Re: [PATCH] btrfs: add a sanity check for btrfs root

From: Qu Wenruo
Date: Fri Oct 25 2024 - 17:06:41 EST




在 2024/10/26 04:33, David Sterba 写道:
On Fri, Oct 25, 2024 at 08:23:07PM +1030, Qu Wenruo wrote:


在 2024/10/25 15:25, Lizhi Xu 写道:
Syzbot report a null-ptr-deref in btrfs_search_slot.
It use the input logical can't find the extent root in extent_from_logical,
and triger the null-ptr-deref in btrfs_search_slot.
Add sanity check for btrfs root before using it in btrfs_search_slot.

Although I'd prefer to explain a little more about why the NULL root
pointer can happen (caused by the rescue=all mount option), which can
cause NULL root pointer for non-critical tree roots, like
uuid/csum/extent or even device trees.

You don't need to bother sending an update.
I can add such info when pushing to the maintainer's tree.


Reported-by: syzbot+3030e17bd57a73d39bd7@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://syzkaller.appspot.com/bug?extid=3030e17bd57a73d39bd7
Signed-off-by: Lizhi Xu <lizhi.xu@xxxxxxxxxxxxx>

Reviewed-by: Qu Wenruo <wqu@xxxxxxxx>

@@ -2023,6 +2023,10 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
int min_write_lock_level;
int prev_cmp;

+ if (!root)
+ return -EINVAL;

The function returns errors indirectly so it's not clear which could be
ultimately returned. I did a quick search over the calls starting in
btrfs_search_slot() and it seems that EINVAL is not used so we'd know if
it ends up in some error report. The ones I found: EAGAIN, EIO, EUCLEAN,
ENOMEM.


If you want, I can add extra (ratelimited though) error/warning message
for such cases.

Considering this is only possible for rescue=all cases, extra error
messages should be fine.

Or do you prefer some more rare return values?

Thanks,
Qu