Re: Another Assertion failure in journal_start()

[Andrew Morton]

Jan Kara <jack@xxxxxxx> wrote:
>
>  > Just hit our main server, while doing kernel compile
>  > (producing a .deb using custom script which does quite
>  > some usage of symlinks - hence sys_symlink() operation
>  > is in call trace) -- that particular filesystem stopped
>  > working, while the rest of the system was still operational.
>  > 
>  > It's 2.6.15.1 kernel running on an athlon-1.3GHz, pretty
>  > old but pretty stable box, with ECC memory.  The filesystem
>  > in question is on top of a raid0 array out of 4 scsi drives
>  > (it's used as a "staging area" for various temporary stuff,
>  > incl. compiles and whatnot).
>  > 
>  > Any clues about this one?
>  > 
>  > Note it's the first time I encountered an error like this
>  > one, but I did quite alot of kernel compiles on this box
>  > already since last boot (I'm experimenting with Xen on
>  > another box, this box is used as a "compiling server").
>  > So I can hardly say the problem is "easily reproduceable".
>    The trace you provided is good enough so that I don't need to
>  reproduce the problem :). The bug is in ext3 - the problem is that we
>  start a transaction and then do something that needs to allocate memory
>  but we don't set GFP_NOFS. As we are low on memory we try to shrink
>  caches and remove some inode on different filesystem from memory. We
>  recurse back into the fs code which finds out we have already started
>  transaction on different fs and BUGs.
>    The right solution probably is to pass gfp flags to page_symlink().
>  I'll write a patch.

That'd be the safest approach.  It'd be nicer to close off the transaction
while running page_symlink(), although we'd need to think hard about the
atomicity implications of that.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/