diff mbox series

[2/2] xfs_io: don't count fsmaps before querying fsmaps

Message ID 162528110051.38807.5958877066692397152.stgit@locust (mailing list archive)
State Superseded
Headers show
Series xfs_io: small fixes to fsmap command | expand

Commit Message

Darrick J. Wong July 3, 2021, 2:58 a.m. UTC 
From: Darrick J. Wong <djwong@kernel.org>

There's a bunch of code in fsmap.c that tries to count the GETFSMAP
records so that it can size the fsmap array appropriately for the
GETFSMAP call.  It's pointless to iterate the entire result set /twice/
(unlike the bmap command where the extent count is actually stored in
the fs metadata), so get rid of the duplicate walk.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
---
 io/fsmap.c |   30 ------------------------------
 1 file changed, 30 deletions(-)

Comments

Christoph Hellwig July 5, 2021, 3:25 p.m. UTC | #1 
On Fri, Jul 02, 2021 at 07:58:20PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@kernel.org>
> 
> There's a bunch of code in fsmap.c that tries to count the GETFSMAP
> records so that it can size the fsmap array appropriately for the
> GETFSMAP call.  It's pointless to iterate the entire result set /twice/
> (unlike the bmap command where the extent count is actually stored in
> the fs metadata), so get rid of the duplicate walk.

In otherwords:  just keep iterating over the records using the default
chunk size instead of doing one call to find the size and then do
a giant allocation and GETFSMAP call.

I find the current commit log a little confusing, but the change itself
looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>
Darrick J. Wong July 6, 2021, 6:30 p.m. UTC | #2 
On Mon, Jul 05, 2021 at 04:25:27PM +0100, Christoph Hellwig wrote:
> On Fri, Jul 02, 2021 at 07:58:20PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > There's a bunch of code in fsmap.c that tries to count the GETFSMAP
> > records so that it can size the fsmap array appropriately for the
> > GETFSMAP call.  It's pointless to iterate the entire result set /twice/
> > (unlike the bmap command where the extent count is actually stored in
> > the fs metadata), so get rid of the duplicate walk.
> 
> In otherwords:  just keep iterating over the records using the default
> chunk size instead of doing one call to find the size and then do
> a giant allocation and GETFSMAP call.

I'll paste this into the commit log, thanks.

--D

> I find the current commit log a little confusing, but the change itself
> looks good:
> 
> Reviewed-by: Christoph Hellwig <hch@lst.de>
diff mbox series

Patch

diff --git a/io/fsmap.c b/io/fsmap.c
index 9f179fa8..f540a7c0 100644
--- a/io/fsmap.c
+++ b/io/fsmap.c
@@ -372,13 +372,11 @@  fsmap_f(
 	char			**argv)
 {
 	struct fsmap		*p;
-	struct fsmap_head	*nhead;
 	struct fsmap_head	*head;
 	struct fsmap		*l, *h;
 	struct xfs_fsop_geom	fsgeo;
 	long long		start = 0;
 	long long		end = -1;
-	int			nmap_size;
 	int			map_size;
 	int			nflag = 0;
 	int			vflag = 0;
@@ -492,34 +490,6 @@  fsmap_f(
 	h->fmr_flags = UINT_MAX;
 	h->fmr_offset = ULLONG_MAX;
 
-	/* Count mappings */
-	if (!nflag) {
-		head->fmh_count = 0;
-		i = ioctl(file->fd, FS_IOC_GETFSMAP, head);
-		if (i < 0) {
-			fprintf(stderr, _("%s: xfsctl(XFS_IOC_GETFSMAP)"
-				" iflags=0x%x [\"%s\"]: %s\n"),
-				progname, head->fmh_iflags, file->name,
-				strerror(errno));
-			exitcode = 1;
-			free(head);
-			return 0;
-		}
-		if (head->fmh_entries > map_size + 2) {
-			map_size = 11ULL * head->fmh_entries / 10;
-			nmap_size = map_size > (1 << 24) ? (1 << 24) : map_size;
-			nhead = realloc(head, fsmap_sizeof(nmap_size));
-			if (nhead == NULL) {
-				fprintf(stderr,
-					_("%s: cannot realloc %zu bytes\n"),
-					progname, fsmap_sizeof(nmap_size));
-			} else {
-				head = nhead;
-				map_size = nmap_size;
-			}
-		}
-	}
-
 	/*
 	 * If this is an XFS filesystem, remember the data device.
 	 * (We report AG number/block for data device extents on XFS).