Received-SPF: pass (google.com: domain of linux-kernel+bounces-143930-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Danny Lin <danny@orbstack.dev>
To: danny@orbstack.dev
Cc: stable@vger.kernel.org,
	Miklos Szeredi <miklos@szeredi.hu>,
	linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: [PATCH] fuse: fix leaked ENOSYS error on first statx call
Date: Sat, 13 Apr 2024 17:34:31 -0700
Message-ID: <20240414003434.2659-1-danny@orbstack.dev>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

FUSE attempts to detect server support for statx by trying it once and
setting no_statx=1 if it fails with ENOSYS, but consider the following
scenario:

- Userspace (e.g. sh) calls stat() on a file
  * succeeds
- Userspace (e.g. lsd) calls statx(BTIME) on the same file
  - request_mask = STATX_BASIC_STATS | STATX_BTIME
  - first pass: sync=true due to differing cache_mask
  - statx fails and returns ENOSYS
  - set no_statx and retry
  - retry sets mask = STATX_BASIC_STATS
  - now mask == cache_mask; sync=false (time_before: still valid)
  - so we take the "else if (stat)" path
  - "err" is still ENOSYS from the failed statx call

Fix this by zeroing "err" before retrying the failed call.

Fixes: d3045530bdd2 ("fuse: implement statx")
Cc: stable@vger.kernel.org # v6.6
Signed-off-by: Danny Lin <danny@orbstack.dev>
---
 fs/fuse/dir.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index de452cdbf3cf..a63125ce70a4 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -1362,6 +1362,7 @@ static int fuse_update_get_attr(struct inode *inode, struct file *file,
 			err = fuse_do_statx(inode, file, stat);
 			if (err == -ENOSYS) {
 				fc->no_statx = 1;
+				err = 0;
 				goto retry;
 			}
 		} else {
-- 
2.44.0