Skip to contentSkip to conversation

fix xfs_copy betwen devices with mismatching sizes

PATCH In review Christoph Hellwig 2 participants 5 messages Jul 29

lore
Share & tools
lore
#0 lore raw

This fixes copy between devices with different alignment requirements. xfs/601 trigges this when TEST_DIR is on a device that requires minimal alignment, but SCRATCH_DEV is on a device that requires more alignment.

#1 lore raw

PATCH 1/2 copy: log the error number when failing to write

Log the string form of the errno value to help debugging a failed write.

---
 copy/xfs_copy.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
copy/xfs_copy.c +2 -2
index 3ba9a07e4469..ad18ceb1ac1b 100644
@@ -891,8 +891,8 @@ main(int argc, char **argv)
 			off -= XFS_MAX_SECTORSIZE;
 			len = pwrite(target[i].fd, lb, XFS_MAX_SECTORSIZE, off);
 			if (len < 0) {
-				do_log(_("%s:  failed to write last block\n"),
-					progname);
+				do_log(_("%s:  failed to write last block: %s\n"),
+					progname, strerror(errno));
 				do_log(_("\tIs target \"%s\" too small?\n"),
 					target[i].name);
 				die_perror();
Signature · 1 line
2.53.0
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] copy: log the error number when failing to write

Quoted 3 lines · Christoph Hellwig

Log the string form of the errno value to help debugging a failed write.

Good idea!

--D

Quoted 23 lines
---
 copy/xfs_copy.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
copy/xfs_copy.c +2 -2
index 3ba9a07e4469..ad18ceb1ac1b 100644
@@ -891,8 +891,8 @@ main(int argc, char **argv)
 			off -= XFS_MAX_SECTORSIZE;
 			len = pwrite(target[i].fd, lb, XFS_MAX_SECTORSIZE, off);
 			if (len < 0) {
-				do_log(_("%s:  failed to write last block\n"),
-					progname);
+				do_log(_("%s:  failed to write last block: %s\n"),
+					progname, strerror(errno));
 				do_log(_("\tIs target \"%s\" too small?\n"),
 					target[i].name);
 				die_perror();
Signature · 1 line
2.53.0
#3 lore raw

PATCH 2/2 copy: always align to at least page size

In reply to Christoph Hellwig

The buffers used by xfs_copy need to fit the alignment of both the source and target. As we can't rely on statx for more detailed information yet, work around this by never going below the default page size alignment.

This fixes copy an image from an xfs file systems on a NVMe device that only reqires dword alignment to a virtio_blk device that requires 512-byte alignment.

---
 copy/xfs_copy.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
copy/xfs_copy.c +2 -2
index ad18ceb1ac1b..5435c87dba38 100644
@@ -673,6 +673,7 @@ main(int argc, char **argv)
 	if (S_ISREG(statbuf.st_mode))
 		source_is_file = 1;
 
+	wbuf_align = getpagesize();
 	if (source_is_file && platform_test_xfs_fd(source_fd))  {
 		if (fcntl(source_fd, F_SETFL, open_flags | O_DIRECT) < 0)  {
 			do_log(_("%s: Cannot set direct I/O flag on \"%s\".\n"),
@@ -685,13 +686,12 @@ main(int argc, char **argv)
 			die_perror();
 		}
 
-		wbuf_align = d.d_mem;
+		wbuf_align = max(wbuf_align, d.d_mem);
 		wbuf_size = min(d.d_maxiosz, 1 * 1024 * 1024);
 		wbuf_miniosize = d.d_miniosz;
 	} else  {
 		/* set arbitrary I/O params, miniosize at least 1 disk block */
 
-		wbuf_align = getpagesize();
 		wbuf_size = 1 * 1024 * 1024;
 		wbuf_miniosize = -1;	/* set after mounting source fs */
 	}
Signature · 1 line
2.53.0
Reviewed-by on v1 2/2
#4 lore raw

Re: [PATCH 2/2] copy: always align to at least page size

Quoted 9 lines · Christoph Hellwig

The buffers used by xfs_copy need to fit the alignment of both the source and target. As we can't rely on statx for more detailed information yet, work around this by never going below the default page size alignment.

This fixes copy an image from an xfs file systems on a NVMe device that only reqires dword alignment to a virtio_blk device that requires 512-byte alignment.

Mmmmmm, fun...

--D

Quoted 35 lines
---
 copy/xfs_copy.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
copy/xfs_copy.c +2 -2
index ad18ceb1ac1b..5435c87dba38 100644
@@ -673,6 +673,7 @@ main(int argc, char **argv)
 	if (S_ISREG(statbuf.st_mode))
 		source_is_file = 1;
 
+	wbuf_align = getpagesize();
 	if (source_is_file && platform_test_xfs_fd(source_fd))  {
 		if (fcntl(source_fd, F_SETFL, open_flags | O_DIRECT) < 0)  {
 			do_log(_("%s: Cannot set direct I/O flag on \"%s\".\n"),
@@ -685,13 +686,12 @@ main(int argc, char **argv)
 			die_perror();
 		}
 
-		wbuf_align = d.d_mem;
+		wbuf_align = max(wbuf_align, d.d_mem);
 		wbuf_size = min(d.d_maxiosz, 1 * 1024 * 1024);
 		wbuf_miniosize = d.d_miniosz;
 	} else  {
 		/* set arbitrary I/O params, miniosize at least 1 disk block */
 
-		wbuf_align = getpagesize();
 		wbuf_size = 1 * 1024 * 1024;
 		wbuf_miniosize = -1;	/* set after mounting source fs */
 	}
Signature · 1 line
2.53.0