[Cluster-devel] [PATCH] gfs2_grow: write one rindex entry and then the rest

[Cluster-devel] [PATCH] gfs2_grow: write one rindex entry and then the rest

[Benjamin Marzinski]

There is a problem with gfs2_grow when the rindex file is stuffed, or the
filesystem blocksize is smaller than the pagesize. When gfs2_grow writes
out all the rindex entries at once, there needs to be enough space left for
it to be able to write the first page of data without doing any allocations.
If there isn't gfs2_grow will fail, even though there is space for
additional resrouce groups in the last block of the rindex file. This patch
makes gfs2_grow write one rindex entry frist, to grow the filesystem by
one resource group's worth of space. Then it writes the rest.  Since
the code guarantees that there will always be enough space in the rindex
file for one more rindex entry, the first write will never require an
allocation.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
---
 gfs2/mkfs/main_grow.c |   14 ++++++++++++--
 1 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/gfs2/mkfs/main_grow.c b/gfs2/mkfs/main_grow.c
index 03eb57b..81f492b 100644
--- a/gfs2/mkfs/main_grow.c
+++ b/gfs2/mkfs/main_grow.c
@@ -244,8 +244,18 @@ static void fix_rindex(struct gfs2_sbd *sdp, int rindex_fd, int old_rg_count)
 		}
 		/* Now write the new RGs to the end of the rindex */
 		lseek(rindex_fd, 0, SEEK_END);
-		count = write(rindex_fd, buf, writelen);
-		if (count != writelen) {
+		count = write(rindex_fd, buf, sizeof(struct gfs2_rindex));
+		if (count != sizeof(struct gfs2_rindex)) {
+			log_crit("Error writing first new rindex entry;"
+				 "aborted.\n");
+			if (count > 0)
+				goto trunc;
+			else
+				goto out;
+		}
+		count = write(rindex_fd, buf + sizeof(struct gfs2_rindex),
+			      writelen - sizeof(struct gfs2_rindex));
+		if (count != writelen - sizeof(struct gfs2_rindex)) {
 			log_crit("Error writing new rindex entries;"
 				 "aborted.\n");
 			if (count > 0)
-- 
1.5.3.3

[Cluster-devel] [PATCH] gfs2_grow: write one rindex entry and then the rest

[Steven Whitehouse]

Hi,

Looks good to me,

Steve.

On Thu, 2011-05-26 at 15:54 -0500, Benjamin Marzinski wrote:
> There is a problem with gfs2_grow when the rindex file is stuffed, or the
> filesystem blocksize is smaller than the pagesize. When gfs2_grow writes
> out all the rindex entries at once, there needs to be enough space left for
> it to be able to write the first page of data without doing any allocations.
> If there isn't gfs2_grow will fail, even though there is space for
> additional resrouce groups in the last block of the rindex file. This patch
> makes gfs2_grow write one rindex entry frist, to grow the filesystem by
> one resource group's worth of space. Then it writes the rest.  Since
> the code guarantees that there will always be enough space in the rindex
> file for one more rindex entry, the first write will never require an
> allocation.
> 
> Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
> ---
>  gfs2/mkfs/main_grow.c |   14 ++++++++++++--
>  1 files changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/gfs2/mkfs/main_grow.c b/gfs2/mkfs/main_grow.c
> index 03eb57b..81f492b 100644
> --- a/gfs2/mkfs/main_grow.c
> +++ b/gfs2/mkfs/main_grow.c
> @@ -244,8 +244,18 @@ static void fix_rindex(struct gfs2_sbd *sdp, int rindex_fd, int old_rg_count)
>  		}
>  		/* Now write the new RGs to the end of the rindex */
>  		lseek(rindex_fd, 0, SEEK_END);
> -		count = write(rindex_fd, buf, writelen);
> -		if (count != writelen) {
> +		count = write(rindex_fd, buf, sizeof(struct gfs2_rindex));
> +		if (count != sizeof(struct gfs2_rindex)) {
> +			log_crit("Error writing first new rindex entry;"
> +				 "aborted.\n");
> +			if (count > 0)
> +				goto trunc;
> +			else
> +				goto out;
> +		}
> +		count = write(rindex_fd, buf + sizeof(struct gfs2_rindex),
> +			      writelen - sizeof(struct gfs2_rindex));
> +		if (count != writelen - sizeof(struct gfs2_rindex)) {
>  			log_crit("Error writing new rindex entries;"
>  				 "aborted.\n");
>  			if (count > 0)