git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH] notes-merge: use O_EXCL to avoid overwriting existing files

From
Jeff King <peff@peff.net>
Date
Jul 8, 2016, 09:37 UTC
Message-ID
<20160708093733.GA20528@sigill.intra.peff.net>
In-Reply-To
<577EB6BE.6090504@web.de>
On Thu, Jul 07, 2016 at 10:08:30PM +0200, René Scharfe wrote:
Show 15 quoted lines
> diff --git a/notes-merge.c b/notes-merge.c
> index b7814c9..2b29fc4 100644
> --- a/notes-merge.c
> +++ b/notes-merge.c
> @@ -298,12 +298,8 @@ static void write_buf_to_worktree(const unsigned char *obj,
>  	char *path = git_pathdup(NOTES_MERGE_WORKTREE "/%s", sha1_to_hex(obj));
>  	if (safe_create_leading_directories_const(path))
>  		die_errno("unable to create directory for '%s'", path);
> -	if (file_exists(path))
> -		die("found existing file at '%s'", path);
>  
> -	fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, 0666);
> -	if (fd < 0)
> -		die_errno("failed to open '%s'", path);
> +	fd = xopen(path, O_WRONLY | O_EXCL | O_CREAT, 0666);

While working on write_file_buf() elsewhere, I wondered if this was a good candidate for conversion. But it's not because of the O_EXCL.

I was tempted by something like:
  write_file_buf(path, O_EXCL, buf, len);

but I think that just makes the interface more cluttered for the other callers. If you are doing something clever with O_EXCL, you probably should just do it by hand.

However, we can make the loop less painful, as below (on top of your patch).

-- >8 --
Subject: [PATCH] notes-merge: use write_or_die()

This output loop is overkill. For one thing, we do not need to loop on write_in_full(); it's entire purpose is to avoid looping, and it even includes the "disk full?" check itself.

Secondly, the check for EPIPE is pointless. We know this is an on-disk file that we just opened, so short of somebody sneaking a FIFO into our notes-merge working tree, it will not be a pipe. And even if it is, we would generally die of SIGPIPE before hitting this code. And even if we somehow ignored SIGPIPE, dying (rather than silently ignoring the error) seems like the only sensible action anyway.

So what we're left with is calling write_in_full() and dying if it fails. And that's exactly what write_or_die() does.

Signed-off-by: Jeff King <peff@peff.net>
---
 notes-merge.c | 16 +---------------
 1 file changed, 1 insertion(+), 15 deletions(-)
diff --git a/notes-merge.c b/notes-merge.c
index 2b29fc4..a659a62 100644
--- a/notes-merge.c
+++ b/notes-merge.c
@@ -300,21 +300,7 @@ static void write_buf_to_worktree(const unsigned char *obj,
 		die_errno("unable to create directory for '%s'", path);
 
 	fd = xopen(path, O_WRONLY | O_EXCL | O_CREAT, 0666);
-
-	while (size > 0) {
-		long ret = write_in_full(fd, buf, size);
-		if (ret < 0) {
-			/* Ignore epipe */
-			if (errno == EPIPE)
-				break;
-			die_errno("notes-merge");
-		} else if (!ret) {
-			die("notes-merge: disk full?");
-		}
-		size -= ret;
-		buf += ret;
-	}
-
+	write_or_die(fd, buf, size);
 	close(fd);
 	free(path);
 }
-- 
2.9.0.393.g704e522
Previous: Junio C Hamano
Message 4 of 4 in “notes-merge: use O_EXCL to avoid overwriting existing files”
  1. notes-merge: use O_EXCL to avoid overwriting existing filesRené Scharfe, Jul 7, 2016
  2. Jeff KingJul 7, 2016
  3. Junio C HamanoJul 7, 2016
  4. Jeff KingJul 8, 2016

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.