Calculating free blocks in ext[234] is surprisingly hard, since
by default we report "bsd" style df which doesn't count filesystem
"overhead" blocks as used.
With a lot of code dedicated to sorting out what to report as
free, things tend to go wrong surprisingly often.
Here's a test to actually try to stop the next regression. ;)
NB: For bsddf, the kernel currently does not count journal blocks
as overhead; it probably should. But the test below looks to have
the result within 1% of perfection, so it still passes even if
the kernel doesn't count the journal against free blocks.
Signed-off-by: Eric Sandeen <[email protected]>
---
diff --git a/289 b/289
new file mode 100755
index 0000000..bf0e897
--- /dev/null
+++ b/289
@@ -0,0 +1,103 @@
+#! /bin/bash
+# FS QA Test No. 286
+#
+# Test overhead & df output for extN filesystems
+#
+#-----------------------------------------------------------------------
+# Copyright (c) 2012 Red Hat, Inc. All Rights Reserved.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation.
+#
+# This program is distributed in the hope that it would be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write the Free Software Foundation,
+# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+#-----------------------------------------------------------------------
+#
+# creator
[email protected]
+
+seq=`basename $0`
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1 # failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+ cd /
+ rm -f $tmp.*
+}
+
+# get standard environment, filters and checks
+. ./common.rc
+. ./common.filter
+
+# real QA test starts here
+
+# Modify as appropriate.
+_supported_fs ext2 ext3 ext4
+_supported_os Linux
+_require_scratch
+
+rm -f $seq.full
+
+_scratch_mkfs >> $seq.full 2>&1
+
+# Get the honest truth about block counts straight from metadata on disk
+TOTAL_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
+ | awk '/Block count:/{print $3}'`
+
+FREE_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
+ | awk '/Free blocks:/{print $3}'`
+
+# nb: kernels today don't count journal blocks as overhead, but should.
+# For most filesystems this will still be within tolerance.
+# Overhead is all the blocks (already) used by the fs itself:
+OVERHEAD=$(($TOTAL_BLOCKS-$FREE_BLOCKS))
+
+# bsddf|minixdf
+# Set the behaviour for the statfs system call. The minixdf
+# behaviour is to return in the f_blocks field the total number of
+# blocks of the filesystem, while the bsddf behaviour (which is
+# the default) is to subtract the overhead blocks used by the ext2
+# filesystem and not available for file storage.
+
+# stat -f output looks like this; we get f_blocks from that, which
+# varies depending on the df mount options used below:
+
+# File: "/mnt/test"
+# ID: affc5f2b2f57652 Namelen: 255 Type: ext2/ext3
+# Block size: 4096 Fundamental block size: 4096
+# Blocks: Total: 5162741 Free: 5118725 Available: 4856465
+# Inodes: Total: 1313760 Free: 1313749
+
+_scratch_mount "-o minixdf"
+MINIX_F_BLOCKS=`stat -f $SCRATCH_MNT | awk '/^Blocks/{print $3}'`
+umount $SCRATCH_MNT
+
+_scratch_mount "-o bsddf"
+BSD_F_BLOCKS=`stat -f $SCRATCH_MNT | awk '/^Blocks/{print $3}'`
+umount $SCRATCH_MNT
+
+# Echo data to $seq.full for analysis
+echo "Overhead is $OVERHEAD blocks out of $TOTAL_BLOCKS ($FREE_BLOCKS free)" >> $seq.full
+echo "MINIX free blocks $MINIX_F_BLOCKS" >> $seq.full
+echo "BSD free blocks $BSD_F_BLOCKS" >> $seq.full
+
+# minix should be exactly equal (hence tolerance of 0)
+_within_tolerance "minix f_blocks" $MINIX_F_BLOCKS $TOTAL_BLOCKS 0 -v
+# bsd should be within ... we'll say 1% for some slop
+_within_tolerance "bsd f_blocks" $BSD_F_BLOCKS $(($TOTAL_BLOCKS-$OVERHEAD)) 1% -v
+
+# success, all done
+status=0
+exit
diff --git a/289.out b/289.out
new file mode 100644
index 0000000..a4de760
--- /dev/null
+++ b/289.out
@@ -0,0 +1,3 @@
+QA output created by 289
+minix f_blocks is in range
+bsd f_blocks is in range
diff --git a/group b/group
index fb0f8eb..a846b60 100644
--- a/group
+++ b/group
@@ -407,3 +407,4 @@ deprecated
286 other
287 auto dump quota quick
288 auto quick ioctl trim
+289 auto quick
On 10/25/2012 12:19 PM, Eric Sandeen wrote:
> Calculating free blocks in ext[234] is surprisingly hard, since
> by default we report "bsd" style df which doesn't count filesystem
> "overhead" blocks as used.
>
> With a lot of code dedicated to sorting out what to report as
> free, things tend to go wrong surprisingly often.
>
> Here's a test to actually try to stop the next regression. ;)
>
> NB: For bsddf, the kernel currently does not count journal blocks
> as overhead; it probably should. But the test below looks to have
> the result within 1% of perfection, so it still passes even if
> the kernel doesn't count the journal against free blocks.
>
> Signed-off-by: Eric Sandeen <[email protected]>
>
> ---
>
>
> diff --git a/289 b/289
> new file mode 100755
> index 0000000..bf0e897
> --- /dev/null
> +++ b/289
> @@ -0,0 +1,103 @@
> +#! /bin/bash
> +# FS QA Test No. 286
^
289
I know this may change at commit time. ;)
> +#
> +# Test overhead & df output for extN filesystems
> +#
> +#-----------------------------------------------------------------------
> +# Copyright (c) 2012 Red Hat, Inc. All Rights Reserved.
> +#
> +# This program is free software; you can redistribute it and/or
> +# modify it under the terms of the GNU General Public License as
> +# published by the Free Software Foundation.
> +#
> +# This program is distributed in the hope that it would be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program; if not, write the Free Software Foundation,
> +# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
> +#-----------------------------------------------------------------------
> +#
> +# creator
> [email protected]
> +
> +seq=`basename $0`
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1 # failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> + cd /
> + rm -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common.rc
> +. ./common.filter# ./check 289
FSTYP -- ext4
PLATFORM -- Linux/x86_64 cxfsxe4 3.7.0-rc2+
MKFS_OPTIONS -- /dev/sdc2
MOUNT_OPTIONS -- -o acl,user_xattr /dev/sdc2 /xfs_scratch
289 - output mismatch (see 289.out.bad)
--- 289.out 2012-10-26 12:33:27.000000000 -0500
+++ 289.out.bad 2012-10-26 12:35:03.000000000 -0500
@@ -1,3 +1,4 @@
QA output created by 289
-minix f_blocks is in range
+minix f_blocks has value of 7208959
+minix f_blocks is NOT in range 7323904 .. 7323904
bsd f_blocks is in range
Ran: 289
Failures: 289
Failed 1 of 1 tests
> +
> +# real QA test starts here
> +
> +# Modify as appropriate.
> +_supported_fs ext2 ext3 ext4
> +_supported_os Linux
> +_require_scratch
> +
> +rm -f $seq.full
> +
> +_scratch_mkfs >> $seq.full 2>&1
> +
> +# Get the honest truth about block counts straight from metadata on disk
> +TOTAL_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
> + | awk '/Block count:/{print $3}'`
> +
> +FREE_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
> + | awk '/Free blocks:/{print $3}'`
> +
> +# nb: kernels today don't count journal blocks as overhead, but should.
> +# For most filesystems this will still be within tolerance.
> +# Overhead is all the blocks (already) used by the fs itself:
> +OVERHEAD=$(($TOTAL_BLOCKS-$FREE_BLOCKS))
> +
> +# bsddf|minixdf
> +# Set the behaviour for the statfs system call. The minixdf
> +# behaviour is to return in the f_blocks field the total number of
> +# blocks of the filesystem, while the bsddf behaviour (which is
> +# the default) is to subtract the overhead blocks used by the ext2
> +# filesystem and not available for file storage.
> +
> +# stat -f output looks like this; we get f_blocks from that, which
> +# varies depending on the df mount options used below:
> +
> +# File: "/mnt/test"
> +# ID: affc5f2b2f57652 Namelen: 255 Type: ext2/ext3
> +# Block size: 4096 Fundamental block size: 4096
> +# Blocks: Total: 5162741 Free: 5118725 Available: 4856465
> +# Inodes: Total: 1313760 Free: 1313749
> +
> +_scratch_mount "-o minixdf"
> +MINIX_F_BLOCKS=`stat -f $SCRATCH_MNT | awk '/^Blocks/{print $3}'`
> +umount $SCRATCH_MNT
> +
> +_scratch_mount "-o bsddf"
> +BSD_F_BLOCKS=`stat -f $SCRATCH_MNT | awk '/^Blocks/{print $3}'`
> +umount $SCRATCH_MNT
> +
> +# Echo data to $seq.full for analysis
> +echo "Overhead is $OVERHEAD blocks out of $TOTAL_BLOCKS ($FREE_BLOCKS free)" >> $seq.full
> +echo "MINIX free blocks $MINIX_F_BLOCKS" >> $seq.full
> +echo "BSD free blocks $BSD_F_BLOCKS" >> $seq.full
> +
This passes for ext[23] but not ext4.
> +# minix should be exactly equal (hence tolerance of 0)
> +_within_tolerance "minix f_blocks" $MINIX_F_BLOCKS $TOTAL_BLOCKS 0 -v
This is what I got when I ran it on an 80G SSD.
Model: ATA INTEL SSDSA2M080 (scsi)
Disk /dev/sdc: 80.0GB
Sector size (logical/physical): 512B/512B
Partition Table: gpt_sync_mbr
Number Start End Size File system Name Flags
1 17.4kB 30.0GB 30.0GB ext4 primary
2 30.0GB 60.0GB 30.0GB ext4 primary
# ./check 289
FSTYP -- ext4
PLATFORM -- Linux/x86_64 cxfsxe4 3.7.0-rc2+
MKFS_OPTIONS -- /dev/sdc2
MOUNT_OPTIONS -- -o acl,user_xattr /dev/sdc2 /xfs_scratch
289 - output mismatch (see 289.out.bad)
--- 289.out 2012-10-26 12:33:27.000000000 -0500
+++ 289.out.bad 2012-10-26 12:35:03.000000000 -0500
@@ -1,3 +1,4 @@
QA output created by 289
-minix f_blocks is in range
+minix f_blocks has value of 7208959
+minix f_blocks is NOT in range 7323904 .. 7323904
bsd f_blocks is in range
Ran: 289
Failures: 289
Failed 1 of 1 tests
> +# bsd should be within ... we'll say 1% for some slop
> +_within_tolerance "bsd f_blocks" $BSD_F_BLOCKS $(($TOTAL_BLOCKS-$OVERHEAD)) 1% -v
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/289.out b/289.out
> new file mode 100644
> index 0000000..a4de760
> --- /dev/null
> +++ b/289.out
> @@ -0,0 +1,3 @@
> +QA output created by 289
> +minix f_blocks is in range
> +bsd f_blocks is in range
> diff --git a/group b/group
> index fb0f8eb..a846b60 100644
> --- a/group
> +++ b/group
> @@ -407,3 +407,4 @@ deprecated
> 286 other
> 287 auto dump quota quick
> 288 auto quick ioctl trim
> +289 auto quick
>
Regards
--Rich
On 10/26/12 1:03 PM, Rich Johnston wrote:
> On 10/25/2012 12:19 PM, Eric Sandeen wrote:
>> Calculating free blocks in ext[234] is surprisingly hard, since
>> by default we report "bsd" style df which doesn't count filesystem
>> "overhead" blocks as used.
>>
>> With a lot of code dedicated to sorting out what to report as
>> free, things tend to go wrong surprisingly often.
>>
>> Here's a test to actually try to stop the next regression. ;)
>>
>> NB: For bsddf, the kernel currently does not count journal blocks
>> as overhead; it probably should. But the test below looks to have
>> the result within 1% of perfection, so it still passes even if
>> the kernel doesn't count the journal against free blocks.
>>
>> Signed-off-by: Eric Sandeen <[email protected]>
>>
>> ---
>>
>>
>> diff --git a/289 b/289
>> new file mode 100755
>> index 0000000..bf0e897
>> --- /dev/null
>> +++ b/289
>> @@ -0,0 +1,103 @@
>> +#! /bin/bash
>> +# FS QA Test No. 286
> ^
> 289
> I know this may change at commit time. ;)
meh, right. Dumb to have it in the file, maybe.
>> +#
>> +# Test overhead & df output for extN filesystems
>> +#
>> +#-----------------------------------------------------------------------
>> +# Copyright (c) 2012 Red Hat, Inc. All Rights Reserved.
>> +#
>> +# This program is free software; you can redistribute it and/or
>> +# modify it under the terms of the GNU General Public License as
>> +# published by the Free Software Foundation.
>> +#
>> +# This program is distributed in the hope that it would be useful,
>> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
>> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
>> +# GNU General Public License for more details.
>> +#
>> +# You should have received a copy of the GNU General Public License
>> +# along with this program; if not, write the Free Software Foundation,
>> +# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
>> +#-----------------------------------------------------------------------
>> +#
>> +# creator
>> [email protected]
>> +
>> +seq=`basename $0`
>> +echo "QA output created by $seq"
>> +
>> +here=`pwd`
>> +tmp=/tmp/$$
>> +status=1 # failure is the default!
>> +trap "_cleanup; exit \$status" 0 1 2 3 15
>> +
>> +_cleanup()
>> +{
>> + cd /
>> + rm -f $tmp.*
>> +}
>> +
>> +# get standard environment, filters and checks
>> +. ./common.rc
>> +. ./common.filter# ./check 289
> FSTYP -- ext4
> PLATFORM -- Linux/x86_64 cxfsxe4 3.7.0-rc2+
> MKFS_OPTIONS -- /dev/sdc2
> MOUNT_OPTIONS -- -o acl,user_xattr /dev/sdc2 /xfs_scratch
>
> 289 - output mismatch (see 289.out.bad)
> --- 289.out 2012-10-26 12:33:27.000000000 -0500
> +++ 289.out.bad 2012-10-26 12:35:03.000000000 -0500
> @@ -1,3 +1,4 @@
> QA output created by 289
> -minix f_blocks is in range
> +minix f_blocks has value of 7208959
> +minix f_blocks is NOT in range 7323904 .. 7323904
> bsd f_blocks is in range
> Ran: 289
> Failures: 289
> Failed 1 of 1 tests
Yep - it's an ext4 bug. I sent a patch to fix it.
[PATCH] ext4: fix overhead calculations in ext4_stats, again
You might want to retest w/ that.
-Eric
>> +
>> +# real QA test starts here
>> +
>> +# Modify as appropriate.
>> +_supported_fs ext2 ext3 ext4
>> +_supported_os Linux
>> +_require_scratch
>> +
>> +rm -f $seq.full
>> +
>> +_scratch_mkfs >> $seq.full 2>&1
>> +
>> +# Get the honest truth about block counts straight from metadata on disk
>> +TOTAL_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
>> + | awk '/Block count:/{print $3}'`
>> +
>> +FREE_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
>> + | awk '/Free blocks:/{print $3}'`
>> +
>> +# nb: kernels today don't count journal blocks as overhead, but should.
>> +# For most filesystems this will still be within tolerance.
>> +# Overhead is all the blocks (already) used by the fs itself:
>> +OVERHEAD=$(($TOTAL_BLOCKS-$FREE_BLOCKS))
>> +
>> +# bsddf|minixdf
>> +# Set the behaviour for the statfs system call. The minixdf
>> +# behaviour is to return in the f_blocks field the total number of
>> +# blocks of the filesystem, while the bsddf behaviour (which is
>> +# the default) is to subtract the overhead blocks used by the ext2
>> +# filesystem and not available for file storage.
>> +
>> +# stat -f output looks like this; we get f_blocks from that, which
>> +# varies depending on the df mount options used below:
>> +
>> +# File: "/mnt/test"
>> +# ID: affc5f2b2f57652 Namelen: 255 Type: ext2/ext3
>> +# Block size: 4096 Fundamental block size: 4096
>> +# Blocks: Total: 5162741 Free: 5118725 Available: 4856465
>> +# Inodes: Total: 1313760 Free: 1313749
>> +
>> +_scratch_mount "-o minixdf"
>> +MINIX_F_BLOCKS=`stat -f $SCRATCH_MNT | awk '/^Blocks/{print $3}'`
>> +umount $SCRATCH_MNT
>> +
>> +_scratch_mount "-o bsddf"
>> +BSD_F_BLOCKS=`stat -f $SCRATCH_MNT | awk '/^Blocks/{print $3}'`
>> +umount $SCRATCH_MNT
>> +
>> +# Echo data to $seq.full for analysis
>> +echo "Overhead is $OVERHEAD blocks out of $TOTAL_BLOCKS ($FREE_BLOCKS free)" >> $seq.full
>> +echo "MINIX free blocks $MINIX_F_BLOCKS" >> $seq.full
>> +echo "BSD free blocks $BSD_F_BLOCKS" >> $seq.full
>> +
>
> This passes for ext[23] but not ext4.
*nod*
>> +# minix should be exactly equal (hence tolerance of 0)
>> +_within_tolerance "minix f_blocks" $MINIX_F_BLOCKS $TOTAL_BLOCKS 0 -v
>
> This is what I got when I ran it on an 80G SSD.
>
> Model: ATA INTEL SSDSA2M080 (scsi)
> Disk /dev/sdc: 80.0GB
> Sector size (logical/physical): 512B/512B
> Partition Table: gpt_sync_mbr
>
> Number Start End Size File system Name Flags
> 1 17.4kB 30.0GB 30.0GB ext4 primary
> 2 30.0GB 60.0GB 30.0GB ext4 primary
>
> # ./check 289
> FSTYP -- ext4
> PLATFORM -- Linux/x86_64 cxfsxe4 3.7.0-rc2+
> MKFS_OPTIONS -- /dev/sdc2
> MOUNT_OPTIONS -- -o acl,user_xattr /dev/sdc2 /xfs_scratch
>
> 289 - output mismatch (see 289.out.bad)
> --- 289.out 2012-10-26 12:33:27.000000000 -0500
> +++ 289.out.bad 2012-10-26 12:35:03.000000000 -0500
> @@ -1,3 +1,4 @@
> QA output created by 289
> -minix f_blocks is in range
> +minix f_blocks has value of 7208959
> +minix f_blocks is NOT in range 7323904 .. 7323904
> bsd f_blocks is in range
> Ran: 289
> Failures: 289
> Failed 1 of 1 tests
>
>
>> +# bsd should be within ... we'll say 1% for some slop
>> +_within_tolerance "bsd f_blocks" $BSD_F_BLOCKS $(($TOTAL_BLOCKS-$OVERHEAD)) 1% -v
>> +
>> +# success, all done
>> +status=0
>> +exit
>> diff --git a/289.out b/289.out
>> new file mode 100644
>> index 0000000..a4de760
>> --- /dev/null
>> +++ b/289.out
>> @@ -0,0 +1,3 @@
>> +QA output created by 289
>> +minix f_blocks is in range
>> +bsd f_blocks is in range
>> diff --git a/group b/group
>> index fb0f8eb..a846b60 100644
>> --- a/group
>> +++ b/group
>> @@ -407,3 +407,4 @@ deprecated
>> 286 other
>> 287 auto dump quota quick
>> 288 auto quick ioctl trim
>> +289 auto quick
>>
>
> Regards
> --Rich
>
On 10/26/2012 01:39 PM, Eric Sandeen wrote:
> On 10/26/12 1:03 PM, Rich Johnston wrote:
>> On 10/25/2012 12:19 PM, Eric Sandeen wrote:
>>> Calculating free blocks in ext[234] is surprisingly hard, since
>>> by default we report "bsd" style df which doesn't count filesystem
>>> "overhead" blocks as used.
>>>
>>> With a lot of code dedicated to sorting out what to report as
>>> free, things tend to go wrong surprisingly often.
>>>
>>> Here's a test to actually try to stop the next regression. ;)
>>>
>>> NB: For bsddf, the kernel currently does not count journal blocks
>>> as overhead; it probably should. But the test below looks to have
>>> the result within 1% of perfection, so it still passes even if
>>> the kernel doesn't count the journal against free blocks.
>>>
>>> Signed-off-by: Eric Sandeen <[email protected]>
>>>
>>> ---
>>>
>
>
> Yep - it's an ext4 bug. I sent a patch to fix it.
>
> [PATCH] ext4: fix overhead calculations in ext4_stats, again
>
> You might want to retest w/ that.
>
> -Eric
>
>>> +
Thanks Eric,
Everything passes now.
Reviewed-by: Rich Johnston <[email protected]>
xfstests: test ext4 statfs
Calculating free blocks in ext[234] is surprisingly hard, since
by default we report "bsd" style df which doesn't count filesystem
"overhead" blocks as used.
With a lot of code dedicated to sorting out what to report as
free, things tend to go wrong surprisingly often.
Here's a test to actually try to stop the next regression. ;)
NB: For bsddf, the kernel currently does not count journal blocks
as overhead; it probably should. But the test below looks to have
the result within 1% of perfection, so it still passes even if
the kernel doesn't count the journal against free blocks.
Eric,
This patch has been committed to git://oss.sgi.com/xfs/cmds/xfstests,
master branch, commit 0b2ab695.
Thanks
--Rich