Skip to contentSkip to conversation

fstests: add helper to canonicalize devices used to enable persistent disks

PATCH Superseded Luis Chamberlain 5 participants 10 messages Jul 20, 2023 – Jul 27, 2023 VFS

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fstests: add helper to canonicalize devices used to enable persistent disks

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

Using a symlink is desirable if you want to enable persistent tests across reboots. For example you may want to use /dev/disk/by-id/nvme-eui.* so to ensure that the same drives are used even after reboot. This is very useful if you are testing for example with a virtualized environment and are using PCIe passthrough with other qemu NVMe drives with one or many NVMe drives.

To enable support just add a helper to canonicalize devices prior to running the tests.

This allows one test runner, kdevops, which I just extended with support to use real NVMe drives. The drives it uses for the filesystem configuration optionally is with NVMe eui symlinks so to allow the same drives to be used over reboots.

---
 check         |  1 +
 common/config | 44 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
check +1 -0
index 89e7e7bf20df..d063d3f498fd 100755
@@ -734,6 +734,7 @@ function run_section()
 	fi
 
 	get_next_config $section
+	_canonicalize_devices
 
 	mkdir -p $RESULT_BASE
 	if [ ! -d $RESULT_BASE ]; then
common/config +43 -1

61 lines not on this page · raw message

Signature · 1 line
2.39.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 14 lines · Luis Chamberlain

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

Using a symlink is desirable if you want to enable persistent tests across reboots. For example you may want to use /dev/disk/by-id/nvme-eui.* so to ensure that the same drives are used even after reboot. This is very useful if you are testing for example with a virtualized environment and are using PCIe passthrough with other qemu NVMe drives with one or many NVMe drives.

To enable support just add a helper to canonicalize devices prior to running the tests.

Quoted 7 lines

This allows one test runner, kdevops, which I just extended with support to use real NVMe drives. The drives it uses for the filesystem configuration optionally is with NVMe eui symlinks so to allow the same drives to be used over reboots.

---

Looks good to me:

Signature · 1 line
- Andrey
#2 lore raw
In reply to Luis Chamberlain
Quoted 14 lines · Luis Chamberlain

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

Using a symlink is desirable if you want to enable persistent tests across reboots. For example you may want to use /dev/disk/by-id/nvme-eui.* so to ensure that the same drives are used even after reboot. This is very useful if you are testing for example with a virtualized environment and are using PCIe passthrough with other qemu NVMe drives with one or many NVMe drives.

To enable support just add a helper to canonicalize devices prior to running the tests.

Quoted 7 lines

This allows one test runner, kdevops, which I just extended with support to use real NVMe drives. The drives it uses for the filesystem configuration optionally is with NVMe eui symlinks so to allow the same drives to be used over reboots.

---

Hi Luis,

Hmmm... this's a default behavior change for fstests, although I'm not sure what will be affect. I'm wondering if we should do this in fstests. I don't want to tell the users that the device names they give to fstests will always be truned to real names from now on.

Generally the users of fstests provide the device names, so the users might need to take care of the name is "/dev/mapper/testvg-testdev" or "/dev/dm-4". The users can deal with the device names when their script prepare to run fstests.

If more developers prefer this change, I'd like to make it to be optional by an option of ./check at least, not always turn devname to realpath. Welcome review points from others.

Quoted 21 lines
 check         |  1 +
 common/config | 44 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
check +1 -0
index 89e7e7bf20df..d063d3f498fd 100755
@@ -734,6 +734,7 @@ function run_section()
 	fi
 
 	get_next_config $section
+	_canonicalize_devices
 
 	mkdir -p $RESULT_BASE
 	if [ ! -d $RESULT_BASE ]; then
common/config +0 -0
index 936ac225f4b1..f5a3815a0435 100644

common/config _canonicalize_mountpoint() · line 662

Quoted 8 lines
common/config +5 -0
@@ -655,6 +655,47 @@ _canonicalize_mountpoint()
 	echo "$parent/$base"
 }
 
+# Enables usage of /dev/disk/by-id/ symlinks to persist target devices
+# over reboots
+_canonicalize_devices()
+{
+	if [ ! -z "$TEST_DEV" ] && [ -L $TEST_DEV ]; then

I think [ -L "$TEST_DEV" ] is enough.

common/config _canonicalize_mountpoint() · line 663

Quoted 1 line
common/config +1 -0
+		TEST_DEV=$(realpath -e $TEST_DEV)

Anyone knows the difference of realpatch and readlink?

Quoted 19 lines
common/config +19 -0
+	fi
+
+	if [ ! -z "$SCRATCH_DEV" ] && [ -L $SCRATCH_DEV ]; then
+		SCRATCH_DEV=$(realpath -e $SCRATCH_DEV)
+	fi
+
+	if [ ! -z "$TEST_LOGDEV" ] && [ -L $TEST_LOGDEV ]; then
+		TEST_LOGDEV=$(realpath -e $TEST_LOGDEV)
+	fi
+
+	if [ ! -z "$TEST_RTDEV" ] && [ -L $TEST_RTDEV ]; then
+		TEST_RTDEV=$(realpath -e $TEST_RTDEV)
+	fi
+
+	if [ ! -z "$SCRATCH_RTDEV" ] && [ -L $SCRATCH_RTDEV ]; then
+		SCRATCH_RTDEV=$(realpath -e $SCRATCH_RTDEV)
+	fi
+
+	if [ ! -z "$LOGWRITES_DEV" ] && [ -L $LOGWRITES_DEV ]; then

common/config _canonicalize_mountpoint() · line 692

Quoted 10 lines
common/config +10 -0
+		LOGWRITES_DEV=$(realpath -e $LOGWRITES_DEV)
+	fi
+
+	if [ ! -z "$SCRATCH_DEV_POOL" ]; then
+		NEW_SCRATCH_POOL=""
+		for i in $SCRATCH_DEV_POOL; do
+			if [ -L $i ]; then
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(realpath -e $i)"
+			else
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i)"
                                                                     ^^^

What's this half ")" for ?

Thanks,
Zorro

Quoted 28 lines
common/config +8 -1
+			fi
+		done
+		SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL"
+	fi
+}
+
 # On check -overlay, for the non multi section config case, this
 # function is called on every test, before init_rc().
 # When SCRATCH/TEST_* vars are defined in config file, config file
@@ -785,7 +826,6 @@ get_next_config() {
 	fi
 
 	parse_config_section $1
-
 	if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then
 		[ -z "$MOUNT_OPTIONS" ] && _mount_opts
 		[ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts
@@ -901,5 +941,7 @@ else
 	fi
 fi
 
+_canonicalize_devices
+
 # make sure this script returns success
 /bin/true
Signature · 1 line
2.39.2
#3 lore raw
Quoted 30 lines · Zorro Lang
Quoted 14 lines · Luis Chamberlain

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

Using a symlink is desirable if you want to enable persistent tests across reboots. For example you may want to use /dev/disk/by-id/nvme-eui.* so to ensure that the same drives are used even after reboot. This is very useful if you are testing for example with a virtualized environment and are using PCIe passthrough with other qemu NVMe drives with one or many NVMe drives.

To enable support just add a helper to canonicalize devices prior to running the tests.

Quoted 7 lines

This allows one test runner, kdevops, which I just extended with support to use real NVMe drives. The drives it uses for the filesystem configuration optionally is with NVMe eui symlinks so to allow the same drives to be used over reboots.

---

Hi Luis,

Hmmm... this's a default behavior change for fstests, although I'm not sure what will be affect. I'm wondering if we should do this in fstests. I don't want to tell the users that the device names they give to fstests will always be truned to real names from now on.

Quoted 8 lines

Generally the users of fstests provide the device names, so the users might need to take care of the name is "/dev/mapper/testvg-testdev" or "/dev/dm-4". The users can deal with the device names when their script prepare to run fstests.

If more developers prefer this change, I'd like to make it to be optional by an option of ./check at least, not always turn devname to realpath. Welcome review points from others.

Hmm. SCRATCH_DEV=/dev/testvg/testlv works right now, it'd be sort of annoying to have "/dev/dm-4" get written into the report and then you've lost the correlation to whatever the user passed in.

Oh wait. My giant mound of ./check wrapper script already does that canonicalization and has for years.

Ok. Sounds good to me then. :P

Quoted 24 lines
Quoted 24 lines
 check         |  1 +
 common/config | 44 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
check +1 -0
index 89e7e7bf20df..d063d3f498fd 100755
@@ -734,6 +734,7 @@ function run_section()
 	fi
 
 	get_next_config $section
+	_canonicalize_devices
 
 	mkdir -p $RESULT_BASE
 	if [ ! -d $RESULT_BASE ]; then
common/config +0 -0
index 936ac225f4b1..f5a3815a0435 100644
@@ -655,6 +655,47 @@ _canonicalize_mountpoint()
 	echo "$parent/$base"
 }
 
Quoted 7 lines

common/config _canonicalize_mountpoint() · line 662

Quoted 5 lines
common/config +5 -0
+# Enables usage of /dev/disk/by-id/ symlinks to persist target devices
+# over reboots
+_canonicalize_devices()
+{
+	if [ ! -z "$TEST_DEV" ] && [ -L $TEST_DEV ]; then

I think [ -L "$TEST_DEV" ] is enough.

I don't think it is.

$ unset moo
$ [ -L $moo ]
$ echo $?
0
$ realpath -e $moo
realpath: missing operand
Try 'realpath --help' for more information.
Quoted 3 lines

common/config _canonicalize_mountpoint() · line 663

Quoted 1 line
common/config +1 -0
+		TEST_DEV=$(realpath -e $TEST_DEV)

Anyone knows the difference of realpatch and readlink?

readlink doesn't follow nested symlinks; realpath does:

$ touch /tmp/a
$ ln -s /tmp/a /tmp/b
$ ln -s /tmp/b /tmp/c
$ readlink /tmp/c
/tmp/b
$ realpath /tmp/c
/tmp/a
$ readlink -m /tmp/c
/tmp/a
Quoted 4 lines

common/config _canonicalize_mountpoint() · line 667

Quoted 4 lines
common/config +4 -0
+	fi
+
+	if [ ! -z "$SCRATCH_DEV" ] && [ -L $SCRATCH_DEV ]; then
+		SCRATCH_DEV=$(realpath -e $SCRATCH_DEV)

Extra question: Shouldn't we be putting theis ^^^ variables in quotes?

Supposing someone starts passing in
SCRATCH_DEV=/dev/disk/by-label/har har"

This expression will barf everywhere:

$ SCRATCH_DEV=$(realpath -e $SCRATCH_DEV)
realpath: /dev/disk/by-label/har: No such file or directory
realpath: har: No such file or directory

Due to the lack of quoting on its way to turning that into /dev/sda3. Granted fstests has historically required no spaces in anything, but still, it's bad hygiene.

[writing anything in bash is bad hygiene, but for the sweet sweet pipe goodness]

Quoted 27 lines
Quoted 15 lines
common/config +15 -0
+	fi
+
+	if [ ! -z "$TEST_LOGDEV" ] && [ -L $TEST_LOGDEV ]; then
+		TEST_LOGDEV=$(realpath -e $TEST_LOGDEV)
+	fi
+
+	if [ ! -z "$TEST_RTDEV" ] && [ -L $TEST_RTDEV ]; then
+		TEST_RTDEV=$(realpath -e $TEST_RTDEV)
+	fi
+
+	if [ ! -z "$SCRATCH_RTDEV" ] && [ -L $SCRATCH_RTDEV ]; then
+		SCRATCH_RTDEV=$(realpath -e $SCRATCH_RTDEV)
+	fi
+
+	if [ ! -z "$LOGWRITES_DEV" ] && [ -L $LOGWRITES_DEV ]; then

common/config _canonicalize_mountpoint() · line 692

Quoted 10 lines
common/config +10 -0
+		LOGWRITES_DEV=$(realpath -e $LOGWRITES_DEV)
+	fi
+
+	if [ ! -z "$SCRATCH_DEV_POOL" ]; then
+		NEW_SCRATCH_POOL=""
+		for i in $SCRATCH_DEV_POOL; do
+			if [ -L $i ]; then
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(realpath -e $i)"
+			else
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i)"
                                                                     ^^^
Quoted 1 line

What's this half ")" for ?

Some kind of typo, I assume...

--D

Quoted 34 lines

Thanks,
Zorro

Quoted 28 lines
common/config +8 -1
+			fi
+		done
+		SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL"
+	fi
+}
+
 # On check -overlay, for the non multi section config case, this
 # function is called on every test, before init_rc().
 # When SCRATCH/TEST_* vars are defined in config file, config file
@@ -785,7 +826,6 @@ get_next_config() {
 	fi
 
 	parse_config_section $1
-
 	if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then
 		[ -z "$MOUNT_OPTIONS" ] && _mount_opts
 		[ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts
@@ -901,5 +941,7 @@ else
 	fi
 fi
 
+_canonicalize_devices
+
 # make sure this script returns success
 /bin/true
Signature · 1 line
2.39.2
#4 lore raw
Quoted 40 lines · Darrick J. Wong
Quoted 30 lines · Zorro Lang
Quoted 14 lines · Luis Chamberlain

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

Using a symlink is desirable if you want to enable persistent tests across reboots. For example you may want to use /dev/disk/by-id/nvme-eui.* so to ensure that the same drives are used even after reboot. This is very useful if you are testing for example with a virtualized environment and are using PCIe passthrough with other qemu NVMe drives with one or many NVMe drives.

To enable support just add a helper to canonicalize devices prior to running the tests.

Quoted 7 lines

This allows one test runner, kdevops, which I just extended with support to use real NVMe drives. The drives it uses for the filesystem configuration optionally is with NVMe eui symlinks so to allow the same drives to be used over reboots.

---

Hi Luis,

Hmmm... this's a default behavior change for fstests, although I'm not sure what will be affect. I'm wondering if we should do this in fstests. I don't want to tell the users that the device names they give to fstests will always be truned to real names from now on.

Quoted 8 lines

Generally the users of fstests provide the device names, so the users might need to take care of the name is "/dev/mapper/testvg-testdev" or "/dev/dm-4". The users can deal with the device names when their script prepare to run fstests.

If more developers prefer this change, I'd like to make it to be optional by an option of ./check at least, not always turn devname to realpath. Welcome review points from others.

Quoted 8 lines

Hmm. SCRATCH_DEV=/dev/testvg/testlv works right now, it'd be sort of annoying to have "/dev/dm-4" get written into the report and then you've lost the correlation to whatever the user passed in.

Oh wait. My giant mound of ./check wrapper script already does that canonicalization and has for years.

Ok. Sounds good to me then. :P

So you hope fstests can do this translation (use real device name) by default?

Quoted 33 lines
Quoted 25 lines
Quoted 25 lines
 check         |  1 +
 common/config | 44 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
check +1 -0
index 89e7e7bf20df..d063d3f498fd 100755
@@ -734,6 +734,7 @@ function run_section()
 	fi
 
 	get_next_config $section
+	_canonicalize_devices
 
 	mkdir -p $RESULT_BASE
 	if [ ! -d $RESULT_BASE ]; then
common/config +0 -0
index 936ac225f4b1..f5a3815a0435 100644
@@ -655,6 +655,47 @@ _canonicalize_mountpoint()
 	echo "$parent/$base"
 }
 
Quoted 7 lines

common/config _canonicalize_mountpoint() · line 662

Quoted 5 lines
common/config +5 -0
+# Enables usage of /dev/disk/by-id/ symlinks to persist target devices
+# over reboots
+_canonicalize_devices()
+{
+	if [ ! -z "$TEST_DEV" ] && [ -L $TEST_DEV ]; then

I think [ -L "$TEST_DEV" ] is enough.

Quoted 4 lines

I don't think it is.

$ unset moo
$ [ -L $moo ]

The double quotation marks "" are needed.

Quoted 12 lines
$ echo $?
0
$ realpath -e $moo
realpath: missing operand
Try 'realpath --help' for more information.
Quoted 3 lines

common/config _canonicalize_mountpoint() · line 663

Quoted 1 line
common/config +1 -0
+		TEST_DEV=$(realpath -e $TEST_DEV)

Anyone knows the difference of realpatch and readlink?

readlink doesn't follow nested symlinks; realpath does:

Quoted 9 lines
$ touch /tmp/a
$ ln -s /tmp/a /tmp/b
$ ln -s /tmp/b /tmp/c
$ readlink /tmp/c
/tmp/b
$ realpath /tmp/c
/tmp/a
$ readlink -m /tmp/c
/tmp/a

The -e option helps:

# readlink -e /tmp/c
/tmp/a
# realpath -e /tmp/c
/tmp/a
Quoted 7 lines
Quoted 5 lines

common/config _canonicalize_mountpoint() · line 667

Quoted 5 lines
common/config +4 -0
+	fi
+
+	if [ ! -z "$SCRATCH_DEV" ] && [ -L $SCRATCH_DEV ]; then
+		SCRATCH_DEV=$(realpath -e $SCRATCH_DEV)

Extra question: Shouldn't we be putting theis ^^^ variables in quotes?

Make sense to me.

Thanks,
Zorro

Quoted 85 lines

Supposing someone starts passing in
SCRATCH_DEV=/dev/disk/by-label/har har"

This expression will barf everywhere:

$ SCRATCH_DEV=$(realpath -e $SCRATCH_DEV)
realpath: /dev/disk/by-label/har: No such file or directory
realpath: har: No such file or directory

Due to the lack of quoting on its way to turning that into /dev/sda3. Granted fstests has historically required no spaces in anything, but still, it's bad hygiene.

[writing anything in bash is bad hygiene, but for the sweet sweet pipe goodness]

Quoted 27 lines
Quoted 15 lines
common/config +15 -0
+	fi
+
+	if [ ! -z "$TEST_LOGDEV" ] && [ -L $TEST_LOGDEV ]; then
+		TEST_LOGDEV=$(realpath -e $TEST_LOGDEV)
+	fi
+
+	if [ ! -z "$TEST_RTDEV" ] && [ -L $TEST_RTDEV ]; then
+		TEST_RTDEV=$(realpath -e $TEST_RTDEV)
+	fi
+
+	if [ ! -z "$SCRATCH_RTDEV" ] && [ -L $SCRATCH_RTDEV ]; then
+		SCRATCH_RTDEV=$(realpath -e $SCRATCH_RTDEV)
+	fi
+
+	if [ ! -z "$LOGWRITES_DEV" ] && [ -L $LOGWRITES_DEV ]; then

common/config _canonicalize_mountpoint() · line 692

Quoted 10 lines
common/config +10 -0
+		LOGWRITES_DEV=$(realpath -e $LOGWRITES_DEV)
+	fi
+
+	if [ ! -z "$SCRATCH_DEV_POOL" ]; then
+		NEW_SCRATCH_POOL=""
+		for i in $SCRATCH_DEV_POOL; do
+			if [ -L $i ]; then
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(realpath -e $i)"
+			else
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i)"
                                                                     ^^^
Quoted 1 line

What's this half ")" for ?

Some kind of typo, I assume...

--D

Quoted 34 lines

Thanks,
Zorro

Quoted 28 lines
common/config +8 -1
+			fi
+		done
+		SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL"
+	fi
+}
+
 # On check -overlay, for the non multi section config case, this
 # function is called on every test, before init_rc().
 # When SCRATCH/TEST_* vars are defined in config file, config file
@@ -785,7 +826,6 @@ get_next_config() {
 	fi
 
 	parse_config_section $1
-
 	if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then
 		[ -z "$MOUNT_OPTIONS" ] && _mount_opts
 		[ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts
@@ -901,5 +941,7 @@ else
 	fi
 fi
 
+_canonicalize_devices
+
 # make sure this script returns success
 /bin/true
Signature · 1 line
2.39.2
#6 lore raw
In reply to Darrick J. Wong
Quoted 5 lines · Darrick J. Wong
Quoted 4 lines · Zorro Lang
Quoted 3 lines · Luis Chamberlain

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

I'm a little confused. Where are these "sanity checks" enforced? I've been using

SCRATCH_DEV=/dev/mapper/xt-vdc

where /dev/mapper/xt-vdc is a symlink to /dev/dm-4 (or some such) without any problems. So I don't quite understand why we need to canonicalize devices?

- Ted

#7 lore raw
Quoted 7 lines · Theodore Ts'o
Quoted 5 lines · Darrick J. Wong
Quoted 4 lines · Zorro Lang
Quoted 3 lines · Luis Chamberlain

The filesystem configuration file does not allow you to use symlinks to devices given the existing sanity checks verify that the target end device matches the source.

Quoted 8 lines

I'm a little confused. Where are these "sanity checks" enforced? I've been using

SCRATCH_DEV=/dev/mapper/xt-vdc

where /dev/mapper/xt-vdc is a symlink to /dev/dm-4 (or some such) without any problems. So I don't quite understand why we need to canonicalize devices?

That might work, but try using /dev/disk/by-id/ stuff, that'll bust. So to keep existing expecations by fstests, it's needed.

Luis

#5 lore raw
In reply to Zorro Lang
Quoted 15 lines · Zorro Lang
Quoted 14 lines · Darrick J. Wong
Quoted 8 lines · Zorro Lang
Quoted 3 lines · Luis Chamberlain

Using a symlink is desirable if you want to enable persistent tests across reboots. For example you may want to use /dev/disk/by-id/nvme-eui.* so to ensure that the same drives are used even after reboot.

If more developers prefer this change, I'd like to make it to be optional by an option of ./check at least, not always turn devname to realpath. Welcome review points from others.

Hmm. SCRATCH_DEV=/dev/testvg/testlv works right now, it'd be sort of annoying to have "/dev/dm-4" get written into the report and then you've lost the correlation to whatever the user passed in.

Quoted 6 lines
Quoted 4 lines

Oh wait. My giant mound of ./check wrapper script already does that canonicalization and has for years.

Ok. Sounds good to me then. :P

So you hope fstests can do this translation (use real device name) by default?

I'm fine if we make this optional too. In my next spin I'll add an option that does this.

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

I think [ -L "$TEST_DEV" ] is enough.

I don't think it is.

$ unset moo
$ [ -L $moo ]

The double quotation marks "" are needed.

Indeed it does. Will go with that for v2.

Quoted 22 lines
Quoted 12 lines
$ echo $?
0
$ realpath -e $moo
realpath: missing operand
Try 'realpath --help' for more information.
Quoted 3 lines

Review

Quoted 1 line
(quoted diff) +1 -0
+		TEST_DEV=$(realpath -e $TEST_DEV)

Anyone knows the difference of realpatch and readlink?

readlink doesn't follow nested symlinks; realpath does:

Quoted 9 lines
$ touch /tmp/a
$ ln -s /tmp/a /tmp/b
$ ln -s /tmp/b /tmp/c
$ readlink /tmp/c
/tmp/b
$ realpath /tmp/c
/tmp/a
$ readlink -m /tmp/c
/tmp/a
Quoted 4 lines

The -e option helps:

# readlink -e /tmp/c
/tmp/a

I can trace readlink -e support back to 2004:

https://github.com/coreutils/coreutils/commit/e0b8973bd4b146b5fb39641a4ee7984e922c3ff5

Quoted 2 lines
# realpath -e /tmp/c
/tmp/a

realpeath is much newer than readlink, it's first commit including support for -e dates back to 2011;

https://github.com/coreutils/coreutils/commit/77ea441f79aa115f79b47d9c1fc9c0004c5c7111

And so canonicalizing with readlink is 7 years older and likely to be supported in more ancient distros with older coreutils. I'll go with readlink -e unless I hear otherwise.

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+	fi
+
+	if [ ! -z "$SCRATCH_DEV" ] && [ -L $SCRATCH_DEV ]; then
+		SCRATCH_DEV=$(realpath -e $SCRATCH_DEV)

Extra question: Shouldn't we be putting theis ^^^ variables in quotes?

Make sense to me.

Done. The only eye sore is how to convert the loop for the SCRATCH_DEV_POOL, here's what I have for a v2:

README +3 -0
index 1ca506492bf0..97ef63d6d693 100644
@@ -268,6 +268,9 @@ Misc:
    this option is supported for all filesystems currently only -overlay is
    expected to run without issues. For other filesystems additional patches
    and fixes to the test suite might be needed.
+ - set CANON_DEVS=yes to canonicalize device symlinks. This will let you
+   for example use something like TEST_DEV/dev/disk/by-id/nvme-* so the
+   device remains persistent between reboots. This is disabled by default.
 
 ______________________
 USING THE FSQA SUITE
check +1 -0
index 0bf5b22e061a..577e09655844 100755
@@ -711,6 +711,7 @@ function run_section()
 	fi
 
 	get_next_config $section
+	_canonicalize_devices
 
 	mkdir -p $RESULT_BASE
 	if [ ! -d $RESULT_BASE ]; then
common/config +31 -1
index 6c8cb3a5ba68..7d74c285ac71 100644
@@ -25,6 +25,9 @@
 # KEEP_DMESG -      whether to keep all dmesg for each test case.
 #                   yes: keep all dmesg
 #                   no: only keep dmesg with error/warning (default)
+# CANON_DEVS -      whether or not to canonicalize device symlinks
+#                   yes: canonicalize device symlinks
+#                   no (default) do not canonicalize device if they are symlinks
 #
 # - These can be added to $HOST_CONFIG_DIR (witch default to ./config)
 #   below or a separate local configuration file can be used (using
@@ -644,6 +647,32 @@ _canonicalize_mountpoint()
 	echo "$parent/$base"
 }
 
+# Enables usage of /dev/disk/by-id/ symlinks to persist target devices
+# over reboots
+_canonicalize_devices()
+{
+	if [ "$CANON_DEVS" != "yes" ]; then
+		return
+	fi
+	[ -L "$TEST_DEV" ]	&& TEST_DEV=$(readlink -e "$TEST_DEV")
+	[ -L $SCRATCH_DEV ]	&& SCRATCH_DEV=$(readlink -e "$SCRATCH_DEV")
+	[ -L $TEST_LOGDEV ]	&& TEST_LOGDEV=$(readlink -e "$TEST_LOGDEV")
+	[ -L $TEST_RTDEV ]	&& TEST_RTDEV=$(readlink -e "$TEST_RTDEV")
+	[ -L $SCRATCH_RTDEV ]	&& SCRATCH_RTDEV=$(readlink -e "$SCRATCH_RTDEV")
+	[ -L $LOGWRITES_DEV ]	&& LOGWRITES_DEV=$(readlink -e "$LOGWRITES_DEV")
+	if [ ! -z "$SCRATCH_DEV_POOL" ]; then
+		NEW_SCRATCH_POOL=""
+		for i in $SCRATCH_DEV_POOL; do
+			if [ -L $i ]; then
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(readlink -e $i)"
+			else
+				NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i)"
+			fi
+		done
+		SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL"
+	fi
+}
+
 # On check -overlay, for the non multi section config case, this
 # function is called on every test, before init_rc().
 # When SCRATCH/TEST_* vars are defined in config file, config file
@@ -774,7 +803,6 @@ get_next_config() {
 	fi
 
 	parse_config_section $1
-
 	if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then
 		[ -z "$MOUNT_OPTIONS" ] && _mount_opts
 		[ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts
@@ -890,5 +918,7 @@ else
 	fi
 fi
 
+_canonicalize_devices
+
 # make sure this script returns success
 /bin/true
#8 lore raw
In reply to Luis Chamberlain
Quoted 11 lines · Luis Chamberlain
Quoted 8 lines

I'm a little confused. Where are these "sanity checks" enforced? I've been using

SCRATCH_DEV=/dev/mapper/xt-vdc

where /dev/mapper/xt-vdc is a symlink to /dev/dm-4 (or some such) without any problems. So I don't quite understand why we need to canonicalize devices?

That might work, but try using /dev/disk/by-id/ stuff, that'll bust. So to keep existing expecations by fstests, it's needed.

What goes wrong, and why? /dev/disk/by-id/<disk-id> is a symlink, just like /dev/mapper/<vg>-<lv> is a symlink.

What am I missing?

Thanks,

- Ted

#9 lore raw
Quoted 10 lines · Theodore Ts'o
Quoted 9 lines · Luis Chamberlain
Quoted 8 lines

I'm a little confused. Where are these "sanity checks" enforced? I've been using

SCRATCH_DEV=/dev/mapper/xt-vdc

where /dev/mapper/xt-vdc is a symlink to /dev/dm-4 (or some such) without any problems. So I don't quite understand why we need to canonicalize devices?

Quoted 7 lines
Quoted 2 lines

That might work, but try using /dev/disk/by-id/ stuff, that'll bust. So to keep existing expecations by fstests, it's needed.

What goes wrong, and why? /dev/disk/by-id/<disk-id> is a symlink, just like /dev/mapper/<vg>-<lv> is a symlink.

What am I missing?

# mkfs.xfs -f /dev/sda
# mount /dev/sda /mnt
# TEST_DIR=/mnt TEST_DEV=/dev/sda FSTYP=xfs ./check generic/110
FSTYP         -- xfs (debug)
PLATFORM      -- Linux/x86_64 flax-mtr01 6.5.0-rc3-djwx #rc3 SMP PREEMPT_DYNAMIC Wed Jul 26 14:26:48 PDT 2023

generic/110 2s
Ran: generic/110
Passed all 1 tests

versus:

# TEST_DIR=/mnt TEST_DEV=/dev/disk/by-id/scsi-0QEMU_RAMDISK_drive-scsi0-0-0-0 FSTYP=xfs ./check generic/110
mount: /mnt: /dev/sda already mounted on /mnt.
common/rc: retrying test device mount with external set
mount: /mnt: /dev/sda already mounted on /mnt.
common/rc: could not mount /dev/disk/by-id/scsi-0QEMU_RAMDISK_drive-scsi0-0-0-0 on /mnt

# umount /mnt
# TEST_DIR=/mnt TEST_DEV=/dev/disk/by-id/scsi-0QEMU_RAMDISK_drive-scsi0-0-0-0 FSTYP=xfs ./check generic/110
TEST_DEV=/dev/disk/by-id/scsi-0QEMU_RAMDISK_drive-scsi0-0-0-0 is mounted but not on TEST_DIR=/mnt - aborting
Already mounted result:
/dev/sda /mnt

(This is not really how I run fstests, it's just the minimum example.)

--D

Quoted 4 lines

Thanks,

- Ted