[PATCH] ubifs: Fix memory leak in do_rename

[PATCH] ubifs: Fix memory leak in do_rename

[Mårten Lindahl]

If renaming a file in an encrypted directory, function
fscrypt_setup_filename allocates memory for a file name. This name is
never used, and before returning to the caller the memory for it is not
freed.

When running kmemleak on it we see that it is registered as a leak. The
report below is triggered by a simple program 'rename' that renames a
file in an encrypted directory:

  unreferenced object 0xffff888101502840 (size 32):
    comm "rename", pid 9404, jiffies 4302582475 (age 435.735s)
    backtrace:
      __kmem_cache_alloc_node
      __kmalloc
      fscrypt_setup_filename
      do_rename
      ubifs_rename
      vfs_rename
      do_renameat2

To fix this we can remove the call to fscrypt_setup_filename as it's not
needed.

Reported-by: Zhihao Cheng <chengzhihao1@huawei.com>
Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
---
 fs/ubifs/dir.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
index 0f29cf201136..4c2dda1346e7 100644
--- a/fs/ubifs/dir.c
+++ b/fs/ubifs/dir.c
@@ -358,7 +358,6 @@ static struct inode *create_whiteout(struct inode *dir, struct dentry *dentry)
 	umode_t mode = S_IFCHR | WHITEOUT_MODE;
 	struct inode *inode;
 	struct ubifs_info *c = dir->i_sb->s_fs_info;
-	struct fscrypt_name nm;
 
 	/*
 	 * Create an inode('nlink = 1') for whiteout without updating journal,
@@ -369,10 +368,6 @@ static struct inode *create_whiteout(struct inode *dir, struct dentry *dentry)
 	dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
 		dentry, mode, dir->i_ino);
 
-	err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &nm);
-	if (err)
-		return ERR_PTR(err);
-
 	inode = ubifs_new_inode(c, dir, mode, false);
 	if (IS_ERR(inode)) {
 		err = PTR_ERR(inode);
@@ -395,7 +390,6 @@ static struct inode *create_whiteout(struct inode *dir, struct dentry *dentry)
 	make_bad_inode(inode);
 	iput(inode);
 out_free:
-	fscrypt_free_filename(&nm);
 	ubifs_err(c, "cannot create whiteout file, error %d", err);
 	return ERR_PTR(err);
 }

---
base-commit: c9c3395d5e3dcc6daee66c6908354d47bf98cb0c
change-id: 20230330-memleak_rename-fix-6f94a6f99350

Best regards,
-- 
Mårten Lindahl <marten.lindahl@axis.com>

Re: [PATCH] ubifs: Fix memory leak in do_rename

[Zhihao Cheng]

> If renaming a file in an encrypted directory, function
> fscrypt_setup_filename allocates memory for a file name. This name is
> never used, and before returning to the caller the memory for it is not
> freed.
> 
> When running kmemleak on it we see that it is registered as a leak. The
> report below is triggered by a simple program 'rename' that renames a
> file in an encrypted directory:
> 
>    unreferenced object 0xffff888101502840 (size 32):
>      comm "rename", pid 9404, jiffies 4302582475 (age 435.735s)
>      backtrace:
>        __kmem_cache_alloc_node
>        __kmalloc
>        fscrypt_setup_filename
>        do_rename
>        ubifs_rename
>        vfs_rename
>        do_renameat2
> 
> To fix this we can remove the call to fscrypt_setup_filename as it's not
> needed.
> 
> Reported-by: Zhihao Cheng <chengzhihao1@huawei.com>
> Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
> ---
>   fs/ubifs/dir.c | 6 ------
>   1 file changed, 6 deletions(-)
> 

I'm okay with this modification. It would be better to add a fix tag. 
(Fixes: 278d9a243635f26 ("ubifs: Rename whiteout atomically"))


Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com>

> diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
> index 0f29cf201136..4c2dda1346e7 100644
> --- a/fs/ubifs/dir.c
> +++ b/fs/ubifs/dir.c
> @@ -358,7 +358,6 @@ static struct inode *create_whiteout(struct inode *dir, struct dentry *dentry)
>   	umode_t mode = S_IFCHR | WHITEOUT_MODE;
>   	struct inode *inode;
>   	struct ubifs_info *c = dir->i_sb->s_fs_info;
> -	struct fscrypt_name nm;
>   
>   	/*
>   	 * Create an inode('nlink = 1') for whiteout without updating journal,
> @@ -369,10 +368,6 @@ static struct inode *create_whiteout(struct inode *dir, struct dentry *dentry)
>   	dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
>   		dentry, mode, dir->i_ino);
>   
> -	err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &nm);
> -	if (err)
> -		return ERR_PTR(err);
> -
>   	inode = ubifs_new_inode(c, dir, mode, false);
>   	if (IS_ERR(inode)) {
>   		err = PTR_ERR(inode);
> @@ -395,7 +390,6 @@ static struct inode *create_whiteout(struct inode *dir, struct dentry *dentry)
>   	make_bad_inode(inode);
>   	iput(inode);
>   out_free:
> -	fscrypt_free_filename(&nm);
>   	ubifs_err(c, "cannot create whiteout file, error %d", err);
>   	return ERR_PTR(err);
>   }
> 
> ---
> base-commit: c9c3395d5e3dcc6daee66c6908354d47bf98cb0c
> change-id: 20230330-memleak_rename-fix-6f94a6f99350
> 
> Best regards,
> 

Re: [PATCH] ubifs: Fix memory leak in do_rename

[Mårten Lindahl]

Hi Zhihao!

On 3/30/23 16:17, Zhihao Cheng wrote:
>> If renaming a file in an encrypted directory, function
>> fscrypt_setup_filename allocates memory for a file name. This name is
>> never used, and before returning to the caller the memory for it is not
>> freed.
>>
>> When running kmemleak on it we see that it is registered as a leak. The
>> report below is triggered by a simple program 'rename' that renames a
>> file in an encrypted directory:
>>
>>    unreferenced object 0xffff888101502840 (size 32):
>>      comm "rename", pid 9404, jiffies 4302582475 (age 435.735s)
>>      backtrace:
>>        __kmem_cache_alloc_node
>>        __kmalloc
>>        fscrypt_setup_filename
>>        do_rename
>>        ubifs_rename
>>        vfs_rename
>>        do_renameat2
>>
>> To fix this we can remove the call to fscrypt_setup_filename as it's not
>> needed.
>>
>> Reported-by: Zhihao Cheng <chengzhihao1@huawei.com>
>> Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
>> ---
>>   fs/ubifs/dir.c | 6 ------
>>   1 file changed, 6 deletions(-)
>>
>
> I'm okay with this modification. It would be better to add a fix tag. 
> (Fixes: 278d9a243635f26 ("ubifs: Rename whiteout atomically"))
>
>
> Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com>

Thanks! I'll send a v2 with the Fixes tag and your Reviewed-by tag.

Kind regards

Mårten

>
>> diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
>> index 0f29cf201136..4c2dda1346e7 100644
>> --- a/fs/ubifs/dir.c
>> +++ b/fs/ubifs/dir.c
>> @@ -358,7 +358,6 @@ static struct inode *create_whiteout(struct inode 
>> *dir, struct dentry *dentry)
>>       umode_t mode = S_IFCHR | WHITEOUT_MODE;
>>       struct inode *inode;
>>       struct ubifs_info *c = dir->i_sb->s_fs_info;
>> -    struct fscrypt_name nm;
>>         /*
>>        * Create an inode('nlink = 1') for whiteout without updating 
>> journal,
>> @@ -369,10 +368,6 @@ static struct inode *create_whiteout(struct 
>> inode *dir, struct dentry *dentry)
>>       dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
>>           dentry, mode, dir->i_ino);
>>   -    err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &nm);
>> -    if (err)
>> -        return ERR_PTR(err);
>> -
>>       inode = ubifs_new_inode(c, dir, mode, false);
>>       if (IS_ERR(inode)) {
>>           err = PTR_ERR(inode);
>> @@ -395,7 +390,6 @@ static struct inode *create_whiteout(struct inode 
>> *dir, struct dentry *dentry)
>>       make_bad_inode(inode);
>>       iput(inode);
>>   out_free:
>> -    fscrypt_free_filename(&nm);
>>       ubifs_err(c, "cannot create whiteout file, error %d", err);
>>       return ERR_PTR(err);
>>   }
>>
>> ---
>> base-commit: c9c3395d5e3dcc6daee66c6908354d47bf98cb0c
>> change-id: 20230330-memleak_rename-fix-6f94a6f99350
>>
>> Best regards,
>>
>