2016-12-20 15:23:30

by Eric Sandeen

Subject: [PATCH] libext2fs: don't ignore fsync errors

Today, if mke2fs experiences IO errors (say, on a thin device
which filled up during mkfs), mke2fs is silent and returns
success even though the filesystem was not properly created.

Catch errors from the io_channel_flush() callchain to
fix this up. Fix formatting of the printed error as
well:

...
Creating journal (262144 blocks): done
Writing superblocks and filesystem accounting information:
Warning, had trouble writing out superblocks.
# echo $?
5

Signed-off-by: Eric Sandeen <[email protected]>
---

I'm not actually sure what's supposed to be returned
from unix_flush; it's an errcode_t; is a standard error
number (EIO) ok or is there a libcom_err for that?
Anyway, seems to work ok.

diff --git a/lib/ext2fs/closefs.c b/lib/ext2fs/closefs.c
index b69fa09..b255759 100644
--- a/lib/ext2fs/closefs.c
+++ b/lib/ext2fs/closefs.c
@@ -429,16 +429,22 @@ write_primary_superblock_only:
if (retval)
return retval;

- if (!(flags & EXT2_FLAG_FLUSH_NO_SYNC))
+ if (!(flags & EXT2_FLAG_FLUSH_NO_SYNC)) {
retval = io_channel_flush(fs->io);
+ if (retval)
+ goto errout;
+ }
retval = write_primary_superblock(fs, super_shadow);
if (retval)
goto errout;

fs->flags &= ~EXT2_FLAG_DIRTY;

- if (!(flags & EXT2_FLAG_FLUSH_NO_SYNC))
+ if (!(flags & EXT2_FLAG_FLUSH_NO_SYNC)) {
retval = io_channel_flush(fs->io);
+ if (retval)
+ goto errout;
+ }
errout:
fs->super->s_state = fs_state;
#ifdef WORDS_BIGENDIAN
diff --git a/lib/ext2fs/unix_io.c b/lib/ext2fs/unix_io.c
index 429ea24..f4e6148 100644
--- a/lib/ext2fs/unix_io.c
+++ b/lib/ext2fs/unix_io.c
@@ -1030,7 +1030,8 @@ static errcode_t unix_flush(io_channel channel)
#ifndef NO_IO_CACHE
retval = flush_cached_blocks(channel, data, 0);
#endif
- fsync(data->dev);
+ if (!retval && fsync(data->dev) != 0)
+ return errno;
return retval;
}

diff --git a/misc/mke2fs.c b/misc/mke2fs.c
index 9f18c83..49c6e94 100644
--- a/misc/mke2fs.c
+++ b/misc/mke2fs.c
@@ -3227,7 +3227,7 @@ no_journal:
retval = ext2fs_close_free(&fs);
if (retval) {
fprintf(stderr, "%s",
- _("\nWarning, had trouble writing out superblocks."));
+ _("\nWarning, had trouble writing out superblocks.\n"));
} else if (!quiet) {
printf("%s", _("done\n\n"));
if (!getenv("MKE2FS_SKIP_CHECK_MSG"))

2017-01-13 17:32:12

by Theodore Ts'o

Subject: Re: [PATCH] libext2fs: don't ignore fsync errors

On Tue, Dec 20, 2016 at 09:23:29AM -0600, Eric Sandeen wrote:
> Today, if mke2fs experiences IO errors (say, on a thin device
> which filled up during mkfs), mke2fs is silent and returns
> success even though the filesystem was not properly created.
>
> Catch errors from the io_channel_flush() callchain to
> fix this up. Fix formatting of the printed error as
> well:
>
> ...
> Creating journal (262144 blocks): done
> Writing superblocks and filesystem accounting information:
> Warning, had trouble writing out superblocks.
> # echo $?
> 5
>
> Signed-off-by: Eric Sandeen <[email protected]>

Applied, thanks.

- Ted