[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20250721170147.GT2672070@frogsfrogsfrogs>
Date: Mon, 21 Jul 2025 10:01:47 -0700
From: "Darrick J. Wong" <djwong@...nel.org>
To: Theodore Ts'o <tytso@....edu>
Cc: Sam James <sam@...too.org>, linux-ext4@...r.kernel.org
Subject: Re: [PATCH 2/5] fuse2fs: stop aliasing stderr with ff->err_fp
On Mon, Jul 21, 2025 at 07:41:17AM -0400, Theodore Ts'o wrote:
> On Sun, Jul 20, 2025 at 10:42:34PM -0700, Darrick J. Wong wrote:
> > Oh wait no it turns out that libfuse obliterates std{in,out,err} in
> > fuse_daemonize() by opening /dev/null and using that exact trick. So
> > the only reason why I was ever getting any FUSE2FS debug output
> > throughout *any* of the fuse+iomap development sprints was that glibc
> > lets you assign stdout/stderr directly.
> >
> > freopen also won't work (at least on glibc) because its freopen
> > implementation uses the dup2 trick which will be undone by libfuse.
> >
> > GREAT! I only got to debug my program because OF A WEIRD GLIBC QUIRK!!
>
> So either we need to find some way to inhibit fuse_daemonize() and
> then have fuse2fs handling doing the daemonization. Or maybe we can
> arrange to have some kind callback to set up stdout/stderr after
> fuse_main() is called? fuse_daemonize() is called from fuse_main(),
> right?
Right. We could just set up the log file again in op_init, which will
fix the daemonize() problem; use dup2 to cover any libraries that output
directly to STDOUT/ERR_FILE; and use freopen on /dev/fd/XX which will
cover any other libraries that call printf or fprintf(stdout/err.
$ diffstat < patches-djwong-dev/019*
fuse2fs.c | 141 +++++++++++++++++++++++++----
1 file changed, 123 insertions(+), 18 deletions(-)
Yuuuuuck.
--D
> - Ted
Powered by blists - more mailing lists