Re: [PATCH v2 02/30] fuse: Clear setuid bit even in cache=never path

[Miklos Szeredi]

On Wed, May 15, 2019 at 03:26:47PM -0400, Vivek Goyal wrote:
> If fuse daemon is started with cache=never, fuse falls back to direct IO.
> In that write path we don't call file_remove_privs() and that means setuid
> bit is not cleared if unpriviliged user writes to a file with setuid bit set.
> 
> pjdfstest chmod test 12.t tests this and fails.

I think better sulution is to tell the server if the suid bit needs to be
removed, so it can do so in a race free way.

Here's the kernel patch, and I'll reply with the libfuse patch.

---
 fs/fuse2/file.c           |    2 ++
 include/uapi/linux/fuse.h |    3 +++
 2 files changed, 5 insertions(+)

--- a/fs/fuse2/file.c
+++ b/fs/fuse2/file.c
@@ -363,6 +363,8 @@ static ssize_t fuse_send_write(struct fu
 		inarg->flags |= O_DSYNC;
 	if (iocb->ki_flags & IOCB_SYNC)
 		inarg->flags |= O_SYNC;
+	if (!capable(CAP_FSETID))
+		inarg->write_flags |= FUSE_WRITE_KILL_PRIV;
 	req->inh.opcode = FUSE_WRITE;
 	req->inh.nodeid = ff->nodeid;
 	req->inh.len = req->inline_inlen + count;
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -125,6 +125,7 @@
  *
  *  7.29
  *  - add FUSE_NO_OPENDIR_SUPPORT flag
+ *  - add FUSE_WRITE_KILL_PRIV flag
  */
 
 #ifndef _LINUX_FUSE_H
@@ -318,9 +319,11 @@ struct fuse_file_lock {
  *
  * FUSE_WRITE_CACHE: delayed write from page cache, file handle is guessed
  * FUSE_WRITE_LOCKOWNER: lock_owner field is valid
+ * FUSE_WRITE_KILL_PRIV: kill suid and sgid bits
  */
 #define FUSE_WRITE_CACHE	(1 << 0)
 #define FUSE_WRITE_LOCKOWNER	(1 << 1)
+#define FUSE_WRITE_KILL_PRIV	(1 << 2)
 
 /**
  * Read flags