Kernel panic during usage + overwrote a file on BFS when paniced.

[xray7224]

Hey,

I was programming on haiku and using running the code (python) when randomly I experanced a kernel panic. I have taken a photo of the panic and have the file with random data in (it looks like just binrary junk been dumped in there).

I am running on the nightly hrev44563, this isn't really reproducable (at least I don't think) but hopefully this info is enough.

[xray7224]

File which was corrupted.

[axeld]

(In #8919) Thanks for the note, duplicate of #8910.

[axeld]

This is probably related to my latest changes to the block cache.

[axeld]

Should be fixed in hrev44585.

[diger]

similar error on hrev44588

[axeld]

Seems I need to have another look, thanks for testing!

[anevilyak]

FYI, ran into the same exact panic here while doing a build of webcore. Anything extra I can grab out of the kernel debugger that'd be of help?

[pulkomandy]

I got the same while running installer from a fresh install of haiku this morning. I think this is easy enough to get into to be an alpha4 blocker, more so if data corruption is involved.

[mmadia]

Script that generates KDL, debug_server exceptions

[mmadia]

hrev44598-gcc2h

[mmadia]

hrev44598-gcc2h

[mmadia]

hrev44598-gcc2h. attachment:install_deps.sh​ has reliably caused KDLs or debug_server errors.

Prior to each event, the partition in question would be cleaned with checkfs.

[axeld]

Please retry with hrev44610; the assert shouldn't have been there.

[mmadia]

With initial testing, hrev44610-2h does not KDL during the install_deps.sh script. Thanks!

[mmadia]

userland crash during install_deps.sh on hrev44610-gcc2h

[mmadia]

serial log of KDL, shortly after userland crash on hrev44610-2h

[mmadia]

Spoke to soon, sort of. While letting a while [ 1 -ge 0 ] ; do rm -rf dependencies ; bash install_deps.sh ; done run, libtool crashed​. While mucking around with gdb, Terminal locked up. Closing the Terminal window, which sparked the crash, Haiku KDL'd​.

To note, libtool did crash on a previous run, but a KDL was not triggered. Upon rebooting Haiku (and later mounting /generated), there were many lines similar to KERN: bfs: replay block run 0:161:1 in log at 1654! . checkfs /generated found and freed some blocks.

[mmadia]

serial-log-of-userland-crashes

[mmadia]

To make sure that the recently rebuilt optional packages are not at fault, I rolled back hrev44593's libtool + sed to libtool-2.4-r1a3-x86-gcc2-2011-05-17.zip and sed-4.2.1-r1a3-x86-gcc2-2011-05-17.zip. After letting while [ 1 -ge 0 ] ; do rm -rf dependencies ; sync ; bash install_deps.sh ; done run for a bit, several programs crashed -- libtool, WonderBrush, and Tracker all within seconds of each other. Here is the serial debug output​ during the crashes. To note, the libpthread.so lines are from pkg-config/glib's configure test.

[axeld]

And you are sure this is related in any way to this ticket?

[axeld]

I mean, does it work fine with a revision before, let's say hrev44000?

[mmadia]

*cough* libtool crashes in hrev43925-2h as well.

[axeld]

Alright, then I assume this is a different issue; at least I can't imagine how those two can be connected unless the block cache is trashing memory.