[PATCH] generic/465: fix a read bug when encounter EOF

[PATCH] generic/465: fix a read bug when encounter EOF

[Su Yanjun]

In nfs test, if the writer has not written enough data for reader reading,
then reader only get partial data and test fails.

We can continue reading until read enough data.

Signed-off-by: Su Yanjun <suyj.fnst@cn.fujitsu.com>
---
 src/aio-dio-regress/aio-dio-append-write-read-race.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/src/aio-dio-regress/aio-dio-append-write-read-race.c b/src/aio-dio-regress/aio-dio-append-write-read-race.c
index 911f272..f28ef3c 100644
--- a/src/aio-dio-regress/aio-dio-append-write-read-race.c
+++ b/src/aio-dio-regress/aio-dio-append-write-read-race.c
@@ -44,14 +44,27 @@ static void usage(const char *prog)
 static void *reader(void *arg)
 {
 	struct io_data *data = (struct io_data *)arg;
+	size_t blksize = data->blksize;
+	size_t offset = data->offset;
+	char   *buf = data->buf;
 
 	memset(data->buf, 'b', data->blksize);
 	reader_ready = 1;
 	do {
-		data->read_sz = pread(data->fd, data->buf, data->blksize,
-				      data->offset);
+		data->read_sz = pread(data->fd, buf, blksize,
+				      offset);
 		if (data->read_sz < 0)
 			perror("read file");
+
+		if (data->read_sz > 0) {
+			blksize -= data->read_sz;
+			offset  += data->read_sz;
+			buf	+= data->read_sz;
+			data->read_sz = 0;
+		}
+
+		if (blksize <= 0)
+			break;
 	} while (data->read_sz <= 0);
 
 	return NULL;
-- 
2.7.4

Re: [PATCH] generic/465: fix a read bug when encounter EOF

[Eryu Guan]

On Mon, Sep 16, 2019 at 01:24:26PM +0800, Su Yanjun wrote:
> In nfs test, if the writer has not written enough data for reader reading,
> then reader only get partial data and test fails.
> 
> We can continue reading until read enough data.
> 
> Signed-off-by: Su Yanjun <suyj.fnst@cn.fujitsu.com>

Would you please provide more info on the actual failure you've seen? I
think the partial read issue should have been addressed by commit
2428c08a6025 ("generic/465: just check the actual read data under dio
read/write")

Thanks,
Eryu

> ---
>  src/aio-dio-regress/aio-dio-append-write-read-race.c | 17 +++++++++++++++--
>  1 file changed, 15 insertions(+), 2 deletions(-)
> 
> diff --git a/src/aio-dio-regress/aio-dio-append-write-read-race.c b/src/aio-dio-regress/aio-dio-append-write-read-race.c
> index 911f272..f28ef3c 100644
> --- a/src/aio-dio-regress/aio-dio-append-write-read-race.c
> +++ b/src/aio-dio-regress/aio-dio-append-write-read-race.c
> @@ -44,14 +44,27 @@ static void usage(const char *prog)
>  static void *reader(void *arg)
>  {
>  	struct io_data *data = (struct io_data *)arg;
> +	size_t blksize = data->blksize;
> +	size_t offset = data->offset;
> +	char   *buf = data->buf;
>  
>  	memset(data->buf, 'b', data->blksize);
>  	reader_ready = 1;
>  	do {
> -		data->read_sz = pread(data->fd, data->buf, data->blksize,
> -				      data->offset);
> +		data->read_sz = pread(data->fd, buf, blksize,
> +				      offset);
>  		if (data->read_sz < 0)
>  			perror("read file");
> +
> +		if (data->read_sz > 0) {
> +			blksize -= data->read_sz;
> +			offset  += data->read_sz;
> +			buf	+= data->read_sz;
> +			data->read_sz = 0;
> +		}
> +
> +		if (blksize <= 0)
> +			break;
>  	} while (data->read_sz <= 0);
>  
>  	return NULL;
> -- 
> 2.7.4
> 
> 
> 

Re: [PATCH] generic/465: fix a read bug when encounter EOF

[Su Yanjun]

在 2019/9/25 17:47, Eryu Guan 写道:
> On Mon, Sep 16, 2019 at 01:24:26PM +0800, Su Yanjun wrote:
>> In nfs test, if the writer has not written enough data for reader reading,
>> then reader only get partial data and test fails.
>>
>> We can continue reading until read enough data.
>>
>> Signed-off-by: Su Yanjun <suyj.fnst@cn.fujitsu.com>
> Would you please provide more info on the actual failure you've seen? I
> think the partial read issue should have been addressed by commit
> 2428c08a6025 ("generic/465: just check the actual read data under dio
> read/write")

I rechecked my patch and my patch is not needed at all.

It's  only workaround of a kernel bug of nfs. In some cases nfs returns 
wrong data to userspace, but the return value is ok.

I will send a kernel patch to nfs mailist.

Thanks a lot.

>
> Thanks,
> Eryu
>
>> ---
>>   src/aio-dio-regress/aio-dio-append-write-read-race.c | 17 +++++++++++++++--
>>   1 file changed, 15 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/aio-dio-regress/aio-dio-append-write-read-race.c b/src/aio-dio-regress/aio-dio-append-write-read-race.c
>> index 911f272..f28ef3c 100644
>> --- a/src/aio-dio-regress/aio-dio-append-write-read-race.c
>> +++ b/src/aio-dio-regress/aio-dio-append-write-read-race.c
>> @@ -44,14 +44,27 @@ static void usage(const char *prog)
>>   static void *reader(void *arg)
>>   {
>>   	struct io_data *data = (struct io_data *)arg;
>> +	size_t blksize = data->blksize;
>> +	size_t offset = data->offset;
>> +	char   *buf = data->buf;
>>   
>>   	memset(data->buf, 'b', data->blksize);
>>   	reader_ready = 1;
>>   	do {
>> -		data->read_sz = pread(data->fd, data->buf, data->blksize,
>> -				      data->offset);
>> +		data->read_sz = pread(data->fd, buf, blksize,
>> +				      offset);
>>   		if (data->read_sz < 0)
>>   			perror("read file");
>> +
>> +		if (data->read_sz > 0) {
>> +			blksize -= data->read_sz;
>> +			offset  += data->read_sz;
>> +			buf	+= data->read_sz;
>> +			data->read_sz = 0;
>> +		}
>> +
>> +		if (blksize <= 0)
>> +			break;
>>   	} while (data->read_sz <= 0);
>>   
>>   	return NULL;
>> -- 
>> 2.7.4
>>
>>
>>
>

Re: [PATCH] generic/465: fix a read bug when encounter EOF

[Su Yanjun]

在 2019/9/25 17:47, Eryu Guan 写道:
> On Mon, Sep 16, 2019 at 01:24:26PM +0800, Su Yanjun wrote:
>> In nfs test, if the writer has not written enough data for reader reading,
>> then reader only get partial data and test fails.
>>
>> We can continue reading until read enough data.
>>
>> Signed-off-by: Su Yanjun <suyj.fnst@cn.fujitsu.com>
> Would you please provide more info on the actual failure you've seen? I
> think the partial read issue should have been addressed by commit
> 2428c08a6025 ("generic/465: just check the actual read data under dio
> read/write")

I rechecked my patch and my patch is not needed at all.

It's  only workaround of a kernel bug of nfs. In some cases nfs returns 
wrong data to userspace, but the return value is ok.

I will send a kernel patch to nfs mailist.

Thanks a lot.

>
> Thanks,
> Eryu
>
>> ---
>>   src/aio-dio-regress/aio-dio-append-write-read-race.c | 17 +++++++++++++++--
>>   1 file changed, 15 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/aio-dio-regress/aio-dio-append-write-read-race.c b/src/aio-dio-regress/aio-dio-append-write-read-race.c
>> index 911f272..f28ef3c 100644
>> --- a/src/aio-dio-regress/aio-dio-append-write-read-race.c
>> +++ b/src/aio-dio-regress/aio-dio-append-write-read-race.c
>> @@ -44,14 +44,27 @@ static void usage(const char *prog)
>>   static void *reader(void *arg)
>>   {
>>   	struct io_data *data = (struct io_data *)arg;
>> +	size_t blksize = data->blksize;
>> +	size_t offset = data->offset;
>> +	char   *buf = data->buf;
>>   
>>   	memset(data->buf, 'b', data->blksize);
>>   	reader_ready = 1;
>>   	do {
>> -		data->read_sz = pread(data->fd, data->buf, data->blksize,
>> -				      data->offset);
>> +		data->read_sz = pread(data->fd, buf, blksize,
>> +				      offset);
>>   		if (data->read_sz < 0)
>>   			perror("read file");
>> +
>> +		if (data->read_sz > 0) {
>> +			blksize -= data->read_sz;
>> +			offset  += data->read_sz;
>> +			buf	+= data->read_sz;
>> +			data->read_sz = 0;
>> +		}
>> +
>> +		if (blksize <= 0)
>> +			break;
>>   	} while (data->read_sz <= 0);
>>   
>>   	return NULL;
>> -- 
>> 2.7.4
>>
>>
>>
>