Hi folks, Sorry for taking such a long time to compose a fix patch set for bcache journal deadlock issue since the first time it was reported. The initial issue was know as several kthreads or kworkers were reported by kernel to hang for quite long time. The reason was a deadlock happened when there is no more journal space avialable for new coming journal request. Finally it turns out to be two conditions that the jouranl no-space deadlock may happen, one is in jouranl replay time, one is in normal journal runtime. During my testing, I find deadlock still exists with my fixes, after a lot of effort other three related bugs are explored and fixed. Now the patch set is testing on two machines for 3+ hours, the journal deadlock does not appear yet. In my previous testings, it may show up within 30 minutes on my machine. (In order to make the jouranl space to be more easier exhuasted, I change SB_JOURNAL_BUCKETS from 256U to 3U both in kernel and bcache-tools code). More testings on more machines will start soon, and the patches are not stable enough for production environment usage. Bbut I think it is time to post the patch set for your review and comments. I will continue to improve the fixes, e.g. remove some BUG_ON() once I am sure they won't happen indeed. If you may help to test the patch set, that will be really cool. Hope we may make this patch set stable soon. Thanks in advance for your help. Coly Li --- Coly Li (6): bcache: acquire c->journal.lock in bch_btree_leaf_dirty() bcache: move definition of 'int ret' out of macro read_bucket() bcache: never set 0 to KEY_PTRS of jouranl key in journal_reclaim() bcache: reload jouranl key information during journal replay bcache: fix journal deadlock during jouranl replay bcache: reserve space for journal_meta() in run time drivers/md/bcache/btree.c | 2 + drivers/md/bcache/journal.c | 244 +++++++++++++++++++++++++++++++++++++++++--- drivers/md/bcache/journal.h | 5 + 3 files changed, 238 insertions(+), 13 deletions(-) -- 2.16.4
