[PATCH] fs/dcache: use write lock in the fallback instead of read lock

[PATCH] fs/dcache: use write lock in the fallback instead of read lock

[Niels Dossche]

Currently, there is a fallback with a WARN that uses down_read_trylock
as a safety measure for when there is no lock taken. The current
callsites expect a write lock to be taken. Moreover, the s_root field
is written to, which is not allowed under a read lock.
This code safety fallback should not be executed unless there is an
issue somewhere else.
The fix is to change the read lock to a write lock in the fallback.

Note:
I am currently working on a static analyser to detect missing locks
using type-based static analysis as my master's thesis
in order to obtain my master's degree.
If you would like to have more details, please let me know.
This was a reported case. I manually verified the report by looking
at the code, so that I do not send wrong information or patches.
After concluding that this seems to be a true positive, I created
this patch. I have both compile-tested this patch and runtime-tested
this patch on x86_64. The effect on a running system could be a
potential race condition in exceptional cases.
This issue was found on Linux v5.17.

Fixes: c636ebdb186bf ("VFS: Destroy the dentries contributed by a superblock on unmounting")
Signed-off-by: Niels Dossche <dossche.niels@gmail.com>
---
 fs/dcache.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/dcache.c b/fs/dcache.c
index c84269c6e8bf..d81f5b9c2bce 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -1692,7 +1692,7 @@ void shrink_dcache_for_umount(struct super_block *sb)
 {
 	struct dentry *dentry;
 
-	WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
+	WARN(down_write_trylock(&sb->s_umount), "s_umount should've been locked");
 
 	dentry = sb->s_root;
 	sb->s_root = NULL;
-- 
2.35.1

Re: [PATCH] fs/dcache: use write lock in the fallback instead of read lock

[Christoph Hellwig]

> @@ -1692,7 +1692,7 @@ void shrink_dcache_for_umount(struct super_block *sb)
>  {
>  	struct dentry *dentry;
>  
> -	WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
> +	WARN(down_write_trylock(&sb->s_umount), "s_umount should've been locked");

This really should be a lockdep_assert_held_write() instead.

Re: [PATCH] fs/dcache: use write lock in the fallback instead of read lock

[Niels Dossche]

On 25/03/2022 17:47, Christoph Hellwig wrote:
>> @@ -1692,7 +1692,7 @@ void shrink_dcache_for_umount(struct super_block *sb)
>>  {
>>  	struct dentry *dentry;
>>  
>> -	WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
>> +	WARN(down_write_trylock(&sb->s_umount), "s_umount should've been locked");
> 
> This really should be a lockdep_assert_held_write() instead.

That's probably a bit nicer indeed.
I can write up a patch that does a lockdep_assert_held_write() if you want.

Re: [PATCH] fs/dcache: use write lock in the fallback instead of read lock

[Christoph Hellwig]

On Fri, Mar 25, 2022 at 05:48:43PM +0100, Niels Dossche wrote:
> >> -	WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
> >> +	WARN(down_write_trylock(&sb->s_umount), "s_umount should've been locked");
> > 
> > This really should be a lockdep_assert_held_write() instead.
> 
> That's probably a bit nicer indeed.
> I can write up a patch that does a lockdep_assert_held_write() if you want.

I would much prefer that.