AUDIT_TRIM is expected to be idempotent, but multiple executions resulted in a
refcount underflow and use-after-free.
git bisect fingered commit fb041bb7c0a918b95c6889fc965cdc4a75b4c0ca (2019-11)
("locking/refcount: Consolidate implementations of refcount_t")
but this patch with its more thorough checking that wasn't in the x86 assembly
code merely exposed a previously existing tree refcount imbalance in the case
of tree trimming code that was refactored with prune_one() to remove a tree
introduced in commit 8432c70062978d9a57bde6715496d585ec520c3e (2018-11)
("audit: Simplify locking around untag_chunk()")
Move the put_tree() to cover only the prune_one() case.
Passes audit-testsuite and 3 passes of "auditctl -t" with at least one
directory watch.
Fixes: 8432c7006297 ("audit: Simplify locking around untag_chunk()")
Signed-off-by: Richard Guy Briggs <[email protected]>
Cc: Jan Kara <[email protected]>
Cc: Will Deacon <[email protected]>
Cc: Alexander Viro <[email protected]>
Cc: Seiji Nishikawa <[email protected]>
---
kernel/audit_tree.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c
index b2be4e978ba3..2cd7b5694422 100644
--- a/kernel/audit_tree.c
+++ b/kernel/audit_tree.c
@@ -593,7 +593,6 @@ static void prune_tree_chunks(struct audit_tree *victim, bool tagged)
spin_lock(&hash_lock);
}
spin_unlock(&hash_lock);
- put_tree(victim);
}
/*
@@ -602,6 +601,7 @@ static void prune_tree_chunks(struct audit_tree *victim, bool tagged)
static void prune_one(struct audit_tree *victim)
{
prune_tree_chunks(victim, false);
+ put_tree(victim);
}
/* trim the uncommitted chunks from tree */
--
2.27.0
On Mon 23-08-21 22:04:09, Richard Guy Briggs wrote:
> AUDIT_TRIM is expected to be idempotent, but multiple executions resulted in a
> refcount underflow and use-after-free.
>
> git bisect fingered commit fb041bb7c0a918b95c6889fc965cdc4a75b4c0ca (2019-11)
> ("locking/refcount: Consolidate implementations of refcount_t")
> but this patch with its more thorough checking that wasn't in the x86 assembly
> code merely exposed a previously existing tree refcount imbalance in the case
> of tree trimming code that was refactored with prune_one() to remove a tree
> introduced in commit 8432c70062978d9a57bde6715496d585ec520c3e (2018-11)
> ("audit: Simplify locking around untag_chunk()")
>
> Move the put_tree() to cover only the prune_one() case.
>
> Passes audit-testsuite and 3 passes of "auditctl -t" with at least one
> directory watch.
>
> Fixes: 8432c7006297 ("audit: Simplify locking around untag_chunk()")
> Signed-off-by: Richard Guy Briggs <[email protected]>
> Cc: Jan Kara <[email protected]>
> Cc: Will Deacon <[email protected]>
> Cc: Alexander Viro <[email protected]>
> Cc: Seiji Nishikawa <[email protected]>
Ah, that was indeed a stupid mistake. Thanks for debugging this! The patch
looks good to me. You can add:
Reviewed-by: Jan Kara <[email protected]>
Honza
> ---
> kernel/audit_tree.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c
> index b2be4e978ba3..2cd7b5694422 100644
> --- a/kernel/audit_tree.c
> +++ b/kernel/audit_tree.c
> @@ -593,7 +593,6 @@ static void prune_tree_chunks(struct audit_tree *victim, bool tagged)
> spin_lock(&hash_lock);
> }
> spin_unlock(&hash_lock);
> - put_tree(victim);
> }
>
> /*
> @@ -602,6 +601,7 @@ static void prune_tree_chunks(struct audit_tree *victim, bool tagged)
> static void prune_one(struct audit_tree *victim)
> {
> prune_tree_chunks(victim, false);
> + put_tree(victim);
> }
>
> /* trim the uncommitted chunks from tree */
> --
> 2.27.0
>
--
Jan Kara <[email protected]>
SUSE Labs, CR
On Mon, Aug 23, 2021 at 10:05 PM Richard Guy Briggs <[email protected]> wrote:
>
> AUDIT_TRIM is expected to be idempotent, but multiple executions resulted in a
> refcount underflow and use-after-free.
>
> git bisect fingered commit fb041bb7c0a918b95c6889fc965cdc4a75b4c0ca (2019-11)
> ("locking/refcount: Consolidate implementations of refcount_t")
> but this patch with its more thorough checking that wasn't in the x86 assembly
> code merely exposed a previously existing tree refcount imbalance in the case
> of tree trimming code that was refactored with prune_one() to remove a tree
> introduced in commit 8432c70062978d9a57bde6715496d585ec520c3e (2018-11)
> ("audit: Simplify locking around untag_chunk()")
>
> Move the put_tree() to cover only the prune_one() case.
>
> Passes audit-testsuite and 3 passes of "auditctl -t" with at least one
> directory watch.
>
> Fixes: 8432c7006297 ("audit: Simplify locking around untag_chunk()")
> Signed-off-by: Richard Guy Briggs <[email protected]>
> Cc: Jan Kara <[email protected]>
> Cc: Will Deacon <[email protected]>
> Cc: Alexander Viro <[email protected]>
> Cc: Seiji Nishikawa <[email protected]>
> ---
> kernel/audit_tree.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
First a quick comment about the commit description, when referencing
specific commits in the description please use the same format that
you used in the "Fixes:" tag; it makes the description easier to read.
No need to resend, I fixed it when I merged your patch, but something
to keep in mind for the future.
As for the patch itself, thanks for finding this and sending a fix.
Normally this is something I would send up to Linus at the end of the
week during the -rcX phase, but since we are currently at -rc7 I'm
going to simply add the -stable marking and merge it into audit/next
to get pushed up to Linus early next week, assuming we see v5.14
released this Sunday. If for some reason we see a v5.14-rc8 next week
I'll adjust things and send it to Linus as a -stable patch.
--
paul moore
http://www.paul-moore.com