I had thought that keeping the reported tail level clearly different
from the module name was a good idea, but I've changed my mind.
'raid5' is better and probably less confusing than 'RAID-5'.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The raid5 stripe cache was recently changed from fixed size (NR_STRIPES) to
variable size (conf->max_nr_stripes). However there are two places that still
use the constant and as a result, reducing the size of the stripe cache can
result in a deadlock.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Who would submit code with a FIXME like that in it !!!!
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If you have an array with a write-intent-bitmap, and you remove a device, then
re-add it, a full recovery isn't needed. We detect a re-add by looking at
saved_raid_disk. For raid1, it doesn't matter which disk it was, only whether
or not it was an active device. The old code being removed set a value of
'mirror' which was then ignored, so it can go. The changed code performs the
correct check.
For raid6, if there are two missing devices, make sure we chose the right slot
on --re-add rather than always the first slot.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If an array is created using set_array_info, default_bitmap_offset isn't set
properly meaning that an internal bitmap cannot be hot-added until the array
is stopped and re-assembled.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When doing a recovery, we need to know whether the array will still be
degraded after the recovery has finished, so we can know whether bits can be
clearred yet or not. This patch performs the required check.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
bitmap_unplug actually writes data (bits) to storage, so we shouldn't be
holding a spinlock...
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
raid10 has two different layouts. One uses near-copies (so multiple
copies of a block are at the same or similar offsets of different
devices) and the other uses far-copies (so multiple copies of a block
are stored a greatly different offsets on different devices). The point
of far-copies is that it allows the first section (normally first half)
to be layed out in normal raid0 style, and thus provide raid0 sequential
read performance.
Unfortunately, the read balancing in raid10 makes some poor decisions
for far-copies arrays and you don't get the desired performance. So
turn off that bad bit of read_balance for far-copies arrays.
With this patch, read speed of an 'f2' array is comparable with a raid0
with the same number of devices, though write speed is ofcourse still
very slow.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The spinlock region_lock is held while calling mark_region which can sleep.
Drop the spinlock before calling that function.
A region's state and inclusion in the clean list are altered by rh_inc and
rh_dec. The state variable is set to RH_CLEAN in rh_dec, but only if
'pending' is zero. It is set to RH_DIRTY in rh_inc, but not if it is already
so. The changes to 'pending', the state, and the region's inclusion in the
clean list need to be atomicly.
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
bio_list_merge() should do nothing if the second list is empty - not oops.
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
do_end_io() can be called without interrupts blocked.
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The linux bitset operators (test_bit, set_bit etc) work on arrays of "unsigned
long". dm-log uses such bitsets but treats them as arrays of uint32_t, only
allocating and zeroing a multiple of 4 bytes (as 'clean_bits' is a uint32_t).
The patch below fixes this problem.
The problem is specific to 64-bit big endian machines such as s390x or ppc-64
and can prevent pvmove terminating.
In the simplest case, if "region_count" were (say) 30, then
bitset_size (below) would be 4 and bitset_uint32_count would be 1.
Thus the memory for this butset, after allocation and zeroing would
be
0 0 0 0 X X X X
On a bigendian 64bit machine, bit 0 for this bitset is in the 8th
byte! (and every bit that dm-log would use would be in the X area).
0 0 0 0 X X X X
^
here
which hasn't been cleared properly.
As the dm-raid1 code only syncs and counts regions which have a 0 in the
'sync_bits' bitset, and only finishes when it has counted high enough, a large
number of 1's among those 'X's will cause the sync to not complete.
It is worth noting that the code uses the same bitsets for in-memory and
on-disk logs. As these bitsets are host-endian and host-sized, this means
that they cannot safely be moved between computers with
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
In some circumstances the LIST_VERSIONS output is truncated because the size
calculation forgets about a 'uint32_t' in each structure - but the inclusion
of the whole of ALIGN_MASK frequently compensates for the omission.
This is a quick workaround to use an upper bound. (The code ought to be fixed
to supply the actual size.)
Running 'dmsetup targets' may demonstrate the problem: when I run it, the last
line comes out as 'erro' instead of 'error'. Consequently, 'lvcreate --type
error' doesn't work.
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
An error path in table_load() forgets to release a table that won't now be
referenced.
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
md needs to monitor the rate of requests to its devices when doing
resync/recovery so that it can back-off when there is non-resync IO. It
does this by comparing resync IO, which it counts, with total IO which is
taken from disk_stats.
disk_stats were recently changed to account sectors when a request
completes instead of when it is queued. This upsets md's calculations.
We could do the sync_io accounting at the end of requests too, but that has
problems. If an underlying device is an md array, the accounting will
still be done when the request is submitted. This could be changed for
some raid levels, but it cannot be changed for raid0 or linear without
substantial code changes.
So instead, we increase the error that is_mddev_idle allows, up to the
maximum amount of resync IO that can be in flight at any time. The
calculation is current fragile as each personality as different limits for
in-flight resync. This should be fixed up.
For now, this simple patch fixes the problem.
Increasing the error margin decreases the sensitivity to non-resync IO. To
partially compensate for this, the time to wait when non-resync IO is
detected is increased so that less steady IO is required to keep the resync
at bay.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Some filesystems go oops.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Despite the fact that md threads don't need to be signalled, and won't
respond to signals anyway, we need to have an 'interruptible' wait, else
they stay in 'D' state and add to the load average.
(akpm: the signal_pending() test is unneeded - we'll fix that up in the next
round. For now, leave it there because that's how the code used to be).
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This was marked deprecated "after 2.6" back in the 2.5 days. But now it
seems there isn't going to be any "after 2.6", and we deprecate by date
now. So set a date.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Document in Documentation/md.txt the files that now appear in sysfs, and make
a couple of small refinements to exactly when 'level' and 'raid_disks' are
empty, to make it match the documentation.
Signed-off-by: Neil Brown <neilb@suse.de>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The current sync_action for an array can be one of
idle - nothing happening
resync - reduncancy being recalcualted
recover - missing device being recoverred to spare
check - user initiated check of redundancy
repair - like resync but user-initiated and ignores
bitmap optimisation.
Each of these strings can also be written to the 'sync_action' file to cause
that action to happen (if appropriate).
While 'sync' is not technically correct, as a recovery is *not* a 'sync', I
think it is the most servicable word here. Also 'action' is a strong word
than 'mode'.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
There are a few loose ends following the conversion of md to use kthreads:
- Some fields in mdk_thread_t that aren't needed (kthreads does it's own
completion and manages it's own name).
- thread->run is now never NULL, so no need to check
- Some tests for signal_pending that aren't needed (As we don't use signals
to stop threads any more)
- Some flush_signals are not needed
- Some waits are interruptible and don't need to be.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The 'auto-readonly' flag (which suppresses resync and superblock updates until
the first write) is not meaningful for personalities that don't support resync
or superblock writes (raid0, linear, etc).
So clear the setting early to avoid it confusing anything - e.g. appearing in
/proc/mdstat
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The introduction of 'resync=PENDING' (for read-only devices) caused that
message to appear for non-syncable arrays like raid0 and linear. Simplest
thing is to not try to print any resync info unless the personality clearly
supports it.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Some, but not all, md array support data redundancy and hence support checking
and restoring that redundancy (resync, rebuild).
Some attributes apply specifically to functions involving this redundancy, and
so should only appear for md arrays for which they are meaningful. i.e. they
should not appear for raid0, linear, multpath, faulty.
This patch separates these into a distinct group and creates the group only if
the personality supports sync_request.
Signed-off-by: Neil Brown <neilb@suse.de>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
1/ I really should be using the __ATTR macros for defining attributes, so
that the .owner field get set properly, otherwise modules can be removed
while sysfs files are open. This also involves some name changes of _show
routines.
2/ Always lock the mddev (against reconfiguration) for all sysfs attribute
access. This easily avoid certain races and is completely consistant with
other interfaces (ioctl and /proc/mdstat both always lock against
reconfiguration).
3/ raid5 attributes must check that the 'conf' structure actually exists
(the array could have been stopped while an attribute file was open).
4/ A missing 'kfree' from when the raid5_conf_t was converted to have a
kobject embedded, and then converted back again.
Signed-off-by: Neil Brown <neilb@suse.de>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
A sync of raid5 usually ignore blocks which the bitmap says are in-sync. But
a user-request check or repair should not ignore these.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Raid1 currently optimises resync using the intent bitmap etc. This
optimisation is not wanted when we explicitly request a repair through sysfs,
so add appropriate checks.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If a block_device is a partition, then it's kobject is
bdev->bd_part->kobj
otherwise (if it is a full device), the kobject is
bdev->bd_disk->kobj
As md wants back-links to the correct object (whether partition or not), we
need to respect this difference... (Thus current code shows a link to the
whole device, whether we are using a partition or not, which is wrong).
Signed-off-by: Neil Brown <neilb@suse.de>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When an md array is started, the superblock will be written, and resync may
commense. This is not good if you want to be completely read-only as, for
example, when preparing to resume from a suspend-to-disk image.
So introduce a module parameter "start_ro" which can be set
to '1' at boot, at module load, or via
/sys/module/md_mod/parameters/start_ro
When this is set, new arrays get an 'auto-ro' mode, which disables all
internal io (superblock updates, resync, recovery) and is automatically
switched to 'rw' when the first write request arrives.
The array can be set to true 'ro' mode using 'mdadm -r' before the first
write request, or resync can be started without a write using 'mdadm -w'.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With version-0.90 superblock, component devices on an md device to not have
any stable name related to the array -(version-1 assigns a fixed index when
a device is added to an array, and this remains despit any hot-swap).
The intial code for making these devices appear in sysfs used dynamic
names, which would change whenever a hot-spare was swapped for a failed or
missing device. This turns out not to be practical in sysfs for a number
of reasons.
This patch changes then naming of component devices to be based on the
result of 'bdevname'. This is stable and should be unique.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
We can only accept BARRIER requests if all slaves handle
barriers, and that can, of course, change with time....
So we keep track of whether the whole array seems safe for barriers,
and also whether each individual rdev handles barriers.
We initially assumes barriers are OK.
When writing the superblock we try a barrier, and if that fails, we flag
things for no-barriers. This will usually clear the flags fairly quickly.
If writing the superblock finds that BIO_RW_BARRIER is -ENOTSUPP, we need to
resubmit, so introduce function "md_super_wait" which waits for requests to
finish, and retries ENOTSUPP requests without the barrier flag.
When writing the real raid1, write requests which were BIO_RW_BARRIER but
which aresn't supported need to be retried. So raid1d is enhanced to do this,
and when any bio write completes (i.e. no retry needed) we remove it from the
r1bio, so that devices needing retry are easy to find.
We should hardly ever get -ENOTSUPP errors when writing data to the raid.
It should only happen if:
1/ the device used to support BARRIER, but now doesn't. Few devices
change like this, though raid1 can!
or
2/ the array has no persistent superblock, so there was no opportunity to
pre-test for barriers when writing the superblock.
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Current bitmaps use set_bit et.al and so are host-endian, which means
not-portable. Oops.
Define a new version number (4) for which bitmaps are little-endian.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This has the advantage of removing the confusion caused by 'rdev_t' and
'mddev_t' both having 'in_sync' fields.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Two refinements to the 'attempt-overwrite-on-read-error' mechanism.
1/ If the array is read-only, don't attempt an over-write.
2/ If there are more than max_nr_stripes read errors on a device with
no success, fail the drive. This will make sure a dead
drive will be eventually kicked even when we aren't trying
to rewrite (which would normally kick a dead drive more quickly.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
There isn't really a need for raid5 attributes to be an a subdirectory,
so this patch moves them from
/sys/block/mdX/md/raid5/attribute
to
/sys/block/mdX/md/attribute
This suggests that all md personalities should co-operate about
namespace usage, but that shouldn't be a problem.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
1/ Use reduce stack usage, because 'gcc' apparently doesn't overlay
different variables that are in separate scopes...
2/ Use test_bit instead of ( .. & 1<< ..) which in this case is buggy.
Thanks to Andrew Morton
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With this, raid5 can be asked to check parity without repairing it. It also
keeps a count of the number of incorrect parity blocks found (mismatches) and
reports them through sysfs.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
You can trigger a 'check' with
echo check > /sys/block/mdX/md/scan_mode
or a check-and-repair errors with
echo repair > /sys/block/mdX/md/scan_mode
and read the current state from the same file.
Note: personalities need to know the different between 'check' and 'repair',
but don't yet. Until they do, 'check' will be the same as 'repair' and will
just do a normal resync pass.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
/sys/block/mdX/md/raid5/
contains raid5-related attributes.
Currently
stripe_cache_size
is number of entries in stripe cache, and is settable.
stripe_cache_active
is number of active entries, and in only readable.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Each device in an md array how has a corresponding
/sys/block/mdX/md/devNN/
directory which can contain attributes. Currently there is only 'state' which
summarises the state, nd 'super' which has a copy of the superblock, and
'block' which is a symlink to the block device.
Also, /sys/block/mdX/md/rdNN represents slot 'NN' in the array, and is a
symlink to the relevant 'devNN'. Obviously spare devices do not have a slot
in the array, and so don't have such a symlink.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Start using kobjects in mddevs, and provide a couple of simple attributes
(level and disks). Attributes live in
/sys/block/mdX/md/attr-name
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch changes the behaviour of raid5 when it gets a read error.
Instead of just failing the device, it tried to find out what should have
been there, and writes it over the bad block. For some media-errors, this
has a reasonable chance of fixing the error. If the write succeeds, and a
subsequent read succeeds as well, raid5 decided the address is OK and
conitnues.
Instead of failing a drive on read-error, we attempt to re-write the block,
and then re-read. If that all works, we allow the device to remain in the
array.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch removes almost all inclusions of linux/version.h. The 3
#defines are unused in most of the touched files.
A few drivers use the simple KERNEL_VERSION(a,b,c) macro, which is
unfortunatly in linux/version.h.
There are also lots of #ifdef for long obsolete kernels, this was not
touched. In a few places, the linux/version.h include was move to where
the LINUX_VERSION_CODE was used.
quilt vi `find * -type f -name "*.[ch]"|xargs grep -El '(UTS_RELEASE|LINUX_VERSION_CODE|KERNEL_VERSION|linux/version.h)'|grep -Ev '(/(boot|coda|drm)/|~$)'`
search pattern:
/UTS_RELEASE\|LINUX_VERSION_CODE\|KERNEL_VERSION\|linux\/\(utsname\|version\).h
Signed-off-by: Olaf Hering <olh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Use schedule_timeout_interruptible() instead of
set_current_state()/schedule_timeout() to reduce kernel size.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Instead of having ->read_sectors and ->write_sectors, combine the two
into ->sectors[2] and similar for the other fields. This saves a branch
several places in the io path, since we don't have to care for what the
actual io direction is. On my x86-64 box, that's 200 bytes less text in
just the core (not counting the various drivers).
Signed-off-by: Jens Axboe <axboe@suse.de>
This patch uses sg_set_buf/sg_init_one in some places where it was
duplicated.
Signed-off-by: David Hardeman <david@2gen.com>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Greg KH <greg@kroah.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
There are still a couple of cases where md threads (the resync/recovery
thread) is not interruptible since the change to use kthreads. All places
there it tests "signal_pending", it should also test kthread_should_stop,
as with this patch.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Neil Brown
Jonathan E Brassow
jblunck@suse.de
Stefan Bader
Alasdair G Kergon
Kiyoshi Ueda
Suzanne Wood
Olaf Hering
Nishanth Aravamudan
Jens Axboe
David Hardeman
Al Viro