[PATCH] ext4/048: Fix hangup due to no free inodes

[PATCH] ext4/048: Fix hangup due to no free inodes

[Ojaswin Mujoo]

We currently mkfs a 128MB filesystem, which gives use ~2048 free inodes
on 64k blocksize. The test then keeps adding new files to a directory to
trigger an htree split. For 64k this takes more than the total free
inodes, which causes touch to return -ENOSPC. This leads to the while
loop in induce_node_split() to never finish.

To fix this:
1. Format a 1G FS which gives us atleast 16K inodes to work with.
2. _fail if there's any error while trying to induce node split, so we
   dont get stuck in loop

Fixes: 466ddbfd1151 ("ext4: add test for ext4_dir_entry2 wipe")
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
---
 tests/ext4/048 | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/tests/ext4/048 b/tests/ext4/048
index 2031c8c8..6343ff3a 100755
--- a/tests/ext4/048
+++ b/tests/ext4/048
@@ -69,6 +69,11 @@ induce_node_split() {
 	while [[ "$(stat --printf="%s" $testdir)" == "$dir_size" ]]; do
 		file_num=$(($file_num + 1))
 		touch $testdir/test"$(printf "%04d" $file_num)"
+		local ret=$?
+		if [[ $ret -ne 0 ]]
+		then
+			_fail "ERROR induce_node_split(): $ret"
+		fi
 	done
 	_scratch_unmount >> $seqres.full 2>&1
 }
@@ -81,7 +86,7 @@ test_file1="test0001"
 test_file2="test0002"
 test_file3="test0003"
 
-_scratch_mkfs_sized $((128 * 1024 * 1024)) >> $seqres.full 2>&1
+_scratch_mkfs_sized $((1 * 1024 * 1024 * 1024)) >> $seqres.full 2>&1
 
 # create scratch dir for testing
 # create some files with no name a substr of another name so we can grep later
-- 
2.51.0

Re: [PATCH] ext4/048: Fix hangup due to no free inodes

[Baokun Li]

On 2025-10-28 15:17, Ojaswin Mujoo wrote:
> We currently mkfs a 128MB filesystem, which gives use ~2048 free inodes
> on 64k blocksize. The test then keeps adding new files to a directory to
> trigger an htree split. For 64k this takes more than the total free
> inodes, which causes touch to return -ENOSPC. This leads to the while
> loop in induce_node_split() to never finish.
>
> To fix this:
> 1. Format a 1G FS which gives us atleast 16K inodes to work with.
> 2. _fail if there's any error while trying to induce node split, so we
>    dont get stuck in loop
>
> Fixes: 466ddbfd1151 ("ext4: add test for ext4_dir_entry2 wipe")
> Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
> ---

Yeah, I also hit this issue when testing LBS — file creation kept failing
without breaking out of the loop, which resulted in the test case spinning
endlessly.

Looks good to me. Feel free to add:

Reviewed-by: Baokun Li <libaokun1@huawei.com>

>  tests/ext4/048 | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/tests/ext4/048 b/tests/ext4/048
> index 2031c8c8..6343ff3a 100755
> --- a/tests/ext4/048
> +++ b/tests/ext4/048
> @@ -69,6 +69,11 @@ induce_node_split() {
>  	while [[ "$(stat --printf="%s" $testdir)" == "$dir_size" ]]; do
>  		file_num=$(($file_num + 1))
>  		touch $testdir/test"$(printf "%04d" $file_num)"
> +		local ret=$?
> +		if [[ $ret -ne 0 ]]
> +		then
> +			_fail "ERROR induce_node_split(): $ret"
> +		fi
>  	done
>  	_scratch_unmount >> $seqres.full 2>&1
>  }
> @@ -81,7 +86,7 @@ test_file1="test0001"
>  test_file2="test0002"
>  test_file3="test0003"
>  
> -_scratch_mkfs_sized $((128 * 1024 * 1024)) >> $seqres.full 2>&1
> +_scratch_mkfs_sized $((1 * 1024 * 1024 * 1024)) >> $seqres.full 2>&1
>  
>  # create scratch dir for testing
>  # create some files with no name a substr of another name so we can grep later

Re: [PATCH] ext4/048: Fix hangup due to no free inodes

[Ojaswin Mujoo]

On Tue, Oct 28, 2025 at 03:57:00PM +0800, Baokun Li wrote:
> On 2025-10-28 15:17, Ojaswin Mujoo wrote:
> > We currently mkfs a 128MB filesystem, which gives use ~2048 free inodes
> > on 64k blocksize. The test then keeps adding new files to a directory to
> > trigger an htree split. For 64k this takes more than the total free
> > inodes, which causes touch to return -ENOSPC. This leads to the while
> > loop in induce_node_split() to never finish.
> >
> > To fix this:
> > 1. Format a 1G FS which gives us atleast 16K inodes to work with.
> > 2. _fail if there's any error while trying to induce node split, so we
> >    dont get stuck in loop
> >
> > Fixes: 466ddbfd1151 ("ext4: add test for ext4_dir_entry2 wipe")
> > Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
> > ---
> 
> Yeah, I also hit this issue when testing LBS — file creation kept failing
> without breaking out of the loop, which resulted in the test case spinning
> endlessly.
> 
> Looks good to me. Feel free to add:
> 
> Reviewed-by: Baokun Li <libaokun1@huawei.com>

Hi Baokun, I was planning to CC you since I thought you might've hit it,
but missed it while sending the mail.

Thanks for the review :)

Regards,
ojaswin

> 
> >  tests/ext4/048 | 7 ++++++-
> >  1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/tests/ext4/048 b/tests/ext4/048
> > index 2031c8c8..6343ff3a 100755
> > --- a/tests/ext4/048
> > +++ b/tests/ext4/048
> > @@ -69,6 +69,11 @@ induce_node_split() {
> >  	while [[ "$(stat --printf="%s" $testdir)" == "$dir_size" ]]; do
> >  		file_num=$(($file_num + 1))
> >  		touch $testdir/test"$(printf "%04d" $file_num)"
> > +		local ret=$?
> > +		if [[ $ret -ne 0 ]]
> > +		then
> > +			_fail "ERROR induce_node_split(): $ret"
> > +		fi
> >  	done
> >  	_scratch_unmount >> $seqres.full 2>&1
> >  }
> > @@ -81,7 +86,7 @@ test_file1="test0001"
> >  test_file2="test0002"
> >  test_file3="test0003"
> >  
> > -_scratch_mkfs_sized $((128 * 1024 * 1024)) >> $seqres.full 2>&1
> > +_scratch_mkfs_sized $((1 * 1024 * 1024 * 1024)) >> $seqres.full 2>&1
> >  
> >  # create scratch dir for testing
> >  # create some files with no name a substr of another name so we can grep later
> 
> 

Re: [PATCH] ext4/048: Fix hangup due to no free inodes

[Baokun Li]

On 2025-10-28 17:39, Ojaswin Mujoo wrote:
> On Tue, Oct 28, 2025 at 03:57:00PM +0800, Baokun Li wrote:
>> On 2025-10-28 15:17, Ojaswin Mujoo wrote:
>>> We currently mkfs a 128MB filesystem, which gives use ~2048 free inodes
>>> on 64k blocksize. The test then keeps adding new files to a directory to
>>> trigger an htree split. For 64k this takes more than the total free
>>> inodes, which causes touch to return -ENOSPC. This leads to the while
>>> loop in induce_node_split() to never finish.
>>>
>>> To fix this:
>>> 1. Format a 1G FS which gives us atleast 16K inodes to work with.
>>> 2. _fail if there's any error while trying to induce node split, so we
>>>    dont get stuck in loop
>>>
>>> Fixes: 466ddbfd1151 ("ext4: add test for ext4_dir_entry2 wipe")
>>> Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
>>> ---
>> Yeah, I also hit this issue when testing LBS — file creation kept failing
>> without breaking out of the loop, which resulted in the test case spinning
>> endlessly.
>>
>> Looks good to me. Feel free to add:
>>
>> Reviewed-by: Baokun Li <libaokun1@huawei.com>
> Hi Baokun, I was planning to CC you since I thought you might've hit it,
> but missed it while sending the mail.

No worries, the mailing list can see it as well.

Thanks for testing and fixing this. 🤝


Cheers,
Baokun