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.
13 Now unix domain sockets behave "as expected" on nullfs(5). Previously
14 nullfs(5) did not pass through all behaviours to the underlying layer,
15 as a result if we bound to a socket on the lower layer we could connect
16 only to the lower path; if we bound to the upper layer we could connect
17 only to the upper path. The new behavior is one can connect to both the
18 lower and the upper paths regardless what layer path one binds to.
21 The acpi_wmi(4) status device /dev/wmistat has been renamed to
25 A new VOP_ADVISE() was added to support posix_fadvise(2). All
26 filesystem modules must be recompiled.
29 The interface of the VOP_VPTOCNP(9) changed, now the returned
30 vnode shall be referenced, previously it was required to be
31 only held. All in-tree filesystems are converted.
37 The broken amd(4) driver has been replaced with esp(4) in the amd64,
38 i386 and pc98 GENERIC kernel configuration files.
41 This commit modifies vfs_register() so that it uses a hash
42 calculation to set vfc_typenum, which is enabled by default.
43 The first time a system is booted after this change, the
44 vfc_typenum values will change for all file systems. The
45 main effect of this is a change to the NFS server file handles
46 for file systems that use vfc_typenum in their fsid, such as ZFS.
47 It will, however, prevent vfc_typenum from changing when file
48 systems are loaded in a different order for subsequent reboots.
49 To disable this, you can set vfs.typenumhash=0 in /boot/loader.conf
50 until you are ready to remount all NFS clients after a reboot.
53 Bump the shared library version numbers for libraries that
54 do not use symbol versioning, have changed the ABI compared
55 to stable/8 and which shared library version was not bumped.
56 Done as part of 9.0-RELEASE cycle.
59 During the merge of Capsicum features, the fget(9) KPI was modified.
60 This may require the rebuilding of out-of-tree device drivers --
61 issues have been reported specifically with the nVidia device driver.
62 __FreeBSD_version is bumped to 900041.
64 Also, there is a period between 20110811 and 20110814 where the
65 special devices /dev/{stdin,stdout,stderr} did not work correctly.
66 Building world from a kernel during that window may not work.
69 The packet filter (pf) code has been updated to OpenBSD 4.5.
70 You need to update userland tools to be in sync with kernel.
71 This update breaks backward compatibility with earlier pfsync(4)
72 versions. Care must be taken when updating redundant firewall setups.
75 The following sysctls and tunables are retired on x86 platforms:
77 machdep.hlt_logical_cpus
78 The following sysctl is retired:
79 machdep.hyperthreading_allowed
80 The sysctls were supposed to provide a way to dynamically offline and
81 online selected CPUs on x86 platforms, but the implementation has not
82 been reliable especially with SCHED_ULE scheduler.
83 machdep.hyperthreading_allowed tunable is still available to ignore
84 hyperthreading CPUs at OS level.
85 Individual CPUs can be disabled using hint.lapic.X.disabled tunable,
86 where X is an APIC ID of a CPU. Be advised, though, that disabling
87 CPUs in non-uniform fashion will result in non-uniform topology and
88 may lead to sub-optimal system performance with SCHED_ULE, which is
92 cpumask_t type is retired and cpuset_t is used in order to describe
96 Changes to ifconfig(8) for dynamic address family detection mandate
97 that you are running a kernel of 20110525 or later. Make sure to
98 follow the update procedure to boot a new kernel before installing
102 Support for sun4v architecture is officially dropped
105 Several KPI breaking changes have been committed to the mii(4) layer,
106 the PHY drivers and consequently some Ethernet drivers using mii(4).
107 This means that miibus.ko and the modules of the affected Ethernet
108 drivers need to be recompiled.
110 Note to kernel developers: Given that the OUI bit reversion problem
111 was fixed as part of these changes all mii(4) commits related to OUIs,
112 i.e. to sys/dev/mii/miidevs, PHY driver probing and vendor specific
113 handling, no longer can be merged verbatim to stable/8 and previous
117 Users of the Atheros AR71xx SoC code now need to add 'device ar71xx_pci'
118 into their kernel configurations along with 'device pci'.
121 The default NFS client is now the new NFS client, so fstype "newnfs"
122 is now "nfs" and the regular/old NFS client is now fstype "oldnfs".
123 Although mounts via fstype "nfs" will usually work without userland
124 changes, it is recommended that the mount(8) and mount_nfs(8)
125 commands be rebuilt from sources and that a link to mount_nfs called
126 mount_oldnfs be created. The new client is compiled into the
127 kernel with "options NFSCL" and this is needed for diskless root
128 file systems. The GENERIC kernel configs have been changed to use
129 NFSCL and NFSD (the new server) instead of NFSCLIENT and NFSSERVER.
130 To use the regular/old client, you can "mount -t oldnfs ...". For
131 a diskless root file system, you must also include a line like:
133 vfs.root.mountfrom="oldnfs:"
135 in the boot/loader.conf on the root fs on the NFS server to make
136 a diskless root fs use the old client.
139 The GENERIC kernels for all architectures now default to the new
140 CAM-based ATA stack. It means that all legacy ATA drivers were
141 removed and replaced by respective CAM drivers. If you are using
142 ATA device names in /etc/fstab or other places, make sure to update
143 them respectively (adX -> adaY, acdX -> cdY, afdX -> daY, astX -> saY,
144 where 'Y's are the sequential numbers starting from zero for each type
145 in order of detection, unless configured otherwise with tunables,
146 see cam(4)). There will be symbolic links created in /dev/ to map
147 old adX devices to the respective adaY. They should provide basic
148 compatibility for file systems mounting in most cases, but they do
149 not support old user-level APIs and do not have respective providers
150 in GEOM. Consider using updated management tools with new device names.
152 It is possible to load devices ahci, ata, siis and mvs as modules,
153 but option ATA_CAM should remain in kernel configuration to make ata
154 module work as CAM driver supporting legacy ATA controllers. Device ata
155 still can be used in modular fashion (atacore + ...). Modules atadisk
156 and atapi* are not used and won't affect operation in ATA_CAM mode.
157 Note that to use CAM-based ATA kernel should include CAM devices
158 scbus, pass, da (or explicitly ada), cd and optionally others. All of
159 them are parts of the cam module.
161 ataraid(4) functionality is now supported by the RAID GEOM class.
162 To use it you can load geom_raid kernel module and use graid(8) tool
163 for management. Instead of /dev/arX device names, use /dev/raid/rX.
165 No kernel config options or code have been removed, so if a problem
166 arises, please report it and optionally revert to the old ATA stack.
167 In order to do it you can remove from the kernel config:
172 , and instead add back:
173 device atadisk # ATA disk drives
174 device ataraid # ATA RAID drives
175 device atapicd # ATAPI CDROM drives
176 device atapifd # ATAPI floppy drives
177 device atapist # ATAPI tape drives
180 The default NFS server has been changed to the new server, which
181 was referred to as the experimental server. If you need to switch
182 back to the old NFS server, you must now put the "-o" option on
183 both the mountd and nfsd commands. This can be done using the
184 mountd_flags and nfs_server_flags rc.conf variables until an
185 update to the rc scripts is committed, which is coming soon.
188 The GNU Objective-C runtime library (libobjc), and other Objective-C
189 related components have been removed from the base system. If you
190 require an Objective-C library, please use one of the available ports.
193 ath(4) has been split into bus- and device- modules. if_ath contains
194 the HAL, the TX rate control and the network device code. if_ath_pci
195 contains the PCI bus glue. For Atheros MIPS embedded systems, if_ath_ahb
196 contains the AHB glue. Users need to load both if_ath_pci and if_ath
197 in order to use ath on everything else.
199 TO REPEAT: if_ath_ahb is not needed for normal users. Normal users only
200 need to load if_ath and if_ath_pci for ath(4) operation.
203 As part of the replacement of sysinstall, the process of building
204 release media has changed significantly. For details, please re-read
205 release(7), which has been updated to reflect the new build process.
208 GNU binutils 2.17.50 (as of 2007-07-03) has been merged to -HEAD. This
209 is the last available version under GPLv2. It brings a number of new
210 features, such as support for newer x86 CPU's (with SSE-3, SSSE-3, SSE
211 4.1 and SSE 4.2), better support for powerpc64, a number of new
212 directives, and lots of other small improvements. See the ChangeLog
213 file in contrib/binutils for the full details.
216 IPsec's HMAC_SHA256-512 support has been fixed to be RFC4868
217 compliant, and will now use half of hash for authentication.
218 This will break interoperability with all stacks (including all
219 actual FreeBSD versions) who implement
220 draft-ietf-ipsec-ciph-sha-256-00 (they use 96 bits of hash for
222 The only workaround with such peers is to use another HMAC
223 algorithm for IPsec ("phase 2") authentication.
226 Remove the uio_yield prototype and symbol. This function has
227 been misnamed since it was introduced and should not be
228 globally exposed with this name. The equivalent functionality
229 is now available using kern_yield(curthread->td_user_pri).
230 The function remains undocumented.
233 A SYSCTL_[ADD_]UQUAD was added for unsigned uint64_t pointers,
234 symmetric with the existing SYSCTL_[ADD_]QUAD. Type checking
235 for scalar sysctls is defined but disabled. Code that needs
236 UQUAD to pass the type checking that must compile on older
237 systems where the define is not present can check against
238 __FreeBSD_version >= 900030.
240 The system dialog(1) has been replaced with a new version previously
241 in ports as devel/cdialog. dialog(1) is mostly command-line compatible
242 with the previous version, but the libdialog associated with it has
243 a largely incompatible API. As such, the original version of libdialog
244 will be kept temporarily as libodialog, until its base system consumers
245 are replaced or updated. Bump __FreeBSD_version to 900030.
248 If you are trying to run make universe on a -stable system, and you get
249 the following warning:
250 "Makefile", line 356: "Target architecture for i386/conf/GENERIC
251 unknown. config(8) likely too old."
252 or something similar to it, then you must upgrade your -stable system
253 to 8.2-Release or newer (really, any time after r210146 7/15/2010 in
254 stable/8) or build the config from the latest stable/8 branch and
255 install it on your system.
257 Prior to this date, building a current universe on 8-stable system from
258 between 7/15/2010 and 1/2/2011 would result in a weird shell parsing
259 error in the first kernel build phase. A new config on those old
260 systems will fix that problem for older versions of -current.
263 The TCP stack has been modified to allow Khelp modules to interact with
264 it via helper hook points and store per-connection data in the TCP
265 control block. Bump __FreeBSD_version to 900029. User space tools that
266 rely on the size of struct tcpcb in tcp_var.h (e.g. sockstat) need to
270 Generic IEEE 802.3 annex 31B full duplex flow control support has been
271 added to mii(4) and bge(4), bce(4), msk(4), nfe(4) and stge(4) along
272 with brgphy(4), e1000phy(4) as well as ip1000phy() have been converted
273 to take advantage of it instead of using custom implementations. This
274 means that these drivers now no longer unconditionally advertise
275 support for flow control but only do so if flow control is a selected
276 media option. This was implemented in the generic support that way in
277 order to allow flow control to be switched on and off via ifconfig(8)
278 with the PHY specific default to typically off in order to protect
279 from unwanted effects. Consequently, if you used flow control with
280 one of the above mentioned drivers you now need to explicitly enable
282 ifconfig bge0 media auto mediaopt flowcontrol
284 Along with the above mentioned changes generic support for setting
285 1000baseT master mode also has been added and brgphy(4), ciphy(4),
286 e1000phy(4) as well as ip1000phy(4) have been converted to take
287 advantage of it. This means that these drivers now no longer take the
288 link0 parameter for selecting master mode but the master media option
289 has to be used instead, for example like in the following:
290 ifconfig bge0 media 1000baseT mediaopt full-duplex,master
292 Selection of master mode now is also available with all other PHY
293 drivers supporting 1000baseT.
296 The TCP stack has received a significant update to add support for
297 modularised congestion control and generally improve the clarity of
298 congestion control decisions. Bump __FreeBSD_version to 900025. User
299 space tools that rely on the size of struct tcpcb in tcp_var.h (e.g.
300 sockstat) need to be recompiled.
303 The man(1) utility has been replaced by a new version that no longer
304 uses /etc/manpath.config. Please consult man.conf(5) for how to
305 migrate local entries to the new format.
308 The copyright strings printed by login(1) and sshd(8) at the time of a
309 new connection have been removed to follow other operating systems and
313 A workaround for a fixed ld bug has been removed in kernel code,
314 so make sure that your system ld is built from sources after
315 revision 210245 from 2010-07-19 (r211583 if building head kernel
316 on stable/8, r211584 for stable/7; both from 2010-08-21).
317 A symptom of incorrect ld version is different addresses for
318 set_pcpu section and __start_set_pcpu symbol in kernel and/or modules.
321 The $ipv6_prefer variable in rc.conf(5) has been split into
322 $ip6addrctl_policy and $ipv6_activate_all_interfaces.
324 The $ip6addrctl_policy is a variable to choose a pre-defined
325 address selection policy set by ip6addrctl(8). A value
326 "ipv4_prefer", "ipv6_prefer" or "AUTO" can be specified. The
329 The $ipv6_activate_all_interfaces specifies whether IFDISABLED
330 flag (see an entry of 20090926) is set on an interface with no
331 corresponding $ifconfig_IF_ipv6 line. The default is "NO" for
332 security reason. If you want IPv6 link-local address on all
333 interfaces by default, set this to "YES".
335 The old ipv6_prefer="YES" is equivalent to
336 ipv6_activate_all_interfaces="YES" and
337 ip6addrctl_policy="ipv6_prefer".
340 DTrace has grown support for userland tracing. Due to this, DTrace is
341 now i386 and amd64 only.
342 dtruss(1) is now installed by default on those systems and a new
343 kernel module is needed for userland tracing: fasttrap.
344 No changes to your kernel config file are necessary to enable
345 userland tracing, but you might consider adding 'STRIP=' and
346 'CFLAGS+=-fno-omit-frame-pointer' to your make.conf if you want
347 to have informative userland stack traces in DTrace (ustack).
350 The acpi_aiboost(4) driver has been removed in favor of the new
351 aibs(4) driver. You should update your kernel configuration file.
354 BSD grep has been imported to the base system and it is built by
355 default. It is completely BSD licensed, highly GNU-compatible, uses
356 less memory than its GNU counterpart and has a small codebase.
357 However, it is slower than its GNU counterpart, which is mostly
358 noticeable for larger searches, for smaller ones it is measurable
359 but not significant. The reason is complex, the most important factor
360 is that we lack a modern and efficient regex library and GNU
361 overcomes this by optimizing the searches internally. Future work
362 on improving the regex performance is planned, for the meantime,
363 users that need better performance, can build GNU grep instead by
364 setting the WITH_GNU_GREP knob.
367 Due to the import of powerpc64 support, all existing powerpc kernel
368 configuration files must be updated with a machine directive like this:
369 machine powerpc powerpc
371 In addition, an updated config(8) is required to build powerpc kernels
375 A new version of ZFS (version 15) has been merged to -HEAD.
376 This version uses a python library for the following subcommands:
377 zfs allow, zfs unallow, zfs groupspace, zfs userspace.
378 For full functionality of these commands the following port must
379 be installed: sysutils/py-zfs
382 'vm_page's are now hashed by physical address to an array of mutexes.
383 Currently this is only used to serialize access to hold_count. Over
384 time the page queue mutex will be peeled away. This changes the size
385 of pmap on every architecture. And requires all callers of vm_page_hold
386 and vm_page_unhold to be updated.
389 WITH_CTF can now be specified in src.conf (not recommended, there
390 are some problems with static executables), make.conf (would also
391 affect ports which do not use GNU make and do not override the
392 compile targets) or in the kernel config (via "makeoptions
394 When WITH_CTF was specified there before this was silently ignored,
395 so make sure that WITH_CTF is not used in places which could lead
396 to unwanted behavior.
399 The kernel option COMPAT_IA32 has been replaced with COMPAT_FREEBSD32
400 to allow 32-bit compatibility on non-x86 platforms. All kernel
401 configurations on amd64 and ia64 platforms using these options must
402 be modified accordingly.
405 The utmp user accounting database has been replaced with utmpx,
406 the user accounting interface standardized by POSIX.
407 Unfortunately the semantics of utmp and utmpx don't match,
408 making it practically impossible to support both interfaces.
409 The user accounting database is used by tools like finger(1),
410 last(1), talk(1), w(1) and ac(8).
412 All applications in the base system use utmpx. This means only
413 local binaries (e.g. from the ports tree) may still use these
414 utmp database files. These applications must be rebuilt to make
417 After the system has been upgraded, it is safe to remove the old
418 log files (/var/run/utmp, /var/log/lastlog and /var/log/wtmp*),
419 assuming their contents is of no importance anymore. Old wtmp
420 databases can only be used by last(1) and ac(8) after they have
421 been converted to the new format using wtmpcvt(1).
424 Introduce the kernel thread "deadlock resolver" (which can be enabled
425 via the DEADLKRES option, see NOTES for more details) and the
426 sleepq_type() function for sleepqueues.
429 The rc.firewall and rc.firewall6 were unified, and
430 rc.firewall6 and rc.d/ip6fw were removed.
431 According to the removal of rc.d/ip6fw, ipv6_firewall_* rc
432 variables are obsoleted. Instead, the following new rc
433 variables are added to rc.d/ipfw:
435 firewall_client_net_ipv6, firewall_simple_iif_ipv6,
436 firewall_simple_inet_ipv6, firewall_simple_oif_ipv6,
437 firewall_simple_onet_ipv6, firewall_trusted_ipv6
439 The meanings correspond to the relevant IPv4 variables.
445 The default terminal emulation for syscons(4) has been changed
446 from cons25 to xterm on all platforms except pc98. This means
447 that the /etc/ttys file needs to be updated to ensure correct
448 operation of applications on the console.
450 The terminal emulation style can be toggled per window by using
451 vidcontrol(1)'s -T flag. The TEKEN_CONS25 kernel configuration
452 options can be used to change the compile-time default back to
455 To prevent graphical artifacts, make sure the TERM environment
456 variable is set to match the terminal emulation that is being
457 performed by syscons(4).
460 The layout of the structure ieee80211req_scan_result has changed.
461 Applications that require wireless scan results (e.g. ifconfig(8))
462 from net80211 need to be recompiled.
464 Applications such as wpa_supplicant(8) may require a full world
465 build without using NO_CLEAN in order to get synchronized with the
469 The iwn(4) driver has been updated to support the 5000 and 5150 series.
470 There's one kernel module for each firmware. Adding "device iwnfw"
471 to the kernel configuration file means including all three firmware
472 images inside the kernel. If you want to include just the one for
473 your wireless card, use the devices iwn4965fw, iwn5000fw or
477 The rc.d/network_ipv6, IPv6 configuration script has been integrated
478 into rc.d/netif. The changes are the following:
480 1. To use IPv6, simply define $ifconfig_IF_ipv6 like $ifconfig_IF
481 for IPv4. For aliases, $ifconfig_IF_aliasN should be used.
482 Note that both variables need the "inet6" keyword at the head.
484 Do not set $ipv6_network_interfaces manually if you do not
485 understand what you are doing. It is not needed in most cases.
487 $ipv6_ifconfig_IF and $ipv6_ifconfig_IF_aliasN still work, but
490 2. $ipv6_enable is obsolete. Use $ipv6_prefer and
491 "inet6 accept_rtadv" keyword in ifconfig(8) instead.
493 If you define $ipv6_enable=YES, it means $ipv6_prefer=YES and
494 all configured interfaces have "inet6 accept_rtadv" in the
495 $ifconfig_IF_ipv6. These are for backward compatibility.
497 3. A new variable $ipv6_prefer has been added. If NO, IPv6
498 functionality of interfaces with no corresponding
499 $ifconfig_IF_ipv6 is disabled by using "inet6 ifdisabled" flag,
500 and the default address selection policy of ip6addrctl(8)
501 is the IPv4-preferred one (see rc.d/ip6addrctl for more details).
502 Note that if you want to configure IPv6 functionality on the
503 disabled interfaces after boot, first you need to clear the flag by
504 using ifconfig(8) like:
506 ifconfig em0 inet6 -ifdisabled
508 If YES, the default address selection policy is set as
511 The default value of $ipv6_prefer is NO.
513 4. If your system need to receive Router Advertisement messages,
514 define "inet6 accept_rtadv" in $ifconfig_IF_ipv6. The rc(8)
515 scripts automatically invoke rtsol(8) when the interface becomes
516 UP. The Router Advertisement messages are used for SLAAC
517 (State-Less Address AutoConfiguration).
520 802.11s D3.03 support was committed. This is incompatible with the
521 previous code, which was based on D3.0.
524 A sysctl variable net.inet6.ip6.accept_rtadv now sets the default value
525 of a per-interface flag ND6_IFF_ACCEPT_RTADV, not a global knob to
526 control whether accepting Router Advertisement messages or not.
527 Also, a per-interface flag ND6_IFF_AUTO_LINKLOCAL has been added and
528 a sysctl variable net.inet6.ip6.auto_linklocal is its default value.
529 The ifconfig(8) utility now supports these flags.
532 ZFS snapshots are now mounted with MNT_IGNORE flag. Use -v option for
533 mount(8) and -a option for df(1) to see them.
536 The old tunable hw.bus.devctl_disable has been superseded by
537 hw.bus.devctl_queue. hw.bus.devctl_disable=1 in loader.conf should be
538 replaced by hw.bus.devctl_queue=0. The default for this new tunable
542 Remove the option STOP_NMI. The default action is now to use NMI only
543 for KDB via the newly introduced function stop_cpus_hard() and
544 maintain stop_cpus() to just use a normal IPI_STOP on ia32 and amd64.
547 The stable/8 branch created in subversion. This corresponds to the
548 RELENG_8 branch in CVS.
551 Bump the shared library version numbers for all libraries that do not
552 use symbol versioning as part of the 8.0-RELEASE cycle. Bump
553 __FreeBSD_version to 800105.
556 Due to changes in the implementation of virtual network stack support,
557 all network-related kernel modules must be recompiled. As this change
558 breaks the ABI, bump __FreeBSD_version to 800104.
561 The TOE interface to the TCP syncache has been modified to remove
562 struct tcpopt (<netinet/tcp_var.h>) from the ABI of the network stack.
563 The cxgb driver is the only TOE consumer affected by this change, and
564 needs to be recompiled along with the kernel. As this change breaks
565 the ABI, bump __FreeBSD_version to 800103.
568 Padding has been added to struct tcpcb, sackhint and tcpstat in
569 <netinet/tcp_var.h> to facilitate future MFCs and bug fixes whilst
570 maintaining the ABI. However, this change breaks the ABI, so bump
571 __FreeBSD_version to 800102. User space tools that rely on the size of
572 any of these structs (e.g. sockstat) need to be recompiled.
575 The NFS_LEGACYRPC option has been removed along with the old kernel
576 RPC implementation that this option selected. Kernel configurations
577 may need to be adjusted.
580 The network interface device nodes at /dev/net/<interface> have been
581 removed. All ioctl operations can be performed the normal way using
582 routing sockets. The kqueue functionality can generally be replaced
583 with routing sockets.
586 The documentation from the FreeBSD Documentation Project (Handbook,
587 FAQ, etc.) is now installed via packages by sysinstall(8) and under
588 the /usr/local/share/doc/freebsd directory instead of /usr/share/doc.
591 The ABI of various structures related to the SYSV IPC API have been
592 changed. As a result, the COMPAT_FREEBSD[456] and COMPAT_43 kernel
593 options now all require COMPAT_FREEBSD7. Bump __FreeBSD_version to
597 Layout of struct vnet has changed as routing related variables were
598 moved to their own Vimage module. Modules need to be recompiled. Bump
599 __FreeBSD_version to 800099.
602 NGROUPS_MAX and NGROUPS have been increased from 16 to 1023 and 1024
603 respectively. As long as no more than 16 groups per process are used,
604 no changes should be visible. When more than 16 groups are used, old
605 binaries may fail if they call getgroups() or getgrouplist() with
606 statically sized storage. Recompiling will work around this, but
607 applications should be modified to use dynamically allocated storage
608 for group arrays as POSIX.1-2008 does not cap an implementation's
609 number of supported groups at NGROUPS_MAX+1 as previous versions did.
611 NFS and portalfs mounts may also be affected as the list of groups is
612 truncated to 16. Users of NFS who use more than 16 groups, should
613 take care that negative group permissions are not used on the exported
614 file systems as they will not be reliable unless a GSSAPI based
615 authentication method is used.
618 The compiling option ADAPTIVE_LOCKMGRS has been introduced. This
619 option compiles in the support for adaptive spinning for lockmgrs
620 which want to enable it. The lockinit() function now accepts the flag
621 LK_ADAPTIVE in order to make the lock object subject to adaptive
622 spinning when both held in write and read mode.
625 The layout of the structure returned by IEEE80211_IOC_STA_INFO has
626 changed. User applications that use this ioctl need to be rebuilt.
629 The layout of struct thread has changed. Kernel and modules need to
633 The layout of structs ifnet, domain, protosw and vnet_net has changed.
634 Kernel modules need to be rebuilt. Bump __FreeBSD_version to 800097.
637 window(1) has been removed from the base system. It can now be
638 installed from ports. The port is called misc/window.
641 The way we are storing and accessing `routing table' entries has
642 changed. Programs reading the FIB, like netstat, need to be
646 A new netisr implementation has been added for FreeBSD 8. Network
647 file system modules, such as igmp, ipdivert, and others, should be
649 Bump __FreeBSD_version to 800096.
652 Remove the tunable/sysctl debug.mpsafevfs as its initial purpose is no
656 Add VOP_ACCESSX(9). File system modules need to be rebuilt.
657 Bump __FreeBSD_version to 800094.
660 Add mnt_xflag field to 'struct mount'. File system modules need to be
662 Bump __FreeBSD_version to 800093.
665 The compiling option ADAPTIVE_SX has been retired while it has been
666 introduced the option NO_ADAPTIVE_SX which handles the reversed logic.
667 The KPI for sx_init_flags() changes as accepting flags:
668 SX_ADAPTIVESPIN flag has been retired while the SX_NOADAPTIVE flag has
669 been introduced in order to handle the reversed logic.
670 Bump __FreeBSD_version to 800092.
673 Add support for hierarchical jails. Remove global securelevel.
674 Bump __FreeBSD_version to 800091.
677 The layout of struct vnet_net has changed, therefore modules
679 Bump __FreeBSD_version to 800090.
682 The newly imported zic(8) produces a new format in the output. Please
683 run tzsetup(8) to install the newly created data to /etc/localtime.
686 The sysctl tree for the usb stack has renamed from hw.usb2.* to
687 hw.usb.* and is now consistent again with previous releases.
690 802.11 monitor mode support was revised and driver api's were changed.
691 Drivers dependent on net80211 now support DLT_IEEE802_11_RADIO instead
692 of DLT_IEEE802_11. No user-visible data structures were changed but
693 applications that use DLT_IEEE802_11 may require changes.
694 Bump __FreeBSD_version to 800088.
697 The layout of the following structs has changed: sysctl_oid,
698 socket, ifnet, inpcbinfo, tcpcb, syncache_head, vnet_inet,
699 vnet_inet6 and vnet_ipfw. Most modules need to be rebuild or
700 panics may be experienced. World rebuild is required for
701 correctly checking networking state from userland.
702 Bump __FreeBSD_version to 800085.
705 MLDv2 and Source-Specific Multicast (SSM) have been merged
706 to the IPv6 stack. VIMAGE hooks are in but not yet used.
707 The implementation of SSM within FreeBSD's IPv6 stack closely
708 follows the IPv4 implementation.
710 For kernel developers:
712 * The most important changes are that the ip6_output() and
713 ip6_input() paths no longer take the IN6_MULTI_LOCK,
714 and this lock has been downgraded to a non-recursive mutex.
716 * As with the changes to the IPv4 stack to support SSM, filtering
717 of inbound multicast traffic must now be performed by transport
718 protocols within the IPv6 stack. This does not apply to TCP and
719 SCTP, however, it does apply to UDP in IPv6 and raw IPv6.
721 * The KPIs used by IPv6 multicast are similar to those used by
722 the IPv4 stack, with the following differences:
723 * im6o_mc_filter() is analogous to imo_multicast_filter().
724 * The legacy KAME entry points in6_joingroup and in6_leavegroup()
725 are shimmed to in6_mc_join() and in6_mc_leave() respectively.
726 * IN6_LOOKUP_MULTI() has been deprecated and removed.
727 * IPv6 relies on MLD for the DAD mechanism. KAME's internal KPIs
728 for MLDv1 have an additional 'timer' argument which is used to
729 jitter the initial membership report for the solicited-node
730 multicast membership on-link.
731 * This is not strictly needed for MLDv2, which already jitters
732 its report transmissions. However, the 'timer' argument is
733 preserved in case MLDv1 is active on the interface.
735 * The KAME linked-list based IPv6 membership implementation has
736 been refactored to use a vector similar to that used by the IPv4
738 Code which maintains a list of its own multicast memberships
739 internally, e.g. carp, has been updated to reflect the new
742 * There is a known Lock Order Reversal (LOR) due to in6_setscope()
743 acquiring the IF_AFDATA_LOCK and being called within ip6_output().
744 Whilst MLDv2 tries to avoid this otherwise benign LOR, it is an
745 implementation constraint which needs to be addressed in HEAD.
747 For application developers:
749 * The changes are broadly similar to those made for the IPv4
752 * The use of IPv4 and IPv6 multicast socket options on the same
753 socket, using mapped addresses, HAS NOT been tested or supported.
755 * There are a number of issues with the implementation of various
756 IPv6 multicast APIs which need to be resolved in the API surface
757 before the implementation is fully compatible with KAME userland
758 use, and these are mostly to do with interface index treatment.
760 * The literature available discusses the use of either the delta / ASM
761 API with setsockopt(2)/getsockopt(2), or the full-state / ASM API
762 using setsourcefilter(3)/getsourcefilter(3). For more information
763 please refer to RFC 3768, 'Socket Interface Extensions for
764 Multicast Source Filters'.
766 * Applications which use the published RFC 3678 APIs should be fine.
768 For systems administrators:
770 * The mtest(8) utility has been refactored to support IPv6, in
771 addition to IPv4. Interface addresses are no longer accepted
772 as arguments, their names must be used instead. The utility
773 will map the interface name to its first IPv4 address as
774 returned by getifaddrs(3).
776 * The ifmcstat(8) utility has also been updated to print the MLDv2
777 endpoint state and source filter lists via sysctl(3).
779 * The net.inet6.ip6.mcast.loop sysctl may be tuned to 0 to disable
780 loopback of IPv6 multicast datagrams by default; it defaults to 1
781 to preserve the existing behaviour. Disabling multicast loopback is
782 recommended for optimal system performance.
784 * The IPv6 MROUTING code has been changed to examine this sysctl
785 instead of attempting to perform a group lookup before looping
786 back forwarded datagrams.
788 Bump __FreeBSD_version to 800084.
791 Implement low-level Bluetooth HCI API.
792 Bump __FreeBSD_version to 800083.
795 The layout of struct malloc_type, used by modules to register new
796 memory allocation types, has changed. Most modules will need to
797 be rebuilt or panics may be experienced.
798 Bump __FreeBSD_version to 800081.
801 Anticipate overflowing inp_flags - add inp_flags2.
802 This changes most offsets in inpcb, so checking v4 connection
803 state will require a world rebuild.
804 Bump __FreeBSD_version to 800080.
807 Add an llentry to struct route and struct route_in6. Modules
808 embedding a struct route will need to be recompiled.
809 Bump __FreeBSD_version to 800079.
812 The size of rt_metrics_lite and by extension rtentry has changed.
813 Networking administration apps will need to be recompiled.
814 The route command now supports show as an alias for get, weighting
815 of routes, sticky and nostick flags to alter the behavior of stateful
817 Bump __FreeBSD_version to 800078.
820 Do not use Giant for kbdmux(4) locking. This is wrong and
821 apparently causing more problems than it solves. This will
822 re-open the issue where interrupt handlers may race with
823 kbdmux(4) in polling mode. Typical symptoms include (but
824 not limited to) duplicated and/or missing characters when
825 low level console functions (such as gets) are used while
826 interrupts are enabled (for example geli password prompt,
827 mountroot prompt etc.). Disabling kbdmux(4) may help.
830 The size of structs vnet_net, vnet_inet and vnet_ipfw has changed;
831 kernel modules referencing any of the above need to be recompiled.
832 Bump __FreeBSD_version to 800075.
835 GEOM_PART has become the default partition slicer for storage devices,
836 replacing GEOM_MBR, GEOM_BSD, GEOM_PC98 and GEOM_GPT slicers. It
837 introduces some changes:
839 MSDOS/EBR: the devices created from MSDOS extended partition entries
840 (EBR) can be named differently than with GEOM_MBR and are now symlinks
841 to devices with offset-based names. fstabs may need to be modified.
843 BSD: the "geometry does not match label" warning is harmless in most
844 cases but it points to problems in file system misalignment with
845 disk geometry. The "c" partition is now implicit, covers the whole
846 top-level drive and cannot be (mis)used by users.
848 General: Kernel dumps are now not allowed to be written to devices
849 whose partition types indicate they are meant to be used for file
850 systems (or, in case of MSDOS partitions, as something else than
853 Most of these changes date approximately from 200812.
856 The uscanner(4) driver has been removed from the kernel. This follows
857 Linux removing theirs in 2.6 and making libusb the default interface
861 The multicast forwarding code has been cleaned up. netstat(1)
862 only relies on KVM now for printing bandwidth upcall meters.
863 The IPv4 and IPv6 modules are split into ip_mroute_mod and
864 ip6_mroute_mod respectively. The config(5) options for statically
865 compiling this code remain the same, i.e. 'options MROUTING'.
868 Support for the IFF_NEEDSGIANT network interface flag has been
869 removed, which means that non-MPSAFE network device drivers are no
870 longer supported. In particular, if_ar, if_sr, and network device
871 drivers from the old (legacy) USB stack can no longer be built or
875 POSIX.1 Native Language Support (NLS) has been enabled in libc and
876 a bunch of new language catalog files have also been added.
877 This means that some common libc messages are now localized and
878 they depend on the LC_MESSAGES environmental variable.
881 The k8temp(4) driver has been renamed to amdtemp(4) since
882 support for Family 10 and Family 11 CPU families was added.
885 IGMPv3 and Source-Specific Multicast (SSM) have been merged
886 to the IPv4 stack. VIMAGE hooks are in but not yet used.
888 For kernel developers, the most important changes are that the
889 ip_output() and ip_input() paths no longer take the IN_MULTI_LOCK(),
890 and this lock has been downgraded to a non-recursive mutex.
892 Transport protocols (UDP, Raw IP) are now responsible for filtering
893 inbound multicast traffic according to group membership and source
894 filters. The imo_multicast_filter() KPI exists for this purpose.
895 Transports which do not use multicast (SCTP, TCP) already reject
896 multicast by default. Forwarding and receive performance may improve
897 as a mutex acquisition is no longer needed in the ip_input()
898 low-level input path. in_addmulti() and in_delmulti() are shimmed
899 to new KPIs which exist to support SSM in-kernel.
901 For application developers, it is recommended that loopback of
902 multicast datagrams be disabled for best performance, as this
903 will still cause the lock to be taken for each looped-back
904 datagram transmission. The net.inet.ip.mcast.loop sysctl may
905 be tuned to 0 to disable loopback by default; it defaults to 1
906 to preserve the existing behaviour.
908 For systems administrators, to obtain best performance with
909 multicast reception and multiple groups, it is always recommended
910 that a card with a suitably precise hash filter is used. Hash
911 collisions will still result in the lock being taken within the
912 transport protocol input path to check group membership.
914 If deploying FreeBSD in an environment with IGMP snooping switches,
915 it is recommended that the net.inet.igmp.sendlocal sysctl remain
916 enabled; this forces 224.0.0.0/24 group membership to be announced
919 The size of 'struct igmpstat' has changed; netstat needs to be
920 recompiled to reflect this.
921 Bump __FreeBSD_version to 800070.
924 libusb20.so.1 is now installed as libusb.so.1 and the ports system
925 updated to use it. This requires a buildworld/installworld in order to
926 update the library and dependencies (usbconfig, etc). Its advisable to
927 rebuild all ports which uses libusb. More specific directions are given
928 in the ports collection UPDATING file. Any /etc/libmap.conf entries for
929 libusb are no longer required and can be removed.
932 A workaround is committed to allow the creation of System V shared
933 memory segment of size > 2 GB on the 64-bit architectures.
934 Due to a limitation of the existing ABI, the shm_segsz member
935 of the struct shmid_ds, returned by shmctl(IPC_STAT) call is
936 wrong for large segments. Note that limits must be explicitly
937 raised to allow such segments to be created.
940 The layout of struct ifnet has changed, requiring a rebuild of all
941 network device driver modules.
944 The /dev handling for the new USB stack has changed, a
945 buildworld/installworld is required for libusb20.
948 The new USB2 stack has now been permanently moved in and all kernel and
949 module names reverted to their previous values (eg, usb, ehci, ohci,
950 ums, ...). The old usb stack can be compiled in by prefixing the name
951 with the letter 'o', the old usb modules have been removed.
952 Updating entry 20090216 for xorg and 20090215 for libmap may still
956 The rc.conf(5) option if_up_delay has been renamed to
957 defaultroute_delay to better reflect its purpose. If you have
958 customized this setting in /etc/rc.conf you need to update it to
962 xorg 7.4 wants to configure its input devices via hald which does not
963 yet work with USB2. If the keyboard/mouse does not work in xorg then
965 Option "AllowEmptyInput" "off"
966 to your ServerLayout section. This will cause X to use the configured
967 kbd and mouse sections from your xorg.conf.
970 The GENERIC kernels for all architectures now default to the new USB2
971 stack. No kernel config options or code have been removed so if a
972 problem arises please report it and optionally revert to the old USB
973 stack. If you are loading USB kernel modules or have a custom kernel
974 that includes GENERIC then ensure that usb names are also changed over,
975 eg uftdi -> usb2_serial_ftdi.
977 Older programs linked against the ports libusb 0.1 need to be
978 redirected to the new stack's libusb20. /etc/libmap.conf can
980 # Map old usb library to new one for usb2 stack
981 libusb-0.1.so.8 libusb20.so.1
984 All USB ethernet devices now attach as interfaces under the name ueN
985 (eg. ue0). This is to provide a predictable name as vendors often
986 change usb chipsets in a product without notice.
989 The ichsmb(4) driver has been changed to require SMBus slave
990 addresses be left-justified (xxxxxxx0b) rather than right-justified.
991 All of the other SMBus controller drivers require left-justified
992 slave addresses, so this change makes all the drivers provide the
996 INET6 statistics (struct ip6stat) was updated.
997 netstat(1) needs to be recompiled.
1000 NTFS has been removed from GENERIC kernel on amd64 to match
1001 GENERIC on i386. Should not cause any issues since mount_ntfs(8)
1002 will load ntfs.ko module automatically when NTFS support is
1003 actually needed, unless ntfs.ko is not installed or security
1004 level prohibits loading kernel modules. If either is the case,
1005 "options NTFS" has to be added into kernel config.
1008 TCP Appropriate Byte Counting (RFC 3465) support added to kernel.
1009 New field in struct tcpcb breaks ABI, so bump __FreeBSD_version to
1010 800061. User space tools that rely on the size of struct tcpcb in
1011 tcp_var.h (e.g. sockstat) need to be recompiled.
1014 ng_tty(4) module updated to match the new TTY subsystem.
1015 Due to API change, user-level applications must be updated.
1016 New API support added to mpd5 CVS and expected to be present
1017 in next mpd5.3 release.
1020 With __FreeBSD_version 800060 the makefs tool is part of
1021 the base system (it was a port).
1024 The afdata and ifnet locks have been changed from mutexes to
1025 rwlocks, network modules will need to be re-compiled.
1028 __FreeBSD_version 800059 incorporates the new arp-v2 rewrite.
1029 RTF_CLONING, RTF_LLINFO and RTF_WASCLONED flags are eliminated.
1030 The new code reduced struct rtentry{} by 16 bytes on 32-bit
1031 architecture and 40 bytes on 64-bit architecture. The userland
1032 applications "arp" and "ndp" have been updated accordingly.
1033 The output from "netstat -r" shows only routing entries and
1034 none of the L2 information.
1037 __FreeBSD_version 800057 marks the switchover from the
1038 binary ath hal to source code. Users must add the line:
1040 options AH_SUPPORT_AR5416
1042 to their kernel config files when specifying:
1046 The ath_hal module no longer exists; the code is now compiled
1047 together with the driver in the ath module. It is now
1048 possible to tailor chip support (i.e. reduce the set of chips
1049 and thereby the code size); consult ath_hal(4) for details.
1052 __FreeBSD_version 800054 adds memory barriers to
1053 <machine/atomic.h>, new interfaces to ifnet to facilitate
1054 multiple hardware transmit queues for cards that support
1055 them, and a lock-less ring-buffer implementation to
1056 enable drivers to more efficiently manage queueing of
1060 A new version of ZFS (version 13) has been merged to -HEAD.
1061 This version has zpool attribute "listsnapshots" off by
1062 default, which means "zfs list" does not show snapshots,
1063 and is the same as Solaris behavior.
1066 dummynet(4) ABI has changed. ipfw(8) needs to be recompiled.
1069 The uhci, ohci, ehci and slhci USB Host controller drivers have
1070 been put into separate modules. If you load the usb module
1071 separately through loader.conf you will need to load the
1072 appropriate *hci module as well. E.g. for a UHCI-based USB 2.0
1073 controller add the following to loader.conf:
1079 The ABI used by the PMC toolset has changed. Please keep
1080 userland (libpmc(3)) and the kernel module (hwpmc(4)) in
1084 atapci kernel module now includes only generic PCI ATA
1085 driver. AHCI driver moved to ataahci kernel module.
1086 All vendor-specific code moved into separate kernel modules:
1087 ataacard, ataacerlabs, ataadaptec, ataamd, ataati, atacenatek,
1088 atacypress, atacyrix, atahighpoint, ataintel, ataite, atajmicron,
1089 atamarvell, atamicron, atanational, atanetcell, atanvidia,
1090 atapromise, ataserverworks, atasiliconimage, atasis, atavia
1093 The TTY subsystem of the kernel has been replaced by a new
1094 implementation, which provides better scalability and an
1095 improved driver model. Most common drivers have been migrated to
1096 the new TTY subsystem, while others have not. The following
1097 drivers have not yet been ported to the new TTY layer:
1100 cy, digi, rc, rp, sio
1106 ng_h4, ng_tty, ppp, sl, snp
1108 Adding these drivers to your kernel configuration file shall
1109 cause compilation to fail.
1112 ntpd has been upgraded to 4.2.4p5.
1115 OpenSSH has been upgraded to 5.1p1.
1117 For many years, FreeBSD's version of OpenSSH preferred DSA
1118 over RSA for host and user authentication keys. With this
1119 upgrade, we've switched to the vendor's default of RSA over
1120 DSA. This may cause upgraded clients to warn about unknown
1121 host keys even for previously known hosts. Users should
1122 follow the usual procedure for verifying host keys before
1123 accepting the RSA key.
1125 This can be circumvented by setting the "HostKeyAlgorithms"
1126 option to "ssh-dss,ssh-rsa" in ~/.ssh/config or on the ssh
1129 Please note that the sequence of keys offered for
1130 authentication has been changed as well. You may want to
1131 specify IdentityFile in a different order to revert this
1135 The sio(4) driver has been removed from the i386 and amd64
1136 kernel configuration files. This means uart(4) is now the
1137 default serial port driver on those platforms as well.
1139 To prevent collisions with the sio(4) driver, the uart(4) driver
1140 uses different names for its device nodes. This means the
1141 onboard serial port will now most likely be called "ttyu0"
1142 instead of "ttyd0". You may need to reconfigure applications to
1143 use the new device names.
1145 When using the serial port as a boot console, be sure to update
1146 /boot/device.hints and /etc/ttys before booting the new kernel.
1147 If you forget to do so, you can still manually specify the hints
1148 at the loader prompt:
1150 set hint.uart.0.at="isa"
1151 set hint.uart.0.port="0x3F8"
1152 set hint.uart.0.flags="0x10"
1153 set hint.uart.0.irq="4"
1157 The gpt(8) utility has been removed. Use gpart(8) to partition
1161 The version that Linuxulator emulates was changed from 2.4.2
1162 to 2.6.16. If you experience any problems with Linux binaries
1163 please try to set sysctl compat.linux.osrelease to 2.4.2 and
1164 if it fixes the problem contact emulation mailing list.
1167 ISDN4BSD (I4B) was removed from the src tree. You may need to
1168 update a your kernel configuration and remove relevant entries.
1171 I have checked in code to support multiple routing tables.
1172 See the man pages setfib(1) and setfib(2).
1173 This is a hopefully backwards compatible version,
1174 but to make use of it you need to compile your kernel
1175 with options ROUTETABLES=2 (or more up to 16).
1178 The 802.11 wireless support was redone to enable multi-bss
1179 operation on devices that are capable. The underlying device
1180 is no longer used directly but instead wlanX devices are
1181 cloned with ifconfig. This requires changes to rc.conf files.
1182 For example, change:
1183 ifconfig_ath0="WPA DHCP"
1186 ifconfig_wlan0="WPA DHCP"
1187 see rc.conf(5) for more details. In addition, mergemaster of
1188 /etc/rc.d is highly recommended. Simultaneous update of userland
1189 and kernel wouldn't hurt either.
1191 As part of the multi-bss changes the wlan_scan_ap and wlan_scan_sta
1192 modules were merged into the base wlan module. All references
1193 to these modules (e.g. in kernel config files) must be removed.
1196 psm(4) has gained write(2) support in native operation level.
1197 Arbitrary commands can be written to /dev/psm%d and status can
1198 be read back from it. Therefore, an application is responsible
1199 for status validation and error recovery. It is a no-op in
1200 other operation levels.
1203 Support for KSE threading has been removed from the kernel. To
1204 run legacy applications linked against KSE libmap.conf may
1205 be used. The following libmap.conf may be used to ensure
1206 compatibility with any prior release:
1208 libpthread.so.1 libthr.so.1
1209 libpthread.so.2 libthr.so.2
1210 libkse.so.3 libthr.so.3
1213 The layout of struct vmspace has changed. This affects libkvm
1214 and any executables that link against libkvm and use the
1215 kvm_getprocs() function. In particular, but not exclusively,
1216 it affects ps(1), fstat(1), pkill(1), systat(1), top(1) and w(1).
1217 The effects are minimal, but it's advisable to upgrade world
1221 The latest em driver no longer has support in it for the
1222 82575 adapter, this is now moved to the igb driver. The
1223 split was done to make new features that are incompatible
1224 with older hardware easier to do.
1227 The new geom_lvm(4) geom class has been renamed to geom_linux_lvm(4),
1228 likewise the kernel option is now GEOM_LINUX_LVM.
1231 The default NFS mount mode has changed from UDP to TCP for
1232 increased reliability. If you rely on (insecurely) NFS
1233 mounting across a firewall you may need to update your
1237 Belatedly note the addition of m_collapse for compacting
1241 The fts(3) structures have been changed to use adequate
1242 integer types for their members and so to be able to cope
1243 with huge file trees. The old fts(3) ABI is preserved
1244 through symbol versioning in libc, so third-party binaries
1245 using fts(3) should still work, although they will not take
1246 advantage of the extended types. At the same time, some
1247 third-party software might fail to build after this change
1248 due to unportable assumptions made in its source code about
1249 fts(3) structure members. Such software should be fixed
1250 by its vendor or, in the worst case, in the ports tree.
1251 FreeBSD_version 800015 marks this change for the unlikely
1252 case that a portable fix is impossible.
1255 To upgrade to -current after this date, you must be running
1256 FreeBSD not older than 6.0-RELEASE. Upgrading to -current
1257 from 5.x now requires a stop over at RELENG_6 or RELENG_7 systems.
1260 The ADAPTIVE_GIANT kernel option has been retired because its
1261 functionality is the default now.
1264 The AT keyboard emulation of sunkbd(4) has been turned on
1265 by default. In order to make the special symbols of the Sun
1266 keyboards driven by sunkbd(4) work under X these now have
1267 to be configured the same way as Sun USB keyboards driven
1268 by ukbd(4) (which also does AT keyboard emulation), f.e.:
1270 Option "XkbLayout" "us"
1271 Option "XkbRules" "xorg"
1272 Option "XkbSymbols" "pc(pc105)+sun_vndr/usb(sun_usb)+us"
1275 It has been decided that it is desirable to provide ABI
1276 backwards compatibility to the FreeBSD 4/5/6 versions of the
1277 PCIOCGETCONF, PCIOCREAD and PCIOCWRITE IOCTLs, which was
1278 broken with the introduction of PCI domain support (see the
1279 20070930 entry). Unfortunately, this required the ABI of
1280 PCIOCGETCONF to be broken again in order to be able to
1281 provide backwards compatibility to the old version of that
1282 IOCTL. Thus consumers of PCIOCGETCONF have to be recompiled
1283 again. As for prominent ports this affects neither pciutils
1284 nor xorg-server this time, the hal port needs to be rebuilt
1288 The misnamed kthread_create() and friends have been renamed
1289 to kproc_create() etc. Many of the callers already
1290 used kproc_start()..
1291 I will return kthread_create() and friends in a while
1292 with implementations that actually create threads, not procs.
1293 Renaming corresponds with version 800002.
1302 Avoid using make -j when upgrading. While generally safe, there are
1303 sometimes problems using -j to upgrade. If your upgrade fails with
1304 -j, please try again without -j. From time to time in the past there
1305 have been problems using -j with buildworld and/or installworld. This
1306 is especially true when upgrading between "distant" versions (eg one
1307 that cross a major release boundary or several minor releases, or when
1308 several months have passed on the -current branch).
1310 Sometimes, obscure build problems are the result of environment
1311 poisoning. This can happen because the make utility reads its
1312 environment when searching for values for global variables. To run
1313 your build attempts in an "environmental clean room", prefix all make
1314 commands with 'env -i '. See the env(1) manual page for more details.
1316 When upgrading from one major version to another it is generally best
1317 to upgrade to the latest code in the currently installed branch first,
1318 then do an upgrade to the new branch. This is the best-tested upgrade
1319 path, and has the highest probability of being successful. Please try
1320 this approach before reporting problems with a major version upgrade.
1324 When upgrading the boot ZFS pool to a new version, always follow
1327 1.) recompile and reinstall the ZFS boot loader and boot block
1328 (this is part of "make buildworld" and "make installworld")
1330 2.) update the ZFS boot block on your boot drive
1332 The following example updates the ZFS boot block on the first
1333 partition (freebsd-boot) of a GPT partitioned drive ad0:
1334 "gpart bootcode -p /boot/gptzfsboot -i 1 ad0"
1336 Non-boot pools do not need these updates.
1340 If you are updating from a prior version of FreeBSD (even one just
1341 a few days old), you should follow this procedure. It is the most
1342 failsafe as it uses a /usr/obj tree with a fresh mini-buildworld,
1344 make kernel-toolchain
1345 make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE
1346 make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE
1348 To test a kernel once
1349 ---------------------
1350 If you just want to boot a kernel once (because you are not sure
1351 if it works, or if you want to boot a known bad kernel to provide
1352 debugging information) run
1353 make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel
1354 nextboot -k testkernel
1356 To just build a kernel when you know that it won't mess you up
1357 --------------------------------------------------------------
1358 This assumes you are already running a CURRENT system. Replace
1359 ${arch} with the architecture of your machine (e.g. "i386",
1360 "arm", "amd64", "ia64", "pc98", "sparc64", "powerpc", "mips", etc).
1362 cd src/sys/${arch}/conf
1363 config KERNEL_NAME_HERE
1364 cd ../compile/KERNEL_NAME_HERE
1369 If this fails, go to the "To build a kernel" section.
1371 To rebuild everything and install it on the current system.
1372 -----------------------------------------------------------
1373 # Note: sometimes if you are running current you gotta do more than
1374 # is listed here if you are upgrading from a really old current.
1376 <make sure you have good level 0 dumps>
1378 make kernel KERNCONF=YOUR_KERNEL_HERE
1380 <reboot in single user> [3]
1388 To cross-install current onto a separate partition
1389 --------------------------------------------------
1390 # In this approach we use a separate partition to hold
1391 # current's root, 'usr', and 'var' directories. A partition
1392 # holding "/", "/usr" and "/var" should be about 2GB in
1395 <make sure you have good level 0 dumps>
1398 make buildkernel KERNCONF=YOUR_KERNEL_HERE
1399 <maybe newfs current's root partition>
1400 <mount current's root partition on directory ${CURRENT_ROOT}>
1401 make installworld DESTDIR=${CURRENT_ROOT}
1402 make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd
1403 make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT}
1404 cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd
1405 <edit ${CURRENT_ROOT}/etc/fstab to mount "/" from the correct partition>
1406 <reboot into current>
1407 <do a "native" rebuild/install as described in the previous section>
1408 <maybe install compatibility libraries from ports/misc/compat*>
1412 To upgrade in-place from 8.x-stable to current
1413 ----------------------------------------------
1414 <make sure you have good level 0 dumps>
1416 make kernel KERNCONF=YOUR_KERNEL_HERE [8]
1418 <reboot in single user> [3]
1425 Make sure that you've read the UPDATING file to understand the
1426 tweaks to various things you need. At this point in the life
1427 cycle of current, things change often and you are on your own
1428 to cope. The defaults can also change, so please read ALL of
1429 the UPDATING entries.
1431 Also, if you are tracking -current, you must be subscribed to
1432 freebsd-current@freebsd.org. Make sure that before you update
1433 your sources that you have read and understood all the recent
1434 messages there. If in doubt, please track -stable which has
1435 much fewer pitfalls.
1437 [1] If you have third party modules, such as vmware, you
1438 should disable them at this point so they don't crash your
1441 [3] From the bootblocks, boot -s, and then do
1446 adjkerntz -i # if CMOS is wall time
1447 Also, when doing a major release upgrade, it is required that
1448 you boot into single user mode to do the installworld.
1450 [4] Note: This step is non-optional. Failure to do this step
1451 can result in a significant reduction in the functionality of the
1452 system. Attempting to do it by hand is not recommended and those
1453 that pursue this avenue should read this file carefully, as well
1454 as the archives of freebsd-current and freebsd-hackers mailing lists
1455 for potential gotchas. The -U option is also useful to consider.
1456 See mergemaster(8) for more information.
1458 [5] Usually this step is a noop. However, from time to time
1459 you may need to do this if you get unknown user in the following
1460 step. It never hurts to do it all the time. You may need to
1461 install a new mergemaster (cd src/usr.sbin/mergemaster && make
1462 install) after the buildworld before this step if you last updated
1463 from current before 20020224 or from -stable before 20020408.
1465 [6] This only deletes old files and directories. Old libraries
1466 can be deleted by "make delete-old-libs", but you have to make
1467 sure that no program is using those libraries anymore.
1469 [8] In order to have a kernel that can run the 4.x binaries needed to
1470 do an installworld, you must include the COMPAT_FREEBSD4 option in
1471 your kernel. Failure to do so may leave you with a system that is
1472 hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is
1473 required to run the 5.x binaries on more recent kernels. And so on
1474 for COMPAT_FREEBSD6 and COMPAT_FREEBSD7.
1476 Make sure that you merge any new devices from GENERIC since the
1477 last time you updated your kernel config file.
1479 [9] When checking out sources, you must include the -P flag to have
1480 cvs prune empty directories.
1482 If CPUTYPE is defined in your /etc/make.conf, make sure to use the
1483 "?=" instead of the "=" assignment operator, so that buildworld can
1484 override the CPUTYPE if it needs to.
1486 MAKEOBJDIRPREFIX must be defined in an environment variable, and
1487 not on the command line, or in /etc/make.conf. buildworld will
1488 warn if it is improperly defined.
1491 This file contains a list, in reverse chronological order, of major
1492 breakages in tracking -current. It is not guaranteed to be a complete
1493 list of such breakages, and only contains entries since October 10, 2007.
1494 If you need to see UPDATING entries from before that date, you will need
1495 to fetch an UPDATING file from an older FreeBSD release.
1497 Copyright information:
1499 Copyright 1998-2009 M. Warner Losh. All Rights Reserved.
1501 Redistribution, publication, translation and use, with or without
1502 modification, in full or in part, in any form or format of this
1503 document are permitted without further permission from the author.
1505 THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR
1506 IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1507 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1508 DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT,
1509 INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
1510 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1511 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1512 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
1513 STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
1514 IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1515 POSSIBILITY OF SUCH DAMAGE.
1517 Contact Warner Losh if you have any questions about your use of