Hi, On Tue, Apr 16, 2002 at 05:02:55PM +0100, Matt Bernstein wrote: > At 14:20 +0100 Stephen C. Tweedie wrote: > > >> >Assertion failure in do_get_write_access() at transaction.c:611: "!(((jh2bh(jh))->b_state & (1UL << BH_Lock)) != 0)" > > > >What were you doing at the time? Writing to the live fs can cause > >this (that's normally illegal, but "tune2fs" can do it safely in some > >cases, eg. to write a new volume label.) > > It was about 2am, so *we* weren't doing anything, but the backups will be > running. > > >More seriously, since 2.4.11, there's a page-cache/buffer-cache > >aliasing problem which can lead to this being hit in ext3 if you > >merely read from the buffered block device on which the fs sits. This > >can happen, for example, if you dump(8) a live filesystem. I've got a > >fix pending for the ext3 side of that problem, and have posted the > >core VFS fix to l-k. > > *blush* We dump our filesystems live. I bet we're not the only ones. Indeed, it's common enough, but it's unsafe for various reasons. With LVM, it should work a bit better. At least with the current patches, it shouldn't oops any more. Cheers, Stephen
