common: give lockdep's group name to OpenSSL's mutexes#28987
common: give lockdep's group name to OpenSSL's mutexes#28987liewegas merged 5 commits intoceph:masterfrom
Conversation
|
I've enjoyed reading https://gist.github.com/rzarzynski/b2e73a4fefd86c1cbbafc76f8e3ee55f . Thanks for the write-up! |
|
I don't understand why openssl is making use of the ceph debug mutex... IMO passing an empty name to the lock ctor is the bug. That shouldn't have gotten past review in BlueFS (my bad there), although to be fair the additional args there are supposed to disable lockdep. |
Because of https://github.com/ceph/ceph/blob/master/src/common/ceph_crypto.cc#L97 |
Ah. We should be able to construct those with names like "ssl-%d" then, right? |
Working on that. There is some pain with non-moveability of mutexes after combining with 1) our abstractions ( Possibly will need to introduce some new boilerplate. |
|
Tested locally on CentOS 7: |
fd6cde9 to
34d0edf
Compare
Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Fixes: http://tracker.ceph.com/issues/40698 Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
This reverts commit d92117d at the rason for it has been fixed. Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
34d0edf to
8ba6775
Compare
|
jenkins retest this please ( |
|
assuming no backport for this? @rzarzynski |
* refs/pull/28987/head: Revert "mgr/dashboard: Add nolockdep option to e2e-script" common: refactor handling of lockdep's group name in debug locks. common: shared_mutex_debug doesn't use empty group name for lockdep. common, crypto: give names to OpenSSL's mutexes. common: tiny_vector – CPU-friendly container for mutexes & co. Reviewed-by: Kefu Chai <kchai@redhat.com>
Yup, no backport needed as #27834 is hasn't been backported as well. |
Fixes: http://tracker.ceph.com/issues/40698
Signed-off-by: Radoslaw Zarzynski rzarzyns@redhat.com