[PATCH] mke2fs: Inform user of ongoing discard

[PATCH] mke2fs: Inform user of ongoing discard

[Lukas Czerner]

For some time now we are doing initial discard of the device prior to
filesystem creation. However, there is no feedback for the user and
hence on some devices with slow TRIM implementation it may appear that
mke2fs is stuck.

This commit introduce new function mke2fs_discard_device(), which is a
wrapper for io_channel_discard(). The discard is done per-partes and
discard progress is being reported back to the user. The discard step
has been set to 2GB size, which works reasonably well on both slow and
fast devices.

I gave up on doing fancy things like align discard according to
discard_alignment, checking for discard granularity and computing
estimate time. First of all, because it would require either new ioctl
to retrieve those information or use of libudev library, none of it
seems to be worth it. Regarding discard_granularity, I doubt there is
any sane device with discard granularity that big it would affect this.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
---
 misc/mke2fs.c |   55 ++++++++++++++++++++++++++++++++++++++++++-------------
 1 files changed, 42 insertions(+), 13 deletions(-)

diff --git a/misc/mke2fs.c b/misc/mke2fs.c
index bc1211d..16b4826 100644
--- a/misc/mke2fs.c
+++ b/misc/mke2fs.c
@@ -72,6 +72,9 @@ extern int optind;
 #define ZAP_BOOTBLOCK
 #endif
 
+#define DISCARD_STEP_MB		(2048)
+#define MB			(1024*1024)
+
 extern int isatty(int);
 extern FILE *fpopen(const char *cmd, const char *mode);
 
@@ -1922,6 +1925,44 @@ static int mke2fs_setup_tdb(const char *name, io_manager *io_ptr)
 	return retval;
 }
 
+static int mke2fs_discard_device(ext2_filsys fs)
+{
+	struct ext2fs_numeric_progress_struct progress;
+	blk64_t blocks = ext2fs_blocks_count(fs->super);
+	blk64_t count = DISCARD_STEP_MB;
+	blk64_t cur = 0;
+	int retval = 0;
+
+	count *= MB;
+	count /= fs->blocksize;
+
+	ext2fs_numeric_progress_init(fs, &progress,
+				     _("Discarding device blocks: "),
+				     blocks);
+	while (cur < blocks) {
+		ext2fs_numeric_progress_update(fs, &progress, cur);
+
+		retval = io_channel_discard(fs->io, cur, count, fs->blocksize);
+		if (retval)
+			break;
+
+		cur += count;
+		if (cur + count > blocks)
+			count = blocks - cur;
+	}
+
+	if (retval) {
+		ext2fs_numeric_progress_close(fs, &progress,
+				      _("Error - "));
+		if (!quiet)
+			printf("%s\n",error_message(retval));
+	} else
+		ext2fs_numeric_progress_close(fs, &progress,
+				      _("done                            \n"));
+
+	return retval;
+}
+
 int main (int argc, char *argv[])
 {
 	errcode_t	retval = 0;
@@ -1982,19 +2023,7 @@ int main (int argc, char *argv[])
 
 	/* Can't undo discard ... */
 	if (discard && (io_ptr != undo_io_manager)) {
-		blk64_t blocks = ext2fs_blocks_count(fs->super);
-		if (verbose)
-			printf(_("Calling BLKDISCARD from 0 to %llu... "),
-			       (unsigned long long) blocks);
-		retval = io_channel_discard(fs->io, 0, blocks, fs->blocksize);
-		if (verbose) {
-			if (retval)
-				printf(_("failed (%s)\n"),
-				       error_message(retval));
-			else
-				printf(_("succeeded\n"));
-		}

Re: [PATCH] mke2fs: Inform user of ongoing discard

[Tao Ma]

Hi Lukas,
On 12/13/2010 05:45 PM, Lukas Czerner wrote:
<snip>
> @@ -1922,6 +1925,44 @@ static int mke2fs_setup_tdb(const char *name, io_manager *io_ptr)
>   	return retval;
>   }
>
> +static int mke2fs_discard_device(ext2_filsys fs)
> +{
> +	struct ext2fs_numeric_progress_struct progress;
> +	blk64_t blocks = ext2fs_blocks_count(fs->super);
> +	blk64_t count = DISCARD_STEP_MB;
> +	blk64_t cur = 0;
> +	int retval = 0;
> +
> +	count *= MB;
> +	count /= fs->blocksize;
> +
> +	ext2fs_numeric_progress_init(fs,&progress,
> +				     _("Discarding device blocks: "),
> +				     blocks);
> +	while (cur<  blocks) {
> +		ext2fs_numeric_progress_update(fs,&progress, cur);
> +
> +		retval = io_channel_discard(fs->io, cur, count, fs->blocksize);
Here we need to limit 'count' within ext4's block range for the 1st 
round, otherwise it will discard the blocks outside our ext4 volume in 
case the size is less than 2GB.

Regards,
Tao

Re: [PATCH] mke2fs: Inform user of ongoing discard

[Lukas Czerner]

On Mon, 13 Dec 2010, Tao Ma wrote:

> Hi Lukas,
> On 12/13/2010 05:45 PM, Lukas Czerner wrote:
> <snip>
> > @@ -1922,6 +1925,44 @@ static int mke2fs_setup_tdb(const char *name,
> > io_manager *io_ptr)
> >   	return retval;
> >   }
> > 
> > +static int mke2fs_discard_device(ext2_filsys fs)
> > +{
> > +	struct ext2fs_numeric_progress_struct progress;
> > +	blk64_t blocks = ext2fs_blocks_count(fs->super);
> > +	blk64_t count = DISCARD_STEP_MB;
> > +	blk64_t cur = 0;
> > +	int retval = 0;
> > +
> > +	count *= MB;
> > +	count /= fs->blocksize;
> > +
> > +	ext2fs_numeric_progress_init(fs,&progress,
> > +				     _("Discarding device blocks: "),
> > +				     blocks);
> > +	while (cur<  blocks) {
> > +		ext2fs_numeric_progress_update(fs,&progress, cur);
> > +
> > +		retval = io_channel_discard(fs->io, cur, count,
> > fs->blocksize);
> Here we need to limit 'count' within ext4's block range for the 1st round,
> otherwise it will discard the blocks outside our ext4 volume in case the size
> is less than 2GB.
> 
> Regards,
> Tao
> 

Oh, of course. Thanks, I'll fix that.

-Lukas

[PATCH] mke2fs: Inform user of ongoing discard

[Lukas Czerner]

For some time now we are doing initial discard of the device prior to
filesystem creation. However, there is no feedback for the user and
hence on some devices with slow TRIM implementation it may appear that
mke2fs is stuck.

This commit introduce new function mke2fs_discard_device(), which is a
wrapper for io_channel_discard(). The discard is done per-partes and
discard progress is being reported back to the user. The discard step
has been set to 2GB size, which works reasonably well on both slow and
fast devices.

I gave up on doing fancy things like align discard according to
discard_alignment, checking for discard granularity and computing
estimate time. First of all, because it would require either new ioctl
to retrieve those information or use of libudev library, none of it
seems to be worth it. Regarding discard_granularity, I doubt there is
any sane device with discard granularity that big it would affect this.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
---
 misc/mke2fs.c |   55 ++++++++++++++++++++++++++++++++++++++++++-------------
 1 files changed, 42 insertions(+), 13 deletions(-)

diff --git a/misc/mke2fs.c b/misc/mke2fs.c
index bc1211d..75b07bb 100644
--- a/misc/mke2fs.c
+++ b/misc/mke2fs.c
@@ -72,6 +72,9 @@ extern int optind;
 #define ZAP_BOOTBLOCK
 #endif
 
+#define DISCARD_STEP_MB		(2048)
+#define MB			(1024*1024)
+
 extern int isatty(int);
 extern FILE *fpopen(const char *cmd, const char *mode);
 
@@ -1922,6 +1925,44 @@ static int mke2fs_setup_tdb(const char *name, io_manager *io_ptr)
 	return retval;
 }
 
+static int mke2fs_discard_device(ext2_filsys fs)
+{
+	struct ext2fs_numeric_progress_struct progress;
+	blk64_t blocks = ext2fs_blocks_count(fs->super);
+	blk64_t count = DISCARD_STEP_MB;
+	blk64_t cur = 0;
+	int retval = 0;
+
+	count *= MB;
+	count /= fs->blocksize;
+
+	ext2fs_numeric_progress_init(fs, &progress,
+				     _("Discarding device blocks: "),
+				     blocks);
+	while (cur < blocks) {
+		ext2fs_numeric_progress_update(fs, &progress, cur);
+
+		if (cur + count > blocks)
+			count = blocks - cur;
+
+		retval = io_channel_discard(fs->io, cur, count, fs->blocksize);
+		if (retval)
+			break;
+		cur += count;
+	}
+
+	if (retval) {
+		ext2fs_numeric_progress_close(fs, &progress,
+				      _("failed - "));
+		if (!quiet)
+			printf("%s\n",error_message(retval));
+	} else
+		ext2fs_numeric_progress_close(fs, &progress,
+				      _("done                            \n"));
+
+	return retval;
+}
+
 int main (int argc, char *argv[])
 {
 	errcode_t	retval = 0;
@@ -1982,19 +2023,7 @@ int main (int argc, char *argv[])
 
 	/* Can't undo discard ... */
 	if (discard && (io_ptr != undo_io_manager)) {
-		blk64_t blocks = ext2fs_blocks_count(fs->super);
-		if (verbose)
-			printf(_("Calling BLKDISCARD from 0 to %llu... "),
-			       (unsigned long long) blocks);
-		retval = io_channel_discard(fs->io, 0, blocks, fs->blocksize);
-		if (verbose) {
-			if (retval)
-				printf(_("failed (%s)\n"),
-				       error_message(retval));
-			else
-				printf(_("succeeded\n"));
-		}

Re: [PATCH] mke2fs: Inform user of ongoing discard

[Lukas Czerner]

On Mon, 13 Dec 2010, Lukas Czerner wrote:

> For some time now we are doing initial discard of the device prior to
> filesystem creation. However, there is no feedback for the user and
> hence on some devices with slow TRIM implementation it may appear that
> mke2fs is stuck.
> 
> This commit introduce new function mke2fs_discard_device(), which is a
> wrapper for io_channel_discard(). The discard is done per-partes and
> discard progress is being reported back to the user. The discard step
> has been set to 2GB size, which works reasonably well on both slow and
> fast devices.
> 
> I gave up on doing fancy things like align discard according to
> discard_alignment, checking for discard granularity and computing
> estimate time. First of all, because it would require either new ioctl
> to retrieve those information or use of libudev library, none of it
> seems to be worth it. Regarding discard_granularity, I doubt there is
> any sane device with discard granularity that big it would affect this.
> 
> Signed-off-by: Lukas Czerner <lczerner@redhat.com>
> ---
>  misc/mke2fs.c |   55 ++++++++++++++++++++++++++++++++++++++++++-------------
>  1 files changed, 42 insertions(+), 13 deletions(-)
> 
> diff --git a/misc/mke2fs.c b/misc/mke2fs.c
> index bc1211d..75b07bb 100644
> --- a/misc/mke2fs.c
> +++ b/misc/mke2fs.c
> @@ -72,6 +72,9 @@ extern int optind;
>  #define ZAP_BOOTBLOCK
>  #endif
>  
> +#define DISCARD_STEP_MB		(2048)
> +#define MB			(1024*1024)
> +
>  extern int isatty(int);
>  extern FILE *fpopen(const char *cmd, const char *mode);
>  
> @@ -1922,6 +1925,44 @@ static int mke2fs_setup_tdb(const char *name, io_manager *io_ptr)
>  	return retval;
>  }
>  
> +static int mke2fs_discard_device(ext2_filsys fs)
> +{
> +	struct ext2fs_numeric_progress_struct progress;
> +	blk64_t blocks = ext2fs_blocks_count(fs->super);
> +	blk64_t count = DISCARD_STEP_MB;
> +	blk64_t cur = 0;
> +	int retval = 0;
> +
> +	count *= MB;
> +	count /= fs->blocksize;
> +
> +	ext2fs_numeric_progress_init(fs, &progress,
> +				     _("Discarding device blocks: "),
> +				     blocks);
> +	while (cur < blocks) {
> +		ext2fs_numeric_progress_update(fs, &progress, cur);
> +
> +		if (cur + count > blocks)
> +			count = blocks - cur;
> +
> +		retval = io_channel_discard(fs->io, cur, count, fs->blocksize);
> +		if (retval)
> +			break;
> +		cur += count;
> +	}
> +
> +	if (retval) {
> +		ext2fs_numeric_progress_close(fs, &progress,
> +				      _("failed - "));
> +		if (!quiet)
> +			printf("%s\n",error_message(retval));
> +	} else
> +		ext2fs_numeric_progress_close(fs, &progress,
> +				      _("done                            \n"));
> +
> +	return retval;
> +}
> +
>  int main (int argc, char *argv[])
>  {
>  	errcode_t	retval = 0;
> @@ -1982,19 +2023,7 @@ int main (int argc, char *argv[])
>  
>  	/* Can't undo discard ... */
>  	if (discard && (io_ptr != undo_io_manager)) {
> -		blk64_t blocks = ext2fs_blocks_count(fs->super);
> -		if (verbose)
> -			printf(_("Calling BLKDISCARD from 0 to %llu... "),
> -			       (unsigned long long) blocks);
> -		retval = io_channel_discard(fs->io, 0, blocks, fs->blocksize);
> -		if (verbose) {
> -			if (retval)
> -				printf(_("failed (%s)\n"),
> -				       error_message(retval));
> -			else
> -				printf(_("succeeded\n"));
> -		}
> -
> +		retval = mke2fs_discard_device(fs);
>  		if (!retval && io_channel_discard_zeroes_data(fs->io)) {
>  			if (verbose)
>  				printf(_("Discard succeeded and will return 0s "
> 

Hi Ted,

what is the status of this patch ?

Thanks!
-Lukas