[LTP] [PATCH v2] syscalls/stat04&lstat03: remove fs block size related code

[LTP] [PATCH v2] syscalls/stat04&lstat03: remove fs block size related code

[Zorro Lang via ltp]

The st_blksize isn't equivalent to the filesystem block size. The
stat(3) manual describe st_blksize as:

  "This field gives the "preferred" block size for efficient filesystem I/O."

So the st_blksize is the "preferred" block size for "efficient" fs
I/O, extN might think the "preferred" block size is fs block size.
But not all filesystems think same with extN. For example, xfs thinks
the "preferred" block size is:

  max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize)

So you might get st_blksize=4096, no matter on 1k blocksize xfs or 4k
blocksize xfs. We shouldn't expect to use a different blocksize mkfs
option to get a different st_blksize. This part of code is useless,
except causing unexpected test failures on other filesystems (e.g.
xfs, btrfs and so on).

Signed-off-by: Zorro Lang <zlang@kernel.org>
---

Hi,

I tried to fix the mkfs problem last year:
  https://lists.linux.it/pipermail/ltp/2024-December/041038.html

Now I got a chance to look back this test failure, I think it's not a mkfs
option problem, but the test case misunderstood the st_blksize.

Except we limit this test only run on extN, or we don't need to make
fs with a different block size, especially shouldn't expect to get
a different st_blksize from that.

Thanks,
Zorro

 testcases/kernel/syscalls/lstat/lstat03.c | 12 +-----------
 testcases/kernel/syscalls/stat/stat04.c   | 11 +----------
 2 files changed, 2 insertions(+), 21 deletions(-)

diff --git a/testcases/kernel/syscalls/lstat/lstat03.c b/testcases/kernel/syscalls/lstat/lstat03.c
index 9438e2920..7d9d01ad1 100644
--- a/testcases/kernel/syscalls/lstat/lstat03.c
+++ b/testcases/kernel/syscalls/lstat/lstat03.c
@@ -34,7 +34,6 @@ static void run(void)
 	TST_EXP_EXPR(file_stat->st_gid != symb_stat->st_gid);
 	TST_EXP_EXPR(file_stat->st_size != symb_stat->st_size);
 	TST_EXP_EXPR(file_stat->st_blocks != symb_stat->st_blocks);
-	TST_EXP_EXPR(file_stat->st_blksize != symb_stat->st_blksize);
 	TST_EXP_EXPR(file_stat->st_atime != symb_stat->st_atime);
 	TST_EXP_EXPR(file_stat->st_mtime != symb_stat->st_mtime);
 	TST_EXP_EXPR(file_stat->st_ctime != symb_stat->st_ctime);
@@ -42,18 +41,9 @@ static void run(void)
 
 static void setup(void)
 {
-	char opt_bsize[32];
-	const char *const fs_opts[] = {opt_bsize, NULL};
-	struct stat sb;
-	int pagesize;
 	int fd;
 
-	/* change st_blksize / st_dev */
-	SAFE_STAT(".", &sb);
-	pagesize = sb.st_blksize == 4096 ? 1024 : 4096;
-
-	snprintf(opt_bsize, sizeof(opt_bsize), "-b %i", pagesize);
-	SAFE_MKFS(tst_device->dev, tst_device->fs_type, fs_opts, NULL);
+	SAFE_MKFS(tst_device->dev, tst_device->fs_type, NULL, NULL);
 	SAFE_MOUNT(tst_device->dev, MNTPOINT, tst_device->fs_type, 0, 0);
 
 	SAFE_TOUCH(FILENAME, 0777, NULL);
diff --git a/testcases/kernel/syscalls/stat/stat04.c b/testcases/kernel/syscalls/stat/stat04.c
index 09a9a6823..3c233df51 100644
--- a/testcases/kernel/syscalls/stat/stat04.c
+++ b/testcases/kernel/syscalls/stat/stat04.c
@@ -41,21 +41,12 @@ static void run(void)
 
 static void setup(void)
 {
-	char opt_bsize[32];
-	const char *const fs_opts[] = {opt_bsize, NULL};
-	struct stat sb;
-	int pagesize;
 	int fd;
 
 	file_path = tst_tmpdir_genpath(FILENAME);
 	symb_path = tst_tmpdir_genpath(SYMBNAME);
 
-	/* change st_blksize / st_dev */
-	SAFE_STAT(".", &sb);
-	pagesize = sb.st_blksize == 4096 ? 1024 : 4096;
-
-	snprintf(opt_bsize, sizeof(opt_bsize), "-b %i", pagesize);
-	SAFE_MKFS(tst_device->dev, tst_device->fs_type, fs_opts, NULL);
+	SAFE_MKFS(tst_device->dev, tst_device->fs_type, NULL, NULL);
 	SAFE_MOUNT(tst_device->dev, MNTPOINT, tst_device->fs_type, 0, 0);
 
 	SAFE_TOUCH(FILENAME, 0777, NULL);
-- 
2.51.0


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] [PATCH v2] syscalls/stat04&lstat03: remove fs block size related code

[Andrea Cervesato via ltp]

Hi!

Sorry for the delay to this review.

On Thu Nov 13, 2025 at 5:22 PM CET, Zorro Lang via ltp wrote:
> The st_blksize isn't equivalent to the filesystem block size. The
> stat(3) manual describe st_blksize as:
>
>   "This field gives the "preferred" block size for efficient filesystem I/O."
>
> So the st_blksize is the "preferred" block size for "efficient" fs
> I/O, extN might think the "preferred" block size is fs block size.
> But not all filesystems think same with extN. For example, xfs thinks
> the "preferred" block size is:
>
>   max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize)
>
> So you might get st_blksize=4096, no matter on 1k blocksize xfs or 4k
> blocksize xfs. We shouldn't expect to use a different blocksize mkfs
> option to get a different st_blksize. This part of code is useless,
> except causing unexpected test failures on other filesystems (e.g.
> xfs, btrfs and so on).
>
> Signed-off-by: Zorro Lang <zlang@kernel.org>
> ---
>
> Hi,
>
> I tried to fix the mkfs problem last year:
>   https://lists.linux.it/pipermail/ltp/2024-December/041038.html
>
> Now I got a chance to look back this test failure, I think it's not a mkfs
> option problem, but the test case misunderstood the st_blksize.
>
> Except we limit this test only run on extN, or we don't need to make
> fs with a different block size, especially shouldn't expect to get
> a different st_blksize from that.

If testing `st_blksize` is an issue only under certain filesystems, we
should probably test it only when the right ones are in use.
For instance, this can be done as following :

	if (!strcmp(tst_device->fs_type, "ext2")) {
		/* test `st_blocks` attribute */
	}

In this way we avoid to disable `st_blksize` testing for all existing
filesystems.

-- 
Andrea Cervesato
SUSE QE Automation Engineer Linux
andrea.cervesato@suse.com


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp