FAT: KDLs on device removal without unmount

[Catmengi]

After disconnecting pen drive it leaves phantom /dev/disk/usb/* node. Partition isn't unmounting which leads to "general protection exception" 0x0 Tested on: x64 laptop (i3 6006u "acer travelmate p259")

Steps to recreate: connect pendrive, let haiku mount partition, disconnect pendrive, /dev/disk/usb/* will stay. If you didn't open mounted or waited seconds (vary on cpu speed) partition before and try to open it you will have kernel panic

This issue doesnt exist in previous versions of haiku. So this issue is from bsd fat driver. It mostly dont proper handle hard unmount event / or dont have proper checks of existence of device!

[Catmengi]

Panic log

[waddlesplash]

If you didn't Unmount/"Eject" before disconnecting the pendrive, then some misbehavior is expected because there may be writes that weren't flushed to the disk, and you may get data loss. The system shouldn't crash, of course, but disconnecting disks without unmounting them isn't good behavior regardless.

[Catmengi]

Note that crash is going even without writing anything to it. It crashes on attempt to read this phantom device

[Catmengi]

Also phantom /dev/disk/usb nodes is quite bad for tty users (and maybe system)

[waddlesplash]

The "phantom nodes" should go away if you unmount the disk properly before removing it. If you don't mount the device at all after plugging it in, they should also go away when you remove it in that case too.

[Catmengi]

Attempt to open right click button leads to gui freeze, then kernel panic. I'll add an attachment

[waddlesplash]

This is with the disk still inserted?

[Catmengi]

Without. With inserted is everything ok, BUT /dev/disk/usb/* still exist

[Catmengi]

NEW INFO: device nodes is deleting but /dev/disk/usb/* is not deleted

[Catmengi]

Additional info: partition disappeared now, in 5 seconds ± but /dev path still exists, kernel's device nodes in this path is deleted

[waddlesplash]

I would expect the /dev file to disappear only after both unmount and then removal of the device, I think.

[Catmengi]

Files in /dev/disk/usb/*/* path is deleting after hot unplug drive. This is ok) path/directory isn't deleting itself

[Catmengi]

This somehow leading that every pen-drive reconnection create new /dev/disk/usb entry. Leading to garbage in /dev after time

[Catmengi]

Please, can you change ticket to kernel/other suitable? This bug isnt related to FAT FS, its related to /dev

[waddlesplash]

All the images you have posted are kernel panics in the FAT driver, so it definitely is. Whether or not the node "disappears" is probably related to this.

[Catmengi]

Ok, i agree with this. Currently the issue that this phantom disk crashes kernel. It's not unmounted on hot unplug. /dev/disk/usb/* path isnt deleted leading to useless garbage in it. Possible, the part of kernel that work with pen drive know that it disconnected, but there isnt any automatic unmounting mechanism(or it is) for hot unplug in FAT driver (befs driver possible have this, it isnt crashing os and partition content is not visible after hot unplug). I think this issue can be solved in this way(adding some kind of signaling from device driver, usb drive in this case to fat driver to force unmount partitions that was on this device). But im not an expert in os architecture and haiku's codebase, so this may not be an ideal solution.

[Catmengi]

Even attempt to open right click menu of this fat partition crashing the whole system. This isnt the case with BeFS, so this possible really fat issue. Befs unmount property automatically.

[waddlesplash]

If you are pulling the device without unmounting it in the right-click menu, you are definitely going to lose data whether you're on BFS or FAT.

[Catmengi]

Doesn't matter system shouldn't crash

[waddlesplash]

Sure, which is why this ticket exists. But I'm just noting that even once the kernel crashes are fixed, pulling devices isn't good behavior anyway and will certainly get you in trouble.

[Catmengi]

Is anything about this issue changed yet?

[Catmengi]

Is it ok if you (waddlesplash) change this bug priority to high? Or it isnt too much critical?

[Catmengi]

Also may this be an issue of the new fat driver only? As i read from the beta 5 change log, it use new fat driver from freebsd which may have another method of handling disconnect of the device. I need to do some test on haiku hrev1 beta 4

[waddlesplash]

Please be patient and don't comment every day. Haiku is a project mostly run by volunteers, things take time.

[Catmengi]

I discovered that this issue only exist in haiku hrev1/beta 5. This issue doesn't exist in beta 4

[Catmengi]

Sorry for yet another comment but this is new info about error. Partition can unmount itself. But if you attempt to "ls" immediately in this disconnected disk you will have crash, if you will wait 5 seconds it will work properly