Failure in reading xfs v4 node directories

[Mashijams]

We are getting segmentation fault in reading node type directories for xfs v4. How many entries we are able to read changes every-time we test.

To reproduce the bug :

• Create XFS image with entries(could be file or directory or both) on range 800-1000.

• Compile image on xfs_shell and list all entries using "ls" command.

• Here is the link for my XFS image ​https://drive.google.com/file/d/1qpg41ZztrtW4ZX-ncIEKiljijE4I4aKe/view?usp=sharing.

• Compile above image on xfs_shell and list all entries in "Node" or "Leaves" directory.

[pulkomandy]

Can you attach a debug report?

If you run under Haiku, there is a dialog when the crash happen to create one.

Under Linux or other systems, you have to use gdb and at least print the stacktrace.

Also instructions to reproduce the problem would be nice:

• How to generate or get an XFS image where this problem happens
• What commands to run to get the issue

[Mashijams]

I didn't tested XFS image on Haiku, all tests are on fs_shell.

We have buildlog with TRACE enabled so can I attach that? (It will be big though).

[pulkomandy]

Yes, any info you have will be helpful.

Also can you give more details on creating the XFS image (commands used, etc). Or just upload your XFS image somewhere.

[Mashijams]

I will add how I created XFS images and testing stuff on documentation part.

Here I will try to attach link to my xfs image upload which will be better :)

[pulkomandy]

Hello,

I built the xfs_shell current version and I did not get any crash. Listing the directories (they are named "node" and "leaves" in lowercase) seems to be working fine.

Is this still a problem or did you already fix it?

[Mashijams]

I think the change I did in reading data extent map in node.cpp file on my recent patch seems to be working (on my side it works sometimes!).

But I think we shouldn't close this for now, I want to do much more testing in node based directory directly inside Haiku and then see if we are good, Or I can make new ticket if problem persists.

[pulkomandy]

Yes, keeping the ticket open for general testing and stabilization is fine. I just wanted to know if you needed my help on investigating a specific crash, but it looks like I am too late for that :D

[Mashijams]

I think we are good to close this issue now. It is fixed in hrev56423