add test for an attribute which contains an operator char

[elogind.git] / RELEASE-NOTES

udev 105
========
Bugfixes.

DRIVER== will match only for devices that actually have a real
driver. DRIVERS== must be used, if parent devices should be
included in the match.

Libvolume_id's "linux_raid" detection needed another fix.

udev 104
========
Bugfixes.

udev 103
========
Add additional check to volume_id detection of via_raid, cause
some company decided to put a matching pattern all over the empty
storage area of their music players.

udev 102
========
Fix path_id for SAS devices.

udev 101
========
The udev daemon can be started with --debug-trace now, which will
execute all events serialized to get a chance to catch a possible
action that crashes the box.

A warning is logged, if PHYSDEV* keys, the "device" link, or a parent
device attribute like $attr{../file} is used, only WAIT_FOR_SYSFS rules
are excluded from the warning. Referencing parent attributes directly
may break when something in the kernel driver model changes. Udev will
just find the attribute by walking up the parent chain.

Udevtrigger now sorts the list of devices depending on the device
dependency, so a "usb" device is triggered after the parent "pci"
device.

udev 100
========
Revert persistent-storage ata-serial '_' '-' replacement.

udev 099
========
Bugfixes.

Udevtrigger can now filter the list of devices to be triggered. Matches
for subsystems or sysfs attributes can be specified.

The entries in /dev/.udev/queue and /dev/.udev/failed have changed to
zero-sized files to avoid pointing to /sys and confuse broken tools which
scan the /dev directory. To retry failed events, udevtrigger --retry-failed
should be used now.

The rules and scripts to create udev rules for persistent network
devices and optical drives are in the extras/rules_generator directory
now. If you use something similar, please consider replacing your own
version with this, to share the support effort. The rule_generator
installs its own rules into /etc/udev/rules.d.

The cdrom_id tool installs its own rule now in /etc/udev/rules.d, cause
the rule_generator depends on cdrom_id to be called in an earlier rule.

udev 098
========
Bugfixes.

Renaming of some key names (the old names still work):
BUS -> SUBSYSTEMS, ID -> KERNELS, SYSFS -> ATTRS, DRIVER -> DRIVERS.
(The behavior of the key DRIVER will change soon in one of the next
releases, to match only the event device, please switch to DRIVERS
instead. If DRIVER is used, it will behave like DRIVERS, but an error
is logged.
With the new key names, we have a more consistent and simpler scheme.
We can match the properties of the event device only, with: KERNEL,
SUBSYSTEM, ATTR, DRIVER. Or include all the parent devices in the match,
with: KERNELS, SUBSYSTEMS, ATTRS, DRIVERS. ID, BUS, SYSFS, DRIVER are no
longer mentioned in the man page and should be switched in the rule
files.

ATTR{file}="value" can be used now, to write to a sysfs file of the
event device. Instead of:
  ..., SYSFS{type}=="0|7|14", RUN+="/bin/sh -c 'echo 60 > /sys$$DEVPATH/timeout'"
we now can do:
  ..., ATTR{type}=="0|7|14", ATTR{timeout}="60"

All the PHYSDEV* keys are deprecated and will be removed from a
future kernel:
  PHYDEVPATH -    is the path of a parent device and should not be
                  needed at all.
  PHYSDEVBUS -    is just a SUBSYSTEM value of a parent, and can be
                  matched with SUBSYSTEMS==
  PHYSDEVDRIVER - for bus devices it is available as ENV{DRIVER}.
                  Newer kernels will have DRIVER in the environment,
                  for older kernels udev puts in. Class device will
                  no longer carry this property of a parent and
                  DRIVERS== can be used to match such a parent value.
Note that ENV{DRIVER} is only available for a few bus devices, where
the driver is already bound at device event time. On coldplug, the
events for a lot devices are already bound to a driver, and they will have
that value set. But on hotplug, at the time the kernel creates the device,
it can't know what driver may claim the device after that, therefore
in most cases it will be empty.

Failed events should now be re-triggered with:
   udevtrigger --retry-failed.
Please switch to this command, so we keep the details of the /dev/.udev/failed/
files private to the udev tools. We may need to switch the current symlink
target, cause some obviously broken tools try to scan all files in /dev
including /dev/.udev/, find the links to /sys and end up stat()'ing sysfs files
million times. This takes ages on slow boxes.

The udevinfo attribute walk (-a) now works with giving a device node
name (-n) instead of a devpath (-p). The query now always works, also when
no database file was created by udev.

The built-in /etc/passwd /etc/group parser is removed, we always depend on
getpwnam() and getgrnam() now. One of the next releases will depend on
fnmatch() and may use getopt_long().

udev 097
========
Bugfixes and small improvements.

udev 096
========
Fix path_id for recent kernels.

udev 095
========
%e is finally gone.

Added support for swapping network interface names, by temporarily
renaming the device and wait for the target name to become free.

udev 094
========
The built-in MODALIAS key and substitution is removed.

udev 093
========
The binary firmware helper is replaced by the usual simple
shell script. Udevsend is removed from the tree.

udev 092
========
Bugfix release.

udev 091
========
Some more keys require the correct use of '==' and '=' depending
on the kind of operation beeing an assignment or a match. Rules
with invalid operations are skipped and logged to syslog. Please
test with udevtest if the parsing of your rules throws errors and
fix possibly broken rules.

udev 090
========
Provide "udevsettle" to wait for all current udev events to finish.
It also watches the current kernel netlink queue by comparing the
even sequence number to make sure that there are no current pending
events that have not already arrived in the daemon.

udev 089
========
Fix rule to skip persistent rules for removable IDE devices, which
also skipped optical IDE drives.

All *_id program are installed in /lib/udev/ by default now.

No binary is stripped anymore as this should be done in the
packaging process and not at build time.

libvolume_id is provided as a shared library now and vol_id is
linked against it. Also one of the next HAL versions will require
this library, and the HAL build process will also require the
header file to be installed. The copy of the same code in HAL will
be removed to have only a single copy left on the system.

udev 088
========
Add persistent links for SCSI tapes. The rules file is renamed
to 60-persistent-storage.rules.

Create persistent path for usb devices. Can be used for all sorts
of devices that can't be distinguished by other properties like
multiple identical keyboards and mice connected to the same box.

Provide "udevtrigger" program to request events on coldplug. The
shell script is much too slow with thousends of devices.

udev 087
========
Fix persistent disk rules to exclude removable IDE drives.

Warn if %e, $modalias or MODALIAS is used.

udev 086
========
Fix queue export, which wasn't correct for subsequent add/remove
events for the same device.

udev 085
========
Fix cramfs detection on big endian.

Make WAIT_FOR_SYSFS usable in "normal" rules and silent if the whole
device goes away.

udev 084
========
If BUS== and SYSFS{}== have been used in the same rule, the sysfs
attributes were only checked at the parent device that matched the
by BUS requested subsystem. Fix it to also look at the device we
received the event for.

Build variable CROSS has changed to CROSS_COMPILE to match the kernel
build name.

udev 083
========
Fix a bug where NAME="" would prevent RUN from beeing executed.

RUN="/bin/program" does not longer automatically add the subsystem
as the first parameter. This is from the days of /sbin/hotplug
which is dead now and it's just confusing to need to add a space at
the end of the program name to prevent this.
If you use rules that need the subsystem as the first parameter,
like the old "udev_run_hotlugd" and "udev_run_devd", add the subsystem
to the key like RUN+="/bin/program $env{SUBSYSTEM}".

udev 082
========
The udev man page has moved to udev(7) as it does not describe a command
anymore. The programs udev, udevstart and udevsend are no longer installed
by default and must be copied manually, if they should be installed or
included in a package.

Fix a bug where "ignore_device" could run earlier collected RUN keys before
the ignore rule was applied.

More preparation for future sysfs changes. usb_id and scsi_id no longer
depend on a magic order of devices in the /devices chain. Specific devices
should be requested by their subsytem.

This will always find the scsi parent device without depending on a specific
path position:
  dev = sysfs_device_get(devpath);
  dev_usb = sysfs_device_get_parent_with_subsystem(dev, "scsi");

The "device" link in the current sysfs layout will be automatically
_resolved_ as a parent and in the new sysfs layout it will just _be_ the
parent in the devpath. If a device is requested by it's symlink, like all
class devices in the new sysfs layout will look like, it gets automatically
resolved and substituted with the real devpath and not the symlink path.

Note:
A similar logic must be applied to _all_ sysfs users, including
scripts, that search along parent devices in sysfs. The explicit use of
the "device" link must be avoided. With the future sysfs layout all
DEVPATH's will start with /devices/ and have a "subsystem" symlink poiting
back to the "class" or the "bus". The layout of the parent devices in
/devices is not necessarily expected to be stable across kernel releases and
searching for parents by their subsystem should make sysfs users tolerant
for changed parent chains.

udev 081
========
Prepare udev to work with the experimental kernel patch, that moves
/sys/class devices to /sys/devices and /sys/block to /sys/class/block.

Clarify BUS, ID, $id usage and fix $id behavior. This prepares for
moving the class devices to /sys/devices.

Thanks again to Marco for help finding a hopefully nice compromise
to make %b simpler and working again.

udev 080
========
Complete removal of libsysfs, replaced by simple helper functions
which are much simpler and a bit faster. The udev daemon operatesentirely
on event parameters and does not use sysfs for simple rules anymore.
Please report any new bugs/problems, that may be caused by this big
change. They will be fixed immediately.

The enumeration format character '%e' is deprecated and will be
removed sometimes from a future udev version. It never worked correctly
outside of udevstart, so we can't use it with the new parallel
coldplug. A simple enumeration is as useless as the devfs naming
scheme, just get rid of both if you still use it.

MODALIAS and $modalias is not needed and will be removed from one of
the next udev versions, replace it in all rules with ENV{MODALIAS} or
the sysfs "modalias" value.

Thanks a lot to Marco for all his help on finding and fixing bugs.

udev 079
========
Let scsi_id request libata drive serial numbers from page 0x80.

Renamed etc/udev/persistent.rules to persistent-disk.rules and
added /dev/disk/by-name/* for device mapper device names.

Removed %e from the man page. It never worked reliably outside
of udevstart and udevstart is no longer recommended to use.

udev 078
========
Symlinks are now exported to the event environment. Hopefully it's no
longer needed to run udevinfo from an event process, like it was
mentioned on the hotplug list:
  UDEV  [1134776873.702967] add@/block/sdb
  ...
  DEVNAME=/dev/sdb
  DEVLINKS=/dev/disk/by-id/usb-IBM_Memory_Key_0218B301030027E8 /dev/disk/by-path/usb-0218B301030027E8:0:0:0

udev 077
========
Fix a problem if udevsend is used as the hotplug handler and tries to use
syslog, which causes a "vc" event loop. 2.6.15 will make udevsend obsolete
and this kind of problems will hopefully go away soon.

udev 076
========
All built-in logic to work around bad sysfs timing is removed with this
version. The need to wait for sysfs files is almost fixed with a kernel
version that doesn't work with this udev version anyway. Until we fix
the timing of the "bus" link creation, the former integrated logic should
be emulated by a rule placed before all other rules:
  ACTION=="add", DEVPATH=="/devices/*", ENV{PHYSDEVBUS}=="?*", WAIT_FOR_SYSFS="bus"

The option "udev_db" does no longer exist. All udev state will be in
/$udev_root/.udev/ now, there is no longer an option to set this
to anything else.
If the init script or something else used this value, just depend on
this hardcoded path. But remember _all_content_ of this directory is
still private to udev and can change at any time.

Default location for rule sripts and helper programs is now: /lib/udev/.
Everything that is not useful on the commandline should go into this
directory. Some of the helpers in the extras folder are installed there
now. The rules need to be changed, to find the helpers there.

Also /lib/udev/devices is recommended as a directory where packages or
the user can place real device nodes, which get copied over to /dev at
every boot. This should replace the various solutions with custom config
files.

Udevsend does no longer start the udev daemon. This must be done with
the init script that prepares /dev on tmpfs and creates the initial nodes,
before starting the daemon.

udev 075
========
Silent a too verbose error logging for the old hotplug.d/ dev.d/
emulation.

The copy of klibc is removed. A systemwide installed version of klibc
should be used to build a klibc udev now.

udev 074
========
NAME="" will not create any nodes, but execute RUN keys. To completely
ignore an event the OPTION "ignore_device" should be used.

After removal of the reorder queue, events with a TIMEOUT can be executed
without any queuing now.

udev 073
========
Fixed bug in udevd, if inotify is not available. We depend on netlink
uevents now, kernels without that event source will not work with that
version of udev anymore.

udev 072
========
The rule parsing happens now in the daemon once at startup, all udev
event processes inherit the already parsed rules from the daemon.
It is shipped with SUSE10.0 and reduces heavily the system load at
startup. The option to save precompiled rules and let the udev process
pick the them up is removed, as it's no longer needed.

Kernel 2.6.15 will have symlinks at /class/input pointing to the real
device. Libsysfs is changed to "translate" the requested link into the
real device path, as it would happen with the hotplug event. Otherwise
device removal and the udev database will not work.

Using 'make STRIPCMD=' will leave the binaries unstripped for debugging
and packaging.

A few improvements for vol_id, the filesytem probing code.

udev 071
========
Fix a stupid typo in extras/run_directory for "make install".

scsi_id creates the temporary devnode now in /dev for usage with a
non-writable /tmp directory.

The uevent kernel socket buffer can carry app. 50.000 events now,
let's see who can break this again. :)

The upcoming kernel will have a new input driver core integration.
Some class devices are now symlinks to the real device. libsysfs
needs a fix for this to work correctly. Udevstart of older udev
versions will _not_ create these devices!

udev 070
========
Fix a 'install' target in the Makefile, that prevents EXTRAS from
beeing installed.

udev 069
========
A bunch of mostly trivial bugfixes. From now on no node name or
symlink name can contain any character than plain whitelisted ascii
characters or validated utf8 byte-streams. This is needed for the
/dev/disk/by-label/* links, because we import untrusted data and
export it to the filesystem.

udev 068
========
More bugfixes. If udevd was started from the kernel, we don't
have stdin/stdout/stderr, which broke the forked tools in some
situations.

udev 067
========
Bugfix. udevstart event ordering was broken for a long time.
The new run_program() uncovered it, because /dev/null was not
available while we try to run external programs.
Now udevstart should create it before we run anything.

udev 066
========
Minor bugfixes and some distro rules updates. If you don't have the
persistent disk rules in /dev/disk/by-*/* on your distro, just
grab it from here. :)

udev 065
========
We can use socket communication now to pass events from udev to
other programs:
  RUN+="socket:/org/freedesktop/hal/udev_event"
will pass the whole udev event to the HAL daemon without the need
for a forked helper. (See ChangeLog for udevmonitor, as an example)

udev 064
========
Mostly bugfixes and see ChangeLog.

The test for the existence of an environment value should be
switched from:
  ENV{KEY}=="*" to ENV{KEY}=="?*"
because "*" will not fail anymore, if the key does not exist or
is empty.

udev 063
========
Bugfixes and a few tweaks described in the ChangeLog.

udev 062
========
Mostly a Bugfix release.

Added WAIT_FOR_SYSFS="<attribute>" to be able to fight against the sysfs
timing with custom rules.

udev 061
========
We changed the  internal rule storage format. Our large rule files took
2 MB of RAM, with the change we are down to 99kB.

If the device-node has been created with default name and no symlink or
options are to remenber, it is not longer stored in the udevdb. HAL will
need to be updated to work correctly with that change.

To overrride optimization flags, OPTFLAGS may be used now.

udev 060
========
Bugfix release.

udev 059
========
Major changes happened with this release. The goal is to take over the
complete kernel-event handling and provide a more efficient way to dispatch
kernel events. Replacing most of the current shell script logic and the
kernel forked helper with a netlink-daemon and a rule-based event handling.

o udevd listens to netlink events now. The first valid netlink event
  will make udevd ignore any message from udevsend that contains a
  SEQNUM, to avoid duplicate events. The forked events can be disabled
  with:
    echo "" > /proc/sys/kernel/hotplug
  For full support, the broken input-subsytem needs to be fixed, not to
  bypass the driver core.

o /etc/dev.d/ + /etc/hotplug.d/ directory multiplexing is completely
  removed from udev itself and must be emulated by calling small
  helper binaries provided in the extras folder:
    make EXTRAS=extras/run_directory/
  will build udev_run_devd and udev_run_hotplugd, which can be called
  from a rule if needed:
    RUN+="/sbin/udev_run_hotplugd"
  The recommended way to handle this is to convert all the calls from
  the directories to explicit udev rules and get completely rid of the
  multiplexing. (To catch a ttyUSB event, you now no longer need to
  fork and exit 300 tty script instances you are not interested in, it
  is just one rule that matches exactly the device.)

o udev handles now _all_ events not just events for class and block
  devices, this way it is possible to control the complete event
  behavior with udev rules. Especially useful for rules like:
    ACTION="add", DEVPATH="/devices/*", MODALIAS=="?*", RUN+="/sbin/modprobe $modalias"

o As used in the modalias rule, udev supports now textual
  substitution placeholder along with the usual format chars. This
  needs to be documented, for now it's only visible in udev_rules_parse.c.

o The rule keys support now more operations. This is documented in the
  man page. It is possible to add values to list-keys like the SYMLINK
  and RUN list with KEY+="value" and to clear the list by assigning KEY="".
  Also "final"-assignments are supported by using KEY:="value", which will
  prevent changing the key by any later rule.

o kernel 2.6.12 has the "detached_state" attribute removed from
  sysfs, which was used to recognize sysfs population. We switched that
  to wait for the "bus" link, which is only available in kernels after 2.6.11.
  Running this udev version on older kernels may cause a short delay for
  some events.

o To provide infrastructure for persistent device naming, the id programs:
  scsi_id, vol_id (former udev_volume_id), and ata_id (new) are able now
  to export the probed data in environment key format:
    pim:~ # /sbin/ata_id --export /dev/hda
    ID_MODEL=HTS726060M9AT00
    ID_SERIAL=MRH401M4G6UM9B
    ID_REVISION=MH4OA6BA

  The following rules:
    KERNEL="hd*[!0-9]", IMPORT="/sbin/ata_id --export $tempnode"
    KERNEL="hd*[!0-9]", ENV{ID_SERIAL}=="?*", SYMLINK+="disk/by-id/$env{ID_MODEL}_$env{ID_SERIAL}"

  Will create:
    kay@pim:~> tree /dev/disk
    /dev/disk
    |-- by-id
    |   |-- HTS726060M9AT00_MRH401M4G6UM9B -> ../../hda
    |   `-- IBM-Memory_Key -> ../../sda
    |-- by-label
    |   |-- swap -> ../../hda1
    |   |-- date -> ../../sda1
    |   `-- home -> ../../hda3
    `-- by-uuid
        |-- 2E08712B0870F2E7 -> ../../hda3
        |-- 9352cfef-7687-47bc-a2a3-34cf136f72e1 -> ../../hda1
        |-- E845-7A89 -> ../../sda1
        `-- b2a61681-3812-4f13-a4ff-920d70604299 -> ../../hda2

  The IMPORT= operation will import these keys in the environment and make
  it available for later PROGRAM= and RUN= executed programs. The keys are
  also stored in the udevdb and can be queried from there with one of the
  next udev versions.

o A few binaries are silently added to the repository, which can be used
  to replay kernel events from initramfs instead of using coldplug. udevd
  can be instructed now to queue-up events while the stored events from
  initramfs are filled into the udevd-queue. This code is still under
  development and there is no documentation now besides the code itself.
  The additional binaries get compiled, but are not installed by default.

o There is also a temporary fix for a performance problem where too many
  events happen in parallel and every event needs to parse the rules.
  udev can now read precompiled rules stored on disk. This is likely to be
  replaced by a more elegant solution in a future udev version.

udev 058
========
With kernel version 2.6.12, the sysfs file "detached_state" was removed.
Fix for libsysfs not to expect this file was added.

udev 057
========
All rules are applied now, but only the first matching rule with a NAME-key
will be applied. All later rules with NAME-key are completely ignored. This
way system supplied symlinks or permissions gets applied to user-defined
naming rules.

Note:
Please check your rules setup, if you may need to add OPTIONS="last_rule"
to some rules, to keep the old behavior.

The rules are read on "remove"-events too. That makes is possible to match
with keys that are available on remove (KERNEL, SUBSYSTEM, ID, ENV, ...) to
instruct udev to ignore an event (OPTIONS="ignore_device").
The new ACTION-key may be used to let a rule act only at a "remove"-event.

The new RUN-key supports rule-based execution of programs after device-node
handling. This is meant as a general replacement for the dev.d/-directories
to give fine grained control over the execution of programs.

The %s{}-sysfs format char replacement values are searched at any of the
devices in the device chain now, not only at the class-device.

We support log priority levels now. The value udev_log in udev.conf is used
to determine what is printed to syslog. This makes it possible to
run a version with compiled-in debug messages in a production environment
which is sometimes needed to find a bug.
It is still possible to supress the inclusion of _any_ syslog usage with
USE_LOG=false to create the smallest possible binaries if needed.
The configured udev_log value can be overridden with the environment variable
UDEV_LOG.

udev 056
========
Possible use of a system-wide klibc:
  make USE_KLIBC=true KLCC=/usr/bin/klcc all
will link against an external klibc and our own version will be ignored.

udev 055
========
We support an unlimited count of symlinks now.

If USE_STATIC=true is passed to a glibc build, we link statically and use
a built-in userdb parser to resolve user and group names.

The PLACE= key is gone. It can be replaced by an ID= for a long time, because
we walk up the chain of physical devices to find a match.

The KEY="<value>" format supports '=', '==', '!=,' , '+=' now. This makes it
easy to skip certain attribute matches without composing rules with weird
character class negations like:
  KERNEL="[!s][!c][!d]*"
this can now be replaced with:
  KERNEL!="scd*"
The current simple '=' is still supported, and should work as it does today,
but existing rules should be converted if possible, to be better readable.

We have new ENV{}== key now, to match against a maximum of 5 environment
variables.

udevstart is its own binary again, because we don't need co carry this araound
with every forked event.