xfstests 251: fix fitrim support test

On my ext4 filesystem, the simple "did fstrim work" test passes, because it asks to free all blocks in the first 10m of the fs, and those 10m are full of filesystem metadata. Because no blocks are free, no blocks are trimmed, and we get success returned.

But then when the test runs I'm flooded with error messages, because it's a hard drive not an ssd...

So we need to step through the fs until we either free a block, or encounter an error.

I think this is ugly bash, if anyone has a better plan I'm all ears.

(also change FSTRIM to FITRIM in the failure message, it seems to be intended to print the ioctl name ...)

Signed-off-by Eric Sandeen <[email protected]>

---

251 +13 -1

index fa3d74a..5ab0a87 100755
@@ -73,7 +73,19 @@ _fail()
 
 _check_fstrim_support()
 {
-	$here/src/fstrim -l 10M $SCRATCH_MNT &> /dev/null
+	# Go until error or until something gets trimmed
+	step=1048576
+	start=0
+	retval=0
+	nonetrimmed=1
+
+	while [ $retval -eq 0 ] && [ $nonetrimmed -ne 0 ]; do
+		result=`$here/src/fstrim -v -s $start -l $step $SCRATCH_MNT 2>&1`
+		retval=$?
+		[ "${result:0:1}" -eq "0" ] && nonetrimmed=1
+		start=$(( $start + $step ))
+	done
+	return $retval
 }
 
 ##

src/fstrim.c +1 -1

index f1f37ec..ad7fd6a 100644
@@ -236,7 +236,7 @@ int main(int argc, char **argv)
 	}
 
 	if (ioctl(fd, FITRIM, opts->range)) {
-		fprintf(stderr, "%s: FSTRIM: %s\n", program_name,
+		fprintf(stderr, "%s: FITRIM %s\n", program_name,
 			strerror(errno));
 		goto free_opts;
 	}

Quoted 8 lines · Eric Sandeen

On my ext4 filesystem, the simple "did fstrim work" test passes, because it asks to free all blocks in the first 10m of the fs, and those 10m are full of filesystem metadata. Because no blocks are free, no blocks are trimmed, and we get success returned.

But then when the test runs I'm flooded with error messages, because it's a hard drive not an ssd...

Quoted 7 lines

So we need to step through the fs until we either free a block, or encounter an error.

I think this is ugly bash, if anyone has a better plan I'm all ears.

(also change FSTRIM to FITRIM in the failure message, it seems to be intended to print the ioctl name ...)

Hi Eric,

this is actually a filesystem bug found unintentionally by this test :) and it is already fixed upstream 4143179218960a70d821a425e3c23ce44aa93dee for ext4.. So I think we better leave it as it is, since this is unwanted behaviour and should be detected.

What we should fix however, is when fstrim fails after successful fist test, so the test exits and report failure, rather than printing tons of error messages.

Thanks!
-Lukas

Quoted 36 lines

Signed-off-by Eric Sandeen <[email protected]>

---

251 +13 -1

index fa3d74a..5ab0a87 100755
@@ -73,7 +73,19 @@ _fail()
 
 _check_fstrim_support()
 {
-	$here/src/fstrim -l 10M $SCRATCH_MNT &> /dev/null
+	# Go until error or until something gets trimmed
+	step=1048576
+	start=0
+	retval=0
+	nonetrimmed=1
+
+	while [ $retval -eq 0 ] && [ $nonetrimmed -ne 0 ]; do
+		result=`$here/src/fstrim -v -s $start -l $step $SCRATCH_MNT 2>&1`
+		retval=$?
+		[ "${result:0:1}" -eq "0" ] && nonetrimmed=1
+		start=$(( $start + $step ))
+	done
+	return $retval
 }
 
 ##

src/fstrim.c +0 -0

index f1f37ec..ad7fd6a 100644
@@ -236,7 +236,7 @@ int main(int argc, char **argv)
 	}
 

src/fstrim.c int main(int argc, char **argv) · line 242

Quoted 7 lines

src/fstrim.c +1 -1

 	if (ioctl(fd, FITRIM, opts->range)) {
-		fprintf(stderr, "%s: FSTRIM: %s\n", program_name,
+		fprintf(stderr, "%s: FITRIM %s\n", program_name,
 			strerror(errno));
 		goto free_opts;
 	}