Some guests (win2008 server for example) do a lot of unnecessary
flushing when underlying media has not changed. This adds additional
overhead on host when calling fsync/fdatasync.
This change introduces a write generation scheme in BlockDriverState.
Current write generation is checked against last flushed generation to
avoid unnessesary flushes.
The problem with excessive flushing was found by a performance test
which does parallel directory tree creation (from 2 processes).
Results improved from 0.424 loops/sec to 0.432 loops/sec.
Each loop creates 10^3 directories with 10 files in each.
This affected some blkdebug testcases that were expecting error logs from
failure-injected flushes which are now skipped entirely
(tests 026 071 089).
This also affects the performance of block jobs and thus BLOCK_JOB_READY
events for driver-mirror and active block-commit commands now arrives
faster, before QMP send successfully returns to caller (tests 141 144).
Signed-off-by: Evgeny Yakovlev <eyakovlev@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1468870792-7411-5-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
CC: Max Reitz <mreitz@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Fam Zheng <famz@redhat.com>
CC: John Snow <jsnow@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Our qapi conventions document that '.' should only be used in
the prefix of downstream names. BlkdebugEvent was a lone
exception to this. Changing this is not backwards compatible
to the 'blockdev-add' QMP command; however, that command is
not yet fully stable. It can also be argued that the testsuite
is the biggest user of blkdebug, and that any other user can
be taught to deal with the change by paying attention to
introspection results.
Done with:
$ for str in \
l1_grow.{alloc,write,activate}_table \
l2_alloc.{cow_read,write} \
refblock_alloc.{hookup,write,write_blocks,write_table,switch_table} \
pwritev_rmw.{head,after_head,tail,after_tail}; do
str1=$(echo "$str" | sed 's/\./\\./')
str2=$(echo "$str" | sed 's/\./_/')
git grep -l "$str1" | xargs -r sed -i "s/$str1/$str2/g"
done
followed by a manual touchup to test 77 to keep the test working.
Reported-by: Markus Armbruster <armbru@redhat.com>
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1447836791-369-21-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Background:
The blkdebug scripts are currently engineered so that when a debug
event occurs, a prefilter browses a master list of parsed rules for a
certain event and adds them to an "active list" of rules to be used for
the forthcoming action, provided the events and state numbers match.
Then, once the request is received, the last active rule is used to
inject an error if certain parameters match.
This active list is cleared every time the prefilter injects a new
rule for the first time during a debug event.
The "once" rule currently causes the error injection, if it is
triggered, to only clear the active list. This is insufficient for
preventing future injections of the same rule.
Remedy:
This patch /deletes/ the rule from the list that the prefilter
browses, so it is gone for good. In V2, we remove only the rule of
interest from the active list instead of allowing the "once" rule to
clear the entire list of active rules.
Impact:
This affects iotests 026. Several ENOSPC tests that used "once" can
be seen to have output that shows multiple failure messages. After
this patch, the error messages tend to be smaller and less severe, but
the injection can still be seen to be working. I have patched the
expected output to expect the smaller error messages.
Signed-off-by: John Snow <jsnow@redhat.com>
Message-id: 1423257977-25630-1-git-send-email-jsnow@redhat.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This is simply:
$ cd tests/qemu-iotests; sed -i -e 's/ *$//' *.out
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1418110684-19528-2-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
qcow2_cache_flush() may fail; if one of the caches failed to be flushed
successfully to disk in qcow2_close() the image should not be marked
clean, and we should emit a warning.
This breaks the (qcow2-specific) iotests 026, 071 and 089; change their
output accordingly.
Cc: qemu-stable@nongnu.org
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
free_cluster_index is only correct if update_refcount() was called from
an allocation function, and even there it's brittle because it's used to
protect unfinished allocations which still have a refcount of 0 - if it
moves in the wrong place, the unfinished allocation can be corrupted.
So not using it any more seems to be a good idea. Instead, use the
first requested cluster to do the calculations. Return -EAGAIN if
unfinished allocations could become invalid and let the caller restart
its search for some free clusters.
The context of creating a snapsnot is one situation where
update_refcount() is called outside of a cluster allocation. For this
case, the change fixes a buffer overflow if a cluster is referenced in
an L2 table that cannot be represented by an existing refcount block.
(new_table[refcount_table_index] was out of bounds)
[Bump the qemu-iotests 026 refblock_alloc.write leak count from 10 to
11.
--Stefan]
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Because l2_allocate now frees the unused L2 cluster on error, the
according test cases in 026 don't result in one leaked cluster anymore.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The reference output for test case 026 hasn't been updated in a long
time and it's one of the "known failing" cases. This patch updates the
reference output so that unintentional changes can be reliably detected
again.
The problem with this test case is that it produces different output
depending on whether -nocache is used or not. The solution of this patch
is to actually have two different reference outputs. If nnn.out.nocache
exists, it is used as the reference output for -nocache; otherwise,
nnn.out stays valid for both cases.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Some image formats do have a cluster size, others don't, but there are
tests that work with both sets of images and currently we get failures
because the qemu-img create output doesn't mention the cluster size for
some formats.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
The error message for leaked clusters has changed. qemu-iotests needs to be
updated to pass 026 again.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Evgeny Yakovlev
Eric Blake
John Snow
Fam Zheng
Max Reitz
Kevin Wolf