that expects this naming scheme in uname(3). Little-endian
30 settings are unchanged.
33 Disable by default the option VFS_ALLOW_NONMPSAFE for all supported
37 Now unix domain sockets behave "as expected" on nullfs(5). Previously
38 nullfs(5) did not pass through all behaviours to the underlying layer,
39 as a result if we bound to a socket on the lower layer we could connect
40 only to the lower path; if we bound to the upper layer we could connect
41 only to the upper path. The new behavior is one can connect to both the
42 lower and the upper paths regardless what layer path one binds to.
45 The getifaddrs upgrade path broken with 20111215 has been restored.
46 If you have upgraded in between 20111215 and 20120209 you need to
47 recompile libc again with your kernel. You still need to recompile
48 world to be able to configure CARP but this restriction already
52 The set_rcvar() function has been removed from /etc/rc.subr. All
53 base and ports rc.d scripts have been updated, so if you have a
54 port installed with a script in /usr/local/etc/rc.d you can either
55 hand-edit the rcvar= line, or reinstall the port.
57 An easy way to handle the mass-update of /etc/rc.d:
58 rm /etc/rc.d/* && mergemaster -i
61 panic(9) now stops other CPUs in the SMP systems, disables interrupts
62 on the current CPU and prevents other threads from running.
63 This behavior can be reverted using the kern.stop_scheduler_on_panic
65 The new behavior can be incompatible with kern.sync_on_panic.
68 The carp(4) facility has been changed significantly. Configuration
69 of the CARP protocol via ifconfig(8) has changed, as well as format
70 of CARP events submitted to devd(8) has changed. See manual pages
71 for more information. The arpbalance feature of carp(4) is currently
72 not supported anymore.
74 Size of struct in_aliasreq, struct in6_aliasreq has changed. User
75 utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8),
76 need to be recompiled.
79 The acpi_wmi(4) status device /dev/wmistat has been renamed to
83 The option VFS_ALLOW_NONMPSAFE option has been added in order to
84 explicitely support non-MPSAFE filesystems.
85 It is on by default for all supported platform at this present
89 The broken amd(4) driver has been replaced with esp(4) in the amd64,
90 i386 and pc98 GENERIC kernel configuration files.
93 sysinstall has been removed
96 The stable/9 branch created in subversion. This corresponds to the
97 RELENG_9 branch in CVS.
100 This commit modifies vfs_register() so that it uses a hash
101 calculation to set vfc_typenum, which is enabled by default.
102 The first time a system is booted after this change, the
103 vfc_typenum values will change for all file systems. The
104 main effect of this is a change to the NFS server file handles
105 for file systems that use vfc_typenum in their fsid, such as ZFS.
106 It will, however, prevent vfc_typenum from changing when file
107 systems are loaded in a different order for subsequent reboots.
108 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
109 until you are ready to remount all NFS clients after a reboot.
112 Bump the shared library version numbers for libraries that
113 do not use symbol versioning, have changed the ABI compared
114 to stable/8 and which shared library version was not bumped.
115 Done as part of 9.0-RELEASE cycle.
118 During the merge of Capsicum features, the fget(9) KPI was modified.
119 This may require the rebuilding of out-of-tree device drivers --
120 issues have been reported specifically with the nVidia device driver.
121 __FreeBSD_version is bumped to 900041.
123 Also, there is a period between 20110811 and 20110814 where the
124 special devices /dev/{stdin,stdout,stderr} did not work correctly.
125 Building world from a kernel during that window may not work.
128 The packet filter (pf) code has been updated to OpenBSD 4.5.
129 You need to update userland tools to be in sync with kernel.
130 This update breaks backward compatibility with earlier pfsync(4)
131 versions. Care must be taken when updating redundant firewall setups.
134 The following sysctls and tunables are retired on x86 platforms:
136 machdep.hlt_logical_cpus
137 The following sysctl is retired:
138 machdep.hyperthreading_allowed
139 The sysctls were supposed to provide a way to dynamically offline and
140 online selected CPUs on x86 platforms, but the implementation has not
141 been reliable especially with SCHED_ULE scheduler.
142 machdep.hyperthreading_allowed tunable is still available to ignore
143 hyperthreading CPUs at OS level.
144 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
145 where X is an APIC ID of a CPU. Be advised, though, that disabling
146 CPUs in non-uniform fashion will result in non-uniform topology and
147 may lead to sub-optimal system performance with SCHED_ULE, which is
151 cpumask_t type is retired and cpuset_t is used in order to describe
155 Changes to ifconfig(8) for dynamic address family detection mandate
156 that you are running a kernel of 20110525 or later. Make sure to
157 follow the update procedure to boot a new kernel before installing
161 Support for sun4v architecture is officially dropped
164 Several KPI breaking changes have been committed to the mii(4) layer,
165 the PHY drivers and consequently some Ethernet drivers using mii(4).
166 This means that miibus.ko and the modules of the affected Ethernet
167 drivers need to be recompiled.
169 Note to kernel developers: Given that the OUI bit reversion problem
170 was fixed as part of these changes all mii(4) commits related to OUIs,
171 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
172 handling, no longer can be merged verbatim to stable/8 and previous
176 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
177 into their kernel configurations along with 'device pci'.
180 The default NFS client is now the new NFS client, so fstype "newnfs"
181 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
182 Although mounts via fstype "nfs" will usually work without userland
183 changes, it is recommended that the mount(8) and mount_nfs(8)
184 commands be rebuilt from sources and that a link to mount_nfs called
185 mount_oldnfs be created. The new client is compiled into the
186 kernel with "options NFSCL" and this is needed for diskless root
187 file systems. The GENERIC kernel configs have been changed to use
188 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
189 To use the regular/old client, you can "mount -t oldnfs ...". For
190 a diskless root file system, you must also include a line like:
192 vfs.root.mountfrom="oldnfs:"
194 in the boot/loader.conf on the root fs on the NFS server to make
195 a diskless root fs use the old client.
198 The GENERIC kernels for all architectures now default to the new
199 CAM-based ATA stack. It means that all legacy ATA drivers were
200 removed and replaced by respective CAM drivers. If you are using
201 ATA device names in /etc/fstab or other places, make sure to update
202 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
203 where 'Y's are the sequential numbers starting from zero for each type
204 in order of detection, unless configured otherwise with tunables,
205 see cam(4)). There will be symbolic links created in /dev/ to map
206 old adX devices to the respective adaY. They should provide basic
207 compatibility for file systems mounting in most cases, but they do
208 not support old user-level APIs and do not have respective providers
209 in GEOM. Consider using updated management tools with new device names.
211 It is possible to load devices ahci, ata, siis and mvs as modules,
212 but option ATA_CAM should remain in kernel configuration to make ata
213 module work as CAM driver supporting legacy ATA controllers. Device ata
214 still can be used in modular fashion (atacore + ...). Modules atadisk
215 and atapi* are not used and won't affect operation in ATA_CAM mode.
216 Note that to use CAM-based ATA kernel should include CAM devices
217 scbus, pass, da (or explicitly ada), cd and optionally others. All of
218 them are parts of the cam module.
220 ataraid(4) functionality is now supported by the RAID GEOM class.
221 To use it you can load geom_raid kernel module and use graid(8) tool
222 for management. Instead of /dev/arX device names, use /dev/raid/rX.
224 No kernel config options or code have been removed, so if a problem
225 arises, please report it and optionally revert to the old ATA stack.
226 In order to do it you can remove from the kernel config:
231 , and instead add back:
232 device atadisk # ATA disk drives
233 device ataraid # ATA RAID drives
234 device atapicd # ATAPI CDROM drives
235 device atapifd # ATAPI floppy drives
236 device atapist # ATAPI tape drives
239 The default NFS server has been changed to the new server, which
240 was referred to as the experimental server. If you need to switch
241 back to the old NFS server, you must now put the "-o" option on
242 both the mountd and nfsd commands. This can be done using the
243 mountd_flags and nfs_server_flags rc.conf variables until an
244 update to the rc scripts is committed, which is coming soon.
247 The GNU Objective-C runtime library (libobjc), and other Objective-C
248 related components have been removed from the base system. If you
249 require an Objective-C library, please use one of the available ports.
252 ath(4) has been split into bus- and device- modules. if_ath contains
253 the HAL, the TX rate control and the network device code. if_ath_pci
254 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
255 contains the AHB glue. Users need to load both if_ath_pci and if_ath
256 in order to use ath on everything else.
258 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
259 need to load if_ath and if_ath_pci for ath(4) operation.
262 As part of the replacement of sysinstall, the process of building
263 release media has changed significantly. For details, please re-read
264 release(7), which has been updated to reflect the new build process.
267 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
268 is the last available version under GPLv2. It brings a number of new
269 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
270 4.1 and SSE 4.2), better support for powerpc64, a number of new
271 directives, and lots of other small improvements. See the ChangeLog
272 file in contrib/binutils for the full details.
275 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
276 compliant, and will now use half of hash for authentication.
277 This will break interoperability with all stacks (including all
278 actual FreeBSD versions) who implement
279 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
281 The only workaround with such peers is to use another HMAC
282 algorithm for IPsec ("phase 2") authentication.
285 Remove the uio_yield prototype and symbol. This function has
286 been misnamed since it was introduced and should not be
287 globally exposed with this name. The equivalent functionality
288 is now available using kern_yield(curthread->td_user_pri).
289 The function remains undocumented.
292 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
293 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
294 for scalar sysctls is defined but disabled. Code that needs
295 UQUAD to pass the type checking that must compile on older
296 systems where the define is not present can check against
297 __FreeBSD_version >= 900030.
299 The system dialog(1) has been replaced with a new version previously
300 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
301 with the previous version, but the libdialog associated with it has
302 a largely incompatible API. As such, the original version of libdialog
303 will be kept temporarily as libodialog, until its base system consumers
304 are replaced or updated. Bump __FreeBSD_version to 900030.
307 If you are trying to run make universe on a -stable system, and you get
308 the following warning:
309 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
310 unknown. config(8) likely too old."
311 or something similar to it, then you must upgrade your -stable system
312 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
313 stable/8) or build the config from the latest stable/8 branch and
314 install it on your system.
316 Prior to this date, building a current universe on 8-stable system from
317 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
318 error in the first kernel build phase. A new config on those old
319 systems will fix that problem for older versions of -current.
322 The TCP stack has been modified to allow Khelp modules to interact with
323 it via helper hook points and store per-connection data in the TCP
324 control block. Bump __FreeBSD_version to 900029. User space tools that
325 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
329 Generic IEEE 802.3 annex 31B full duplex flow control support has been
330 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
331 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
332 to take advantage of it instead of using custom implementations. This
333 means that these drivers now no longer unconditionally advertise
334 support for flow control but only do so if flow control is a selected
335 media option. This was implemented in the generic support that way in
336 order to allow flow control to be switched on and off via ifconfig(8)
337 with the PHY specific default to typically off in order to protect
338 from unwanted effects. Consequently, if you used flow control with
339 one of the above mentioned drivers you now need to explicitly enable
341 ifconfig bge0 media auto mediaopt flowcontrol
343 Along with the above mentioned changes generic support for setting
344 1000baseT master mode also has been added and brgphy(4), ciphy(4),
345 e1000phy(4) as well as ip1000phy(4) have been converted to take
346 advantage of it. This means that these drivers now no longer take the
347 link0 parameter for selecting master mode but the master media option
348 has to be used instead, for example like in the following:
349 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
351 Selection of master mode now is also available with all other PHY
352 drivers supporting 1000baseT.
355 The TCP stack has received a significant update to add support for
356 modularised congestion control and generally improve the clarity of
357 congestion control decisions. Bump __FreeBSD_version to 900025. User
358 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
359 sockstat) need to be recompiled.
362 The man(1) utility has been replaced by a new version that no longer
363 uses /etc/manpath.config. Please consult man.conf(5) for how to
364 migrate local entries to the new format.
367 The copyright strings printed by login(1) and sshd(8) at the time of a
368 new connection have been removed to follow other operating systems and
372 A workaround for a fixed ld bug has been removed in kernel code,
373 so make sure that your system ld is built from sources after
374 revision 210245 from 2010-07-19 (r211583 if building head kernel
375 on stable/8, r211584 for stable/7; both from 2010-08-21).
376 A symptom of incorrect ld version is different addresses for
377 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
380 The $ipv6_prefer variable in rc.conf(5) has been split into
381 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
383 The $ip6addrctl_policy is a variable to choose a pre-defined
384 address selection policy set by ip6addrctl(8). A value
385 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
388 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
389 flag (see an entry of 20090926) is set on an interface with no
390 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
391 security reason. If you want IPv6 link-local address on all
392 interfaces by default, set this to "YES".
394 The old ipv6_prefer="YES" is equivalent to
395 ipv6_activate_all_interfaces="YES" and
396 ip6addrctl_policy="ipv6_prefer".
399 DTrace has grown support for userland tracing. Due to this, DTrace is
400 now i386 and amd64 only.
401 dtruss(1) is now installed by default on those systems and a new
402 kernel module is needed for userland tracing: fasttrap.
403 No changes to your kernel config file are necessary to enable
404 userland tracing, but you might consider adding 'STRIP=' and
405 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
406 to have informative userland stack traces in DTrace (ustack).
409 The acpi_aiboost(4) driver has been removed in favor of the new
410 aibs(4) driver. You should update your kernel configuration file.
413 BSD grep has been imported to the base system and it is built by
414 default. It is completely BSD licensed, highly GNU-compatible, uses
415 less memory than its GNU counterpart and has a small codebase.
416 However, it is slower than its GNU counterpart, which is mostly
417 noticeable for larger searches, for smaller ones it is measurable
418 but not significant. The reason is complex, the most important factor
419 is that we lack a modern and efficient regex library and GNU
420 overcomes this by optimizing the searches internally. Future work
421 on improving the regex performance is planned, for the meantime,
422 users that need better performance, can build GNU grep instead by
423 setting the WITH_GNU_GREP knob.
426 Due to the import of powerpc64 support, all existing powerpc kernel
427 configuration files must be updated with a machine directive like this:
428 machine powerpc powerpc
430 In addition, an updated config(8) is required to build powerpc kernels
434 A new version of ZFS (version 15) has been merged to -HEAD.
435 This version uses a python library for the following subcommands:
436 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
437 For full functionality of these commands the following port must
438 be installed: sysutils/py-zfs
441 'vm_page's are now hashed by physical address to an array of mutexes.
442 Currently this is only used to serialize access to hold_count. Over
443 time the page queue mutex will be peeled away. This changes the size
444 of pmap on every architecture. And requires all callers of vm_page_hold
445 and vm_page_unhold to be updated.
448 WITH_CTF can now be specified in src.conf (not recommended, there
449 are some problems with static executables), make.conf (would also
450 affect ports which do not use GNU make and do not override the
451 compile targets) or in the kernel config (via "makeoptions
453 When WITH_CTF was specified there before this was silently ignored,
454 so make sure that WITH_CTF is not used in places which could lead
455 to unwanted behavior.
458 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
459 to allow 32-bit compatibility on non-x86 platforms. All kernel
460 configurations on amd64 and ia64 platforms using these options must
461 be modified accordingly.
464 The utmp user accounting database has been replaced with utmpx,
465 the user accounting interface standardized by POSIX.
466 Unfortunately the semantics of utmp and utmpx don't match,
467 making it practically impossible to support both interfaces.
468 The user accounting database is used by tools like finger(1),
469 last(1), talk(1), w(1) and ac(8).
471 All applications in the base system use utmpx. This means only
472 local binaries (e.g. from the ports tree) may still use these
473 utmp database files. These applications must be rebuilt to make
476 After the system has been upgraded, it is safe to remove the old
477 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
478 assuming their contents is of no importance anymore. Old wtmp
479 databases can only be used by last(1) and ac(8) after they have
480 been converted to the new format using wtmpcvt(1).
483 Introduce the kernel thread "deadlock resolver" (which can be enabled
484 via the DEADLKRES option, see NOTES for more details) and the
485 sleepq_type() function for sleepqueues.
488 The rc.firewall and rc.firewall6 were unified, and
489 rc.firewall6 and rc.d/ip6fw were removed.
490 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
491 variables are obsoleted. Instead, the following new rc
492 variables are added to rc.d/ipfw:
494 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
495 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
496 firewall_simple_onet_ipv6, firewall_trusted_ipv6
498 The meanings correspond to the relevant IPv4 variables.
504 The default terminal emulation for syscons(4) has been changed
505 from cons25 to xterm on all platforms except pc98. This means
506 that the /etc/ttys file needs to be updated to ensure correct
507 operation of applications on the console.
509 The terminal emulation style can be toggled per window by using
510 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
511 options can be used to change the compile-time default back to
514 To prevent graphical artifacts, make sure the TERM environment
515 variable is set to match the terminal emulation that is being
516 performed by syscons(4).
519 The layout of the structure ieee80211req_scan_result has changed.
520 Applications that require wireless scan results (e.g. ifconfig(8))
521 from net80211 need to be recompiled.
523 Applications such as wpa_supplicant(8) may require a full world
524 build without using NO_CLEAN in order to get synchronized with the
528 The iwn(4) driver has been updated to support the 5000 and 5150 series.
529 There's one kernel module for each firmware. Adding "device iwnfw"
530 to the kernel configuration file means including all three firmware
531 images inside the kernel. If you want to include just the one for
532 your wireless card, use the devices iwn4965fw, iwn5000fw or
536 The rc.d/network_ipv6, IPv6 configuration script has been integrated
537 into rc.d/netif. The changes are the following:
539 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
540 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
541 Note that both variables need the "inet6" keyword at the head.
543 Do not set $ipv6_network_interfaces manually if you do not
544 understand what you are doing. It is not needed in most cases.
546 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
549 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
550 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
552 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
553 all configured interfaces have "inet6 accept_rtadv" in the
554 $ifconfig_IF_ipv6. These are for backward compatibility.
556 3. A new variable $ipv6_prefer has been added. If NO, IPv6
557 functionality of interfaces with no corresponding
558 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
559 and the default address selection policy of ip6addrctl(8)
560 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
561 Note that if you want to configure IPv6 functionality on the
562 disabled interfaces after boot, first you need to clear the flag by
563 using ifconfig(8) like:
565 ifconfig em0 inet6 -ifdisabled
567 If YES, the default address selection policy is set as
570 The default value of $ipv6_prefer is NO.
572 4. If your system need to receive Router Advertisement messages,
573 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
574 scripts automatically invoke rtsol(8) when the interface becomes
575 UP. The Router Advertisement messages are used for SLAAC
576 (State-Less Address AutoConfiguration).
579 802.11s D3.03 support was committed. This is incompatible with the
580 previous code, which was based on D3.0.
583 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
584 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
585 control whether accepting Router Advertisement messages or not.
586 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
587 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
588 The ifconfig(8) utility now supports these flags.
591 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
592 mount(8) and -a option for df(1) to see them.
595 The old tunable hw.bus.devctl_disable has been superseded by
596 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
597 replaced by hw.bus.devctl_queue=0. The default for this new tunable
601 Remove the option STOP_NMI. The default action is now to use NMI only
602 for KDB via the newly introduced function stop_cpus_hard() and
603 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
606 The stable/8 branch created in subversion. This corresponds to the
607 RELENG_8 branch in CVS.
610 Bump the shared library version numbers for all libraries that do not
611 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
612 __FreeBSD_version to 800105.
615 Due to changes in the implementation of virtual network stack support,
616 all network-related kernel modules must be recompiled. As this change
617 breaks the ABI, bump __FreeBSD_version to 800104.
620 The TOE interface to the TCP syncache has been modified to remove
621 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
622 The cxgb driver is the only TOE consumer affected by this change, and
623 needs to be recompiled along with the kernel. As this change breaks
624 the ABI, bump __FreeBSD_version to 800103.
627 Padding has been added to struct tcpcb, sackhint and tcpstat in
628 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
629 maintaining the ABI. However, this change breaks the ABI, so bump
630 __FreeBSD_version to 800102. User space tools that rely on the size of
631 any of these structs (e.g. sockstat) need to be recompiled.
634 The NFS_LEGACYRPC option has been removed along with the old kernel
635 RPC implementation that this option selected. Kernel configurations
636 may need to be adjusted.
639 The network interface device nodes at /dev/net/<interface> have been
640 removed. All ioctl operations can be performed the normal way using
641 routing sockets. The kqueue functionality can generally be replaced
642 with routing sockets.
645 The documentation from the FreeBSD Documentation Project (Handbook,
646 FAQ, etc.) is now installed via packages by sysinstall(8) and under
647 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
650 The ABI of various structures related to the SYSV IPC API have been
651 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
652 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
656 Layout of struct vnet has changed as routing related variables were
657 moved to their own Vimage module. Modules need to be recompiled. Bump
658 __FreeBSD_version to 800099.
661 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
662 respectively. As long as no more than 16 groups per process are used,
663 no changes should be visible. When more than 16 groups are used, old
664 binaries may fail if they call getgroups() or getgrouplist() with
665 statically sized storage. Recompiling will work around this, but
666 applications should be modified to use dynamically allocated storage
667 for group arrays as POSIX.1-2008 does not cap an implementation's
668 number of supported groups at NGROUPS_MAX+1 as previous versions did.
670 NFS and portalfs mounts may also be affected as the list of groups is
671 truncated to 16. Users of NFS who use more than 16 groups, should
672 take care that negative group permissions are not used on the exported
673 file systems as they will not be reliable unless a GSSAPI based
674 authentication method is used.
677 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
678 option compiles in the support for adaptive spinning for lockmgrs
679 which want to enable it. The lockinit() function now accepts the flag
680 LK_ADAPTIVE in order to make the lock object subject to adaptive
681 spinning when both held in write and read mode.
684 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
685 changed. User applications that use this ioctl need to be rebuilt.
688 The layout of struct thread has changed. Kernel and modules need to
692 The layout of structs ifnet, domain, protosw and vnet_net has changed.
693 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
696 window(1) has been removed from the base system. It can now be
697 installed from ports. The port is called misc/window.
700 The way we are storing and accessing `routing table' entries has
701 changed. Programs reading the FIB, like netstat, need to be
705 A new netisr implementation has been added for FreeBSD 8. Network
706 file system modules, such as igmp, ipdivert, and others, should be
708 Bump __FreeBSD_version to 800096.
711 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
715 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
716 Bump __FreeBSD_version to 800094.
719 Add mnt_xflag field to 'struct mount'. File system modules need to be
721 Bump __FreeBSD_version to 800093.
724 The compiling option ADAPTIVE_SX has been retired while it has been
725 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
726 The KPI for sx_init_flags() changes as accepting flags:
727 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
728 been introduced in order to handle the reversed logic.
729 Bump __FreeBSD_version to 800092.
732 Add support for hierarchical jails. Remove global securelevel.
733 Bump __FreeBSD_version to 800091.
736 The layout of struct vnet_net has changed, therefore modules
738 Bump __FreeBSD_version to 800090.
741 The newly imported zic(8) produces a new format in the output. Please
742 run tzsetup(8) to install the newly created data to /etc/localtime.
745 The sysctl tree for the usb stack has renamed from hw.usb2.* to
746 hw.usb.* and is now consistent again with previous releases.
749 802.11 monitor mode support was revised and driver api's were changed.
750 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
751 of DLT_IEEE802_11. No user-visible data structures were changed but
752 applications that use DLT_IEEE802_11 may require changes.
753 Bump __FreeBSD_version to 800088.
756 The layout of the following structs has changed: sysctl_oid,
757 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
758 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
759 panics may be experienced. World rebuild is required for
760 correctly checking networking state from userland.
761 Bump __FreeBSD_version to 800085.
764 MLDv2 and Source-Specific Multicast (SSM) have been merged
765 to the IPv6 stack. VIMAGE hooks are in but not yet used.
766 The implementation of SSM within FreeBSD's IPv6 stack closely
767 follows the IPv4 implementation.
769 For kernel developers:
771 * The most important changes are that the ip6_output() and
772 ip6_input() paths no longer take the IN6_MULTI_LOCK,
773 and this lock has been downgraded to a non-recursive mutex.
775 * As with the changes to the IPv4 stack to support SSM, filtering
776 of inbound multicast traffic must now be performed by transport
777 protocols within the IPv6 stack. This does not apply to TCP and
778 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
780 * The KPIs used by IPv6 multicast are similar to those used by
781 the IPv4 stack, with the following differences:
782 * im6o_mc_filter() is analogous to imo_multicast_filter().
783 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
784 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
785 * IN6_LOOKUP_MULTI() has been deprecated and removed.
786 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
787 for MLDv1 have an additional 'timer' argument which is used to
788 jitter the initial membership report for the solicited-node
789 multicast membership on-link.
790 * This is not strictly needed for MLDv2, which already jitters
791 its report transmissions. However, the 'timer' argument is
792 preserved in case MLDv1 is active on the interface.
794 * The KAME linked-list based IPv6 membership implementation has
795 been refactored to use a vector similar to that used by the IPv4
797 Code which maintains a list of its own multicast memberships
798 internally, e.g. carp, has been updated to reflect the new
801 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
802 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
803 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
804 implementation constraint which needs to be addressed in HEAD.
806 For application developers:
808 * The changes are broadly similar to those made for the IPv4
811 * The use of IPv4 and IPv6 multicast socket options on the same
812 socket, using mapped addresses, HAS NOT been tested or supported.
814 * There are a number of issues with the implementation of various
815 IPv6 multicast APIs which need to be resolved in the API surface
816 before the implementation is fully compatible with KAME userland
817 use, and these are mostly to do with interface index treatment.
819 * The literature available discusses the use of either the delta / ASM
820 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
821 using setsourcefilter(3)/getsourcefilter(3). For more information
822 please refer to RFC 3768, 'Socket Interface Extensions for
823 Multicast Source Filters'.
825 * Applications which use the published RFC 3678 APIs should be fine.
827 For systems administrators:
829 * The mtest(8) utility has been refactored to support IPv6, in
830 addition to IPv4. Interface addresses are no longer accepted
831 as arguments, their names must be used instead. The utility
832 will map the interface name to its first IPv4 address as
833 returned by getifaddrs(3).
835 * The ifmcstat(8) utility has also been updated to print the MLDv2
836 endpoint state and source filter lists via sysctl(3).
838 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
839 loopback of IPv6 multicast datagrams by default; it defaults to 1
840 to preserve the existing behaviour. Disabling multicast loopback is
841 recommended for optimal system performance.
843 * The IPv6 MROUTING code has been changed to examine this sysctl
844 instead of attempting to perform a group lookup before looping
845 back forwarded datagrams.
847 Bump __FreeBSD_version to 800084.
850 Implement low-level Bluetooth HCI API.
851 Bump __FreeBSD_version to 800083.
854 The layout of struct malloc_type, used by modules to register new
855 memory allocation types, has changed. Most modules will need to
856 be rebuilt or panics may be experienced.
857 Bump __FreeBSD_version to 800081.
860 Anticipate overflowing inp_flags - add inp_flags2.
861 This changes most offsets in inpcb, so checking v4 connection
862 state will require a world rebuild.
863 Bump __FreeBSD_version to 800080.
866 Add an llentry to struct route and struct route_in6. Modules
867 embedding a struct route will need to be recompiled.
868 Bump __FreeBSD_version to 800079.
871 The size of rt_metrics_lite and by extension rtentry has changed.
872 Networking administration apps will need to be recompiled.
873 The route command now supports show as an alias for get, weighting
874 of routes, sticky and nostick flags to alter the behavior of stateful
876 Bump __FreeBSD_version to 800078.
879 Do not use Giant for kbdmux(4) locking. This is wrong and
880 apparently causing more problems than it solves. This will
881 re-open the issue where interrupt handlers may race with
882 kbdmux(4) in polling mode. Typical symptoms include (but
883 not limited to) duplicated and/or missing characters when
884 low level console functions (such as gets) are used while
885 interrupts are enabled (for example geli password prompt,
886 mountroot prompt etc.). Disabling kbdmux(4) may help.
889 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
890 kernel modules referencing any of the above need to be recompiled.
891 Bump __FreeBSD_version to 800075.
894 GEOM_PART has become the default partition slicer for storage devices,
895 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
896 introduces some changes:
898 MSDOS/EBR: the devices created from MSDOS extended partition entries
899 (EBR) can be named differently than with GEOM_MBR and are now symlinks
900 to devices with offset-based names. fstabs may need to be modified.
902 BSD: the "geometry does not match label" warning is harmless in most
903 cases but it points to problems in file system misalignment with
904 disk geometry. The "c" partition is now implicit, covers the whole
905 top-level drive and cannot be (mis)used by users.
907 General: Kernel dumps are now not allowed to be written to devices
908 whose partition types indicate they are meant to be used for file
909 systems (or, in case of MSDOS partitions, as something else than
912 Most of these changes date approximately from 200812.
915 The uscanner(4) driver has been removed from the kernel. This follows
916 Linux removing theirs in 2.6 and making libusb the default interface
920 The multicast forwarding code has been cleaned up. netstat(1)
921 only relies on KVM now for printing bandwidth upcall meters.
922 The IPv4 and IPv6 modules are split into ip_mroute_mod and
923 ip6_mroute_mod respectively. The config(5) options for statically
924 compiling this code remain the same, i.e. 'options MROUTING'.
927 Support for the IFF_NEEDSGIANT network interface flag has been
928 removed, which means that non-MPSAFE network device drivers are no
929 longer supported. In particular, if_ar, if_sr, and network device
930 drivers from the old (legacy) USB stack can no longer be built or
934 POSIX.1 Native Language Support (NLS) has been enabled in libc and
935 a bunch of new language catalog files have also been added.
936 This means that some common libc messages are now localized and
937 they depend on the LC_MESSAGES environmental variable.
940 The k8temp(4) driver has been renamed to amdtemp(4) since
941 support for Family 10 and Family 11 CPU families was added.
944 IGMPv3 and Source-Specific Multicast (SSM) have been merged
945 to the IPv4 stack. VIMAGE hooks are in but not yet used.
947 For kernel developers, the most important changes are that the
948 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
949 and this lock has been downgraded to a non-recursive mutex.
951 Transport protocols (UDP, Raw IP) are now responsible for filtering
952 inbound multicast traffic according to group membership and source
953 filters. The imo_multicast_filter() KPI exists for this purpose.
954 Transports which do not use multicast (SCTP, TCP) already reject
955 multicast by default. Forwarding and receive performance may improve
956 as a mutex acquisition is no longer needed in the ip_input()
957 low-level input path. in_addmulti() and in_delmulti() are shimmed
958 to new KPIs which exist to support SSM in-kernel.
960 For application developers, it is recommended that loopback of
961 multicast datagrams be disabled for best performance, as this
962 will still cause the lock to be taken for each looped-back
963 datagram transmission. The net.inet.ip.mcast.loop sysctl may
964 be tuned to 0 to disable loopback by default; it defaults to 1
965 to preserve the existing behaviour.
967 For systems administrators, to obtain best performance with
968 multicast reception and multiple groups, it is always recommended
969 that a card with a suitably precise hash filter is used. Hash
970 collisions will still result in the lock being taken within the
971 transport protocol input path to check group membership.
973 If deploying FreeBSD in an environment with IGMP snooping switches,
974 it is recommended that the net.inet.igmp.sendlocal sysctl remain
975 enabled; this forces 224.0.0.0/24 group membership to be announced
978 The size of 'struct igmpstat' has changed; netstat needs to be
979 recompiled to reflect this.
980 Bump __FreeBSD_version to 800070.
983 libusb20.so.1 is now installed as libusb.so.1 and the ports system
984 updated to use it. This requires a buildworld/installworld in order to
985 update the library and dependencies (usbconfig, etc). Its advisable to
986 rebuild all ports which uses libusb. More specific directions are given
987 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
988 libusb are no longer required and can be removed.
991 A workaround is committed to allow the creation of System V shared
992 memory segment of size > 2 GB on the 64-bit architectures.
993 Due to a limitation of the existing ABI, the shm_segsz member
994 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
995 wrong for large segments. Note that limits must be explicitly
996 raised to allow such segments to be created.
999 The layout of struct ifnet has changed, requiring a rebuild of all
1000 network device driver modules.
1003 The /dev handling for the new USB stack has changed, a
1004 buildworld/installworld is required for libusb20.
1007 The new USB2 stack has now been permanently moved in and all kernel and
1008 module names reverted to their previous values (eg, usb, ehci, ohci,
1009 ums, ...). The old usb stack can be compiled in by prefixing the name
1010 with the letter 'o', the old usb modules have been removed.
1011 Updating entry 20090216 for xorg and 20090215 for libmap may still
1015 The rc.conf(5) option if_up_delay has been renamed to
1016 defaultroute_delay to better reflect its purpose. If you have
1017 customized this setting in /etc/rc.conf you need to update it to
1021 xorg 7.4 wants to configure its input devices via hald which does not
1022 yet work with USB2. If the keyboard/mouse does not work in xorg then
1024 Option "AllowEmptyInput" "off"
1025 to your ServerLayout section. This will cause X to use the configured
1026 kbd and mouse sections from your xorg.conf.
1029 The GENERIC kernels for all architectures now default to the new USB2
1030 stack. No kernel config options or code have been removed so if a
1031 problem arises please report it and optionally revert to the old USB
1032 stack. If you are loading USB kernel modules or have a custom kernel
1033 that includes GENERIC then ensure that usb names are also changed over,
1034 eg uftdi -> usb2_serial_ftdi.
1036 Older programs linked against the ports libusb 0.1 need to be
1037 redirected to the new stack's libusb20. /etc/libmap.conf can
1039 # Map old usb library to new one for usb2 stack
1040 libusb-0.1.so.8 libusb20.so.1
1043 All USB ethernet devices now attach as interfaces under the name ueN
1044 (eg. ue0). This is to provide a predictable name as vendors often
1045 change usb chipsets in a product without notice.
1048 The ichsmb(4) driver has been changed to require SMBus slave
1049 addresses be left-justified (xxxxxxx0b) rather than right-justified.
1050 All of the other SMBus controller drivers require left-justified
1051 slave addresses, so this change makes all the drivers provide the
1055 INET6 statistics (struct ip6stat) was updated.
1056 netstat(1) needs to be recompiled.
1059 NTFS has been removed from GENERIC kernel on amd64 to match
1060 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1061 will load ntfs.ko module automatically when NTFS support is
1062 actually needed, unless ntfs.ko is not installed or security
1063 level prohibits loading kernel modules. If either is the case,
1064 "options NTFS" has to be added into kernel config.
1067 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1068 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1069 800061. User space tools that rely on the size of struct tcpcb in
1070 tcp_var.h (e.g. sockstat) need to be recompiled.
1073 ng_tty(4) module updated to match the new TTY subsystem.
1074 Due to API change, user-level applications must be updated.
1075 New API support added to mpd5 CVS and expected to be present
1076 in next mpd5.3 release.
1079 With __FreeBSD_version 800060 the makefs tool is part of
1080 the base system (it was a port).
1083 The afdata and ifnet locks have been changed from mutexes to
1084 rwlocks, network modules will need to be re-compiled.
1087 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1088 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1089 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1090 architecture and 40 bytes on 64-bit architecture. The userland
1091 applications "arp" and "ndp" have been updated accordingly.
1092 The output from "netstat -r" shows only routing entries and
1093 none of the L2 information.
1096 __FreeBSD_version 800057 marks the switchover from the
1097 binary ath hal to source code. Users must add the line:
1099 options AH_SUPPORT_AR5416
1101 to their kernel config files when specifying:
1105 The ath_hal module no longer exists; the code is now compiled
1106 together with the driver in the ath module. It is now
1107 possible to tailor chip support (i.e. reduce the set of chips
1108 and thereby the code size); consult ath_hal(4) for details.
1111 __FreeBSD_version 800054 adds memory barriers to
1112 <machine/atomic.h>, new interfaces to ifnet to facilitate
1113 multiple hardware transmit queues for cards that support
1114 them, and a lock-less ring-buffer implementation to
1115 enable drivers to more efficiently manage queueing of
1119 A new version of ZFS (version 13) has been merged to -HEAD.
1120 This version has zpool attribute "listsnapshots" off by
1121 default, which means "zfs list" does not show snapshots,
1122 and is the same as Solaris behavior.
1125 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1128 The uhci, ohci, ehci and slhci USB Host controller drivers have
1129 been put into separate modules. If you load the usb module
1130 separately through loader.conf you will need to load the
1131 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1132 controller add the following to loader.conf:
1138 The ABI used by the PMC toolset has changed. Please keep
1139 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1143 atapci kernel module now includes only generic PCI ATA
1144 driver. AHCI driver moved to ataahci kernel module.
1145 All vendor-specific code moved into separate kernel modules:
1146 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1147 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1148 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1149 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1152 The TTY subsystem of the kernel has been replaced by a new
1153 implementation, which provides better scalability and an
1154 improved driver model. Most common drivers have been migrated to
1155 the new TTY subsystem, while others have not. The following
1156 drivers have not yet been ported to the new TTY layer:
1159 cy, digi, rc, rp, sio
1165 ng_h4, ng_tty, ppp, sl, snp
1167 Adding these drivers to your kernel configuration file shall
1168 cause compilation to fail.
1171 ntpd has been upgraded to 4.2.4p5.
1174 OpenSSH has been upgraded to 5.1p1.
1176 For many years, FreeBSD's version of OpenSSH preferred DSA
1177 over RSA for host and user authentication keys. With this
1178 upgrade, we've switched to the vendor's default of RSA over
1179 DSA. This may cause upgraded clients to warn about unknown
1180 host keys even for previously known hosts. Users should
1181 follow the usual procedure for verifying host keys before
1182 accepting the RSA key.
1184 This can be circumvented by setting the "HostKeyAlgorithms"
1185 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1188 Please note that the sequence of keys offered for
1189 authentication has been changed as well. You may want to
1190 specify IdentityFile in a different order to revert this
1194 The sio(4) driver has been removed from the i386 and amd64
1195 kernel configuration files. This means uart(4) is now the
1196 default serial port driver on those platforms as well.
1198 To prevent collisions with the sio(4) driver, the uart(4) driver
1199 uses different names for its device nodes. This means the
1200 onboard serial port will now most likely be called "ttyu0"
1201 instead of "ttyd0". You may need to reconfigure applications to
1202 use the new device names.
1204 When using the serial port as a boot console, be sure to update
1205 /boot/device.hints and /etc/ttys before booting the new kernel.
1206 If you forget to do so, you can still manually specify the hints
1207 at the loader prompt:
1209 set hint.uart.0.at="isa"
1210 set hint.uart.0.port="0x3F8"
1211 set hint.uart.0.flags="0x10"
1212 set hint.uart.0.irq="4"
1216 The gpt(8) utility has been removed. Use gpart(8) to partition
1220 The version that Linuxulator emulates was changed from 2.4.2
1221 to 2.6.16. If you experience any problems with Linux binaries
1222 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1223 if it fixes the problem contact emulation mailing list.
1226 ISDN4BSD (I4B) was removed from the src tree. You may need to
1227 update a your kernel configuration and remove relevant entries.
1230 I have checked in code to support multiple routing tables.
1231 See the man pages setfib(1) and setfib(2).
1232 This is a hopefully backwards compatible version,
1233 but to make use of it you need to compile your kernel
1234 with options ROUTETABLES=2 (or more up to 16).
1237 The 802.11 wireless support was redone to enable multi-bss
1238 operation on devices that are capable. The underlying device
1239 is no longer used directly but instead wlanX devices are
1240 cloned with ifconfig. This requires changes to rc.conf files.
1241 For example, change:
1242 ifconfig_ath0="WPA DHCP"
1245 ifconfig_wlan0="WPA DHCP"
1246 see rc.conf(5) for more details. In addition, mergemaster of
1247 /etc/rc.d is highly recommended. Simultaneous update of userland
1248 and kernel wouldn't hurt either.
1250 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1251 modules were merged into the base wlan module. All references
1252 to these modules (e.g. in kernel config files) must be removed.
1255 psm(4) has gained write(2) support in native operation level.
1256 Arbitrary commands can be written to /dev/psm%d and status can
1257 be read back from it. Therefore, an application is responsible
1258 for status validation and error recovery. It is a no-op in
1259 other operation levels.
1262 Support for KSE threading has been removed from the kernel. To
1263 run legacy applications linked against KSE libmap.conf may
1264 be used. The following libmap.conf may be used to ensure
1265 compatibility with any prior release:
1267 libpthread.so.1 libthr.so.1
1268 libpthread.so.2 libthr.so.2
1269 libkse.so.3 libthr.so.3
1272 The layout of struct vmspace has changed. This affects libkvm
1273 and any executables that link against libkvm and use the
1274 kvm_getprocs() function. In particular, but not exclusively,
1275 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1276 The effects are minimal, but it's advisable to upgrade world
1280 The latest em driver no longer has support in it for the
1281 82575 adapter, this is now moved to the igb driver. The
1282 split was done to make new features that are incompatible
1283 with older hardware easier to do.
1286 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1287 likewise the kernel option is now GEOM_LINUX_LVM.
1290 The default NFS mount mode has changed from UDP to TCP for
1291 increased reliability. If you rely on (insecurely) NFS
1292 mounting across a firewall you may need to update your
1296 Belatedly note the addition of m_collapse for compacting
1300 The fts(3) structures have been changed to use adequate
1301 integer types for their members and so to be able to cope
1302 with huge file trees. The old fts(3) ABI is preserved
1303 through symbol versioning in libc, so third-party binaries
1304 using fts(3) should still work, although they will not take
1305 advantage of the extended types. At the same time, some
1306 third-party software might fail to build after this change
1307 due to unportable assumptions made in its source code about
1308 fts(3) structure members. Such software should be fixed
1309 by its vendor or, in the worst case, in the ports tree.
1310 FreeBSD_version 800015 marks this change for the unlikely
1311 case that a portable fix is impossible.
1314 To upgrade to -current after this date, you must be running
1315 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1316 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1319 The ADAPTIVE_GIANT kernel option has been retired because its
1320 functionality is the default now.
1323 The AT keyboard emulation of sunkbd(4) has been turned on
1324 by default. In order to make the special symbols of the Sun
1325 keyboards driven by sunkbd(4) work under X these now have
1326 to be configured the same way as Sun USB keyboards driven
1327 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1329 Option "XkbLayout" "us"
1330 Option "XkbRules" "xorg"
1331 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1334 It has been decided that it is desirable to provide ABI
1335 backwards compatibility to the FreeBSD 4/5/6 versions of the
1336 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1337 broken with the introduction of PCI domain support (see the
1338 20070930 entry). Unfortunately, this required the ABI of
1339 PCIOCGETCONF to be broken again in order to be able to
1340 provide backwards compatibility to the old version of that
1341 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1342 again. As for prominent ports this affects neither pciutils
1343 nor xorg-server this time, the hal port needs to be rebuilt
1347 The misnamed kthread_create() and friends have been renamed
1348 to kproc_create() etc. Many of the callers already
1349 used kproc_start()..
1350 I will return kthread_create() and friends in a while
1351 with implementations that actually create threads, not procs.
1352 Renaming corresponds with version 800002.
1361 Avoid using make -j when upgrading. While generally safe, there are
1362 sometimes problems using -j to upgrade. If your upgrade fails with
1363 -j, please try again without -j. From time to time in the past there
1364 have been problems using -j with buildworld and/or installworld. This
1365 is especially true when upgrading between "distant" versions (eg one
1366 that cross a major release boundary or several minor releases, or when
1367 several months have passed on the -current branch).
1369 Sometimes, obscure build problems are the result of environment
1370 poisoning. This can happen because the make utility reads its
1371 environment when searching for values for global variables. To run
1372 your build attempts in an "environmental clean room", prefix all make
1373 commands with 'env -i '. See the env(1) manual page for more details.
1375 When upgrading from one major version to another it is generally best
1376 to upgrade to the latest code in the currently installed branch first,
1377 then do an upgrade to the new branch. This is the best-tested upgrade
1378 path, and has the highest probability of being successful. Please try
1379 this approach before reporting problems with a major version upgrade.
1383 When upgrading the boot ZFS pool to a new version, always follow
1386 1.) recompile and reinstall the ZFS boot loader and boot block
1387 (this is part of "make buildworld" and "make installworld")
1389 2.) update the ZFS boot block on your boot drive
1391 The following example updates the ZFS boot block on the first
1392 partition (freebsd-boot) of a GPT partitioned drive ad0:
1393 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1395 Non-boot pools do not need these updates.
1399 If you are updating from a prior version of FreeBSD (even one just
1400 a few days old), you should follow this procedure. It is the most
1401 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1403 make kernel-toolchain
1404 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1405 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1407 To test a kernel once
1408 ---------------------
1409 If you just want to boot a kernel once (because you are not sure
1410 if it works, or if you want to boot a known bad kernel to provide
1411 debugging information) run
1412 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1413 nextboot -k testkernel
1415 To just build a kernel when you know that it won't mess you up
1416 --------------------------------------------------------------
1417 This assumes you are already running a CURRENT system. Replace
1418 ${arch} with the architecture of your machine (e.g. "i386",
1419 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1421 cd src/sys/${arch}/conf
1422 config KERNEL_NAME_HERE
1423 cd ../compile/KERNEL_NAME_HERE
1428 If this fails, go to the "To build a kernel" section.
1430 To rebuild everything and install it on the current system.
1431 -----------------------------------------------------------
1432 # Note: sometimes if you are running current you gotta do more than
1433 # is listed here if you are upgrading from a really old current.
1435 <make sure you have good level 0 dumps>
1437 make kernel KERNCONF=YOUR_KERNEL_HERE
1439 <reboot in single user> [3]
1447 To cross-install current onto a separate partition
1448 --------------------------------------------------
1449 # In this approach we use a separate partition to hold
1450 # current's root, 'usr', and 'var' directories. A partition
1451 # holding "/", "/usr" and "/var" should be about 2GB in
1454 <make sure you have good level 0 dumps>
1457 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1458 <maybe newfs current's root partition>
1459 <mount current's root partition on directory ${CURRENT_ROOT}>
1460 make installworld DESTDIR=${CURRENT_ROOT}
1461 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1462 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1463 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1464 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1465 <reboot into current>
1466 <do a "native" rebuild/install as described in the previous section>
1467 <maybe install compatibility libraries from ports/misc/compat*>
1471 To upgrade in-place from 8.x-stable to current
1472 ----------------------------------------------
1473 <make sure you have good level 0 dumps>
1475 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1477 <reboot in single user> [3]
1484 Make sure that you've read the UPDATING file to understand the
1485 tweaks to various things you need. At this point in the life
1486 cycle of current, things change often and you are on your own
1487 to cope. The defaults can also change, so please read ALL of
1488 the UPDATING entries.
1490 Also, if you are tracking -current, you must be subscribed to
1491 freebsd-current@freebsd.org. Make sure that before you update
1492 your sources that you have read and understood all the recent
1493 messages there. If in doubt, please track -stable which has
1494 much fewer pitfalls.
1496 [1] If you have third party modules, such as vmware, you
1497 should disable them at this point so they don't crash your
1500 [3] From the bootblocks, boot -s, and then do
1505 adjkerntz -i # if CMOS is wall time
1506 Also, when doing a major release upgrade, it is required that
1507 you boot into single user mode to do the installworld.
1509 [4] Note: This step is non-optional. Failure to do this step
1510 can result in a significant reduction in the functionality of the
1511 system. Attempting to do it by hand is not recommended and those
1512 that pursue this avenue should read this file carefully, as well
1513 as the archives of freebsd-current and freebsd-hackers mailing lists
1514 for potential gotchas. The -U option is also useful to consider.
1515 See mergemaster(8) for more information.
1517 [5] Usually this step is a noop. However, from time to time
1518 you may need to do this if you get unknown user in the following
1519 step. It never hurts to do it all the time. You may need to
1520 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1521 install) after the buildworld before this step if you last updated
1522 from current before 20020224 or from -stable before 20020408.
1524 [6] This only deletes old files and directories. Old libraries
1525 can be deleted by "make delete-old-libs", but you have to make
1526 sure that no program is using those libraries anymore.
1528 [8] In order to have a kernel that can run the 4.x binaries needed to
1529 do an installworld, you must include the COMPAT_FREEBSD4 option in
1530 your kernel. Failure to do so may leave you with a system that is
1531 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1532 required to run the 5.x binaries on more recent kernels. And so on
1533 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1535 Make sure that you merge any new devices from GENERIC since the
1536 last time you updated your kernel config file.
1538 [9] When checking out sources, you must include the -P flag to have
1539 cvs prune empty directories.
1541 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1542 "?=" instead of the "=" assignment operator, so that buildworld can
1543 override the CPUTYPE if it needs to.
1545 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1546 not on the command line, or in /etc/make.conf. buildworld will
1547 warn if it is improperly defined.
1550 This file contains a list, in reverse chronological order, of major
1551 breakages in tracking -current. It is not guaranteed to be a complete
1552 list of such breakages, and only contains entries since October 10, 2007.
1553 If you need to see UPDATING entries from before that date, you will need
1554 to fetch an UPDATING file from an older FreeBSD release.
1556 Copyright information:
1558 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1560 Redistribution, publication, translation and use, with or without
1561 modification, in full or in part, in any form or format of this
1562 document are permitted without further permission from the author.
1564 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1565 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1566 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1567 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1568 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1569 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1570 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1571 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1572 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1573 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1574 POSSIBILITY OF SUCH DAMAGE.
1576 Contact Warner Losh if you have any questions about your use of