Skip to contentSkip to conversation

generic/471: adapt test when running on btrfs to avoid failure on RWF_NOWAIT write

PATCH In review [email protected] 3 participants 5 messages Jun 12, 2020 – Jun 15, 2020 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 generic/471: adapt test when running on btrfs to avoid failure on RWF_NOWAIT write

From: Filipe Manana <[email protected]>

This test currently always fails on btrfs:

generic/471 2s ... - output mismatch (see ...results//generic/471.out.bad)
--- tests/generic/471.out 2020-06-10 19:29:03.850519863 +0100
+++ /home/fdmanana/git/hub/xfstests/results//generic/471.out.bad ...
@@ -2,12 +2,10 @@
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took seconds

This is because btrfs is a COW filesystem and an attempt to write into a previously written file range allocating a new extent (or multiple). The only exceptions are when attempting to write to a file range with a preallocated/unwritten extent or when writing to a NOCOW file that has extents allocated in the target range already.

The test currently expects that writing into a previously written file range succeeds, but that is not true on btrfs since we are not dealing with a NOCOW file. So to make the test pass on btrfs, set the NOCOW bit on the file when the filesystem is btrfs.

---
 tests/generic/471 | 11 +++++++++++
 1 file changed, 11 insertions(+)
tests/generic/471 +11 -0
index 7513f023..e9856b52 100755
@@ -37,6 +37,17 @@ fi
 
 mkdir $testdir
 
+# Btrfs is a COW filesystem, so a RWF_NOWAIT write will always fail with -EAGAIN
+# when writing to a file range except if it's a NOCOW file and an extent for the
+# range already exists or if it's a COW file and preallocated/unwritten extent
+# exists in the target range. So to make sure that the last write succeeds on
+# all filesystems, use a NOCOW file on btrfs.
+if [ $FSTYP == "btrfs" ]; then
+	_require_chattr C
+	touch $testdir/f1
+	$CHATTR_PROG +C $testdir/f1
+fi
+
 # Create a file with pwrite nowait (will fail with EAGAIN)
 $XFS_IO_PROG -f -d -c "pwrite -N -V 1 -b 1M 0 1M" $testdir/f1
 
Signature · 1 line
2.26.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 22 lines · On 6/12/20 10:06 AM, [email protected] wrote:

From: Filipe Manana <[email protected]>

This test currently always fails on btrfs:

generic/471 2s ... - output mismatch (see ...results//generic/471.out.bad)
--- tests/generic/471.out 2020-06-10 19:29:03.850519863 +0100
+++ /home/fdmanana/git/hub/xfstests/results//generic/471.out.bad ...
@@ -2,12 +2,10 @@
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took seconds

This is because btrfs is a COW filesystem and an attempt to write into a previously written file range allocating a new extent (or multiple). The only exceptions are when attempting to write to a file range with a preallocated/unwritten extent or when writing to a NOCOW file that has extents allocated in the target range already.

Quoted 6 lines

The test currently expects that writing into a previously written file range succeeds, but that is not true on btrfs since we are not dealing with a NOCOW file. So to make the test pass on btrfs, set the NOCOW bit on the file when the filesystem is btrfs.

Thanks,

Josef

#2 lore raw
In reply to [email protected]
Quoted 22 lines · On 2020/6/12 下午10:06, [email protected] wrote:

From: Filipe Manana <[email protected]>

This test currently always fails on btrfs:

generic/471 2s ... - output mismatch (see ...results//generic/471.out.bad)
--- tests/generic/471.out 2020-06-10 19:29:03.850519863 +0100
+++ /home/fdmanana/git/hub/xfstests/results//generic/471.out.bad ...
@@ -2,12 +2,10 @@
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took seconds

This is because btrfs is a COW filesystem and an attempt to write into a previously written file range allocating a new extent (or multiple). The only exceptions are when attempting to write to a file range with a preallocated/unwritten extent or when writing to a NOCOW file that has extents allocated in the target range already.

Quoted 4 lines

The test currently expects that writing into a previously written file range succeeds, but that is not true on btrfs since we are not dealing with a NOCOW file. So to make the test pass on btrfs, set the NOCOW bit on the file when the filesystem is btrfs.

Completely agree with the point for btrfs.

Quoted 11 lines
---
 tests/generic/471 | 11 +++++++++++
 1 file changed, 11 insertions(+)
tests/generic/471 +0 -0
index 7513f023..e9856b52 100755

tests/generic/471 fi · line 45

Quoted 9 lines
tests/generic/471 +6 -0
@@ -37,6 +37,17 @@ fi
 
 mkdir $testdir
 
+# Btrfs is a COW filesystem, so a RWF_NOWAIT write will always fail with -EAGAIN
+# when writing to a file range except if it's a NOCOW file and an extent for the
+# range already exists or if it's a COW file and preallocated/unwritten extent
+# exists in the target range. So to make sure that the last write succeeds on
+# all filesystems, use a NOCOW file on btrfs.
+if [ $FSTYP == "btrfs" ]; then

Although I'm not sure if really only specific to btrfs.
XFS has its always_cow sysfs interface to make data write to always do COW, just like what btrfs do by default.

Thus I believe this may be needed for all fses, and just ignore the error if the fs doesn't support COW.

Thanks,
Qu

Quoted 9 lines
tests/generic/471 +5 -0
+	_require_chattr C
+	touch $testdir/f1
+	$CHATTR_PROG +C $testdir/f1
+fi
+
 # Create a file with pwrite nowait (will fail with EAGAIN)
 $XFS_IO_PROG -f -d -c "pwrite -N -V 1 -b 1M 0 1M" $testdir/f1
 
#3 lore raw
Quoted 58 lines · Qu Wenruo
Quoted 22 lines · On 2020/6/12 下午10:06, [email protected] wrote:

From: Filipe Manana <[email protected]>

This test currently always fails on btrfs:

generic/471 2s ... - output mismatch (see ...results//generic/471.out.bad)
--- tests/generic/471.out 2020-06-10 19:29:03.850519863 +0100
+++ /home/fdmanana/git/hub/xfstests/results//generic/471.out.bad ...
@@ -2,12 +2,10 @@
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took seconds

This is because btrfs is a COW filesystem and an attempt to write into a previously written file range allocating a new extent (or multiple). The only exceptions are when attempting to write to a file range with a preallocated/unwritten extent or when writing to a NOCOW file that has extents allocated in the target range already.

Quoted 4 lines

The test currently expects that writing into a previously written file range succeeds, but that is not true on btrfs since we are not dealing with a NOCOW file. So to make the test pass on btrfs, set the NOCOW bit on the file when the filesystem is btrfs.

Completely agree with the point for btrfs.

Quoted 11 lines
---
 tests/generic/471 | 11 +++++++++++
 1 file changed, 11 insertions(+)
tests/generic/471 +0 -0
index 7513f023..e9856b52 100755

tests/generic/471 fi · line 45

Quoted 9 lines
tests/generic/471 +6 -0
@@ -37,6 +37,17 @@ fi
 
 mkdir $testdir
 
+# Btrfs is a COW filesystem, so a RWF_NOWAIT write will always fail with -EAGAIN
+# when writing to a file range except if it's a NOCOW file and an extent for the
+# range already exists or if it's a COW file and preallocated/unwritten extent
+# exists in the target range. So to make sure that the last write succeeds on
+# all filesystems, use a NOCOW file on btrfs.
+if [ $FSTYP == "btrfs" ]; then

Although I'm not sure if really only specific to btrfs.
XFS has its always_cow sysfs interface to make data write to always do COW, just like what btrfs do by default.

Quoted 2 lines

Thus I believe this may be needed for all fses, and just ignore the error if the fs doesn't support COW.

I don't understand your point.
If that flag is enabled on xfs (iirc it's a debug flag), the test would fail on xfs even if we attempt chattr +C (afaics xfs doesn't support +C).
So I don't see the benefit of always doing it.

Thanks.

Quoted 14 lines

Thanks,
Qu

Quoted 9 lines
tests/generic/471 +5 -0
+     _require_chattr C
+     touch $testdir/f1
+     $CHATTR_PROG +C $testdir/f1
+fi
+
 # Create a file with pwrite nowait (will fail with EAGAIN)
 $XFS_IO_PROG -f -d -c "pwrite -N -V 1 -b 1M 0 1M" $testdir/f1
 
#4 lore raw
Quoted 59 lines · Filipe Manana
Quoted 58 lines · Qu Wenruo
Quoted 22 lines · On 2020/6/12 下午10:06, [email protected] wrote:

From: Filipe Manana <[email protected]>

This test currently always fails on btrfs:

generic/471 2s ... - output mismatch (see ...results//generic/471.out.bad)
--- tests/generic/471.out 2020-06-10 19:29:03.850519863 +0100
+++ /home/fdmanana/git/hub/xfstests/results//generic/471.out.bad ...
@@ -2,12 +2,10 @@
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took seconds

This is because btrfs is a COW filesystem and an attempt to write into a previously written file range allocating a new extent (or multiple). The only exceptions are when attempting to write to a file range with a preallocated/unwritten extent or when writing to a NOCOW file that has extents allocated in the target range already.

Quoted 4 lines

The test currently expects that writing into a previously written file range succeeds, but that is not true on btrfs since we are not dealing with a NOCOW file. So to make the test pass on btrfs, set the NOCOW bit on the file when the filesystem is btrfs.

Completely agree with the point for btrfs.

Quoted 11 lines
---
 tests/generic/471 | 11 +++++++++++
 1 file changed, 11 insertions(+)
tests/generic/471 +0 -0
index 7513f023..e9856b52 100755

tests/generic/471 fi · line 45

Quoted 9 lines
tests/generic/471 +6 -0
@@ -37,6 +37,17 @@ fi
 
 mkdir $testdir
 
+# Btrfs is a COW filesystem, so a RWF_NOWAIT write will always fail with -EAGAIN
+# when writing to a file range except if it's a NOCOW file and an extent for the
+# range already exists or if it's a COW file and preallocated/unwritten extent
+# exists in the target range. So to make sure that the last write succeeds on
+# all filesystems, use a NOCOW file on btrfs.
+if [ $FSTYP == "btrfs" ]; then

Although I'm not sure if really only specific to btrfs.
XFS has its always_cow sysfs interface to make data write to always do COW, just like what btrfs do by default.

Quoted 7 lines
Quoted 2 lines

Thus I believe this may be needed for all fses, and just ignore the error if the fs doesn't support COW.

I don't understand your point.
If that flag is enabled on xfs (iirc it's a debug flag), the test would fail on xfs even if we attempt chattr +C (afaics xfs doesn't support +C).

Oh, my fault. Didn't know that xfs won't support chattr +C.

Then forget the comment.

Thanks,
Qu

Quoted 19 lines

So I don't see the benefit of always doing it.

Thanks.

Quoted 14 lines

Thanks,
Qu

Quoted 9 lines
tests/generic/471 +5 -0
+     _require_chattr C
+     touch $testdir/f1
+     $CHATTR_PROG +C $testdir/f1
+fi
+
 # Create a file with pwrite nowait (will fail with EAGAIN)
 $XFS_IO_PROG -f -d -c "pwrite -N -V 1 -b 1M 0 1M" $testdir/f1