[PATCH 0/2] Add some msg for io error

[PATCH 0/2] Add some msg for io error

[zhanchengbin]

If there is an EIO during the process of fsck, the user can be notified of it.

zhanchengbin (2):
  lib/ext2fs: add error handle in unix_flush and unix_write_byte
  e2fsck: add sync error handle to e2fsck.

 e2fsck/ehandler.c    | 24 ++++++++++++++++++++++++
 lib/ext2fs/ext2_io.h |  2 ++
 lib/ext2fs/unix_io.c | 37 ++++++++++++++++++++++++++-----------
 3 files changed, 52 insertions(+), 11 deletions(-)

-- 
2.31.1

[PATCH 1/2] lib/ext2fs: add error handle in unix_flush and unix_write_byte

[zhanchengbin]

As you can see, a new error handling has been added for fsync, and the error
handling for unix_write_byte function has reused the error handling of
write_blk.

Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
---
 lib/ext2fs/ext2_io.h |  2 ++
 lib/ext2fs/unix_io.c | 37 ++++++++++++++++++++++++++-----------
 2 files changed, 28 insertions(+), 11 deletions(-)

diff --git a/lib/ext2fs/ext2_io.h b/lib/ext2fs/ext2_io.h
index 679184e3..becd7078 100644
--- a/lib/ext2fs/ext2_io.h
+++ b/lib/ext2fs/ext2_io.h
@@ -56,6 +56,8 @@ struct struct_io_channel {
 				       size_t size,
 				       int actual_bytes_written,
 				       errcode_t error);
+	errcode_t       (*sync_error)(io_channel channel,
+	                               errcode_t error);
 	int		refcount;
 	int		flags;
 	long		reserved[14];
diff --git a/lib/ext2fs/unix_io.c b/lib/ext2fs/unix_io.c
index 3171c736..283b4eb6 100644
--- a/lib/ext2fs/unix_io.c
+++ b/lib/ext2fs/unix_io.c
@@ -1250,7 +1250,8 @@ static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
 #ifdef ALIGN_DEBUG
 		printf("unix_write_byte: O_DIRECT fallback\n");
 #endif
-		return EXT2_ET_UNIMPLEMENTED;
+		retval = EXT2_ET_UNIMPLEMENTED;
+		goto error_out;
 	}
 
 #ifndef NO_IO_CACHE
@@ -1258,19 +1259,30 @@ static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
 	 * Flush out the cache completely
 	 */
 	if ((retval = flush_cached_blocks(channel, data, FLUSH_INVALIDATE)))
-		return retval;
+		goto error_out;
 #endif
 
-	if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0)
-		return errno;
+	if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0) {
+		retval = errno;
+		goto error_out;
+	}
 
 	actual = write(data->dev, buf, size);
-	if (actual < 0)
-		return errno;
-	if (actual != size)
-		return EXT2_ET_SHORT_WRITE;
-
+	if (actual < 0) {
+		retval = errno;
+		goto error_out;
+	}
+	if (actual != size) {
+		retval = EXT2_ET_SHORT_WRITE;
+		goto error_out;
+	}
 	return 0;
+error_out:
+	if (channel->write_error)
+		retval = (channel->write_error)(channel,
+					offset / channel->block_size, 0, buf,
+					size, actual, retval);
+	return retval;
 }
 
 /*
@@ -1289,8 +1301,11 @@ static errcode_t unix_flush(io_channel channel)
 	retval = flush_cached_blocks(channel, data, 0);
 #endif
 #ifdef HAVE_FSYNC
-	if (!retval && fsync(data->dev) != 0)
-		return errno;
+	if (!retval && fsync(data->dev) != 0) {
+		if (channel->sync_error)
+			retval = (channel->sync_error)(channel, errno);
+		return retval;
+	}
 #endif
 	return retval;
 }
-- 
2.31.1

[PATCH 2/2] e2fsck: add sync error handle to e2fsck.

[zhanchengbin]

If fsync fails during fsck, it is silent and users will not perceive it, so
a function to handle fsync failure should be added to fsck.

Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
---
 e2fsck/ehandler.c | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/e2fsck/ehandler.c b/e2fsck/ehandler.c
index 71ca301c..ae35f3ef 100644
--- a/e2fsck/ehandler.c
+++ b/e2fsck/ehandler.c
@@ -118,6 +118,29 @@ static errcode_t e2fsck_handle_write_error(io_channel channel,
 	return error;
 }
 
+static errcode_t e2fsck_handle_sync_error(io_channel channel,
+                                            errcode_t error)
+{
+	ext2_filsys fs = (ext2_filsys) channel->app_data;
+	e2fsck_t ctx;
+
+	ctx = (e2fsck_t) fs->priv_data;
+	if (ctx->flags & E2F_FLAG_EXITING)
+		return 0;
+	
+	if (operation)
+		printf(_("Error sync (%s) while %s.  "),
+		       error_message(error), operation);
+	else
+		printf(_("Error sync (%s).  "),
+		       error_message(error));
+	preenhalt(ctx);
+	if (ask(ctx, _("Ignore error"), 1))
+		return 0;
+
+	return error;
+}
+
 const char *ehandler_operation(const char *op)
 {
 	const char *ret = operation;
@@ -130,4 +153,5 @@ void ehandler_init(io_channel channel)
 {
 	channel->read_error = e2fsck_handle_read_error;
 	channel->write_error = e2fsck_handle_write_error;
+	channel->sync_error = e2fsck_handle_sync_error;
 }
-- 
2.31.1

Re: [PATCH 2/2] e2fsck: add sync error handle to e2fsck.

[Darrick J. Wong]

On Sat, Mar 25, 2023 at 02:56:52PM +0800, zhanchengbin wrote:
> If fsync fails during fsck, it is silent and users will not perceive it, so
> a function to handle fsync failure should be added to fsck.
> 
> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
> ---
>  e2fsck/ehandler.c | 24 ++++++++++++++++++++++++
>  1 file changed, 24 insertions(+)
> 
> diff --git a/e2fsck/ehandler.c b/e2fsck/ehandler.c
> index 71ca301c..ae35f3ef 100644
> --- a/e2fsck/ehandler.c
> +++ b/e2fsck/ehandler.c
> @@ -118,6 +118,29 @@ static errcode_t e2fsck_handle_write_error(io_channel channel,
>  	return error;
>  }
>  
> +static errcode_t e2fsck_handle_sync_error(io_channel channel,
> +                                            errcode_t error)
> +{
> +	ext2_filsys fs = (ext2_filsys) channel->app_data;
> +	e2fsck_t ctx;
> +
> +	ctx = (e2fsck_t) fs->priv_data;
> +	if (ctx->flags & E2F_FLAG_EXITING)
> +		return 0;
> +	

Nit: ^^^ unnecessary indentation

> +	if (operation)
> +		printf(_("Error sync (%s) while %s.  "),

I think we should be more explicit that *fsync* failed:

"Error during fsync of dirty metadata while %s: %s",
	operation, error_message(...)?


> +		       error_message(error), operation);
> +	else
> +		printf(_("Error sync (%s).  "),
> +		       error_message(error));
> +	preenhalt(ctx);
> +	if (ask(ctx, _("Ignore error"), 1))

ask_yn()?

Not sure what we're asking about here, or what happens if you answer NO?
Unless we're using a redo file, dirty metadata flush has failed so we
might as well keep going, right?

--D

> +		return 0;
> +
> +	return error;
> +}
> +
>  const char *ehandler_operation(const char *op)
>  {
>  	const char *ret = operation;
> @@ -130,4 +153,5 @@ void ehandler_init(io_channel channel)
>  {
>  	channel->read_error = e2fsck_handle_read_error;
>  	channel->write_error = e2fsck_handle_write_error;
> +	channel->sync_error = e2fsck_handle_sync_error;
>  }
> -- 
> 2.31.1
> 

Re: [PATCH 2/2] e2fsck: add sync error handle to e2fsck.

[zhanchengbin]

Thank you for your advice. I will modify patches.
  - bin.

On 2023/3/26 1:13, Darrick J. Wong wrote:
> On Sat, Mar 25, 2023 at 02:56:52PM +0800, zhanchengbin wrote:
>> If fsync fails during fsck, it is silent and users will not perceive it, so
>> a function to handle fsync failure should be added to fsck.
>>
>> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
>> ---
>>   e2fsck/ehandler.c | 24 ++++++++++++++++++++++++
>>   1 file changed, 24 insertions(+)
>>
>> diff --git a/e2fsck/ehandler.c b/e2fsck/ehandler.c
>> index 71ca301c..ae35f3ef 100644
>> --- a/e2fsck/ehandler.c
>> +++ b/e2fsck/ehandler.c
>> @@ -118,6 +118,29 @@ static errcode_t e2fsck_handle_write_error(io_channel channel,
>>   	return error;
>>   }
>>   
>> +static errcode_t e2fsck_handle_sync_error(io_channel channel,
>> +                                            errcode_t error)
>> +{
>> +	ext2_filsys fs = (ext2_filsys) channel->app_data;
>> +	e2fsck_t ctx;
>> +
>> +	ctx = (e2fsck_t) fs->priv_data;
>> +	if (ctx->flags & E2F_FLAG_EXITING)
>> +		return 0;
>> +	
> 
> Nit: ^^^ unnecessary indentation
> 
>> +	if (operation)
>> +		printf(_("Error sync (%s) while %s.  "),
> 
> I think we should be more explicit that *fsync* failed:
> 
> "Error during fsync of dirty metadata while %s: %s",
> 	operation, error_message(...)?
> 
> 
>> +		       error_message(error), operation);
>> +	else
>> +		printf(_("Error sync (%s).  "),
>> +		       error_message(error));
>> +	preenhalt(ctx);
>> +	if (ask(ctx, _("Ignore error"), 1))
> 
> ask_yn()?
> 
> Not sure what we're asking about here, or what happens if you answer NO?
> Unless we're using a redo file, dirty metadata flush has failed so we
> might as well keep going, right?
> 
> --D
> 
>> +		return 0;
>> +
>> +	return error;
>> +}
>> +
>>   const char *ehandler_operation(const char *op)
>>   {
>>   	const char *ret = operation;
>> @@ -130,4 +153,5 @@ void ehandler_init(io_channel channel)
>>   {
>>   	channel->read_error = e2fsck_handle_read_error;
>>   	channel->write_error = e2fsck_handle_write_error;
>> +	channel->sync_error = e2fsck_handle_sync_error;
>>   }
>> -- 
>> 2.31.1
>>
> .
> 

Re: [PATCH 0/2] Add some msg for io error

[Theodore Ts'o]

On Sat, Mar 25, 2023 at 02:56:50PM +0800, zhanchengbin wrote:
> If there is an EIO during the process of fsck, the user can be notified of it.

Can you identify a code path where the user is *not* getting notified
while e2fsck is running without this patch series?

The unix_io.c module calls fsync() through unix_flush() only.  When
unix_write_byte() calls flush_cached blocks(), if the read or write
system call fails, the error will be returned to the caller of
flush_cached_byte(), and the unix_write_byte() will return the error
back to the caller (in this case, e2fsck).

So in both cases, e2fsck checks the error return from ext2fs_flush()
(which is the only place where write_byte gets called) and
io_channel->flush(), and so the user will get some kind of error
report already.

The error message might not identify exactly what I/O failed, but the
"Error sync" message that this commit series provides is not going to
be much better.

						- Ted

Re: [PATCH 0/2] Add some msg for io error

[zhanchengbin]

On 2023/3/26 22:31, Theodore Ts'o wrote:
> On Sat, Mar 25, 2023 at 02:56:50PM +0800, zhanchengbin wrote:
>> If there is an EIO during the process of fsck, the user can be notified of it.
> 
> Can you identify a code path where the user is *not* getting notified
> while e2fsck is running without this patch series?
> 
> The unix_io.c module calls fsync() through unix_flush() only.  When
> unix_write_byte() calls flush_cached blocks(), if the read or write
> system call fails, the error will be returned to the caller of
> flush_cached_byte(), and the unix_write_byte() will return the error
> back to the caller (in this case, e2fsck).
> 

io_channel_flush and io_channel_write_byte do have return values, but
they may not necessarily be checked at their calling points. As in the
following path:

e2fsck_run_ext3_journal
  ext2fs_flush // Ignore errors.
   ext2fs_flush2
    io_channel_flush
  ext2fs_mmp_stop // Ignore errors.
   ext2fs_mmp_write
    io_channel_flush

ext2fs_flush // Many calls ignore errors.
  ext2fs_flush2
   write_primary_superblock
    io_channel_write_byte

Thanks,
  - bin.

> So in both cases, e2fsck checks the error return from ext2fs_flush()
> (which is the only place where write_byte gets called) and
> io_channel->flush(), and so the user will get some kind of error
> report already.
> 
> The error message might not identify exactly what I/O failed, but the
> "Error sync" message that this commit series provides is not going to
> be much better.
> 
> 						- Ted
> 
> .
>