Skip to contentSkip to conversation

fstests: btrfs/253: fix false alert due to _set_fs_sysfs_attr changes

PATCH Superseded Qu Wenruo 2 participants 3 messages Apr 20, 2025 – Apr 21, 2025 Btrfs file system

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH 1/1 fstests: btrfs/253: fix false alert due to _set_fs_sysfs_attr changes

[FALSE FAILURE]
Test btrfs/253 now fails like the following:

btrfs/253 2s ... - output mismatch (see ~/xfstests/results//btrfs/253.out.bad)
--- tests/btrfs/253.out 2022-05-11 11:25:30.753333331 +0930
+++ ~/xfstests/results//btrfs/253.out.bad 2025-04-20 17:28:39.139180394 +0930
@@ -5,6 +5,7 @@
Calculate request size so last memory allocation cannot be completely fullfilled.
Third allocation.
Force allocation of system block type must fail.
+./common/rc: line 5213: echo: write error: No space left on device
Verify first allocation.
Verify second allocation.
Verify third allocation.
...
(Run 'diff -u ~/xfstests/tests/btrfs/253.out ~/xfstests/results//btrfs/253.out.bad' to see the entire diff)

[CAUSE]
Since commit 0a9011ae6a36 ("fstests: common/rc: set_fs_sysfs_attr:
redirect errors to stdout") the function _set_fs_sysfs_attr() always output everything into stdout, thus the stderr redirection makes no sense anymore.

And the expected failure will cause output difference and fail the test.

[FIX]
Instead of the useless re-direct of stderr, save the stdout and check if it contains the word "error" to determine if it failed.

Fixes 0a9011ae6a36 ("fstests: common/rc: set_fs_sysfs_attr: redirect errors to stdout")
---
 tests/btrfs/253 | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tests/btrfs/253 +6 -1
index adbc6bfb..ad69dfe1 100755
@@ -228,7 +228,12 @@ alloc_size "Data" FOURTH_DATA_SIZE_MB
 # Force chunk allocation of system block type must fail.
 #
 echo "Force allocation of system block type must fail."
-_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1 2>/dev/null
+output=$(_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1)
+
+if ! echo "$output" | grep -q "error" ; then
+	echo "Force allocation succeeded unexpectedly."
+	echo "$output" >> $seqres.full
+fi
 
 #
 # Verification of initial allocation.
Signature · 1 line
2.47.1
#1 lore raw
Quoted 40 lines · Qu Wenruo

[FALSE FAILURE]
Test btrfs/253 now fails like the following:

btrfs/253 2s ... - output mismatch (see ~/xfstests/results//btrfs/253.out.bad)
--- tests/btrfs/253.out 2022-05-11 11:25:30.753333331 +0930
+++ ~/xfstests/results//btrfs/253.out.bad 2025-04-20 17:28:39.139180394 +0930
@@ -5,6 +5,7 @@
Calculate request size so last memory allocation cannot be completely fullfilled.
Third allocation.
Force allocation of system block type must fail.
+./common/rc: line 5213: echo: write error: No space left on device
Verify first allocation.
Verify second allocation.
Verify third allocation.
...
(Run 'diff -u ~/xfstests/tests/btrfs/253.out ~/xfstests/results//btrfs/253.out.bad' to see the entire diff)

[CAUSE]
Since commit 0a9011ae6a36 ("fstests: common/rc: set_fs_sysfs_attr:
redirect errors to stdout") the function _set_fs_sysfs_attr() always output everything into stdout, thus the stderr redirection makes no sense anymore.

And the expected failure will cause output difference and fail the test.

[FIX]
Instead of the useless re-direct of stderr, save the stdout and check if it contains the word "error" to determine if it failed.

Fixes 0a9011ae6a36 ("fstests: common/rc: set_fs_sysfs_attr: redirect errors to stdout")
---
 tests/btrfs/253 | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tests/btrfs/253 +0 -0
index adbc6bfb..ad69dfe1 100755

tests/btrfs/253 alloc_size "Data" FOURTH_DATA_SIZE_MB · line 236

Quoted 10 lines
tests/btrfs/253 +6 -1
@@ -228,7 +228,12 @@ alloc_size "Data" FOURTH_DATA_SIZE_MB
 # Force chunk allocation of system block type must fail.
 #
 echo "Force allocation of system block type must fail."
-_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1 2>/dev/null
+output=$(_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1)
+
+if ! echo "$output" | grep -q "error" ; then
+	echo "Force allocation succeeded unexpectedly."
+	echo "$output" >> $seqres.full
+fi

Can't we use _set_sysfs_policy_must_fail() instead? Apparently, that function does not looks only for "policy" setting. So, renaming the function would also be appreciated.

Quoted 3 lines
tests/btrfs/253 +0 -0
 #
 # Verification of initial allocation.
author
#2 lore raw
Quoted 47 lines · Naohiro Aota
Quoted 46 lines · Qu Wenruo

[FALSE FAILURE]
Test btrfs/253 now fails like the following:

btrfs/253 2s ... - output mismatch (see ~/xfstests/results//btrfs/253.out.bad)
--- tests/btrfs/253.out 2022-05-11 11:25:30.753333331 +0930
+++ ~/xfstests/results//btrfs/253.out.bad 2025-04-20 17:28:39.139180394 +0930
@@ -5,6 +5,7 @@
Calculate request size so last memory allocation cannot be completely fullfilled.
Third allocation.
Force allocation of system block type must fail.
+./common/rc: line 5213: echo: write error: No space left on device
Verify first allocation.
Verify second allocation.
Verify third allocation.
...
(Run 'diff -u ~/xfstests/tests/btrfs/253.out ~/xfstests/results//btrfs/253.out.bad' to see the entire diff)

[CAUSE]
Since commit 0a9011ae6a36 ("fstests: common/rc: set_fs_sysfs_attr:
redirect errors to stdout") the function _set_fs_sysfs_attr() always output everything into stdout, thus the stderr redirection makes no sense anymore.

And the expected failure will cause output difference and fail the test.

[FIX]
Instead of the useless re-direct of stderr, save the stdout and check if it contains the word "error" to determine if it failed.

Fixes 0a9011ae6a36 ("fstests: common/rc: set_fs_sysfs_attr: redirect errors to stdout")
---
  tests/btrfs/253 | 7 ++++++-
  1 file changed, 6 insertions(+), 1 deletion(-)
tests/btrfs/253 +2 -1
index adbc6bfb..ad69dfe1 100755
@@ -228,7 +228,12 @@ alloc_size "Data" FOURTH_DATA_SIZE_MB
  # Force chunk allocation of system block type must fail.
  #
  echo "Force allocation of system block type must fail."
-_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1 2>/dev/null
+output=$(_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1)
+
Quoted 8 lines

tests/btrfs/253 alloc_size "Data" FOURTH_DATA_SIZE_MB · line 236

Quoted 4 lines
tests/btrfs/253 +4 -0
+if ! echo "$output" | grep -q "error" ; then
+	echo "Force allocation succeeded unexpectedly."
+	echo "$output" >> $seqres.full
+fi

Can't we use _set_sysfs_policy_must_fail() instead? Apparently, that function does not looks only for "policy" setting. So, renaming the function would also be appreciated.

Thanks a lot for pointing out the _must_fail() version.

Didn't notice that since the functions are split into common/rc and common/sysfs.

I have to say the split and naming is really bad, especially when none of the policy related helpers are not utilized by anyone.

I'll keep the fix simple for now, and leave the rename for Anand, as he would have a more clear view why those helpers are introduced and may have a better naming scheme.

Thanks,
Qu

Quoted 4 lines
Quoted 4 lines
tests/btrfs/253 +0 -0
  #
  # Verification of initial allocation.