From: Al Viro viro@zeniv.linux.org.uk
[ Upstream commit a890a2e339b929dbd843328f9a92a1625404fe63 ]
Theoretically it's an oopsable race, but I don't believe one can manage to hit it on real hardware; might become doable on a KVM, but it still won't be easy to attack.
Anyway, it's easy to deal with - since xdr_encode_hyper() is just a call of put_unaligned_be64(), we can put that under ->d_lock and be done with that.
Signed-off-by: Al Viro viro@zeniv.linux.org.uk Signed-off-by: Anna Schumaker anna.schumaker@oracle.com Signed-off-by: Sasha Levin sashal@kernel.org ---
LLM Generated explanations, may be completely bogus:
YES - The race is in `fs/nfs/nfs4proc.c:395`, where `NFS_FILEID(d_inode(dentry->d_parent))` relied on `d_parent` staying stable while another thread might be executing `__d_move`/rename and mutating that field under `dentry->d_lock`. Without taking that lock we can observe a transient or freed `d_parent`, leading to an oops while synthesising the “..” entry during readdir. - The fix simply wraps the dereference in `spin_lock(&dentry->d_lock)` / `spin_unlock(&dentry->d_lock)` (`fs/nfs/nfs4proc.c:394-396`), matching dcache locking rules so we never race with rename/unhash. The critical section only covers `xdr_encode_hyper` (a `put_unaligned_be64`), so there is virtually no new latency or deadlock risk. - No behavioural changes beyond closing this bug: the rest of `nfs4_setup_readdir()` is untouched, there are no dependency patches required, and this aligns with the long-standing expectation that anyone walking `d_parent` holds the dentry lock. - Because the bug can crash clients performing `readdir()` while directories are being renamed/unlinked—even if the window is narrow—this is a correctness fix with minimal risk and well within the stable backport guidelines.
fs/nfs/nfs4proc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 4de3e4bd724b7..b76da06864e53 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -391,7 +391,9 @@ static void nfs4_setup_readdir(u64 cookie, __be32 *verifier, struct dentry *dent *p++ = htonl(attrs); /* bitmap */ *p++ = htonl(12); /* attribute buffer length */ *p++ = htonl(NF4DIR); + spin_lock(&dentry->d_lock); p = xdr_encode_hyper(p, NFS_FILEID(d_inode(dentry->d_parent))); + spin_unlock(&dentry->d_lock);
readdir->pgbase = (char *)p - (char *)start; readdir->count -= readdir->pgbase;