gfs2 was deadlocking because most code paths acquire the sd_log_flush_lock and
then the page locks, while code paths going through gfs2_writepage acquire the
page locks and then the log flush lock.

This patch makes gfs2_writepage do a trylock on the sd_log_flush_lock. If it
fails, writepage simply redirtys the page and gives up. This should keep the
deadlock from happening, however it might cause a performance hit.

Signed-off-by: Benjamin E. Marzinski <[EMAIL PROTECTED]>

Reply via email to