On Mon, Dec 03, 2018 at 07:53:42AM +1100, Dave Chinner wrote:
> From: Dave Chinner <dchinner@xxxxxxxxxx>
>
> So that O_RDONLY open commands (such as from copy_range) do not
> block forever waiting on a non-existent writer.
>
> Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>
> ---
> io/open.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/io/open.c b/io/open.c
> index 6ea3e9a2019f..b1d9a0fa317c 100644
> --- a/io/open.c
> +++ b/io/open.c
> @@ -56,6 +56,7 @@ openfile(
> struct fs_path *fs_path)
> {
> struct fs_path *fsp;
> + struct stat st;
> int fd;
> int oflags;
>
> @@ -79,6 +80,18 @@ openfile(
> if (flags & IO_NOFOLLOW)
> oflags |= O_NOFOLLOW;
>
> + /*
> + * if we've been passed a pipe to open, don't block waiting for a
> + * reader or writer to appear. We want to either succeed or error out
> + * immediately.
> + */
> + if (stat(path, &st) < 0 && errno != ENOENT) {
> + perror("stat");
> + return -1;
> + }
> + if (S_ISFIFO(st.st_mode))
Hmm... does stat() set st.st_mode even if it returns ENOENT? Or are we
possibly branching on an uninitialized st here?
struct stat st = { 0 }; ?
The rest seems fine to me.
--D
> + oflags |= O_NONBLOCK;
> +
> fd = open(path, oflags, mode);
> if (fd < 0) {
> if (errno == EISDIR &&
> --
> 2.19.1
>
