[PATCH, RFC] libxfs: use FALLOC_FL_ZERO_RANGE in libxfs_device_zero

[Eric Sandeen <sandeen@sandeen.net>]

I had a request from someone who cared about mkfs speed(!)
over a slower network block device to look into using faster
zeroing methods, particularly for the log, during mkfs.xfs.

e2fsprogs already does this, thanks to some guy named Darrick:

/*
 * If we know about ZERO_RANGE, try that before we try PUNCH_HOLE because
 * ZERO_RANGE doesn't unmap preallocated blocks.  We prefer fallocate because
 * it always invalidates page cache, and libext2fs requires that reads after
 * ZERO_RANGE return zeroes.
 */
static int __unix_zeroout(int fd, off_t offset, off_t len)
{
        int ret = -1;

#if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_ZERO_RANGE)
        ret = fallocate(fd, FALLOC_FL_ZERO_RANGE, offset, len);
        if (ret == 0)
                return 0;
#endif
#if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
        ret = fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
                        offset,  len);
        if (ret == 0)
                return 0;
#endif
        errno = EOPNOTSUPP;
        return ret;
}

and nobody has exploded so far, AFAIK.  :)  So, floating this idea
for xfsprogs.  I'm a little scared of the second #ifdef block above, but
if that's really ok/consistent/safe we could add it too.

The patch moves some defines around too, I could split that up and resend
if this isn't laughed out of the room.

Thanks,
-Eric

=====

libxfs: use FALLOC_FL_ZERO_RANGE in libxfs_device_zero

I had a request from someone who cared about mkfs speed(!)
over a slower network block device to look into using faster
zeroing methods, particularly for the log, during mkfs.

Using FALLOC_FL_ZERO_RANGE is faster in this case than writing
a bunch of zeros across a wire.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
---

diff --git a/include/linux.h b/include/linux.h
index 8f3c32b0..425badb5 100644
--- a/include/linux.h
+++ b/include/linux.h
@@ -113,6 +113,26 @@ static __inline__ void platform_uuid_copy(uuid_t *dst, uuid_t *src)
 	uuid_copy(*dst, *src);
 }
 
+#ifndef FALLOC_FL_PUNCH_HOLE
+#define FALLOC_FL_PUNCH_HOLE	0x02
+#endif
+
+#ifndef FALLOC_FL_COLLAPSE_RANGE
+#define FALLOC_FL_COLLAPSE_RANGE 0x08
+#endif
+
+#ifndef FALLOC_FL_ZERO_RANGE
+#define FALLOC_FL_ZERO_RANGE 0x10
+#endif
+
+#ifndef FALLOC_FL_INSERT_RANGE
+#define FALLOC_FL_INSERT_RANGE 0x20
+#endif
+
+#ifndef FALLOC_FL_UNSHARE_RANGE
+#define FALLOC_FL_UNSHARE_RANGE 0x40
+#endif
+
 #ifndef BLKDISCARD
 #define BLKDISCARD	_IO(0x12,119)
 #endif
diff --git a/io/prealloc.c b/io/prealloc.c
index 6d452354..0b4efc45 100644
--- a/io/prealloc.c
+++ b/io/prealloc.c
@@ -12,26 +12,6 @@
 #include "init.h"
 #include "io.h"
 
-#ifndef FALLOC_FL_PUNCH_HOLE
-#define FALLOC_FL_PUNCH_HOLE	0x02
-#endif
-
-#ifndef FALLOC_FL_COLLAPSE_RANGE
-#define FALLOC_FL_COLLAPSE_RANGE 0x08
-#endif
-
-#ifndef FALLOC_FL_ZERO_RANGE
-#define FALLOC_FL_ZERO_RANGE 0x10
-#endif
-
-#ifndef FALLOC_FL_INSERT_RANGE
-#define FALLOC_FL_INSERT_RANGE 0x20
-#endif
-
-#ifndef FALLOC_FL_UNSHARE_RANGE
-#define FALLOC_FL_UNSHARE_RANGE 0x40
-#endif
-
 static cmdinfo_t allocsp_cmd;
 static cmdinfo_t freesp_cmd;
 static cmdinfo_t resvsp_cmd;
diff --git a/libxfs/Makefile b/libxfs/Makefile
index fbcc963a..b4e8864b 100644
--- a/libxfs/Makefile
+++ b/libxfs/Makefile
@@ -105,6 +105,10 @@ CFILES = cache.c \
 #
 #LCFLAGS +=
 
+ifeq ($(HAVE_FALLOCATE),yes)
+LCFLAGS += -DHAVE_FALLOCATE
+endif
+
 FCFLAGS = -I.
 
 LTLIBS = $(LIBPTHREAD) $(LIBRT)
diff --git a/libxfs/rdwr.c b/libxfs/rdwr.c
index 0d9d7202..94f63bbf 100644
--- a/libxfs/rdwr.c
+++ b/libxfs/rdwr.c
@@ -4,6 +4,9 @@
  * All Rights Reserved.
  */
 
+#if defined(HAVE_FALLOCATE)
+#include <linux/falloc.h>
+#endif
 
 #include "libxfs_priv.h"
 #include "init.h"
@@ -60,9 +63,21 @@ int
 libxfs_device_zero(struct xfs_buftarg *btp, xfs_daddr_t start, uint len)
 {
 	xfs_off_t	start_offset, end_offset, offset;
-	ssize_t		zsize, bytes;
+	ssize_t		zsize, bytes, len_bytes;
 	char		*z;
-	int		fd;
+	int		ret, fd;
+
+	fd = libxfs_device_to_fd(btp->dev);
+	start_offset = LIBXFS_BBTOOFF64(start);
+	end_offset = LIBXFS_BBTOOFF64(start + len) - start_offset;
+
+#if defined(HAVE_FALLOCATE)
+	/* try to use special zeroing methods, fall back to writes if needed */
+	len_bytes = LIBXFS_BBTOOFF64(len);
+	ret = fallocate(fd, FALLOC_FL_ZERO_RANGE, start_offset, len_bytes);
+	if (ret == 0)
+		return 0;
+#endif
 
 	zsize = min(BDSTRAT_SIZE, BBTOB(len));
 	if ((z = memalign(libxfs_device_alignment(), zsize)) == NULL) {
@@ -73,9 +88,6 @@ libxfs_device_zero(struct xfs_buftarg *btp, xfs_daddr_t start, uint len)
 	}
 	memset(z, 0, zsize);
 
-	fd = libxfs_device_to_fd(btp->dev);
-	start_offset = LIBXFS_BBTOOFF64(start);
-
 	if ((lseek(fd, start_offset, SEEK_SET)) < 0) {
 		fprintf(stderr, _("%s: %s seek to offset %llu failed: %s\n"),
 			progname, __FUNCTION__,
@@ -83,7 +95,6 @@ libxfs_device_zero(struct xfs_buftarg *btp, xfs_daddr_t start, uint len)
 		exit(1);
 	}
 
-	end_offset = LIBXFS_BBTOOFF64(start + len) - start_offset;
 	for (offset = 0; offset < end_offset; ) {
 		bytes = min((ssize_t)(end_offset - offset), zsize);
 		if ((bytes = write(fd, z, bytes)) < 0) {