[LTP] [PATCH v2 2/2] aio_tio: determine alignment based on target filesystem

Steve Muckle smuckle@google.com
Tue Jan 29 19:19:34 CET 2019


Reviewed-by: Steve Muckle <smuckle@google.com>

On 01/11/2019 12:53 AM, Matthias Maennich wrote:
> The alignment for O_DIRECT operations has to match the blocksize of the
> underlying filesystem. Determine the alignment from the target file's
> file system.
> 
> aio_tio test cases 1 and 2 failed (nondeterministic) on aarch64 when run
> on a 4096 byte blocksize filesystem and with an alignment of 512 bytes.
> 
> Determining the blocksize from the file system and using it for the
> alignment resolves the test issue.
> 
> Signed-off-by: Matthias Maennich <maennich@google.com>
> Reviewed-by: Alessio Balsini <balsini@google.com>
> ---
>   testcases/kernel/io/aio/aio02/aio_tio.c | 8 +++++++-
>   1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/testcases/kernel/io/aio/aio02/aio_tio.c b/testcases/kernel/io/aio/aio02/aio_tio.c
> index 623c81a5e..db6fa2688 100644
> --- a/testcases/kernel/io/aio/aio02/aio_tio.c
> +++ b/testcases/kernel/io/aio/aio02/aio_tio.c
> @@ -34,6 +34,7 @@
>   #include "config.h"
>   #include "common.h"
>   #include "test.h"
> +#include "safe_macros.h"
>   #include <string.h>
>   #include <errno.h>
>   
> @@ -42,7 +43,6 @@
>   #define AIO_MAXIO 32
>   #define AIO_BLKSIZE (64*1024)
>   
> -static int alignment = 512;
>   static int wait_count = 0;
>   
>   /*
> @@ -94,6 +94,8 @@ int io_tio(char *pathname, int flag, int n, int operation)
>   	void *bufptr = NULL;
>   	off_t offset = 0;
>   	struct timespec timeout;
> +	struct stat fi_stat;
> +	size_t alignment;
>   
>   	io_context_t myctx;
>   	struct iocb iocb_array[AIO_MAXIO];
> @@ -105,6 +107,10 @@ int io_tio(char *pathname, int flag, int n, int operation)
>   		return -1;
>   	}
>   
> +	/* determine the alignment from the blksize of the underlying fs */
> +	SAFE_FSTAT(NULL, fd, &fi_stat);
> +	alignment = fi_stat.st_blksize;
> +
>   	res = io_queue_init(n, &myctx);
>   	//printf (" res = %d \n", res);
>   
> 



More information about the ltp mailing list