Comment 23 for bug 1349711

Repost of what I sent to the mailing list just now:

My current interpretation of this problem is that it is some
pathological condition caused by not rebalancing and being nearly out
of space for allocating more metadata and hence it is rarely being
seen by anyone else (because most users are regularly doing
rebalances).

See this thread for details about rebalancing and out of space:
"ENOSPC with mkdir and rename" on 2014-08-02:

http://thread.gmane.org/gmane.comp.file-systems.btrfs/37415

I haven't had the lockups in production since July and I'm now trialling a
nightly rebalance:

$ btrfs filesystem balance start -dusage=50 -musage=10 "$mount"

I'll report back if I encounter further problems.