[PATCH] blkid: open device in nonblock mode.

[PATCH] blkid: open device in nonblock mode.

[Michal Suchanek]

When autoclose is set (kernel default but many distributions reverse the
setting) opening a CD-rom device causes the tray to close.

The function of blkid is to report the current state of the device and
not to change it. Hence it should use O_NONBLOCK when opening the
device to avoid closing a CD-rom tray.

blkid is used liberally in scripts so it can potentially interfere with
the user operating the CD-rom hardware.

Signed-off-by: Michal Suchanek <msuchanek@suse.de>
---
 libblkid/src/verify.c | 2 +-
 misc-utils/blkid.c    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/libblkid/src/verify.c b/libblkid/src/verify.c
index a78c9f8f2aa4..4bda3fd40cee 100644
--- a/libblkid/src/verify.c
+++ b/libblkid/src/verify.c
@@ -126,7 +126,7 @@ blkid_dev blkid_verify(blkid_cache cache, blkid_dev dev)
 		}
 	}
 
-	fd = open(dev->bid_name, O_RDONLY|O_CLOEXEC);
+	fd = open(dev->bid_name, O_RDONLY|O_CLOEXEC|O_NONBLOCK);
 	if (fd < 0) {
 		DBG(PROBE, ul_debug("blkid_verify: error %m (%d) while "
 					"opening %s", errno,
diff --git a/misc-utils/blkid.c b/misc-utils/blkid.c
index f2583d2b883f..0df9f6b6f29e 100644
--- a/misc-utils/blkid.c
+++ b/misc-utils/blkid.c
@@ -499,7 +499,7 @@ static int lowprobe_device(blkid_probe pr, const char *devname,
 	int rc = 0;
 	static int first = 1;
 
-	fd = open(devname, O_RDONLY|O_CLOEXEC);
+	fd = open(devname, O_RDONLY|O_CLOEXEC|O_NONBLOCK);
 	if (fd < 0) {
 		warn(_("error: %s"), devname);
 		return BLKID_EXIT_NOTFOUND;
-- 
2.23.0

Re: [PATCH] blkid: open device in nonblock mode.

[Karel Zak]

On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
> When autoclose is set (kernel default but many distributions reverse the
> setting) opening a CD-rom device causes the tray to close.
> 
> The function of blkid is to report the current state of the device and
> not to change it. Hence it should use O_NONBLOCK when opening the
> device to avoid closing a CD-rom tray.

I can imagine this as optional solution (command line option), but I
have doubts to use O_NONBLOCK by default for all block devices. I have
no example, but it sounds like a way how to introduce regressions in
libblkid behavior. (Any kernel guy around?) Is it really only cdrom
driver(s) where O_NONBLOCK has any impact? What about USB, some random
SCSI, ... I don't know.

The another problem is that the library does not have to open the device,
you can use already open file descriptor (blkid_probe_set_device()). 
So, in many cases the patch will have no effect.

> blkid is used liberally in scripts so it can potentially interfere with
> the user operating the CD-rom hardware.

It's better to use lsblk in script, it reads info from udev -- call
blindly blkid(8) is usually overkill.

    Karel


-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: [PATCH] blkid: open device in nonblock mode.

[Michal Suchánek]

On Tue, Nov 05, 2019 at 12:41:22PM +0100, Karel Zak wrote:
> On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
> > When autoclose is set (kernel default but many distributions reverse the
> > setting) opening a CD-rom device causes the tray to close.
> > 
> > The function of blkid is to report the current state of the device and
> > not to change it. Hence it should use O_NONBLOCK when opening the
> > device to avoid closing a CD-rom tray.
> 
> I can imagine this as optional solution (command line option), but I

That defeats the purpose of this change. You cannot use the option with
old blkid, so using the option is broken and not using it is also broken.

> have doubts to use O_NONBLOCK by default for all block devices. I have
> no example, but it sounds like a way how to introduce regressions in
> libblkid behavior. (Any kernel guy around?) Is it really only cdrom
> driver(s) where O_NONBLOCK has any impact? What about USB, some random
> SCSI, ... I don't know.
> 
> The another problem is that the library does not have to open the device,
> you can use already open file descriptor (blkid_probe_set_device()). 
> So, in many cases the patch will have no effect.

If some random program using libblkid closes the tray I don't care that
much. However, many system scripts use blkid, probably to find a device
with particular ID:

/usr/bin/dracut:            dev=$(blkid -l -t UUID=${dev#UUID=} -o
device)
/usr/bin/dracut:            dev=$(blkid -l -t LABEL=${dev#LABEL=} -o
device)
/usr/bin/dracut:            dev=$(blkid -l -t PARTUUID=${dev#PARTUUID=}
-o device)
/usr/bin/dracut:            dev=$(blkid -l -t
PARTLABEL=${dev#PARTLABEL=} -o device)

/usr/bin/linux-boot-prober:	partition=$(blkid | grep "$UUID" | cut
-d ':' -f 1 | tr '\n' ' ' | cut -d ' ' -f 1)
/usr/bin/os-prober:	blkid | grep btrfs | cut -d ':' -f 1
/usr/bin/os-prober:	type=$(blkid -o value -s TYPE $mapped || true)
/usr/bin/os-prober:		uuid=$(blkid -o value -s UUID $mapped)

> 
> > blkid is used liberally in scripts so it can potentially interfere with
> > the user operating the CD-rom hardware.
> 
> It's better to use lsblk in script, it reads info from udev -- call
> blindly blkid(8) is usually overkill.

First off you need to explain it to all authors of all random scripts
out there.

Secondly udev is not guaranteed to exist/run on every system.

Thanks

Michal

Re: [PATCH] blkid: open device in nonblock mode.

[Michal Suchánek]

On Tue, Nov 05, 2019 at 06:13:57PM +0100, Michal Suchánek wrote:
> On Tue, Nov 05, 2019 at 12:41:22PM +0100, Karel Zak wrote:
> > On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
> > > When autoclose is set (kernel default but many distributions reverse the
> > > setting) opening a CD-rom device causes the tray to close.
> > > 
> > > The function of blkid is to report the current state of the device and
> > > not to change it. Hence it should use O_NONBLOCK when opening the
> > > device to avoid closing a CD-rom tray.
> > 
> > I can imagine this as optional solution (command line option), but I
> 
> That defeats the purpose of this change. You cannot use the option with
> old blkid, so using the option is broken and not using it is also broken.
> 
> > have doubts to use O_NONBLOCK by default for all block devices. I have
> > no example, but it sounds like a way how to introduce regressions in
> > libblkid behavior. (Any kernel guy around?) Is it really only cdrom
> > driver(s) where O_NONBLOCK has any impact? What about USB, some random
Yes, it affect floppies as well:

drivers/block/ataflop.c:        if (mode & FMODE_NDELAY)
drivers/block/floppy.c: if (!(mode & FMODE_NDELAY)) {
drivers/block/pktcdvd.c:        ret = blkdev_get(bdev, FMODE_READ |
FMODE_NDELAY, NULL);
drivers/block/pktcdvd.c:                blkdev_put(bdev, FMODE_READ |
FMODE_NDELAY);
drivers/block/pktcdvd.c:        blkdev_put(bdev, FMODE_READ |
FMODE_NDELAY);
drivers/block/pktcdvd.c:        blkdev_put(pd->bdev, FMODE_READ |
FMODE_NDELAY);
drivers/block/swim.c:   if (mode & FMODE_NDELAY)
drivers/block/swim3.c:  if (err == 0 && (mode & FMODE_NDELAY) == 0
drivers/cdrom/cdrom.c:  if ((mode & FMODE_NDELAY) && (cdi->options &
CDO_USE_FFLAGS)) {
drivers/cdrom/cdrom.c:          !(mode & FMODE_NDELAY);
drivers/ide/ide-gd.c:           if (ret && (mode & FMODE_NDELAY) == 0) {
drivers/scsi/sd.c:      if (sdev->removable && !sdkp->media_present &&
!(mode & FMODE_NDELAY))
drivers/scsi/sd.c:                      (mode & FMODE_NDELAY) != 0);
drivers/scsi/sd.c:                      (mode & FMODE_NDELAY) != 0);
drivers/scsi/sr.c:      return __sr_block_open(bdev, mode |
FMODE_NDELAY);
drivers/scsi/sr.c:      if ((ret == -ENOMEDIUM) && !(mode &
FMODE_NDELAY))
drivers/scsi/sr.c:                      (mode & FMODE_NDELAY) != 0);


> > SCSI, ... I don't know.
> > 
> > The another problem is that the library does not have to open the device,
> > you can use already open file descriptor (blkid_probe_set_device()). 
> > So, in many cases the patch will have no effect.
> 
> If some random program using libblkid closes the tray I don't care that
> much. However, many system scripts use blkid, probably to find a device
> with particular ID:
> 
> /usr/bin/dracut:            dev=$(blkid -l -t UUID=${dev#UUID=} -o
> device)
> /usr/bin/dracut:            dev=$(blkid -l -t LABEL=${dev#LABEL=} -o
> device)
> /usr/bin/dracut:            dev=$(blkid -l -t PARTUUID=${dev#PARTUUID=}
> -o device)
> /usr/bin/dracut:            dev=$(blkid -l -t
> PARTLABEL=${dev#PARTLABEL=} -o device)
> 
> /usr/bin/linux-boot-prober:	partition=$(blkid | grep "$UUID" | cut
> -d ':' -f 1 | tr '\n' ' ' | cut -d ' ' -f 1)
> /usr/bin/os-prober:	blkid | grep btrfs | cut -d ':' -f 1
> /usr/bin/os-prober:	type=$(blkid -o value -s TYPE $mapped || true)
> /usr/bin/os-prober:		uuid=$(blkid -o value -s UUID $mapped)
> 
> > 
> > > blkid is used liberally in scripts so it can potentially interfere with
> > > the user operating the CD-rom hardware.
> > 
> > It's better to use lsblk in script, it reads info from udev -- call
> > blindly blkid(8) is usually overkill.
which uses the udev identifiers which already use O_NONBLOCK:
src/udev/ata_id/ata_id.c:        fd = open(node,
O_RDONLY|O_NONBLOCK|O_CLOEXEC);
src/udev/cdrom_id/cdrom_id.c:                fd = open(node,
O_RDONLY|O_NONBLOCK|O_CLOEXEC);
src/udev/scsi_id/scsi_serial.c:        fd = open(devname, O_RDONLY |
O_NONBLOCK | O_CLOEXEC);
src/udev/scsi_id/scsi_serial.c:                fd = open(devname,
O_RDONLY | O_NONBLOCK | O_CLOEXEC);

> 
> First off you need to explain it to all authors of all random scripts
> out there.
> 
> Secondly udev is not guaranteed to exist/run on every system.
> 
> Thanks
> 
> Michal

Re: [PATCH] blkid: open device in nonblock mode.

[Karel Zak]

On Wed, Nov 06, 2019 at 09:02:56AM +0100, Michal Suchánek wrote:
> > If some random program using libblkid closes the tray I don't care that
> > much. However, many system scripts use blkid, probably to find a device

One of the random programs is udev:

   src/udev/udev-builtin-blkid.c:

       fd = open(devnode, O_RDONLY|O_CLOEXEC);
       if (fd < 0)
               return log_device_debug_errno(dev, errno, "Failed to open block device %s: %m", devnode);

       errno = 0;
       r = blkid_probe_set_device(pr, fd, offset, 0);

> > First off you need to explain it to all authors of all random scripts
> > out there.

That's what I'm doing for years :)

Anyway, I'm going to apply the patch, thanks for your investigation.

Let's hope it will not introduce any change on something else than
on cdroms and floppies.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: [PATCH] blkid: open device in nonblock mode.

[Michal Suchánek]

On Wed, Nov 06, 2019 at 09:48:33AM +0100, Karel Zak wrote:
> On Wed, Nov 06, 2019 at 09:02:56AM +0100, Michal Suchánek wrote:
> > > If some random program using libblkid closes the tray I don't care that
> > > much. However, many system scripts use blkid, probably to find a device
> 
> One of the random programs is udev:
> 
>    src/udev/udev-builtin-blkid.c:
> 
>        fd = open(devnode, O_RDONLY|O_CLOEXEC);

Yes, I have this on my radar. It does not seem to be used in the distro
build of systemd I use but should be addressed as well.

Thanks

Michal

Re: [PATCH] blkid: open device in nonblock mode.

[Anatoly Pugachev]

On Wed, Nov 6, 2019 at 11:52 AM Karel Zak <kzak@redhat.com> wrote:
>> On Wed, Nov 06, 2019 at 09:02:56AM +0100, Michal Suchánek wrote:
> > > First off you need to explain it to all authors of all random scripts
> > > out there.
>
> That's what I'm doing for years :)

considering lsblk / findmnt , how they are supposed to replace blkid
in the following example (taken from os-prober):

                output="$(blkid -o device -t LABEL="$1")" || return 1

i wasn't able to find lsblk alternative to match things (label, uuid,
partition, whatever) - maybe i just don't carefully read manual page
for it ? Of course it is possible to use grep in pipe after lsblk, but
anyway.

Thanks.

Re: [PATCH] blkid: open device in nonblock mode.

[Karel Zak]

On Tue, Nov 12, 2019 at 11:27:28AM +0300, Anatoly Pugachev wrote:
> On Wed, Nov 6, 2019 at 11:52 AM Karel Zak <kzak@redhat.com> wrote:
> >> On Wed, Nov 06, 2019 at 09:02:56AM +0100, Michal Suchánek wrote:
> > > > First off you need to explain it to all authors of all random scripts
> > > > out there.
> >
> > That's what I'm doing for years :)
> 
> considering lsblk / findmnt , how they are supposed to replace blkid
> in the following example (taken from os-prober):
> 
>                 output="$(blkid -o device -t LABEL="$1")" || return 1
> 
> i wasn't able to find lsblk alternative to match things (label, uuid,
> partition, whatever) - maybe i just don't carefully read manual page
> for it ? Of course it is possible to use grep in pipe after lsblk, but
> anyway.

Yes, it's UN*X :-) ... grep or so

    lsblk -rpo LABEL,NAME | awk '/'$1'/ { print $2 }'


 Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: [PATCH] blkid: open device in nonblock mode.

[Karel Zak]

On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
>  libblkid/src/verify.c | 2 +-
>  misc-utils/blkid.c    | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)

Applied, but I have also added O_NONBLOCK to another places in the
library and to wipefs.

Note that we still use it without O_NONBLOCK in libfdisk, mkswap and
partx ... but I guess it's no so big problem as the primary goal are
scripts with blkid(8).

Thanks.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: [PATCH] blkid: open device in nonblock mode.

[Karel Zak]

On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
> When autoclose is set (kernel default but many distributions reverse the
> setting) opening a CD-rom device causes the tray to close.

I found unwanted side effect, open() with O_NONBLOCK also successes
when there is no medium. Unfortunately, we ignore I/O errors for
CDROMs to support some crazy hybrid media -- so overall result is many
I/O warnings in system logs.

I have added CDROM_DRIVE_STATUS to the logic to stop probing when
there is no medium. Let's hope it will be enough:

https://github.com/karelzak/util-linux/commit/dc30fd4383e57a0440cdb0e16ba5c4336a43b290


    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: [PATCH] blkid: open device in nonblock mode.

[Michal Suchánek]

On Tue, Jan 07, 2020 at 05:04:43PM +0100, Karel Zak wrote:
> On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
> > When autoclose is set (kernel default but many distributions reverse the
> > setting) opening a CD-rom device causes the tray to close.
> 
> I found unwanted side effect, open() with O_NONBLOCK also successes
> when there is no medium. Unfortunately, we ignore I/O errors for
> CDROMs to support some crazy hybrid media -- so overall result is many
> I/O warnings in system logs.
> 
> I have added CDROM_DRIVE_STATUS to the logic to stop probing when
> there is no medium. Let's hope it will be enough:
> 
> https://github.com/karelzak/util-linux/commit/dc30fd4383e57a0440cdb0e16ba5c4336a43b290

It it probably safer to do
switch (ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT)) {
to support some crazy CD changers.

Thanks

Michal

Re: [PATCH] blkid: open device in nonblock mode.

[Karel Zak]

On Tue, Jan 07, 2020 at 05:19:18PM +0100, Michal Suchánek wrote:
> On Tue, Jan 07, 2020 at 05:04:43PM +0100, Karel Zak wrote:
> > On Mon, Nov 04, 2019 at 09:23:15PM +0100, Michal Suchanek wrote:
> > > When autoclose is set (kernel default but many distributions reverse the
> > > setting) opening a CD-rom device causes the tray to close.
> > 
> > I found unwanted side effect, open() with O_NONBLOCK also successes
> > when there is no medium. Unfortunately, we ignore I/O errors for
> > CDROMs to support some crazy hybrid media -- so overall result is many
> > I/O warnings in system logs.
> > 
> > I have added CDROM_DRIVE_STATUS to the logic to stop probing when
> > there is no medium. Let's hope it will be enough:
> > 
> > https://github.com/karelzak/util-linux/commit/dc30fd4383e57a0440cdb0e16ba5c4336a43b290
> 
> It it probably safer to do
> switch (ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT)) {
> to support some crazy CD changers.

Good point, I have seen it in kernel docs, but forgot to use it.
Fixed. Thanks for review.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com