Re: [PATCH] btrfs: tests: Fix memory leak in btrfs_test_qgroups()

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

在 2025/12/26 07:19, Qu Wenruo 写道:
> 
> 
> 在 2025/12/25 20:57, Zilin Guan 写道:
>> btrfs_alloc_dummy_root() allocates a root with a reference count of 1.
>> Then btrfs_insert_fs_root() is used to insert the root into the fs_info.
>> On success, it increments the reference count. On failure, it does not.
>>
>> Currently, if btrfs_insert_fs_root() fails, the error handling path
>> jumps to the out label immediately without decrementing the reference
>> count of tmp_root, leading to a memory leak.
>>
>> Fix this by calling btrfs_put_root() unconditionally after
>> btrfs_insert_fs_root(). This correctly handles both cases: on success,
>> it drops the local reference, leaving the root with the reference held
>> by fs_info; on failure, it drops the sole reference, freeing the root.
>>
>> Fixes: 4785e24fa5d23 ("btrfs: don't take an extra root ref at 
>> allocation time")
>> Signed-off-by: Zilin Guan <zilin@seu.edu.cn>
>> ---
>>   fs/btrfs/tests/qgroup-tests.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/btrfs/tests/qgroup-tests.c b/fs/btrfs/tests/qgroup- 
>> tests.c
>> index e9124605974b..0d51e0abaeac 100644
>> --- a/fs/btrfs/tests/qgroup-tests.c
>> +++ b/fs/btrfs/tests/qgroup-tests.c
>> @@ -517,11 +517,11 @@ int btrfs_test_qgroups(u32 sectorsize, u32 
>> nodesize)
>>       tmp_root->root_key.objectid = BTRFS_FS_TREE_OBJECTID;
>>       root->fs_info->fs_root = tmp_root;
>>       ret = btrfs_insert_fs_root(root->fs_info, tmp_root);
>> +    btrfs_put_root(tmp_root);
>>       if (ret) {
>>           test_err("couldn't insert fs root %d", ret);
>>           goto out;
> 
> This will lead to double free.
> 
> If btrfs_insert_fs_root() failed, btrfs_put_root() will do the cleaning 
> and free the root.
> 
> Then btrfs_free_dummy_root() will call btrfs_put_root() again on the 
> root, cause use-after-free.
> 
> So your analyze is completely wrong.

And considering you're sending quite some patches and most of them are 
rejected, next time if you believe you find some 
memory-leak/use-after-free, please hit them in the real world with 
kmemleak/kasan enabled with the call trace.

I believe kmemleak/kasan more than you, and that will save us a lot of time.

> 
> Thanks,
> Qu
> 
>>       }
>> -    btrfs_put_root(tmp_root);
>>       tmp_root = btrfs_alloc_dummy_root(fs_info);
>>       if (IS_ERR(tmp_root)) {
>> @@ -532,11 +532,11 @@ int btrfs_test_qgroups(u32 sectorsize, u32 
>> nodesize)
>>       tmp_root->root_key.objectid = BTRFS_FIRST_FREE_OBJECTID;
>>       ret = btrfs_insert_fs_root(root->fs_info, tmp_root);
>> +    btrfs_put_root(tmp_root);
>>       if (ret) {
>>           test_err("couldn't insert fs root %d", ret);
>>           goto out;
>>       }
>> -    btrfs_put_root(tmp_root);
>>       test_msg("running qgroup tests");
>>       ret = test_no_shared_qgroup(root, sectorsize, nodesize);
> 
>